The Evolution of Agonistic Behavior: Basic Processes and Models

J.M.G. van der Dennen


Behavior patterns in intra-specific aggression

In agonistic encounters the enemy provokes fear as well as aggression; the resulting activities of the sympathetic and the counterbalancing activities of the parasympathetic autonomic nervous system, and the conflict behavior (ambivalence) have been elaborated as aggressive signals and countersignals during ritualized combat (Kommentkampf, Turnier).

Ritualized Combat; Repertoire
I. Autonomic signals, intensity of the outward signs of physiological change
    i. defecation and urination : territorial scent-marking devices
    ii. circulatory disturbances (red flush / extreme pallor) : bare patches of skin
    iii. respiratory disturbances : aggressive vocalizations
    iv. pilo-erection: crests, capes, manes, fringes: inflation-displays
II. Postural and muscular signals, elaborated into intimidation postures
    i. aggressive intention movements: stylized (stamping feet, fist making, chest-beating etc.), facial expression; direct stare.
    ii. ambivalent actions (mosaic) : formalized into rhythmic jerkings
    iii. conflict postures: threat rituals
    iv. general agitation: combat dances
    v. displacement activities: exaggerations
    vi. redirection activities,
All these intention movements act as communication signals and combine effectively with the autonomic signals to provide a precise picture of the intensity of the aggression aroused, and an exact indication of the fight/flight balance. If these signal systems fail, then real fighting follows.
Submissive displays, appeasement signals, non-aggression signals
I. Gross inactivity, frequently combined with threat-opposites (Darwin's principle of antithesis)
    i. crouching; opposite of expanding body
    ii. cowering; opposite of aggressive vocalization
    iii. facing away (cut-off); opposite of posture of frontal attack and direct stare
    iv. offering vulnerable area to attacker
    v. others (species-specific)
II. Remotivating devices
    i. adoption of juvenile food begging posture
    ii. adoption of female sexual posture (presenting); dominant male will mount and pseudo-copulate
    iii. arousal of mood to groom or to be groomed, or self-grooming
III. Flight


Basic Processes of Expression and Signalling

The effector systems in and through which expression takes place in all multicellular animals consist of muscles, glands, and the chromatophore systems (which in many lower orders of vertebrates and nonvertebrates help bring about a rapid change of color), together with the connecting nerves and in some cases certain ancillary systems. Nothing else is involved - even vocalization is, from the point of view of the effectors, the result of muscular contractions (Leyhausen, 1967).
The neural supplies of the effectors belong partly to the autonomic, partly to the motor system. The latter are, so to speak, under the direct control of the propensities; the former have a relative amount of autonomy which is, however, within certain limits governed by the demands of the motor system. Any particular expression is usually a mixture of both effects. The motivational minority achieving expression has various possibilities for exerting influence on the effectors:
(1) It can 'take over' the ones not, or not fully, claimed by the motivational combination currently dominant. This is above all true in the case of autonomic functions, such as pupil reactions, blushing, and perspiring, which betray so much that is not in keeping with the rest of the behavior of the individual concerned. Often, therefore, they are particularly good indicators of an imminent change of mood; for example, a sudden narrowing of pupils, which announces imminent attack, and dilation of pupils which indicates a readiness for escape or for defense, depending on the situation, while all other behavior still simulates complete composure.
(2) The motivational group striving for expression can force a direct influence on the effectors already claimed by the majority, and it can do this in three ways:
(a) by changes of tonus;
(b) by changes in threshold of the motor neurons in the spinal cord;
(c) by motor superimposition.
Through the superimposition of the motor systems of two or more conflicting propensities, mixed motor patterns occur from which it is possible to tell the relative strength of the propensities involved with quantitative precision. An early example of this, provided by Lorenz (1952), was the facial expression of the dog torn between escape and attack. It can be shown in even greater detail in the facial expression, body posture and movement of the cat (Leyhausen, 1956), and the elephant (Kuehme, 1963). The arched back of a cat produced by this process, for example, appears phenomenologically to be a thoroughly uniform expressive (threatening) posture, but experiment and analysis reveal that it consists of several movement patterns superimposed on each other. Tembrock (1963) analyzed corresponding superimpositions in 'mixed calls' by the fox. This principle of ambivalence has been found by practically all vertebrates investigated in this connection (Lorenz, 1952; 1963; Eibl-Eibesfeldt, 1953; 1957; 1967; Leyhausen, 1956; Morris, 1956a,b; 1958; Tembrock, 1957; 1969; 1961; 1962; 1963; Walther, 1958; 1960; Kuehme, 1961; 1963; Peirce & Nuttall, 1961; Moynihan, 1962; van Hooff, 1962; Rowell, 1962; Hinde & Rowell, 1962; Rowell & Hinde, 1962; Geist, 1963; Ewer, 1963; Grant, 1963, Grant & Mackintosh, 1963; Barlow, 1974). Leonhard (1949) described this principle in detail in connection with the 'social miens' of humans.

Fig. 1: Canine facial expressions. Top left: neutral expression, no agonistic tendencies activated. Reading to right across top row: increasing readiness to attack. Reading down left hand column: increasing submission. The other pictures show combinations of fear and aggression corresponding to their positions on the two axes (After Lorenz, 1963).

Fig. 2: Moods expressed by body position in the cat. Increasing attacking and defensive tendencies shown along horizontal and vertical axes (After Leyhausen, 1956).

Fig. 3: Threat and fear expressions in cats. In each section fear increases from above downwards, and aggressiveness from left to right (From Leyhausen, 1956).

Fig. 4: Positions of head, trunk and ears shown by elephants in aggression, fear, and threat (After Kuehme, 1963).

Intra-organismic conflict in courtship and threat
Much of the time a tendency to show one type of behavior is in conflict with tendencies to show others. If tendencies for two incompatible types of behavior are simultaneously present, we may speak of them as being in conflict (Hinde 1966).
When the two tendencies in conflict involve approaching two different objects some distance apart, we can label the situation an 'approach/approach conflict', without, of course implying that the incompatibility is merely a physical one. In such a case the animal may reach a point in between where the tendencies to approach each object are in balance. Its position will then be unstable, for the tendency to approach either goal increases with its proximity. On the other hand, an animal placed between two objects both of which it strives to avoid ('avoidance/avoidance conflict') is in a relatively stable situation.
A more interesting case occurs when the animal has simultaneous tendencies to approach and avoid a given object or situation ('approach/avoidance conflict'). This situation has been the subject of a rigid analysis by Miller (1959) and Brown (1957) in terms of approach and avoidance gradients.

The so-called 'conflict hypothesis' developed by Tinbergen (1952), Morris (1956), and Baerends (1974) is based on the fact that at each encounter between animals of the same species, and also often when animals of different species meet, indications of the activation of different and mutually conflicting tendencies can be observed. These are at least the tendencies to approach and to avoid, but usually the more complicated tendencies to attack and to retreat or escape. In correspondence, detailed observations reveal in a great many signal activities the presence of incomplete or incipient movements of attack as well as of escape. These elements may be parts of the actual fixed action patterns, or the components orienting these patterns ('taxes'), or both (Lorenz & Tinbergen, 1938). The phenomenon is called ambivalence; the opposite components may either be shown successively in alternation with each other, or simultaneously as a mosaic. An example of successive ambivalence is the zigzag-dance of the three-spined stickleback, a signal activity by which the male tries to induce a ripe female to follow it towards the nest in the center of its territory. The zigzag-course is the result of the regular alternation of an orientation towards and away from the female. An example of simultaneous ambivalence is the upright threat display of the black-headed gull in which the stretched neck is an escape component whereas the downward pointing bill and the lifted wingbows are incipient movements of attack (Tinbergen, 1959).
The conflict between an orientation towards and away from the opponent often leads to a compromise in the direction of the movement: it may, for instance, be directed sideways, as in the lateral display circling of the jungle fowl cock. A jungle fowl cock, performing its side display towards an opponent, circles around it while showing the latter its inclined back. This circling can be interpreted as a result of a compromised orientation. The laterally inclined body posture is caused by the part nearest to the opponent making an incipient crouching (= escape) movement, whereas simultaneously the contralateral part of the body is stretched as a preparation for attack (Kruijt, 1964).
In a wide range of species pair formation is initiated by aggressive behavior directed by a member of one sex (usually the male) to other individuals. Changes in the subsequent courtship behavior can be understood in terms of changes in tendencies to attack, flee from, and behave sexually towards the sex partner (Hinde, 1966). The complicated activities shown by fishes and birds during threat and courtship can be understood on the view that tendencies for two or more incompatible types of behavior are present. Hinde (1966) presents the following representative selection of studies: Fish: Baerends, Brouwer & Waterbolk (1955); Barlow (1962, 1963, 1974); Forselius (1957); Keenleyside & Yamamoto (1962); Morris (1954, 1958); Oehlert (1958); Wickler (1958). Birds: Andrew (1957, 1961); Baeumer (1959); Baggerman et al., (1956); Blurton-Jones (1958-9); Brockway (1963); Crook (1960, 1963); Delius (1963); Hinde (1952, 1953, 1954, 1955-6); Immelman (1962); Kruijt (1964); Kunkel (1959, 1962); Lind (1961, 1962); McKinney (1961); Marler (1956); Morris (1954, 1957, 1958); Moynihan (1955, 1956, 1958); Moynihan & Hall (1954); Rueppell (1962); Stamm (1962); Stokes (1962); Tinbergen (1952, 1953, 1959);Wood-Gush (1956). The extent to which such principles are applicable also among the amphibia is not so clear (e.g., Gauss, 1961), but the courtship of one species has been interpreted in terms of an 'interaction of drives' by Rabb & Rabb (1963). In some reptiles, also, courtship clearly involves aggressive components (e.g., Eibl-Eibesfeldt, 1955; Kästle, 1964; MacLean, 1990).
Sufficient work has been done in mammals to show that ambivalence plays a large part in many of their displays. The threat posture and facial expressions of the domestic cat (Leyhausen, 1956), dog (Lorenz, 1952), and the African elephant (Kühme, 1961, 1963) change with tendencies to attack and flee from the rival; the precise nature of the posture changes with the absolute and relative strengths of the two tendencies. Tembrock,(1962) regards the threat displays of foxes as due to a superposition of attacking and defensive tendencies under the influence of social inhibition, and Walther (1958, 1960) found dominance changes to play an important role in the premating behavior of okapi. All the rodent species studied by Eibl-Eibesfeldt (1953), Grant (1963) and Grant & Mackintosh (1963) showed ambivalence between fleeing, aggression, and mating tendencies in social encounters (see also Peirce & Nuttall, 1961). Other mammalian display movements which have been interpreted in terms of ambivalence include those of the moose (Geist, 1963) and various primates (Hinde & Rowell. 1962; Rowell & Hinde, 1962; Rowell, 1962; van Hooff 1962).
Hinde (1966), Leyhausen (1967), and Baerends (1975) have enumerated the various types of behavior which occur in conflict situations, from which the following account is adapted:

(1) Inhibition of all but one response. Undoubtedly the commonest consequence of the simultaneous action of factors for two or more types of behavior is the suppression of all but one of them.
Behavioral inhibition involves suppression of one or more activities while another predominates, but what are the consequences of the conflict for the behavior which does appear? This depends in part on the criteria by which it is assessed, but in terms of frequency measures it too usually suffers a decrement. Indeed, an acute conflict may result in the inhibition of all possible responses. The paralysis of fear is well enough known in human subjects; a comparable phenomenon has been studied by Rigby (1954). He trained rats to obtain food by performing one response in the presence of a light, and to avoid shock by another response in the presence of a buzzer. When both light and buzzer were presented simultaneously, many of the animals sat motionless.
In some cases, however, there is evidence that the behavior not suppressed may actually be augmented. If sticklebacks are fed in only one part of their aquaria, and are then given one or two electric shocks there, they subsequently spend less time in the danger area, but while there they feed more intensively.
The concept of behavioral inhibition is admittedly a loose one, and diverse types of mechanisms may be involved: competition for the mechanisms of selective attention, competition for the effectors, and perhaps competition for more than one type of intervening mechanism.

(2) Intention movements. Intention movements are beginnings of behavior patterns which as a result of only slight motivation or simultaneous, strong inhibition exerted by conflicting tendencies, cannot (yet) be completed. Various simultaneous intentions alternate rhythmically or are superimposed on one another (ambivalent movements: Bastock, Morris & Moynihan, 1953). A related phenomenon is what Leyhausen (1967) calls 'Partial anticipation of an action'. When cats are sitting close to desirable prey without being able to reach it, one may observe that in expectation or anticipation of eating they not only salivate and lick their lips but often open and close their jaws spasmodically as well ('teeth-chattering'), which is interpreted as the rhythmic, mimically exaggerated anticipation of predatory biting. Koehler (1921) described the 'sympathetic movements' which a chimpanzee performed as it watched another piling boxes under bananas which had been hung up in the neighboring cage. Similar sympathetic movements can easily be observed in sportsmen or in children watching a Punch and Judy show. With all these movements it is not a matter of imitation, but quite simply of an often exaggerated performance, in facial mime and pantomime, of movements or parts of movements which one would perform if one were in that situation oneself or could only get at the object. Closely related with these phenomena are

(3) Actions using an alternative object (redirected activities; Bastock, Morris & Moynihan, 1953), some of which can become expressive movements: the fist which dares not hit the opponent pounds the table top and so becomes an expression of violent but helpless or barely controlled rage. During a boundary dispute, herring gulls may direct their fierce attacks not against each other but against grass tufts, which may be pulled at as if they were the opponent's plumage (Tinbergen, 1952).

(4) Alternation. Just before copulation, a male chaffinch has conflicting tendencies to approach and avoid the female. He does not, however, stay still at the point of balance, but alternates between the two patterns, sometimes edging a little closer and then fleeing away. Another well-known example of such alternation is provided by the zigzag courtship dance of the three-spined stickleback. When a pregnant female enters a male's territory he swims in a zigzag course. One leg of this course is an intention movement of attacking the female, the other of leading her to the nest (Tinbergen, 1951; van Iersel, 1953). Similar alternation behavior is to be seen in a rat placed in an approach/avoidance conflict: the animal vacillates about a point some distance from the goal.

(5) Ambivalent behavior. Sometimes intention movements appropriate to the two tendencies are combined into a single pattern which contains components appropriate to both. Thus, a half-tame moorhen (Gallinula chloropus) offered food may make incipient pecks towards it, and even swallowing movements, while simultaneously keeping its distance or even edging away: components of both feeding and fleeing are shown simultaneously. A number of examples of ambivalent postures have already been considered: they are often a patchwork of components, each of which depends on only one of the conflicting tendencies. Since both tendencies are influencing the effectors, the primary competition between them cannot be for the mechanisms of selective attention, but must be nearer the final motor pathways: indeed such postures occur especially when both tendencies are elicited by the same object.

(6) Compromise behavior. Compromise behavior (Andrew, 1956) differs from ambivalent behavior in that one element expresses both tendencies, whereas ambivalent behavior is a patchwork of elements from both. The approach/avoidance conflicts involved in courtship or threat often result in a circling movement round the object of the display.

(7) Autonomic responses. Perhaps because so many cases of conflict are associated with agonistic tendencies, they are frequently accompanied by autonomic responses. Defecation and urination in situations involving conflict or frustration are indeed often used as measures of 'anxiety' or 'emotionality'. Like other types of behavior in conflict situations, these autonomic responses may become ritualized in evolution to serve as social signals.

(8) Displacement movements. Animals in conflict situations sometimes show behavior which appears to be irrelevant to any of the tendencies which are in conflict. For instance, in many aggressive or sexual situations passerine birds may wipe their beaks or preen their feathers, show drinking or feeding behavior, or engage in some other activity which seems unrelated to the context. A stickleback wavering between attack and escape digs in the sand; an avocet in the same situation adopts the sleeping posture; a cat which would like to get up but is too lazy to do so licks its shoulder; a father, in two minds whether he should give his child a kiss or a slap for a pertinent but untimely remark in the presence of relatives, scratches himself behind the ear. Comparable examples have been described in many other groups, including arthropods, fishes and mammals (e.g. Tinbergen, 1940; 1952; Kortlandt, 1940; Crane, 1957). First labeled by Kirkman (1937) as 'substitute activities', they are now usually grouped under the heading of 'displacement activities'. The term displacement is thus used here in a sense different from that found in the psychoanalytic literature.
Displacement is thought to be due to inhibiting interactions between simultaneously stimulated antagonistic behavior systems in the animal (disinhibition hypothesis). Displacement movements are very characteristic of certain internal and external conflict and inhibition situations. In the circumstances concerned they are very conspicuous because they do not seem to fit into the framework of the situation. It was principally expressive movements derived from displacement movements, such as laughing, which defied Darwin's (1872) efforts to derive them from a 'serviceable movement' in accordance with his 'First Principle of Expression' (The Principle of Serviceable Associated Habits). In the framework of the situation at any given time displacement movement have never had a purpose, but some have acquired one secondarily in the service of social communication.

(9) Redirection activities. In redirected activities (Bastock, Morris & Moynihan, 1953), the motor patterns appropriate to one of the conflicting tendencies are shown, but are directed on to an object other than that which initially elicited them. This is often seen in winter flocks of birds: one individual is supplanted at a food source by a superior, and, instead of retaliating, turns its aggressiveness on to an inferior bird. Similarly male blackheaded gulls, whose tendency to attack their mates is inhibited, may attack other birds (Moynihan, 1955). During a boundary dispute, herring gulls may direct their fierce attacks not against each other but against grass tufts, which may be pulled at as if they were the opponent's plumage (Tinbergen, 1952; 1959). Redirection of aggression is a conspicuous phenomenon in primates and man (and a bottomless source of inspiration for cartoons). Actions using an alternative object may become expressive movements: the fist which dares not hit the opponent pounds the table top and so becomes an expression of violent but helpless or barely controlled rage.

(10) Sexual abnormalities. Certain types of sexual inversion among animals appear characteristically in conflict situations. For example, adoption of the female copulatory posture by the male occurs in the zebra finch when the female is unresponsive to the male's courtship (Morris, 1954), and after copulation attempts in carduelines (Hinde, 1955-6). This can be understood on the assumption that sexual arousal in either sex involves an increased tendency to show both male and female patterns, though male patterns normally have a higher priority in males and vice versa. If the behavior of the characteristic type is thwarted, the other may be shown. Similar principles apply to other species (Morris, 1955; Barraud, 1955).

(11) Regression. The literature on regression in animals is now fairly extensive, but there has been little attempt to analyze the processes underlying it. Indeed, the phenomena commonly lumped under this heading are probably physiologically heterogeneous. Some cases of juvenile behavior in adults imply that the mechanisms controlling the juvenile pattern remain present but latent in the adult, and become active when the adult pattern is thwarted (e.g. Holzapfel, 1949).

(12) Immobility responses (thanatosis, hypnoid-stuporous states: Portielje, 1948). When wild-caught great and blue tits are held in the hand, they sometimes lie on the outstretched palm for a minute or longer, their eyes open but bodies limp. They fly immediately at a sharp noise, or if they are thrown into the air. Similar responses to handling or stroking occur in a wide range of animals, and are often accompanied by analgesia: they have been ascribed to frustration or conflict (e.g. Armstrong, 1947; Ratner & Thompson, 1960). The relation of such responses to the 'freezing' postures used when escaping from a predator has not been investigated.

(13) Aggressive behavior. The definition of what constitutes 'frustrating conditions' can be somewhat flexible (e.g. Berkowitz, 1963), and it would be easy to coin a definition such that a territory-owning great tit was 'frustrated' by an intruder. The inclusion of incompatible response tendencies as a source of frustration (Brown & Farber, 1951) indicates that aggression should be mentioned as a possible response to a conflict situation, though the evidence for its importance in animals other than man is largely anecdotal.

(14) Responses to frustration. When an animal, engaged in a sequence of behavior, is unable to complete it because of a physical barrier, the absence of an appropriate stimulus link, or for similar reasons, its behavior may take a number of forms. One possibility is the appearance of investigatory behavior or trial and error, which may lead the animal into a situation which makes possible the completion of the chain. Another is a response to a normally inadequate stimulus situation, making possible the completion of the sequence albeit possibly in a nonfunctional manner. In other cases the animal may show one of the types of behavior already discussed as appearing in conflict situations, such as displacement or aggressive behavior. One feature which seems more frequent in frustration that in conflict situations is an increase in the vigor of the behavior which is being shown at the time.

(15) Severing contact. Chance (1962) was the first to point out how noticeably often an animal breaks off agonistic encounters with a conspecific by means of movements and postures which render it impossible for it to see its opponent arty longer ('cut-off acts'). Mostly these are escape intentions such as closing the eyes or turning the head, or displacement movements: many birds preen the underside of their wings ('displacement preening'). Intention movements of escape are perfectly understandable in such a situation, but the curious fact that displacement movements of the kind mentioned are so frequent when the animal actually ought to be keeping a careful watch on its opponent led Chance to the following conclusion: The sight of the opponent acts as a stimulus on the animal which through stimulus summation would eventually be bound to arouse either its escape or its fighting behavior if the stimulus flow were not interrupted. By turning away from the opponent and thus severing contact with the stimulus, the animal gives itself, so to speak, a respite in which to await developments, or a pause in fighting, and thus often the possibility of avoiding a fight or further fighting without leaving the scene or stimulating the opponent to a chase by running away. Chance quotes examples of this from his own observation of rats and from ornithological literature. Leyhausen (1956) described this 'cut-off' in cats as 'looking around'; this is of practical use to an inferior animal in strained social situations, precisely as Chance assumed, and also, above all, when the partners approach one another in courtship and mating. Such behavior, however, indicates that, on the one hand, an animal is not prepared to yield but also that, on the other, it is not for its part in an aggressive mood. Such a gesture of severing contact contains an offer of peace as well as a warning to the other not to push matters to the limit, and this is the effect it often produces, i.e., in many animals there are appropriate receptive IRMs (innate releasing mechanisms). According to the case, one aspect or another may be emphasized; the sleeping posture of the avocet became a pure gesture of threat, 'looking around' by the cat a gesture of inferiority to inhibit attack - but without being a 'submissive gesture', for if the cat is nonetheless attacked it defends itself energetically.
In at least partial accord with Darwin's (1872) Second Principle of Expression (the Principle of Antithesis), failure to look away, that is express staring, comes to mean an active challenge to the opponent, as Schenkel (1947) discovered in wolves, and as is also true in the case of cats and quite certainly a large number of other mammals, including man (Leyhausen, 1967).

Ritualization and Emancipation
The phenomena of ambivalence, redirection and displacement are not restricted to the occurrence of communication behavior. They are likely to produce 'interruptive behavior' whenever the ongoing behavior is thwarted because of simultaneous activation of incompatible behavior systems. However, exactly those occasions are the moments where communication between the opponents may be of considerable survival value. We must assume that in the course of evolution selection has favored the occurrence of these interruptive activities at these occasions. Moreover selective forces will have made them more suitable for their task by selecting for a constancy in form for each functionally different signal and often by making it more conspicuous through exaggerations of movements or addition of structures and colors (Baerends, 1975). The hypothetical process by which an activity becomes more suited to its new communicative task is called 'ritualization' (Huxley, 1914; 1923; 1967).
Ritualization, as for instance appearing in typical intensity (Morris, 1957) or typical frequency of an activity and in the regular alternation of components in successive ambivalence, must be made possible by special neurological changes. During such new developments, which seem to make the activity less dependent on the original causal factors and the old functional context and which, therefore, are often referred to as 'emancipation' (Tinbergen, 1952), a ritualized activity can, theoretically, become completely independent of the conflict from which it originated. However, according to Baerends, in all signal activities so far studied from this point of view, ties with the antagonistic tendencies still seem to be effective. For instance, in the so called 'enticing display' of the shelduck (Tadorna tadorna) - a display by which in this species the pair bond is formed and maintained - the details in the form in which the pattern was carried out are shown to be correlated with the relative frequencies of overt attack and of fleeing behavior preceding or following this display in the performing bird. The form correlated with a relatively high fleeing ratio was the one predominantly shown in the beginning of the season when the pairs were formed, the form correlated with a relatively high aggression rate chiefly occurred later in the season in females of already established pairs (Baerends 1974; 1975). As Lorenz (1958) has pointed out earlier, in the shelduck the attack component is always directed against a rival female, in other ducks, e.g., the mallard, the orientation of this movement has become emancipated, the attack component no longer being directed with respect to other females but only with respect to the body of the performing bird. For the agonistic displays of the black-headed gull Tinbergen (1959) has shown that transition stages between the distinct display types are much more rare than one would expect if the gradually shifting tendencies to attack and to flee would be entirely free in determining the form of the display. Although it has been made likely that each of these gull displays corresponds to a relative and absolute ratio of these tendencies (Moynihan, 1955), an intercalated emancipation mechanism must have developed promoting the appearance of different standardized displays, each with their own information content. The much higher continuity between different facial expressions or body postures with communicative function in some animals and man suggests that in some cases such an emancipation mechanism is absent or much less dominant (Baerends, 1975).
The influence of the aggressive and the escape systems on agonistic and courtship displays can be considered an important cause for the radiation in form of these displays in animals of closely related taxonomic groups (Baerends, 1975).
The fact that the form of the displays is species-specific and that differences between homologous displays are due to differences in genes does not necessarily imply that in the ontogenetical development of the displays no learning process would be involved. Kruijt (1964) has described such an interactional development for the red jungle fowl. He found that behavior patterns of escape were the first to appear, aggressive patterns came later and courtship patterns were the last to develop in the chicks. Kruijt obtained arguments that the development of aggressive behavior is facilitated by the occasional encounters the chicks have with each other during random escape jumping in the first phase. As soon as the aggressive patterns have appeared interaction with escape becomes possible and leads to displays. The development of the sexual factors later on seems to have a stabilizing affect on the aggression-escape interaction and the performance of the displays. Experiments in which the cocks during periods of different length (6-24 months) beginning at different ages (0-12 weeks) were deprived of the company of congener showed experience with other fowl to be essential for the developmental process. Cocks isolated at less than 10 weeks old for more than 10 months were seriously disturbed in their mating behavior; isolation at later ages had no such effects.

The Semantics of Signalling and Agonistic Displays
Animal signals can be partitioned roughly into two structural categories; discrete and graded, or, as Sebeok (1962) designated them, digital and analog. Discrete signals are dichotomous or binary (signals that can be represented in a simple off-or-on manner, signifying yeas or no, present or absent, here or there, and similar dichotomies), and are thought to have come about through the evolution of 'typical intensity' (Morris, 1957). That is, the intensity and duration of a behavior becomes less variable, so that no matter how weak or strong the stimulus evoking it, the behavior always stays about the same.
In contrast, graded (analog) signals have evolved in a way that increases variability. As a rule the greater the motivation of the animal or the action about to be performed, the more intense and prolonged the signal given (Wilson, 1975). Graded communication is strikingly developed in aggressive displays among animals. In rhesus monkeys, for example, a low-intensity aggressive display is a simple stare. They threaten not only with stares but also with additional displays on an ascending scale of intensity. The new components are added one by one or in combination: the mouth opens, the head bobs up and down, characteristic sounds are uttered, and the hand slaps the ground. By the time the monkey combines all these components, and perhaps begins to make little forward lunges as well, it is likely to carry through with an actual attack (Altmann, 1962). Its opponent responds by retreating or by escalating its own displays. These hostile exchanges play a key role in maintaining dominance relationships in the rhesus society. Squirrels reveal gradually rising hostility by tail movements that increase from a slow waving back and forth to violent twitching. Birds often indicate aggressive tendencies by ruffling their feathers or spreading their wings, movements which create the temporary illusion that they are larger than they really are. Many kinds of fish achieve the same deception by spreading their fins or extending their gill covers. Lizards raise their crests, lower their dewlaps, or flatten the side of their bodies to give an impression of greater depth. In short, the more hostile the animal, the more likely it is to attack and the bigger it seems to become. Such exhibitions are often accompanied by graded changes in both color and vocalization, and even by the scaled release of characteristic odors (Wilson, 1975). One of the most general principles of animal communication was first recognized by Darwin (1872) in The Expression of the Emotions in Man and Animals. Labeled by him the 'Principle of Antithesis', it can be expressed in an oversimplified manner as the following duality: When an animal reverses its intentions, it reverses the signal. The signal antitheses are most sharply defined in aggressive interactions. An animal that approaches another in a conciliatory mood, or else has lost a fight and is trying to appease the victor, uses postures and movements that are the opposite of aggressive displays. Darwin's own description of antithetic signaling in dogs is graphic and precise: “When a dog approaches a strange dog or man in a savage or hostile frame of mind he walks upright and very stiffly; his head is slightly raised, or not much lowered, the tail is held erect and quite rigid; the hairs bristle, especially along the neck and back; the pricked ears are directed forwards, and the eyes have a fixed stare. These actions, as will hereafter be explained, follow from the dog's intention to attack his enemy, and are thus to a large extent intelligible. As he prepares to spring with a savage growl on his enemy, the canine teeth are uncovered, and the ears are pressed close backwards on the head; but with these latter actions we are not here concerned. Let us now suppose that the dog suddenly discovers that the man he is approaching, is not a stranger, but his master; and let it be observed how completely and instantaneously his whole bearing is reversed. Instead of walking upright, the body sinks downwards or even crouches, and is thrown into flexuous movements; his tail, instead of being stiff and upright, is lowered and wagged from side to side; his hair instantly becomes smooth; his ears are depressed and drawn backwards, but not closely to the head; and his lips hang loosely. From the drawing back of the ears, the eyelids become elongated, and the eyes no longer appear round and staring” (Darwin, 1872).
When displaying aggressively, a gull stretches its head forward, the ritualized intention movement by which the bird indicates it is ready to peck at its enemy. But in order to appease an opponent, a gull turns its head 90° to the side. Two gulls attempting to conciliate reciprocally will stand side by side, or face each other with their bodies, but they will be momentarily gazing in opposite directions (Tinbergen, 1960). Dominant male rhesus monkeys raise their tails and display their testicles by lowering them; subordinates lower their tails and heads and raise their testicles. The dominant males also mount their subordinates in ritual pseudocopulation; the subordinates present themselves in a pseudofemale posture to be mounted.
Although such examples can be multiplied at length not all displays opposite in meaning are also antithetical in appearance to the human observer. Even appeasement displays sometimes incorporate wholly new elements unrelated to hostile signaling. Hyenas. for example, rely heavily on penis displays to conciliate one another; even the females are equipped with pseudopenes which they use with convincing skill (Kruuk, 1972). Rodents and primates routinely utilize grooming, while some birds and mammals revert to begging and other juvenile postures (Wickler, 1972).
Even though much convergence of signals exists in aggressive interactions, there is no universal code to which all species of a group subscribe. In the mammals, for example, we find appeasement behavior following much the same form in species after species: the animal tends to crouch, often rolling over to expose its flank or belly. Lorenz (1966) suggested that in some mammals such as the dog the exposure of these most vulnerable parts cancels the aggressive impulse of the opponent. However, the belly-up posture does not invariably mean submission. Among shrews it signifies hostility and dominance - with good reason, since the shrew's best fighting position is on its back (Ewer, 1968).
By combining signals it is possible to give them new meanings. The theoretical upper limit of a combinatorial message is the 'power set' of all of its components, or the set of all possible combinations of subsets. Thus, if A, B, and C are three discrete signals, each with a different meaning, and each combination produces still one more message, the total ensemble of messages possible is the power set consisting of seven elements: A, B, C, AB, AC, BC, and ABC. No animal species communicates in just this way, but many impressive examples have been found in which conspicuous signals are used effectively in different combinations to provide different meanings. A case from the horse family (Equidae) embraces both discrete and graded signals. A zebra or other equid shows hostility by flattening its ears back and friendliness by pointing them upward (discrete signals). In both postures the intensity is indicated by the degree to which the mouth is opened simultaneously (a graded signal). The mare is able to produce a third message by adding two more components: when ready to mate, she presents the stallion with the threat face but at the same time raises her hindquarters and moves her tail aside (Trumler, 1959).
Among vertebrates especially, signals transmitted through different sensory channels are often combined in ways that increase information. In some instances the signals are simply redundant: the simultaneous hissing and body jerking of a chameleon, for example, and the stretch display of a male snowy egret delivered with its courtship call; in both cases the precision of the message is increased, although the combinations add no new meaning not already present in the separate elements.
Components in different modalities can be added as part of the graded intensification of a signal. In closely grouped societies of primates, such as the dense troops of macaques and baboons, the threats of lowest intensity are typically visual in nature. When these visual signs are intensified, characteristic sounds are added for reinforcement (Wilson, 1975).
It is possible for hostile and submissive displays to be combined orthogonally to generate new messages. In other words, the displays do not form a simple spectrum ranging from most hostile at one end to most submissive at the other, but rather constitute two sets of signals that can be presented either separately or in combination. When combined, the signals create a message containing a high level of ambiguity. In the domestic cat a high intensity threat combined with a high-intensity fear-display produces the 'halloween cat' posture: body raised on fully extended limbs and mouth closed (threat); also, body arched and ears flattened (fear). This mosaic of postures is ambiguous with reference to the basic signals but provides new information in a different category. The cat can be interpreted by the human observer, and presumably by other cats as well, as being in a highly excited state, ready to be tipped into either a violent fight or precipitous flight. This message is distinct from the high-intensity states of the purely aggressive or purely submissive postures; it is also different in meaning from the more relaxed mosaic posture of a cat displaying low-intensity aggression and fear (Leyhausen, 1956), This rather involved interpretation has received some neurophysiological support. Using implanted electrodes, Brown, Hunsberger & Rosvold (1969) elicited composite aggressive and flight behavior in cats by simultaneously stimulating hypothalamic centers previously shown to control the responses independently. Similar combinations of signals, more or less orthogonal in nature, have been described in the wolf and dog (Schenkel, 1947). (Wilson, 1975).

A peculiar form of composite signaling is metacommunication, or communication about the meaning of other acts of communication (Bateson, 1955). An animal engaged in metacommunication alters the meaning of signals belonging to categories other than the original signals that are being transmitted either simultaneously or immediately afterward. Altmann (1962), who first applied this concept extensively to the behavior of nonhuman primates, recognized two circumstances in which metacommunication occurs. The first is status signaling. A dominant male rhesus monkey can be recognized by his brisk, striding gait; his lowered conspicuous testicles; the posture of his tail, which is held erect and curled back at the tip; and his calm 'major-domo' posture. during which he gazes in a confident, unhurried manner at any other monkey catching his attention. A subordinate male displays the opposite set of signals. Similar signaling has been recorded in other species of macaques and baboons. Altmann's hypothesis is that the displaying animal communicates his own knowledge of its status and therefore the likelihood that it will attack or retreat if confronted. Since the individual troop members know one another personally, they can judge for themselves whether particular rivals are prepared to alter the dominance order. They evaluate the general 'attitude' of the other members of the society. This explanation is eminently plausible but has not yet been subjected to any convincing test (E.O.Wilson, 1975).
The second form of primate metacommunication is play invitation. The play of rhesus monkeys, like that of most other mammals, is devoted largely to mutual chasing and mock-fighting. The invitation signals consist of gamboling and gazing at playmates from between or beside their own legs with their heads upside down. in the play that ensues, the monkeys wrestle and mouth one another vigorously. Although easily capable of hurting one another, the monkeys seldom do. Real damage will result later from escalated versions of the same behavior during bouts of intense aggression. Play signaling says approximately the same thing as the simple human message: 'What I am doing, or about to do, is for fun; don't take it seriously. In fact - join me' (Wilson 1975). Metacommunicative play signaling in dogs was first described by Darwin (1872): “When my terrier bites my hand in play, often snarling at the same time, if he bites and I say 'gently, gently', he goes on biting, but answers me by a few wags of the tail, which seems to say 'Never mind, it is all fun'”. Dogs initiate play with one another by abruptly crouching down with their forelegs extended stiffly forward and by barking. Both signals appear to be ritualized aggression intention movements (Loizos, 1967; Bekofft 1972).
Where do the communication codes of animals come from in the first place? Any evolutionary change that adds to the communicative function has been called 'semanticization' by Wickler (1967). At one conceivable extreme of the semanticizing process, only the response evolves. Thus the sensory apparatus and behavior of the species is altered in such a way as to provide a more adaptive response to some odor, movement, or anatomical feature that already exists and that itself does not change. Male lobsters and decapod crabs, for example, respond to the molting hormone (crustecdysone) of the female as if it were a sex attractant. It is possible, although not yet proven, that crustecdysone has assumed a signaling function entirely through an evolved change in male behavior.
The vast majority of known cases of semantic alteration, however, involve 'ritualization', the evolutionary process by which a behavior pattern changes to become increasingly effective as a signal. Commonly and perhaps invariably, the process begins with some movement, anatomical feature, or physiological trait that is functional in quite another context acquires a secondary value as a signal. For example, members of a species can begin by recognizing an open mouth as a threat or by interpreting the turning away of an opponent's body in the midst of conflict as an intention to flee. During ritualization such movements are altered in a way that makes their communicative function still more effective. Typically, they acquire morphological support in the form of additional anatomical structures that enhance the conspicuousness of the movement. They also tend to become simplified, stereotyped, and exaggerated in form.
Such ritualized biological traits are referred to as displays. A special form of display is the 'ceremony', a highly evolved set of behaviors used to conciliate and to establish and maintain social bonds (Wilson, 1975). According to Schuster (1978), the adjective 'stylized' (Ewer, 1968) seems more appropriate for fighting that is shaped mostly by the advantages of noninjury rather than for communication.
The ritualization of vertebrate behavior often begins in circumstances of conflict, particularly when an animal is undecided whether to complete an act. Hesitation in behavior communicates, to onlooking members of the same species the animal's state of mind or, to be more precise, its probable future cause of action. The advertisement may begin its evolutionary transformation as a simple intention movement. Birds intending to fly typically crouch, raise their tails, and spread their wings slightly just before taking off. Many species have independently ritualized one or more of these components into effective signals (Daanie, 1950; Andrew, 1956). In some species white rump feathers produce a conspicuous flash when the tail is raised. In others the wing tips are flicked repeatedly downward to uncover conspicuous areas on the primary feathers of the wings. In their more elementary forms the signals serve to coordinate the movement of flock members and perhaps also to warn of approaching predators. When hostile components are added, such as thrusting the head forward or spreading the wings as the bird faces its opponent, flight intention movements become ritualized into threat signals. But the more elaborate and extreme manifestations of this form of ritualization occur where the basic movements are incorporated into courtship displays (Wilson, 1975).
Signals also evolve from the ambivalence created by the conflict between two or more behavioral tendencies (Tinberqen, 1952). When a male faces an opponent, undecided whether to attack or to flee, or approaches a potential mate with strong tendencies both to intimidate and to court, he may at first choose neither course of action. Instead he performs a third, seemingly irrelevant act. He redirects his aggression at some object nearby such as a pebble, a blade of grass, or a bystander, who then serves as a scapegoat. Or the animal may abruptly switch to a 'displacement activity': a behavior pattern with no relevance whatever to the circumstance in which the animal finds itself. The animal preens itself, for example, or launches into ineffectual nest-building movements, or pantomimes feeding and drinking. Such redirected and displacement activities have often been ritualized into strikingly clear signals in courtship. As Tinberqen (1952) expressed the matter, these new signals are derived from preexisting motor patterns - they have been 'emancipated' in evolution from the old functional context.
The concept of ritualization was originated by Julian Huxley (1914; 1923; 1966). Van Hooff, (1972) believes that smiling and laughter in human beings can be homologized in a straightforward way with similar and equally complex displays used by the other higher primates. Smiling, according to van Hooff's hypothesis, was derived in evolution from the 'bared-teeth display', one of the phylogenetical most primitive social signals. The members of most primate species assume this expression when they are confronted with an aversive stimulus and have a moderate to strong tendency to flee. The display intensifies when escape is thwarted. In higher primates the bared-teeth display is commonly silent in expression. Among chimpanzees it is furthermore graded in intensity and is used flexibly to establish friendly contacts within the troop. The 'relaxed open-mouth display', often accompanied by a short expirated vocalization, is a signal ordinarily associated with play. In man these two signals, the silent bared-teeth display and the relaxed open-mouth display, appear to have converged to form two poles in a new, graded series ranging from the general friendly response (smile) to play (laughter).
A third kind of signal that developed from the archaic facial expressions is the bared-teeth scream display. This behavior, which is widespread in primates but missing in man, indicates extreme fear and submission, as well as readiness to attack if the animal is pressed further.
The complex social signals involved in ritualized aggression are well-illustrated by Goodall's (1968) classic study of wild chimpanzees in the Gombe Stream Reserve in Tanzania. Chimps are relatively peaceful, but order is maintained by the threats of high ranking males. There is no single stereotyped threat gesture, but rather a number of rituals which may be employed in varying combinations, including the following: 1. Glaring or staring an opponent into submission. 2. Tipping or jerking the head. 3. Raising an arm high above the head or hunching the shoulders. 4. Arm waving while running bipedally. 5. Swaggering back and forth from foot to foot. 6. Stamping the ground with the feet. 7. Branching, a practice which consists of grabbing a small tree and vigorously shaking it, possibly even uprooting it and dragging it around. 8. Aimed throwing of rocks at an opponent. 9. Hair erection. 10. Hooting and screaming.
Opponents usually flee, but if they do not a fight may erupt which consists of rolling on the ground with an opponent and pounding him with both feet and hands while also biting, slapping, and pulling hair. In some cases fights may take place instantly without warning as if it were the result of a previous quarrel.
In order to head off an attack, a threatened chimp may adopt one of several possible submissive postures: 1. sexual presenting (raising rump toward opponent); 2, whimpering or screaming with an open mouth with lips retracted exposing the teeth and gums 3. bowing, bobbing, or crouching; 4. kissing on the lips or groin; 5. reaching out to be touched; 6, submissive mounting and pelvic thrusting, regardless of the sex of the combatants; this may be accompanied by a grasping of the scrotum if the dominant individual is a male.
Another set of gestures, sometimes called reassurance gestures, are used by a dominant animal to indicate he will not harm a subordinate. They include: patting on the head, touching the subordinate with a finger or hand, inspection or touching of the genitals, hand holding, embracing, mounting the submissive individual, rump-turning, and kissing. If a dominant individual withholds reassurance, the subordinate may become very upset and have a 'temper tantrum'. Sometimes a subordinate female may attempt to calm a dominant male who has just been in a squabble by touching him, grooming his fur, or fondling his scrotum. In summary, there are a number of gestures and rituals, and their precise meanings are partly contextually determined. For example, mounting and thrusting in a non-sexual situation may be an expression of either dominance or submission depending on the context.
It should also be pointed out that squabbles are often peacefully settled through elaborate gestures, but sometimes these rituals fail and a subordinate chimp may take a merciless beating.
An example of a typical encounter is the following, described by Goodall (1968): “An adolescent male (Pepe) approached a mature male (Goliath) who was feeding on bananas. Pepe crouched and whimpered, looking at Goliath, and meekly extended his hand. Goliath patted Pepe, and Pepe slowly reached for a banana, but pulled back screaming in fear. Goliath patted him again and finally Pepe gathered a few bananas and hastily retreated”.
The preponderance of ritualized combat over physical combat is illustrated by a recent study of agonistic behavior among elephant seals living off the coast of California (LeBoeuf & Peterson, 1969). It was reported that for every actual fight there were 67 aggressive encounters which never went beyond ritualistic threats.
A considerable amount of fighting in primates involves bluffing an opponent into intimidation. Kummer (1968) reports that baboons often threaten to bite their opponents an the neck, but hundreds of such threats result in only a few actual bites. The display of threatening or submissive social signals without actual fighting is by far the most common form of agonistic behavior, and most animals quickly recognize and respond to such gestures. It is because these rituals are so effective that overt fighting becomes unnecessary.
Such aggressive behavior has become so ritualized that it often has the appearance of a ceremony rather than a confrontation. Aggressive displays and gestures are easy to recognize because they generally involve marked physical changes which make the combatants appear larger and more threatening. A cat arches its back, raises its fur, bares its teeth, and extends its claws. Birds may fluff their feathers and spread their wings (e.g., Weltyt 1962).
A baboon displays its large canine teeth and erects the fur around its shoulder, which instantly gives it a gorilla-sized appearance (DeVore, 1965). Some fish intensify their color, swish their tail, and extend their fine to maximum size. The agonistic behavior of the Hawaiian coral reef fish (Pomacentris jenkinsi) consists almost entirely of displays, and fights are won and lost by visual cues alone (Rasa, 1969). The normally yellow eye changes to a grayish-black to indicate aggressive motivation, and the raising of a dorsal fin is correlated with fear. In many animals threat displays are accompanied by menacing vocalizations such as hisses, snarls, screeches, or shrieks. Even fish have been found to grunt in an effort to amplify their threats (Brawn, 1961).

Inclusive Fitness and the Optimal Level of Aggressiveness
Tinbergen (1968) has suggested that the problem of causation must be approached with the knowledge that one is dealing with four subquestions which are related to each other by a temporal dimension. In asking the question: 'What makes an animal fight?', one is asking the following: (1) what are the immediate, effective, internal and external factors? (immediate instance causation); (2) what are the particular social experiences in this animal's developmental history that predisposed him toward aggressive behavior? (developmental or ontogenetic causation); (3) what are the historical selection pressures that fostered the evolution of aggressive behavior in this species? (evolutionary or phylogenetic causation); and (4) in what way does this behavior influence the survival, the success of the animal? (survival value).
The answer to each of these subquestions will be quite different, yet they are obviously interdependent. The immediate factors (instigators) always operate on an animal which has had a particular developmental history, and to which he has brought a species-typical genetic heritage. Although his individual complement of genes is unique, they are drawn from a species' gene pool that determines within general limits, his inborn neurological, chemical, and behavioral characteristics (Boelkins & Heiser, 1970).
Tinbergen (1956) has noted that aggression is a time and energy consuming activity (and sometimes injury-producing or fatal). So that it is not the sort of behavior a species should engage in casually. An even more important constraint operating against any tendency to extreme aggressiveness is that the 'costs' in evolutionary terms may outweigh the benefits (Wilson, 1970;
It might be argued that ideally there should be no aggressive behavior at all, but this reasoning is based on the assumption that all conflict is intrinsically bad and without adaptive value. However, agonistic behavior may be either constructive or destructive; it is mainly destructive fighting which is maladaptive. In the long run, agonistic behavior helps assure social stability and prevents serious destructive behavior. A total lack of agonistic behavior would be valuable only in an unchanging world where there were no competition for survival (R. Johnson, 1972).
This is not equivalent to asserting that aggression is always and everywhere adaptive, or that the mere existence of any behavior represents 'a priori' evidence that it is eufunctional in evolutionary terms. There are many instances of once adaptive behaviors that have become maladaptive or relics as the environmental context has changed.
Aggressiveness is a behavioral category of only limited usefulness in relation to the total survival and reproduction problem of a species, and for most species there is a sort of optimal level of aggressiveness (in many species, the optimal level is, of course, zero) (Corning, 1971).
For each species, depending on the details of its life cycle, its food preferences, and its courtship rituals, there exists some optimal level of aggressiveness above which individual fitness is lowered. For some species this level must be zero, in other words the animals should be wholly nonaggressive. There are in fact at least three kinds of constraints on the evolutionary increase of aggressiveness.
First, as Tinbergen (1956) and others have pointed out, an aggressor spends time and energy that could be invested in courtship, nest building, and the feeding and rearing of young. The adverse effects of such 'aggressive neglect' on reproduction have been documented, for example, in the gannet ('Sula bassana') by Nelson (1964; 1965), and in sunbirds and honey eaters by Ripley (1961). Its theoretical implications have been explored by Hutchinson & MacArthur (1959).
Second, both participants in an aggressive encounter are likely to be hurt, for combat brings risk of injury to both parties (Hinde, 1974). For the aggressor there also exists a danger that its hostility will be directed against unrecognized relatives. If the rates of survival and reproduction among relatives are thereby lowered, then the replacement rate of the genes held in common descent between the aggressor and its relatives will also be lowered. Since these genes will include the ones responsible for aggressive behavior, such a reduction in the 'inclusive fitness' (the summed fitness of the aggressor and genes held in common descent) will work against aggressive behavior as well. This process will continue until the difference between the advantage and disadvantage, measured in units of inclusive fitness, is maximized. The theory of the subject has been developed mathematically by Hamilton (1964). The constraints on aggression are such that even when aggression occurs as a genetically determined trait it can be expected to be programmed in such a way as to be brought into play only when it gives a momentary advantage. Such episodes in the life of an animal may be few and far between, yet their rarity in a particular case must not mislead us into assuming that the behavior is not 'natural', i.e., adaptive and genetically programmed (Wilson, 1970; 1975). Wilson holds that for any given species the aggressive responses vary in what can properly be called a genetically programmed manner. At low population densities, to take one conceivable example, all aggressive behavior may be suspended. At moderate densities it may take a mild form such as intermittent territorial defense. At high densities territorial defense may be sharp, while joint occupancy of land is also permitted under the regime of dominance hierarchies. Finally, at extremely high densities, the system may break down almost completely, transforming the pattern of aggressive encounters into 'social pathology'. Whatever the specific program that slides individual responses up and down the aggression scale, however, each of the various degrees of aggressiveness is adaptive at an appropriate level of population density - short of the rarely recurring pathological levels. In sum, it is the total pattern of responses that is adaptive and has been selected for in the course of evolution (Wilson, 1970; 1975).
For example, investigations by Parzefall (1974) on epigeal and cave forms of the fish Poecilia sphenops suggest that in the absence of selectional pressures aggression may be reduced because this behavior lost its significance in the particular cave biotope.

Is Aggression Adaptive?
Although ethologists stress the adaptive significance of aggression, they oversimplify animal aggression as ritualized peacefulness. Rituals are explained as a convergent evolutionary solution to the problem of minimizing injury and death when animals compete for food and mates. Animal aggression, however, is a collection of varied kinds of behaviors arising under many ecological pressures. Apart from suggesting that some animal societies are intrinsically peaceful while others are held in check by strong inhibitions, ethologists have largely ignored quantitative differences in the frequency of aggression and qualitative differences in the degrees of ritualization, fighting, and killing. They do not explain why social species as different as gorilla (Schaller, 1963) and herring gull (Lorenz, 1966) are peaceful while others, such as many cichlid fish, are very aggressive. Nor is it obvious why intraspecific killing is recognized as a normal occurrence in several highly successful species (Wilson, 1975), including the allegedly peaceful lion (Schaller, 1972).
The issue is far from trivial, since Lorenz stresses the prevalence of human violence and murder as evidence for a unique case of nonadaptiveness. The recognition of adaptive violence and the great variation in aggressiveness among animals can counter the assumption that human aggression is necessarily a sign of defect in society (Montagu, 1973). Much of the diversity in aggressiveness among animal species can be traced to the social structure of animal life. Eisenberg (1966), Hinde (1974), and Wilson (1975) have surveyed the many ways in which animals live in groups and the biological significance of this diversity.
Nearly all conditions known to increase aggressiveness are found on a lek (Schuster, 1978). Sexual selection, in general, is most extreme in species that compete only for mating partners (Wilson, 1975). The males of lekking birds - including birds of paradise, manakins, grouse, and pheasants - combine decoration and aggressiveness to a high degree (Darwin, 1871; Lack, 1968; Wilson, 1975). The white-bearded manakin (M. manacus trinitatis) regularly attacks other species when defending a territory (Lill, 1974).
Innate aggressiveness is complemented by lek conditions. Territories are small and clustered, combining frequent territorial interactions with a crowding effect. Because of turnover at the mating center and consequent reshuffling of some other positions, occupants do not hold fixed locations for long periods; thus, the dear-enemy effect is reduced. When females move through peripheral territories to reach the mating center, they are courted and chased by many males along the route, who penetrate adjacent territories. There is an overall centrifugal force exerted by the mating center, so that more central territories are smaller and yet the most coveted and contested. Sexual frustration could also be a factor, since all lek occupants are presumably in rutting condition. Hunger might also be present, since lek occupants, particularly near the center, are continuously displaying and rarely feed or drink (Schuster, 1978). The lek organization of the lechwe includes a number of unique features which probably contribute still further to their aggressive potential, making them possibly the most aggressive of territorial species (Schuster, 1976). During the peak of the breeding season, an overall sense of tension and instability prevails, especially during periods of organization. A single unsettling influence, like a female running to or from the mating center, can spark off a period of complete breakdown, when fights and chases seem to be erupting everywhere. In other respects, lechwe aggression is a model example of the ethologists' rule that aggression evolves to minimize the chance of injury (Schuster, 1978).
Injury and killing, however, seem to be normal and adaptive in a few species. Lorenz's impression of the bucolic life among social carnivores is not supported by detailed studies of lion (Panthera leo) (Schaller, 1972), spotted hyena (Crocuta crocuta) (Kruuk, 1972), and wolf (Canis lupus) (Mech, 1970). In these animals, a small group inhabits a territory within which it hunts and reproduces. Within the group, social interactions include many dominance and submissive behaviors to control aggression. But violence and death are features of territorial clashes between neighboring groups of hyenas and wolves and between male lions for possession of a female group and its territory. There is also a high rate of infant mortality among lions, since the adults sometimes kill the cubs or else the female abandons them or does not feed them. The effect is to keep down the population size and density, permitting individual prides to hunt over areas sometimes more than 100 square miles. This is also a feature of wolf society and is believed to be necessary for large predators.
In general, the antelope and carnivore examples illustrate that the adaptiveness of aggression in animals, including species with frequent aggression or killing, can be understood as a feature of social organization for coping with particular environments. Lack (1969) and Marler (1976) have also emphasized that all competitive social systems, even the most aggression-free, can sentence losers to peripheral areas where isolation, poor food, predation, and a failure to breed can lead to a final solution as surely as violent death (Schuster, 1978).
Is aggression in man really adaptive? From the biologist's point of view it certainly seems to be. It is hard to believe that any characteristic so widespread and easily evoked in a species as aggressive behavior in man could be neutral or negative in its effects on individual survival and reproduction. To be sure, overt aggressiveness is not a trait in all or even the majority of human cultures. But in order to be adaptive it is enough that aggressive patterns be invoked only under certain conditions of stress such as those that might arise during food shortages and periodic high population densities. It also does not matter whether the aggression is wholly innate or is acquired in part or wholly by learning. We are now sophisticated enough to know that the capacity to learn certain behaviors is itself a genetically controlled and therefore evolved trait (Wilson, 1970; 1975; Cf. Lorenz, 1965; Eibl-Eibesfeldt & Wickler, 1968; Nash, 1970; Seligman & Hager, 1972; Hinde & Stevenson-Hinde, 1973). Such an interpretation, which follows from our information on patterned aggression in other animal species, is at the same time very far removed from the sanguinary view of innate aggressiveness which was expressed by Dart (1953) and had so much influence on subsequent authors:
“The blood-bespattered, slaughter-gutted archives of human history from the earliest Egyptian and Sumerian records to the most recent atrocities of the Second World War accord with early universal cannibalism, with animal and human sacrificial practices or their substitutes in formalized religions and with the worldwide scalping, head-hunting, body-mutilating and necrophiliac practices of mankind in proclaiming this common blood lust differentiator, this mark of Cain that separates man dietetically from his anthropoidal relatives and allies him rather to the deadliest of Carnivora”.
This is, as Wilson (1970) comments, obviously bad anthropology, bad ethology, and even bad genetics. It is equally wrong, however, to accept cheerfully the extreme opposite view, espoused by many psychologists, that aggressiveness is only a neurosis brought about by abnormal circumstances and hence, by implication, nonadaptive for the species.

Behavioral Scaling
Troops of vervet (Cercopithecus aethiops) observed by Struhsaker (1967) at the Amboseli Masai Game Reserve, Kenya, are strictly territorial and maintain rigid dominance hierarchies by frequent bouts of fighting. In contrast, those studied by Gartlan (1966) in Uganda had no visible dominance structure at the time of observation, exchange of males occurred frequently between troops, and fighting was rare. In some cases differences of this sort are probably due to geographic variation of a genetic nature, originating in the adaptation of local populations to peculiarities in their immediate environment. Some fraction can undoubtedly also be credited to tradition drift. But a substantial percentage of cases do not represent permanent differences between populations at all: the societies are just temporarily at different points on the same 'behavioral scale'. Behavioral scaling is variation in the magnitude or in the qualitative state of a behavior which is correlated with the life cycle, population density, or certain parameters of the environment. It is a useful working hypothesis to suppose that in each case the scaling is adaptive, meaning that it is genetically programmed to provide the individual with the particular response more or less precisely appropriate to its situation at any moment in time. In other words, the entire scale, not isolated points on it, is the genetically based trait that has been fixed by natural selection (Wilson, 1970; 1975).
In the published cases of scaled social response, the most frequently reported governing parameter is population density. Aggressive encounters among adult hippopotami, for example, are rare where populations are low to moderate. However, when populations in the Upper Semliki near Lake Edward became so dense that there was an average of one animal to every five meters of riverbank, males began to fight viciously, sometimes even to the death (Verheyen, 1954). When snowy owls (Nyctea scandiaca) live at normal population densities, each bird maintains a territory about 5,000 hectares in extent and it does not engage in territorial defense. But when the owls are crowded together, particularly during the times of lemming highs in the Arctic, they are forced to occupy areas covering as few as 120 hectares. Under these conditions they defend the territories overtly (Pitelka in Schoener, 1968). A similar density threshold for the expression of territorial defense has been reported in European weasels (Mustela nivalis) by Lockie (1966).
A second class of aggression scaling effects is the transitions that occur in many vertebrate species from territorial to dominance behavior as the density passes a critical value. Perhaps the first explicit description of this kind of scaling was that of Shoemaker (1939), who found that canaries forced together in small spaces become organized into dominance orders. Given more space, they establish territories (the natural condition for Serinus canaria in the wild), even though low-ranking individuals continue to be dominated around bath bowls, feeding areas, and other nonterritorial public space.
The phenomenon has been subsequently documented In other birds (review in Armstrong, 1947), sunfishes and char (Greenberg, 1947; Fabricius & Gustafson, 1953), iguanid lizards (Evans, 1951, 1953), house mice (Davis, 1958), Norway rats (Barnett, 1958; Calhoun, 1962), 'Neotoma' wood rats (Kinsey, 1971), woodchucks (Bronson, 1963), and cats (Leyhausen, 1956). Kummer (1971) developed the concept with special reference to the social evolution of primates. The clearest cases are found in species, such as certain lizards and rodents, in which the normal state is for solitary individuals or pairs to occupy territories. When forced together, groups of these individuals shift dramatically to despotisms or somewhat more complex dominance orders (Evans, 1953). In most such cases the shift from territoriality to a dominance system is really superficial in nature. In the case of despotism, one individual in effect retains its territory while merely tolerating the existence of the others. Such transitions are not limited to laboratory experiments. In Mexico, Evans (1951) found a crowded colony of the large black lizard Ctenosaura pectinata living on a cemetary wall, which provided shelter from which the lizards ventured to feed in a nearby cultivated field. At least eight adult males made up a dominance hierarchy, with one individual playing the role of a strong tyrant.
Not all density-dependent social responses consist of aggressive behavior. When populations of European voles (Microtus) reach certain high densities, the females join in little nest communities, defend a common territory, and raise their young together (Frank, 1957). In a basically similar way, flocks of wild turkeys (Meleagris gallopavo) gradually increase in size as population density increases (Leopold, 1944). When salmon (Salmo salar) and trout (S. trutta) are crowded in hatcheries enough to disrupt territorial behavior, they shift not to hierarchies but to schools (Kalleberg, 1958). The same transition occurs in crowded natural populations of the ayu (Plecoglossus altivelis), a Japanese salmonid (Kawanabe, 1958). The banded knife-fish (Gymnotus carapo), one of the species that orient and communicate by electric discharges, displays the reverse behavioral scaling from all other known vertebrates: dominance hierarchies at low densities and territories at higher densities (Black-Cleworth, 1970).
Group size itself can affect the intensity of aggressive behavior in ways that can be reliably dissociated from the parallel influence of population density. Blue monkeys (Cercopithecus mitis) of the Budongo Forest, Uganda, are organized in troops of highly variable size. When one large group encounters another at a rich food source such as the fruiting fig tree, the adults threaten and chase one another until one group retreats from the area. Small parties, however, coalesce peacefully when they meet at feeding places (Aldrich-Blake, 1970). It is tempting to speculate that territorial behavior develops in the troops only when their size becomes so large that they have to compete with other large troops for sufficient food. In other words, they resort to aggression only if it is profitable. Aggressiveness also increases as a function of group size in colonies of many kinds of social insects (Wilson, 1975).
The availability and quality of food can also move groups along behavioral scales. Even the way food is distributed in the environment can evoke strong variation in social behavior. For example, Chalmers (1968) found that mangabeys (Cercocebus albigena) have more aggressive interactions when they are feeding on large fruit growing singly than when the food is evenly dispersed in the trees. According to Rowell (1969) forest-dwelling baboons (Papio anubis) in Uganda display a parallel scale of aggression. In most cases their food is spread out and abundant, and aggressive behavior is rare. But when they encounter piles of elephant dung old enough to contain sprouting seedlings of the kind prized as food, they exchange threats in attempts to gain possession.
Although some species display phenotypic variability that covers a substantial portion of the gradients, in other words utilize true behavioral scaling, many others are fixed at a single point. The males of sea lions, elephant seals, and other harem-forming pinnipeds maintain territories with about the same intensity regardless of population density. The adaptive significance of such rigidity is clear. Aggressive behavior in these animals serves the single function of acquiring harems. The means by which this goal is reached and its value to genetic fitness are unaffected by changes in the density of animals on the hauling grounds. In such cases, shifts from one point to another on the behavioral gradients occur by evolution, but probably only when changed environmental circumstances alter the optimum social strategy (Wilson, 1975).

Social Systems, Aggression, and Evolution
The social system of a species is clearly one of its most important biological characteristics. If we make the reasonable assumption that even complex animal behavior reflects at least some component of genotype, then we must expect social systems to have evolved, just as structure and physiology have evolved. As the products of evolution, social systems should thus be adaptive; that is, they should be peculiarly adjusted to each environment in which they occur so as to confer maximum reproductive success upon their practitioners. Thus, the kaleidoscopic diversity of animal social systems takes on new order and significance when viewed as part of the adaptive mechanics of a species, no less 'real' or amenable to scientific analysis than metabolic adaptations for water conservation among desert rodents or stereoscopic vision among arboreal primates (Barash, 1974).
In recognition of this, the new field of 'socioecology' has emphasized comparative analyses of social organization in free-living animals (Crook, 1970). Illustrative is a study of the evolution of marmot societies by Barash (1974) which will be described in some detail. The woodchuck (Marmota monax), is best known east of the Mississippi, where it most commonly inhabits fields and forest ecotones at low elevations. In such environments, the woodchuck generally experiences a relatively long vegetative growing season, exceeding 150 days in southern Pennsylvania, for example, where the species has been intensively studied. In this comparatively equitable environment, woodchucks are solitary and aggressive; the association between adult male and female is essentially limited to copulation, the only lasting social tie being the mother-young nexus, which itself terminates at weaning when the young disperse. By contrast, the Olympic marmot (M. olympus) is highly social, living in distinct, closely organized colonies usually composed of several adults (most commonly one male and two females), 2 year olds, yearlings, and the young of the year. This species is highly tolerant and playful, commonly feeding in social groups of 3 to 6 individuals. No territories or even distinct individual home ranges are maintained; all parts of the colony are equally available to all colony members. Dominance relationships are generally indistinct and nonpunitive. Olympic marmot social life is characterized by a high frequency of active 'greeting' behaviors, previously described for other sciurids and apparently associated with individual recognition. Social organization in the Olympic marmot is thus entirely different from that in woodchucks. The environments of these two species are also at opposite extremes: Olympic marmots inhabit exclusively the alpine meadows in Olympic National Park, Washington. Their colonies are located at or above timberline where they experience very short growing seasons of 40 to 70 days. These differences between the two species therefore suggest a possible correlation between length of the growing season and type of social system, the shorter growing season being associated with increased sociality.
Fortunately, marmots living at intermediate elevations have also been studied. The yellow-bellied marmot (M. flaviventris) inhabits a wide range of environments, from sea-level prairies in eastern Washington through a variety of montane meadows in the Sierras, southern Rockies, and southern Cascades. This species has been intensively studied at intermediate elevation in Yellowstone National Park, Wyoming, where it experiences an appropriately intermediate growing season of approximately 70 to 100 days. It is significant that in this 'intermediate' environment, the social system of the yellow-bellied marmot is also intermediate. This marmot lives in recognizable colonies composed of numerous adults, and is thus considerably less solitary than the woodchuck; but individually distinct home ranges are maintained, physical spacing between individuals is relatively great, and infrequent social interactions are overlain with considerable aggressiveness, indicating that M. flaviventris has a less highly socialized behavioral system than the Olympic marmot.
Barash (1974) offers the following hypothesis: “Given an apparent causative relation between aggressiveness and the onset of dispersal, selection acting in each environment may have favored the degree of aggressiveness which results in optimum age at dispersal”. Barash, in fact, suggests that marmot social evolution in progressively more severe environments involves corresponding decrements in aggressive behavior in order to accommodate the animals' need to disperse at an increased age. Early dispersal among woodchucks is assumed to be a function of the adults' aggression toward the young and, if one considers the woodchuck's generally intolerant nature, this assumption seems reasonable. In nutria (Myocastor coypus) populations, Ryszkowksi (1966) found that in one set of circumstances the level of aggressiveness between individuals becomes high and dispersal occurs before damage to the vegetation. In other circumstances high population densities are tolerated and deterioration of the vegetation ultimately forces emigration to occur. It is not clear, however, whether these circumstances are purely ecological or not.
In a survey of primate social systems, Crook & Gartlan (1966) suggest that the 'hamadryas-patas-gelada' type of social system is basically an adaptation to sparse feeding resources seasonally more limiting than Papio cynocephalus and Macaca habitats. The one-male group in such conditions is considered the most efficient social unit for a sexually dimorphic population in which females are both smaller than males and more numerous. This is because a smaller proportion of available food goes to the individuals least concerned with infant nutrition and reproduction generally. In multimale units the large males must take a much greater proportion of the food. The large size of the male in both hamadryas and gelada and the development of shoulder capes and canines enhancing the appearance of size and strength are considered the result of intense intra-sexual selection. Sparse food supplies correlate with a scattering of one-male groups whereas conditions of abundance favor herd formation. Heavier predation likewise probably exerts positive selection for a multimale assembly with corporate male defense.

From Inter- to Intraspecific Competition
According to Ayala (1972) as selection for interspecific competition decreases, selection for intraspecific competition increases. Hence, as populations diversify and exploit their environments in special ways the need to compete with other species decreases while competition between individuals of the same species becomes more problematical. We can see a possible sequence of evolutionary steps from (1) generalized competition for common areas and resources to (2) habitat specialization and isolation of gene pools and finally (3) greater intraspecific competition (Thiessen, 1976).
Once genetic populations adapt to particular habitats, become food specialists and evolve species-specific social signals any further gene flow into the population is generally disruptive. Thus, any mechanisms that thwart gene flow and reinforce reproductive isolation will be favored by natural selection. Mayr (1970) has classified isolating mechanisms as premating or postmating, depending on the time at which genetic union is interrupted. A modification of this scheme by Thiessen (1976) is summarized in Table X.

Table 2: Behavioral Isolating Mechanisms in Sympatric Populations

Mechanisms that prevent interspecific crosses (premating barriers to gametic wastage)
A. Differential habitat selection and niche specialization
B. Assortative mating (homogamy)
C. Pair formation and internal fertilization
D. Incongruous social signals
E. Uncoordinated mating patterns (including seasonal)
F. Social exclusion
    1. Competition for resources
    2. Social class differences
    3. Territorial barriers
Mechanisms that reduce viability of interspecific crosses (postmating barriers with gametic wastage)
A. Lack of maternal care
B. Failure of young to imprint
C. Agonistic reactions between individuals
D. Lack of ecological adaptation (including incongruous social signals)

Since it is important to minimize maladaptive interactions there is a premium on premating isolating devices which preclude mating, gametic wastage and inviable offspring. Social signals and behavioral displays are often the messages that seal avenues of potential gene flow. Interspecific signals that erect the final reproductive barriers set the stage for individuals within a population to refine conspecific competitive mechanisms. Selection pressures become more individualized. The sea anemone, Anthropleura elegantissima, offers an instructive example of genetic aggregation of like genotypes within a single species and competition between unlike genotypes. In this situation interspecific competition is not the primary issue. Yet there is competition between clones of identical genotypes, giving the appearance of an intermediate form of competition that is neither purely interspecific nor intraspecific. The sea anemone, found on the California coast, lives as a solitary or aggregating form. The aggregating form is unique. Individuals reproduce by clonal replication and group together in separate and homogeneous genotypes. All members of a clone are either male or female. If they are separated and mixed together with individuals of other clones they will reaggregate into isolated uniclonal groups (Francis, 1973a). Different clones remain isolated. Occasionally, however, individuals of different clones do come into contact, either as a result of locomotion and the search for food of because of jostling during tidal shifts. When this happens a very extraordinary pattern of behaviors occurs. Typically the tentacles are extended and the nonclonemates touch. This is followed by a rapid withdrawal of the involved tentacles. The process repeats itself until the next stage of the behavior. During this period of mutual exploration the acrorhagi (also called marginal spherules) on the stalk of the anemone inflate, elongate and become turgid. If the encounter continues 4 to 12 acrorhagi become turgid. The column of the anemone elongates and assumes a more upright appearance. This is followed by a downward sweep of the inflated acrorhagi directed at the opponent. The full movement takes from 30 to 120 seconds and may be repeated one or more times. Neither the acrorhagi response nor the rapid withdrawal of the tentacles occurs when clonemates meet. After an initial exploration with the tentacles the anemones cease their activity and the tentacles may interlace. With individuals from different clones, however, it is clear that the behavior of the tentacles and especially the acrorhagi are part of an act designed to repulse.
In an elegant series of experiments, Francis (1973b) has shown that when an acrorhagus makes contact with a victim, a white surface layer is pulled loose from the acrorhagus and adheres to the body of the nonclonemate. The victim may respond with a similar transfer of ectoderm, but the usual response is to retract its tentacles, shorten its column and either lean away from the opponent or release its hold from the substratum and float away to resettle elsewhere. The tissue in the area of the applied acrorhagial ectoderm begins to deteriorate within a day. This is especially true of the tentacles. After a few days the necrotic tissue is sloughed off along with the foreign ectoderm. Ordinarily the recipient of this treatment repairs itself completely, but if escape is impossible the anemone may be killed. In the natural environment injured individuals can be seen within a single clone; they are always on the periphery of the population, indicating that clonal interaction takes place frequently. These common features of inter- and intraspecific competition provide evidence that intraspecific competition may sometimes be an extension of interspecific competition, both of which may serve similar functions, e.g., spacing and food utilization.
If a species must compete with another species as well as with individuals of its own kind, it is likely that similar if not identical mechanisms will be used in both forms of competition. This should be especially true when the species resemble each other and social signals overlap. For instance, if reproduction and competitive displays are regulated by gonadal secretions in tropical and equatorial birds, it would be expected that species radiating out from this zone would depend on the same secretions, rather than developing new biochemical methods (Thiessen, 1976). Evidence for this is not lacking. For males, at least, sixty North-temperate species of widely separate avian orders have reproductive and aggressive cycles dependent on steroid secretions from the gonads (Lofts & Murton, 1968). What differs among these species are the factors that initiate gonad recrudescence and seasonal activities. Tropical and equatorial birds are not faced with annual fluctuations in photostimulation. As a result, reproductive and territorial synchrony among members of a species is related to proximal factors such as rainfall. The point of these observations will reappear in many discussions of competition and overt aggression. Once a mechanism is genetically fixed because of its success, it will not change its fundamental characteristics in any derivative species or isolated gene pools. The reasons for this seem clear: First, a successful trait will have been selected which is free of genetic variation and thus lacking in selective potential. Second, a successful trait, be it structural or biochemical, becomes coadapted with other bodily processes and cannot be changed easily without disrupting a wide array of integrated mechanisms. And third, radiation of genotypes can only occur in environments that share major similarities; hence, basic traits are not challenged by radically new selection pressures. What does change during diversification of genotypes are modulators of fundamental processes. There is, therefore, reason to believe that competitive abilities among closely related species will depend on similar if not identical biological processes. The closer the relationship the more likely that this will be the case. Similarly, a basic mechanism used for interspecific competition will probably be adapted for intraspecific competition, or the reverse. Evolution may be opportunistic and take advantage of any variation to build adaptations, but it is also conservative and will not reconstruct mechanisms when simpler modifications will suffice (Thiessen, 1976).

Concerning the evolution of aggression, Andrew (1977) hypothesized: “As soon as dangerous bites became possible for feeding in vertebrate evolution, there was strong selection for their use in defense, and the facilitation of attack by pain or startle may be as ancient as this selection. The use of prey-catching behaviour to drive off conspecifics from areas offering important resources (e.g. food, mate, nest) would also be strongly selected, for, once the pattern was available, it would require initially only a slight change in evoking stimuli (e.g., response to rather large moving objects) to allow this result. The crossopterygian ancestors of the tetrapods and their immediate amphibian descendants were dominant predators, so that prey catching and biting in defense can be assumed to have been present. Comparison with present-day inhabitants of muddy shores, both air-breathing fish (e.g., mud skipper) and crabs, suggests that conspicuous displays given to the approach of a conspecific has probably also already evolved. Causally, then, as now, some or all of such displays might have been independent of attack. No doubt they already could be followed either by copulation attempts or by attack. There would be strong selection for appropriate hormones to confine the displays to breeding periods; this selection has happened with frog breeding calls and with bird songs, as well as with bird territorial displays, all of which are now facilitated by testosterone. The fact that this is true also of attack and male copulation has been taken mistakenly in the past to indicate causation of the former categories of bird behaviour by aggression and sex... Perhaps the least advanced group in which behaviour analogous with attack has been described is that of polychaete worms. Predatory ragworms (Nereis) defend their burrows against conspecifics with biting attacks. In other instances, biting is the analogue of vertebrate defensive threat, which has evolved from prey-catching behaviour; most crabs, when disturbed, extend and open their chelipeds (large claws), which evolved for feeding.
The function of defending resources against conspecifics is also served by a variety of adaptations that should not be termed aggressive (e.g., Orthoptera stridulation)” (Andrew, 1977). This latter theme has been elaborated in more detail by R.N. Johnson (1972) in the next paragraph.

Specialized Adaptations for Defensive Behavior
It is not obvious that aggressive behavior is categorically maladaptive, for if it were it would have become extinct eons ago. Perhaps there were organisms that fought indiscriminately; if so, they have disappeared into oblivion. Because of the competition for survival, animals which were skilled at securing and defending the necessities of life tended to be successful. But those that survived were not necessarily the most aggressive, for evolution most likely favored species which were selective in their aggressive behavior. This can be illustrated by the evolution of the canine complex in primates (Washburn & Hamburg, 1968).
At one time large canine teeth and powerful jaws provided a distinct advantage, and many of our relatives, such as the baboon still possess such features. Biting attack is a well-developed skill in primates, largely because of years of infant play in which biting is relatively harmless. But in late adolescence the canine teeth erupt and the temporal muscles more than double in size, and a single bite now may lead to swift death. In the evolution of early man such destructive weapons may have been maladaptive so that selective pressures operated toward the reduction in the size of the dangerous canine teeth. This reduction in canine teeth was completed several million years ago and reflects a general shift away from savage fighting toward the reliance on tools and language. We might conclude that evolution tended to favor the best hunters and killers, just as long as killing was selective and limited to nonconspecifics. While such deadliness may have had advantages, certainly it was not the only nor even the best path to survival (R.N. Johnson, 1972).
Many organisms survived the struggle for existence by outfighting their opponents, but a great many others competed successfully by doing just the opposite: avoiding conflict. Behavioral adaptations which have proved useful include speedy withdrawal, ability to hide, or tonic immobility (lethosis, catalepsy, 'playing possum'). Distraction is also another effective defense, as in the case of many snakes which possess a tail closely resembling their head. When threatened they hide their vulnerable head and wag their tail (Maier & Maier, 1970). Another kind of distraction is that of 'autotomy', in which part of the body is shed to confuse opponents and aid escape (e.g., a lizard dropping its tail). Autotomy is usually associated with withdrawal, but in some cases it may be combined with an attack to strengthen defensive behavior (Robinson, Abele & Robinson, 1970). In such attack autotomy, crabs may pinch their adversary and then autotomize (detach) the cheliped, leaving it attached to their opponent. While the attacker battles the autotomized cheliped, the crab slips away.
Another adaptation which favors survival without fighting is chemical adaptation. A by-product of such chemical defense for other species is that of protective mimicry. Other animals, including skunks, snakes, and many insects depend heavily on chemicals which repel or kill opponents (Whittaker & Feeny, 1971). Skunks treat their opponents to butyl mercaptan, and bombardier beetles squirt quinone secretions. Minute quantities of neurotoxins and cardiotoxins secreted from animals such as snakes and wasps quickly kill or paralyze their opponents. Such adaptations may not be without significance in the evolution of man, for it has been suggested that early man survived, not because of his strength or cunning, but simply because carnivores judged him to be foul-tasting (Leakey, 1967).
Equally impressive are structural adaptations which favor survival. Animals as widely separated as the sea urchin and the porcupine do not need great offensive strength because they are well protected by spiny projections. But protective coverings need not be menacing, as turtles and clams have discovered, and they need not be physically protective. Coloration alone provides a valuable defense for many species from insects to zebras. Disguises based on color and structure are perhaps most successfully utilized by insects. Often they blend in with the background or resort to fake appendages that confuse and scare opponents. For example, the hornworm of India retracts its vulnerable head when threatened and exposes a pair of enormous but fake red and black 'eyes' which make it appear large and dangerous. (For 'fake eyes' as evolved signals mimicking the intimidating stare, see e.g. D. Morris, 1967).
Behavioral adaptations other than fighting have also paved a path to survival with success accompanying the capacity for climbing, flying, swimming, burrowing, crawling, walking, grasping, seeing at night, and so on. The catfish Hassar orestis (Steindachner) survives attacks by piranha fish, not by fleeing or fighting back, but by following his opponent. The catfish swims under the piranha and duplicates every movement of his dangerous rival and thereby avoids being bitten (Markl, 1969). Some moths emit high pitched clicks when touched or exposed to ultrasonic pulses of bats, their main predators, and these clicks are aversive enough to repel some of their enemies (Dunning, 1968).
More complex adaptations have also taken place in social behavior, particularly cooperative social behavior. In a troop of baboons, for example, a small group of adult males will stand together and fight a common opponent while the rest of the troop flees to safety. This cooperative behavior will lead to the defeat of a dangerous opponent such as a leopard which could easily overpower an isolated baboon (Hall & DeVore, 1965). Antelopes cooperate by displaying a white rump to warn other members of their herd of impending danger (Etkin, 1964).
Equally impressive is the schooling behavior of fish and the mobbing behavior of birds. Birds such as starlings or chaffinches will flock together when danger threatens, and the distress call of a single member adds a margin of safety to the entire group (Hinde, 1954).
While this kind of social behavior has obvious survival value, the exact reasons for it remain unknown. It may be that there is safety in numbers because a distress call can warn the entire group, or it may be that individuals congregate in mobs so that others will be eaten first (Goss-Custard, 1970). Sometimes social cues are utilized for 'selfish' rather than 'altruistic' reasons. Arctic foxes have been observed to make fake warning calls to frighten away other foxes (even their own cubs) so they don't have to share a tasty meal (Rueppell, 1969).
In short, it is not necessary to assume that the process of evolution automatically favored the development of aggressive behavior, for fighting ability is only one of many routes to survival. Sometimes it is suggested that more 'advanced' species relied on aggressive ability while more primitive forms of life survived because of defensive rather than offensive adaptations. But this, too, is probably an oversimplification, and may reflect nothing more than the lack of scientific research on the social behavior of simpler organisms (R.N. Johnson, 1972).

Aggression, Reproduction, and Resources
Overt aggression can be selected for only where a resource essential to reproduction is defendable (J.L. Brown, 1964). If a species exploits resources none of which is defendable, then clearly that species will evolve neither agonistic adaptations nor courtship. We find such species in the marine environment feeding on plankton or sifting detritus while practicing broadcast fertilization. In such situations, for species beyond a certain body size, the cost of displacing a conspecific from a plankton or a detritus particle exceeds the benefits derived from it, and individuals compete passively by means of better food-gathering organs and a greater efficiency in converting food to gametes. In broadcast fertilization there is no individual fertilization, and hence no point in expending energy displacing conspecifics from potential partners, nor is there any gain in expending energy in courtship to facilitate the female's choice of partners. Thus we expect neither combat nor courtship in many animals that practice filter feeding and broadcast fertilization. Lorenz (1966) was not quite wrong when he intimated that without aggression there would be no love either.
In the terrestrial environment, broadcast fertilization is not possible; the egg clusters of females are concentrated and, given this condition, competition to fertilize the eggs becomes possible among males. Localized unfertilized eggs are a resource essential to the male for reproduction, and are a defendable resource at that. Similarly, an unfertilized female is a resource essential to the male for reproduction. Both combat and courtship are a mandatory consequence of internal fertilization or fertilization of localized egg clusters (Geist, 1978).
However, when a resource is not only concentrated but also of large size relative to the size of the defending individual and attracts many conspecifics, defense of the resource may not be a viable strategy. This is nowhere better recognized than in observing hyenas about a kill (Kruuk, 1972), when individuals gorge themselves as rapidly as possible without wasting precious time on fighting. Here the best strategy is 'scramble' not 'contest', as E.O. Wilson (1975) has put it.
It is important to recognize that a contest for scarce resources for reproduction aims at maximizing the reproductive fitness of the successful individual. An individual enhances his reproductive fitness, however, not only by successfully competing for resources, but also by directly reducing the reproductive fitness of other individuals, as has been pointed out especially by Geist (1978).
There are weapons and defenses, and strategies of attack and defense. Since in combat each individual attempts to maximize the efficiency of attacks, and at other times the efficiency of defenses, it is evident that, granted the size and armament of a species, there are only a limited number of effective strategies of attack and defense. The interplay of these limited strategies of attack and defense give rise to the 'ritualized' fight. Geist (1966; 1971a; 1974a; 1978) has criticized the conception held by the ethologists of ritual evolving to save the lives of combatants, and he also pointed out that 'conventional' combat is no more and no less than the exercise of species-specific attack and defense tactics. Natural selection maximizes strategies in the attack, and, in weapons, the amount of pain and trauma inflicted. Since we know from medical evidence (Guyton, 1971) that wounds on the body surface tend to be more painful than deep cuts to the body core, pain is maximized by maximizing surface damage. Weapons evolved to maximize the amount of damage to the body surface thus are inappropriate for deep body penetration, which maximizes the chance of killing an opponent. Weapons maximizing surface damage may be clubs that maximize bruising, or organs that cause multiple punctures and lacerations of the skin. The head armaments of giraffes may serve as an example of the former (Backhaus, 1961; Spinage, 1968), and the claws of cats used to rake opponents (Leyhausen, 1956) as an example of the latter.
It is important to note that a weapons which penetrates the body surface in all likelihood achieves about as much pain as a weapon penetrating into vital organs. Thus a weapon achieving deep penetration generates little more pain than a weapon achieving only shallow penetration, but places the aggressor in danger, since the withdrawal of a weapon that penetrates deeply is more difficult than withdrawal of a short weapon. Thus long horns stuck deep in an opponent are likely to cause a skull or neck fracture to their owner as the victim struggles violently to disengage itself (Geist, 1978).
Trauma is inflicted by making the opponent lose control over its body. As a prerequisite, it requires that combatants lock their bodies together so that each can exert a maximum force in moving the opponent's body. Teeth, claws, grasping organs, or locking horns and antlers are essential to this manner of fighting. In ungulates, wrestling with locked horns or antlers arises secondarily out of the defense strategy of catching the opponent's weapons. The primary function of the points, bumps, twists, and surface texture of horns and antlers is to allow a grip on the opponent's weapons (Walther, 1966; Geist, 1966).
Defense organs and defense strategies are selected to neutralize attack. These include evasion of attacks, reducing the target area exposed to attack, catching the opponent's blows against heavy dermal shields or armored heads that are equipped with special hinge mechanisms to absorb the force of the attack, as we find in the caprids, deflecting the opponent's weapon, denying the opponent the use of his weapon, or grasping the opponent's weapon and holding on to it, effectively disarming him (Geist, 1978).
An important factor limiting the evolution of weapons and defenses is body size. The effect of increasing mass on acceleration is such that huge-bodied animals would fracture bones, tear ligaments, and rupture organs if they should collide in attacks and roll on the ground like small mammals. Therefore we find that in large-bodied forms, appendages are used and transformed into weapons; the very mass of these appendages, accelerated rapidly, is sufficient to cause large surface bruises or penetration of the body surface by hornlike organs. Small-bodied animals have opted for combat with weapons that readily penetrate the skin; these are primarily teeth or tusks. This armament strategy is understandable in view of their low body mass and the difficulty of generating accurate blows of sufficient force to cause pain or surface damage to opponents (Geist, 1978).
An examination of weapons and defense among mammals makes it evident at once that these are arranged in recognizable syndromes. Short, sharp tusks and horns in medium-sized and large mammals are paired with thick, tough, and often large, dermal armor on the body. We see this in the wild boar, the mountain goat, the giraffe, the Indian rhino, and the hippopotamus (Harkness, 1971; Fraedrich, 1967). An exception to body size here is the chevrotain, which has tusks paired against a dorsal body armor (Dubost & Terrade, 1970).
Elaborate horns, antlers, and tusks tend to be associated with frontal combat in which opponents parry the attacks and skillfully block the weapons from their vitals or hold on the opponent's weapons. In these forms, dermal shields are either of lesser importance (e.g., Capra and Ovis, Geist, 1971a; Aepyceroc, Jarman, 1972) or absent (e.g., Phacochoerus, Cumming, 1975).
In very small ungulates, in most small mammals, and in most carnivores except the ursids, we find weapons capable of inflicting serious injury, but we find no morphological defenses against them. Defense against lethal weapons has here obviously taken some form other than morphological defenses combined with blocking attacks. In the bears, however, we do find evidence of a dermal armor on the neck, as well as for jaw-grasping as a means of disarming opponents, as evidence by the high incidence of fragmented canines in the males as compared to females, and the relatively more frequent incidence of such injuries in the larger and probably more aggressive species such as brown and polar bears as opposed to the black bear (Herrero & Jonkel, pers. comm. in Geist, 1978).
One can also recognize a relationship between weapons and defenses and the reproductive strategy of a species. In forms in which individuals live for only one short reproductive season, such as various salmon (Oncorhynchus) do we find in the males very elaborate toothed jaws, which appear to inflict significant damage, but no obvious means of defense. We find something similar in the short-lived saiga antelope (Saiga tatarica) in which combat is often lethal during the rutting season (Bannikov et al., 1961). Conversely, in long-lived species that live through numerous reproductive seasons, we find well-developed defenses, nowhere better illustrated than in the genus Ovis (Geist, 1971a).
Some of the differences in the agonistic behavior between species are a direct consequence of their weapons and defenses. The heavy armor on the skulls of caprid males permits frequent long-lasting, or even severe, fighting without noticeable injury to the opponents; it permits 'sporting' combat that flourishes when forage conditions are favorable (Petocz, 1973). Yet in an animal with weapons capable of causing severe wounds, with poor defenses, such as the mountain goat, the incidence of fighting is negligible and even threats decline with improved forage conditions (Petocz, 1973); in this species we find no sparring matches, that is, 'sporting combat'. The rare instance of fighting in species with sharp weapons can be explained as a consequence of retaliation (Geist, 1966; 1971a; 1974a); it is also an explanation of how individual selection acts to limit the development of arms and defenses (Geist, 1978).
It is known that pain causes various birds and mammals to attack companions or inanimate objects; the frequency of attacks, however, is an inverted U-shaped function of the intensity of the stimulus causing attacks (Azrin, 1964). Thus a blow, unless so severe as to be totally debilitating, will trigger an attack on the aggressor by his victim. Granted damaging weapons, a counterattack by the victim is likely to inflict injury on the aggressor, particularly in an all-out attack. Therefore the aggressor and the victim are likely to suffer damage, and the aggressor is not likely to be successful in reproduction, particularly in long-lived species in which individuals reproduce in more than one reproductive season. Only the instant death of the victim, or its inability to retaliate, or some means of escaping retaliation, will permit damaging aggression to flourish. If the biological weapons are selected to maximize pain through surface damage, retaliation is certain and there is selection against frequent combat in species with poor defenses, since the victim punishes the aggressor.
As Geist pointed out in earlier publications, cultural weapons and defenses in humans have broken the feedback loop of retaliation which controls overt aggression in animals; in humans, clearly, it must be cultural means that regulate overt aggression.

Threats vs. Displays
Although we have no data yet on the bioenergetic costs of combat, of living subordinate to a victor, of healing wounds, of the shorter life expectancy as a consequence of higher susceptibility to predation when wounded, of the loss of mating opportunities and the cost of gaining access to resources to restore dominance, all observations from the field and the meager bioenergetic data on activity suggest that combat as an activity, and its consequences, are very expensive indeed. It is self-evident that there is value in searching for alternatives to combat that have much the same ultimate effects but not the same consequences. It was recognized repeatedly that threats and displays probably evolved to substitute for combat (Collias, 1944; Walter, 1964; Geist, 1966 et seq.; Schaller, 1967).
Threats and displays are not the same. Threats are iconic signals indicating incipient attack with a specific weapon system (weapon threats), or they are based on triggering withdrawal to the stimulus termed 'looming' (a rapidly enlarging object) (Gibson, 1970), (rush threats), or - in combination with weapon threats - there is a severe distortion of body form, so that the 'novelty' of the stimulus in combination with the iconic danger signal ought to elicit arousal and withdrawal (scare threats). It is common to all forms of threat that they are clearly directed at an individual, and one can predict the consequences of staying or coming closer to the threatening individual; an attack is likely to ensue (Geist, 1978).
Dominance displays, however, are abstract signals from which one cannot predict the actor's action. They appear to aim at arousal by generating uncertainty in an opponent, as well as acting as reminders of past defeats, and thus condition the opponent to withdraw. The displayer emphasizes the size of weapons or body, does not address the opponent directly, but usually averts his head, changes his movements from the normal to act exaggeratedly slow or fast, does not present weapons in intention posture to strike but displays his armament, and acts in such a fashion as to maximize stimuli that catch the opponent's attention. Thus, dominance displays are very different from threats, as has been emphasized by Walther (1958, 1974) and Geist (1965 et seq.). Of the two, threats are probably the more expensive since they gear up the individual physiologically for combat, and they increase the risk - and hence the consequences - of a counterattack by the threatened opponent. We can therefore assume that combat elements are most costly, since they do lead to visible exertion of the opponents, threats are next in costliness, and dominance displays are least costly per display (Geist, 1978).

Weapons, Defenses, and Displays in Relation to Ecological Adaptive Strategies
Animal weapons are not independently selected for, but form a part of the adaptive syndrome containing strategies of attack and defense, morphological defenses, threats, displays, and internal control mechanisms that regulate the relative frequency of use of each of the components of agonistic adaptations. Geist (1978) offers the following set of hypotheses to explain how different strategies of resource exploitation control the evolution of the agonistic syndrome as a whole.

(1) Defense of resources needed for maintenance, reproduction, and growth selects in both sexes for weapons maximizing surface damage and for defense by evasion unless body mass or circumstances preclude evasion. In such a case dermal armor is selected for.
When resources are defendable, selection must maximize the combat potential of the defenders per unit of defendable resource. This selects for low sexual dimorphism among adults, probably to confuse intruders, and to intensify cooperation among the defenders. Resources for reproduction are scarce since defense is primary. Therefore combat must be minimized in frequency and duration. This can in part be achieved by deterrents such as group displays and olfactory markings, by surprise attacks and mass attacks that catch intruders unprepared, reduce retaliation, and condition the opponent negatively to the area of attack. Weapons that maximize surface damage, and instantaneous violent attacks inflicting massive pain, well beyond the threshold of maximum retaliation (see Azrin, 1964), are selected for. We expect to find sharp tusks, teeth, and short horns such as can be rapidly withdrawn and used again. Since opponents struck by painful blows are likely to retaliate, there will be selection for defenses. If the weapons are used indiscriminately, and the body size is small, then evasion is the best strategy since a small body mass can be readily accelerated. If the weapons preclude evasion and limit the target area, we expect the evolution of dermal armor in small-bodied mammals. However, if the resource on the territory is of low nutritional value requiring much time to assimilate it, and the species has a long extended cycle of reproduction, then the cost of removing competitors from the resource may be too high in relation to the amount of energy freed by digestion for maintenance. In this case, the territory owner may forego all competition save for females, and the territory becomes an exclusive area for breeding (lek).

(2) When resource defense is not possible due to low intensity or significant annual fluctuations of the resource in space and time, then females tend to compete passively and lose or reduce their weapons, while males evolve strong defenses, elaborate displays, or low-intensity interactions.
We expect the loss of weapons by a female to be linked with passive competition, both in habitats which have a low productivity and those of high productivity, provided the resource is undefendable by virtue of low density or other forms of instability. The former habitat is exemplified by a dry forest or savannah with short rainy seasons, the latter by floodplain communities generated by frequent floods. In both instances, reproduction cannot be continuous but must be seasonal. This selects for longevity, and thus against damaging combat. This suggests the evolution both of strong defenses and strong displays. Elaborate display organs, however, cannot be selected for unless the annual pulse of surplus energy and resources is high and long enough to permit it. In the dry, low-productivity habitat, we expect a small body size. In the productive riparian community, we expect an increase in body size over closely related forest forms for reasons explained by Geist (1971b), as well as due to the large annual productivity pulse that favors larger body size (Margalef, 1963).
High productivity and large body size lead to more frequent and intense combat. Therefore, in the species from riparian communities, we expect an emphasis on defenses. An increase in body size reduces maneuverability, since it reduces the speed of acceleration and increases its costs. Thus with increasing body size evasion as a means of defense is becoming increasingly more difficult. This in turn selects for dermal shields to catch the weapons of the opponent, or it selects for mechanisms with which to catch and hold the opponent's weapons. Therefore, in the larger-bodied forms opportunistically exploiting rich sources of food, we expect males with a heavy emphasis on defenses and weapons that can inflict surface damage.

(3) If the productivity of forage is so low as to preclude territoriality, the males of a species may utilize the low flow of net energy for reproduction by a strategy of saving and intermittent expenditure. They accumulate sufficient resources to enter into a state of dominance, and drop from this state of dominance to a period of quiescence once these resources are exhausted. Although such males may remain capable of fertilizing females all year round, they enter into 'rut' only periodically.

(4) If gregariousness evolves as a strategy against predation, weapons are selected for that minimize surface damage and retaliation. Weapons and strategies that permit contests of strength, such as wrestling, are selected for. Dermal shields evolve only around the neck and head to protect them against horns in wrestling. Ungulates exploiting open terrain with little or no cover, be it on an annual or a seasonal basis, tend to clump into herds. This can best be explained as follows: individually, herbivores have very little time for scanning for predators, since they must feed selectively much of the time in order to maximize forage intake. Predators spend little time feeding and have a great deal of time to scan for, spot, and stalk prey. The prey species overcome the handicap through gregariousness; individuals join each other in company so that collectively they maximize the time in which predators may be spotted, while ensuring adequate time for feeding. This is essentially a refinement of the view that many eyes are better than one pair, as reviewed by Treisman (1975). In addition, individuals in the center of the herd are relatively safe from predation, since predators are likely to kill a peripheral individual once they reach the group (Cf. Hamilton, 1971; Treisman, 1975).
Granted these hypotheses, then intraspecific agonistic encounters, which either cause the victim of overt aggression to leave the group or spark damaging retaliation, select against the aggressor. The aggressor that causes a companion to leave increases the risk of being killed by a predator by removal of a guardian and potential buffer against predators. Therefore, weapons causing surface damage and pain, and likely to cause retaliation, will be selected against, since they defeat gregariousness as an antipredator strategy. There will be selection for weapons that permit a testing of strength, and in a serious fight may inflict trauma, rather than pain. Weapons will be such as to permit wrestling by locking opponents together in a match of strength. Heavy predation will ensure that any wounded opponent is quickly eliminated (Estes, 1969); hence an aggressor may not spark retaliation lest by being wounded he becomes predation prone. Therefore, in large-bodied herbivores from coverless habitats faced with many predators, we expect the application of weapons in such a manner that little damage is done; we expect wrestling bouts testing physical strength; we expect that opponents attack into the defenses of opponents, rather than aim attacks at their bodies, and we expect 'wrestling weapons', and therefore little dermal armor. We also expect an elaboration of dominance displays as a substitute for overt aggression, all in the interest of maintaining group cohesion despite active competition by members of the group.
If productivity is reasonably continuous in the coverless plains, territoriality, in which resources are defended for a part of the year, may arise; where resources cannot be readily predicted and defended, rank hierarchies are expected among males. Where the energy pulse is low, such as is expected with long vegetation seasons, little energy and few nutrients will be diverted into horns and display organs and vice versa. Thus we expect larger horns and display organs in the hierarchical large-bodied, dry steppe and desert forms than in forms from more mesic regions, in which the pulse of annual biological productivity is relatively low.
Sexual dimorphism in armament and display ornaments decreases in the highly gregarious forms, probably through male mimicry by the female as the female competes against males, primarily to save energy by reducing harassment caused by frequent approaches of young males. Sexual dimorphism is greatest in forms that form only small groups in which large males can readily protect females from smaller ones, be it in a mobile harem or on a territory; we find small group sizes in species exploiting the ecotone between steppe and forest.

(5) Short reproductive seasons, preceded by seasons of superabundant resources, intensify sexual competition and select for severe combat and showy displays. Exhaustion from the rut increases predation on males. They adopt antipredation strategies of mimicking females, hiding in areas with few resources and few predators, or moving off to separate home ranges that they exploit in bands as an antipredator strategy. In the last two cases, the former strategy may select for weapons causing surface damage, and the latter for wrestling or clashing type weapons.
In the forms from temperate and cold climates, new ecological conditions arise that shape social behavior. As botanical productivity declines latitudinally toward the poles, so do herbivore and carnivore density and diversity, except in some locally highly productive environments such as the periglacial loess steppes. With declining predation pressure we expect combat to become more damaging, and there is some evidence that it is. Seasons of superabundance alternate with seasons of resource scarcity, and these seasons are highly predictable chronologically. During seasons of abundance, males can fatten and grow armaments for the short rutting season, in which males from polygynous species maximize time available for rutting by reducing time spent feeding. This they can do by living off fat stores accumulated during summer and fall. Since the rut is short, agonistic behavior between males is intense. There is intense selection for effective weapons, defenses, and display organs, and there are ample resources during the season of superabundance to grow the required body structures.
The short, but very intense, rutting season generates a very serious problem to the rutting males. They lose a high percentage of their body weight and their mortality increases from caloric bankruptcy, in part to greater susceptibility to predation. In order to reduce predation, males may adopt a number of strategies: they may cast off diagnostic features of maleness and use females as a camouflage to escape detection; they may segregate from the females and search out areas where predators are not likely to search, or band together with other males in habitats different from those occupied by females.
If the food habits of males and females are very broad, they become competitors for food in the same area. It is in the interest of the female to remove males from the area in order to lessen competition and maximize resources for reproduction. It will be in the interest of the male to leave, provided it occupies the ranges of females it inseminated. In mountain goats this conflict may become open and females remove males from what appears to be prime habitat, so that males assume residence on poor habitat close to the females. Exploitation of small patches of habitat peripheral to areas occupied by the female is possible only where forest of broken terrain preclude good visibility. Thus, in species in which males occupy areas peripheral to large female home ranges after the rutting season, we expect the evolution of weapons that can cause severe surface wounds and damage. This system selects for sharp weapons maximally in small populations on poor habitat that would cause a maximum dispersion of individuals, and it assumes that food shortages would at least initially enhance overt aggression. If the males segregate from females on landscapes with excellent visibility, then grouping as an antipredator strategy becomes adaptive. This would select against damaging weapons and for sparring and weapon play among males. The retention of weapons and the size of weapons probably regulates access to resources during periods of resource shortage, so that we can expect males to be well armed if they form separate herds from females following the rut.

Ritualized Fighting
The preponderance of ritualized combat over physical combat is illustrated by a study of agonistic behavior among elephant seals living off the coast of California (LeBoeuf & Peterson, 1969). It was reported that for every actual fight there were 67 aggressive encounters which never went beyond ritualistic threats.
A considerable amount of fighting in primates involves bluffing an opponent into intimidation. Kummer (1968) reports that baboons often threaten to bite their opponents in the neck, but hundreds of such threats result in only a few actual bites. The display of threatening or submissive social signals without actual fighting is by far the most common form of agonistic behavior, and most animals quickly recognize and respond to such gestures. It is because these rituals are so effective that overt fighting becomes unnecessary.
Such aggressive behavior has become so ritualized that if often has the appearance of a ceremony rather than a confrontation. Aggressive displays and gestures are easy to recognize because they generally involve marked physical changes which make the combatants appear larger and more threatening. A cat arches its back, raises its fur, bares its teeth, and extends its claws. Birds may fluff their feathers and spread their wings. A baboon displays its large canine teeth and erects the fur around its shoulder, which instantly gives it a gorilla-sized appearance (DeVore, 1965).
Some fish intensify their color, swish their tail, and extend their fins to maximum size. The agonistic behavior of the Hawaiian coral reef fish (Pomacentris jenkinsi) consists almost entirely of displays, and fights are won and lost by visual cues alone (Rasa, 1969). The normally yellow eye changes to a greyish-black to indicate aggressive motivation, and the raising of a dorsal fin is correlated with fear. Examples of ritualized combat and threat can be multiplied almost 'ad infinitum'.
Evolutionary, ritualized fighting poses a considerable theoretical difficulty, as pointed out by Wilson (1975): Why not always try to kill or maim the enemy outright? And when an opponent is beaten in a ritual encounter, why not go ahead and kill him then? Allowed to run away, to paraphrase the childhood rhyme, the opponent may live to fight another day - and win next time. So in a sense the kindness shown an enemy seems altruistic, an unnecessary risk of personal fitness. In order to understand the solutions proposed to solve this problem, a brief digression to the Levels-of-Selection problem, games theory, and the concept of 'Evolutionarily Stable Strategy' is in order.

Games Theory and the Concept of 'Evolutionarily Stable Strategy' (ESS)
In its application to animal behavior, and especially the analysis of the evolution of agonistic behavior, games theory assumes that behavior has costs and benefits which can be quantified in units based on the contribution to the individual's reproductive fitness. For instance, the victor in a dispute might acquire a food item, a potential territory, or a mate, and in each case the result is assumed to produce a benefit, V, which can be quantified in units of reproductive fitness. The disadvantageous consequences of fighting (ranging from death or serious injury to exhaustion or mere waste of time) are subsumed under the general term 'costs', again quantified in units of fitness (Caryl, 1981). One of the most useful discussions of the reasons for the choice of this unit as a 'common currency' is given by McCleery (1978).
Each model also includes assumptions about tactics that an individual is allowed to adopt in a dispute, and about the chances of victory or of incurring costs while using these tactics. The benefits that accrue to an individual adopting particular tactics will depend on what tactics are adopted by other members of the population, and for some models, this frequency dependence leads to perpetual change in the proportion of individuals adopting particular tactics. But Maynard Smith (1974; 1976), Maynard Smith & Price (1973), and Maynard Smith & Parker (1976) have shown that for certain models of aggressive interaction, there exist 'Strategies' (combinations of tactics - gambits might be a more appropriate term - in specified proportions) which are 'Evolutionarily Stable' in the sense that once they have arisen in a population, they cannot be displaced by a mutant using the same gambits in different proportions. (If the theorist changes the rules of the game by allowing the mutant to adopt different tactics, it may be able to displace the members of an evolutionarily stable strategy (ESS) population. The statement that a particular strategy is an ESS is only relevant within the specific universe of possibilities considered by the particular model) (Caryl, 1981).

Natural selection is ultimately differential survival of alleles in gene pools. We can talk about the Darwinian evolution of behavior only if we are prepared to visualize genetically determined behavioral alternatives in the population. Each genetically determined behavioral alternative is referred to as a 'strategy'. A strategy in this sense can be defined only by contrast with at least one alternative. It emphatically does not have to be something the animal works out in a cognitive or purposive sense. A strategy stands to an animal in the same relation as a program to a computer. It is an unconscious behavior program, a candidate for natural selection in competition with alternative strategies. A strategy is said to be evolutionarily stable against specified list of alternatives if, given that more than a critical portion of the population adopts it, none of the alternative strategies does better.
The defining characteristic of an ESS, according to Dawkins (1980), is not that it is the optimum strategy that could be devised for all individuals. Rather, it is immune to cheating. An ESS, then, is not the 'best' strategy, but it is a strategy that is uninvadable by any of a specified list of alternative minority strategies. Rational decisions do not come into ESS theory. Rather, each animal is assumed to be provided with a nervous system which is wired up in advance so that it performs in a certain way, programmed, in other words. Then we ask which program or combination of programs will be stable against evolutionary invasion by alternative minority programs which might arise in the population by mutation or immigration (Dawkins, 1980).
In simpler phrasing, an ESS is: a strategy with the property that if most of the members of a large population adopt it, then no mutant strategy can invade the population. In other words, a strategy is evolutionarily stable if there is no mutant strategy that gives higher Darwinian fitness to the individuals adopting it (Maynard Smith, 1978). Mathematically, an ESS, which may be a pure or mixed strategy, is defined as follows: A strategy I is an ESS if the expected utility of I played against itself is greater than the utility of any other strategy J played against I. This can be written Ei (I) > Ei (J), where E gives the expected utility of the strategy in parentheses played against the strategy indicated by the subscript.
The relevance of this definition is as follows. In a population consisting entirely of individuals adopting strategy I, rare variants arising by mutation which adopted a different strategy J would not increase in frequency, and hence the population would be stable under mutation and selection. In the definition, we have required that Ei (I) > Ei (J); difficulties arise if Ei (I) = Ei (J). In this case, I is an equilibrium strategy, but it need not be stable. To determine the stability, we need to know Ej (J) and Ej (I). Thus in a population of which a fraction p adopt I and (1 - p) adopt J, the expected 'fitnesses' are

    E(I) = pEi (1) + (1 - p) Ej (I),
    E(J) = pEi (J) + (1 - p) Ej (J).

It will be evolutionary stable if Ej (I) > Ej (J). Thus we can extend our definition, and say that I is an ESS if, for all alternative strategies J, either Ei (I) > Ei (J), or Ei (1) = Ei (J) and Ej (I) > Ej (J). The stability need not be global. There may be more than one ESS for a given game. If so, a population would evolve to a different ESS according to its initial composition (Maynard Smith, 1974).
An ESS may be a 'mixed' strategy; that is, it may consist of adopting one out of a set of 'pure' strategies according to a set of preassigned probabilities. If so, a stable population could either be genetically polymorphic, with appropriate frequencies of individuals adopting different pure strategies, or it could be monomorphic, the behavior of all individuals being random in an appropriate way.
In an appendix to Maynard Smith's (1974) paper, Haigh shows that if there are only two pure strategies it is always possible to find an ESS. If there are three or more pure strategies, there may be no ESS. So far, however, computer-simulated games which were thought of as being models of some biological process have proved to have at least one ESS.
Riley (1979) has shown, for any finite population, that a strategy which is stable in the sense of Maynard Smith may have a lower fitness than a mutant strategy, regardless of the proportion of contestants using the latter. He then proposes two alternative concepts of evolutionary stability. A strategy is described as being strongly stable if no mutant is able to invade because of its higher fitness and weakly stable if it has a higher fitness whenever the contestants using any particular mutant strategy become sufficiently numerous. In the case of natural informational asymmetry, it is shown that there is a strategy which is strongly stable with respect to any feasible mutant as long as the population is sufficiently large. Strategies elected by mutants to counter an established ESS can affect the frequencies with which encounters occur as well as the outcomes of these encounters. The analysis of the corresponding, more complex situation (Hines, 1977) leads to the conclusion that the strategy determined by Maynard Smith does provide stability of composition for a population of normals or N's in a stable environment, but that for certain environments the stability provided is incomplete. In an adequately severe environment, mutants, M, employing a considerably different, meek and unaggressive strategy can establish themselves as a sizeable component of the population.
This two component population might in turn be vulnerable to invasion by new invaders using yet a different strategy. In a subsequent paper, Hines (1978) considered the effect of evolution on such populations, and found that the occurrences of slight variations in strategy from generation to generation render a population using strategies similar to the original ESS-like strategy stable, and all others unstable.

Levels of Selection
Darwinian evolution through natural selection embodies three basic principles: (1) phenotypic variation exists among the members of a population; (2) phenotypes exhibit differential fitness; and (3) fitness is inheritable in some degree.
Whereas Darwin saw natural selection as acting primarily on individuals within a breeding population, group selection theorists, by contrast, view populations and sometimes entire species, or even ecosystems, as the units of selection. Group selectionists assume (usually implicitly) that populations or species show variation between groups, exhibit differential fitness from one to the next, and pass on fitness to succeeding groups. Furthermore, group selection theory holds that certain social behaviors have evolved primarily to assist in population or species survival (Fry, 1980).
The crux of the theoretical difference between researchers adhering to a group selection paradigm and those following an updated Darwinian perspective lies in explaining the existence of altruism. In evolutionary terms, altruism may be defined as any act that increases the fitness of others while simultaneously decreasing the fitness of the actor. A frequently noted aspect of animal conflict is a lack of serious injury to the participants. Among elephant seals, for instance, threats are about sixty times as common as fights, and losers almost never receive fatal wounds (LeBoeuf, 1971).
In some species, fighting is ritualized or patterned in such a way that the chance of injury to the participants is minimized. Commonly cited examples include species of ungulates that engage in bloodless contests of strength by butting horns or locking antlers, and rattlesnakes that push each other with their necks and bodies but do not use their deadly fangs on conspecifics. The fact that the winner usually 'follows the rules' of ritualized fighting and does not seriously harm its rival has been seen by some as an act of altruism. Allowed to live, so this line of reasoning goes, the adversary continues to compete for resources with the winner and in the future might again fight with the winner - and possibly win. If this reasoning is correct - that is, if such behaviors really are altruistic - then Darwinian theory faces a serious problem. How could traits that lower an individual's fitness have evolved if, as Darwin proposed, selection occurs primarily at the level of the individual? Truly altruistic individuals should have been selected out of the gene pool, since by definition they lower their own fitness in relation to other members of the breeding population (Fry, 1980).
To get around this apparent theoretical difficulty for Darwinian natural selection, Wynne-Edwards (1962), Lorenz (1966), Eibl-Eibesfeldt (1961; 1975), and others have proposed group selection explanations. They argue that rivals are not killed or seriously injured in within-species aggression because this would lead to the extinction of the group and thus that ritualized aggression and limited combat - which supposedly encompass some altruistic actions - are widespread in the animal kingdom because they are good for groups and species.
“The wholesale wounding and killing of members by one another is generally damaging to the group and has consequently been suppressed by natural selection... The conventions governing social competition can only have been evolved by group selection, and any immediate advantage accruing to the individual by-killing and thus disposing of his rivals forever must in the long run be overridden by the prejudicial effect of continuous bloodshed on the survival of the group as a whole” (Wynne-Edwards, 1962).
Wynne-Edwards (1962 et seq.) holds that the central characteristic of sociality is the transformation of the bitter struggle for existence into a ritualized contest, with rules to protect the contestants from both getting mortally hurt. He states:

    “The right to feed is the primary object of competition between the rival members of a population; but in fact the members are sidetracked into competing only for token substitutes, like territories and personal status; they do not fight for the food itself. They contest for conventional rewards, and these, once they have been won, confer the right to feed on the successful competitors; those that are unsuccessful forfeit their rights. Sometimes the same competition confers also the right to breed, but in other cases this right is separately contested. Because the rewards sought are only conventional tokens, which confer rights to the things that really matter, namely feeding and breeding, the competition itself becomes conventionalized. Stags roar, antelopes butt and wrestle, skylarks sing and peacocks display their finery, but little or no blood is shed in the process. The bitterness of the struggle for existence has been sublimated and ritualized; the outcome is still as vital as ever, rewarding the successful with the right to a full life, and condemning the losers to barrenness and often to premature death” (Wynne-Edwards, 1971).

In accordance with the above conception of sociality, Wynne-Edwards is an ardent advocate of the group selection thesis: “There are evolutionists who still deny the possibility of group selection. They believe it is possible to explain the evolution of all adaptations by the single process of selecting for fitness among individuals. I hope I have made my view clear, that the social group is an organic entity with properties of its own, properties that could not be vested in separate and independent individuals. I am thinking of the existence within it of hierarchies of individuals, of the customs that dictate the collective behavior and social interrelationships of its members, and secure its collective rights. I have attempted to show that groups with such characters exist, especially in the higher vertebrates and arthropods. If there are many groups within the geographical area of a single species and they differ as they must in survival potential, nothing can prevent selection from occurring between them” (Wynne-Edwards, 1971).
Keith (1916; 1948), in what he called his group theory of the evolution of man, already emphasized the mosaic or cellular structure of populations, and the important part it plays in evolution. His thesis was that right down to the dawn of civilization the habitable earth had formed a mosaic of home areas each belonging to an isolated local community, and that such a grouping had favored rapid evolutionary change. He knew, and it has since been fully confirmed, that some of the non-human primates show the same mosaic pattern. Presumably then it had already been in existence when man's immediate ancestors began to spread. “The area of distribution”, he wrote, “was extended by older successful groups giving off broods which formed new groups or communities. The size of a local group depended on the natural fertility of its territory; in primitive peoples which still retain the original mosaic form a local group varies from 50 to 150 individuals men, women and children. Such local, interbreeding, competitive groups I shall speak of as 'evolutionary units'; they represent the original teams which were involved in the intergroup struggle for survival... Far from speech tending to break down the barriers between local groups, it had an opposite effect, for we know that speech changes quickly when primitive peoples become separated”. In the later stages of human evolution the tendency has, he said, always been towards the production of larger and more powerful 'evolutionary units'; but his own conclusion was that evolutionary change proceeds fastest when the competing units are small and of great number. His enquiries left him no doubt that every one of the units, whether a local community, a tribe or a nation, “inhabited and claimed the sole ownership of a demarcated tract of country; all were bound to their homeland by a strong affection; and life was willingly sacrificed to maintain its integrity”. He came to regard the territorial sense - a conscious ownership of the homeland, one charged with a deep emotion - as a highly important factor in human evolution. Every such territory, he wrote, “served as an evolutionary cradle” (Keith, 1948). So here we find in one theoretical framework the concepts of pseudo-speciation, group territoriality and group selection. Also Scott's (1981) theory of evolution allows for interactions among different levels of selection. The role of group selection in human evolution is still a matter of controversy (Cf. Wilson, 1975; Masters, 1983).

Selfish Gene Theory
Perhaps the single most important discovery of ecology and evolutionary biology is that the genetic traits of an organism usually have adaptive significance for the individual in its environment. To this day, there are few, if any, convincing examples of biological traits that persist at high frequency in a population, in the face of alternative forms with greater individual benefits, because of their advantages at some other level of organization (e.g. deme, species, or ecosystem). It now appears that most genetic aspects of an organism's phenotype have evolved through the selective retention of traits that maximize the long-term representation of an individual's genes in a population (its 'inclusive fitness' in the sense of Hamilton, 1964).
The inclusive fitness of an individual is the sum of its personal (Darwinian) component of fitness and its kin component (Hamilton, 1964; West-Eberhard, 1975; Chapais & Schulman, 1980).
In biology, this phenomenon is explained simply and elegantly by the theory of evolution by natural selection. Genes producing traits that endow their bearers with an advantage in terms of survival and reproduction relative to alternative alleles will, through time, increase their proportional representation in the population's gene pool (Durham, 1979; E.O. Wilson, 1975; Lewontin, 1970; Williams, 1966). In principle, the differential propagation of genes can occur at several levels through the differential reproduction of (1) individuals (classical Darwinian selection), (2) groups of close relatives (kin selection after Hamilton, 1964), (3) social groups (E.O. Wilson, 1975), and (4) demes (interdeme selection as in E.O. Wilson, 1973a), species, communities, and even ecosystems (See also Lewontin, 1970).
Concurrent differential reproduction on various levels may have complementary or opposing effects on gene propagation, so that interdeme selection, for example, may sometimes greatly accelerate or greatly reduce gene frequency changes produced at levels 1 and 2 (Darlington, 1972). However, the differential reproduction of social groups, demes, or higher units is thought to occur too slowly under most conditions to counter the trends of individual and kin selection - i.e., to favor genes whose net phenotypic effect gives conspecifics with other genotypes higher individual fitness (E.O. Wilson, 1973a). It is therefore held to be unlikely that real 'genotypic altruism' is widespread (Cf. Alexander, 1974) and, as a result, the genetic traits of most species probably evolved because they maximize the inclusive fitness of individuals (Williams, 1966).
Despite the simplicity of this theory, several of its major implications have only been appreciated in recent years. We now realize, for example, that underlying gene competition ironically promotes cooperation among conspecifics and mutualism among interspecifics in any circumstances where these relationships can result in mutually enhanced fitness (e.g., Hamilton, 1964; Lin & Michener, 1972). In addition, a number of recent theories describe ways by which altruistic or self-sacrificing attributes can evolve by natural selection even though they may superficially appear to have more individual costs than benefits (Trivers, 1971; Darlington, 1972; Alexander, 1974; D.S. Wilson, 1975; Axelrod, 1984). With these new insights and explanations, it now appears that many of even the most complex behaviors of social organisms may be 'genotypically selfish', as expected on the basis of natural selection (Alexander, 1974; Durham, 1976).
Dawkins (1977) has popularized Selfish Gene Theory, considering all organismic behavior and morphology to be the result of each gene, anthropomorphically speaking, attempting to spread as many copies of itself as possible. The basic unit of natural selection, then, is the gene (or rather a replicator in general, e.g., a block of genes; see Dawkins, 1978; Graham & Instock, 1979) and not the species, the group, the family, the individual, the chromosome or even parts of the chromosome (see e.g., Gilham, 1978). The individual organisms, then, are merely 'throw-away survival machines' for the selfish genes. This Selfish Gene Theory has its roots in earlier mathematical-biological works such as Fisher (1930), Hamilton (1964), G.C. Williams (1966), Trivers (1972) and Alexander (1974), and it has a sound basis in modern biology, especially ethology and biochemistry (Wind, 1980).

There is little evidence that selection operates between groups. Attempts to explain the evolution of traits which benefit the group but decrease the fitness of the individuals carrying them face a major theoretical problem: such altruistic traits will be eliminated by natural selection operating between members of the same group. Even if situations occur where an altruistic genotype drifts to fixation within particular groups, selfish mutants or migrants will be at a selective advantage and the situation will be evolutionary unstable (Hamilton, 1964 et seq.; Dawkins, 1976; Clutton-Brock & Harvey, 1976; Fry, 1980). While it is probable that group selection will have some slight effect where populations are divided into small, isolated groups (Maynard Smith, 1964; Gilpin, 1975) the degree of isolation required for an altruistic genotype to spread through a population is so great as to be improbable in primate societies (Maynard Smith, 1976). A more recent model of group selection (D.S. Wilson, 1975) has been claimed by some (May, 1975; E.O. Wilson, 1975) to have wider generality than previous models. This seems unlikely to be the case since the conditions under which the model could operate are highly specialized (Maynard Smith, 1976). In short, natural selection has nothing whatever to do with the survival of species. Natural selection is the differential survival of alternative fragments of genetic material, or, in Hamilton's words, it is all about individuals maximizing their inclusive fitness. Traits may evolve which increase their carriers' fitness to the detriment of the reproductive rate of the species (See Hamilton, 1971b).

There is much in favor of viewing a great deal of animal behavior as optimum strategies for maximizing the rate of extraction of 'fitness gain' from the available series of 'fitness gain parameters' (resources) present in its environment. One consequence of the occurrence of discontinuously distributed resources is that they may be in short supply. Animal aggression (in the form of resource guarding) will be favored by selection when there are less resources than competitors and where an individual can achieve an immediate gain in fitness by forcibly ousting one of its fellows. Selection for aggression will be more intense the more discrete the resource (i.e., the easier it is to guard) and the higher its yield as a fitness gain parameter (a function both of its absolute effect and its shortness of supply). It is not surprising therefore that most of animal aggression relates to food fighting and especially to mating. Territoriality is often merely an adjunct to these two situations - e.g., an area is guarded because it has a high probable yield of food or mates, or both (Parker, 1974; Cf. Barash, 1982). Fighting tendency will be much modified by the probable relatedness of the two competing individuals, an effect studied by Hamilton (1964). Darwin (1871) was very well aware of the individual advantages of aggression when he founded the theory of sexual selection.

Group selection, kin selection and frequency-dependent selection have been proposed as explanations for the comparative rarity of dangerous weapons or tactics in intraspecific animal conflicts. Thus Huxley (1956) probably expressed the most commonly held view when he argued that the use of dangerous weapons is rare because “it would militate against the survival of the species”. Hamilton (1964; 1971) has emphasized the evolutionary importance of the fact that excessively aggressive individuals may injure their close relatives. Maynard Smith & Price (1973) proposed a model of the evolution of conflict behavior in which selection acts entirely at the individual level, but in which the success of any particular strategy depends on what strategies are adopted by other members of the population. The main conclusion reached by Maynard Smith & Price was that in a species capable either of 'ritualized' or 'escalated' fighting - the latter being capable of seriously injuring an opponent - the evolutionary stable strategy is to adopt the ritualized level, but to respond to escalation from an opponent by escalating in return. The importance of retaliation in the evolution of animal conflict was emphasized earlier by Geist (1966). In a population adopting such a 'retaliation' strategy, a mutant which adopted escalation tactics too readily would be more likely to get seriously injured than the typical members of the population, who would usually settle conflicts without escalation. This conclusion, however, rested on the assumption that two individuals adopting purely ritualized methods could satisfactorily settle a contest. This assumption was investigated further by Maynard Smith (1974). He distinguished two types of ritualized contest: 'tournaments' and 'displays'. An example of a tournament is a fight between two male deer, in which the antlers interlock and a pushing match ensues. The structure of antlers and the behavior of the contestants is adapted to prevent serious injury. Physical contact does take place, however, and victory goes to the larger, stronger and healthier individuals. Tournaments of this kind are common. In such cases, no special difficulty arises in understanding how a ritualized contest can be settled; the model considered by Maynard Smith & Price (1973) seems adequate to explain why more dangerous weapons or tactics do not evolve. In a 'display', no physical contact takes place, or if it does so it does not settle the contest and provides little or no information about which contestant would win an escalated contest. In such a contest, the winner is the contestant who continues for longer, and the loser the one who first gives way. It is the logic of contests of this kind that is considered in the paper by Maynard Smith (1974).
From an evolutionary point of view, tournaments and displays have something in common, in that the winner is the individual which continues the contest longer, and we would therefore expect natural selection to favor characteristics (size and strength in one case and behavioral persistence in the other) enabling an individual to continue; there would be an ultimate balance between the advantages and disadvantages of increased size or persistence. There are, however, important differences. First, size and strength may change with age, but cannot be varied from day to day or from contest to contest. Second, the corresponding disadvantages of greater size would arise, not in the contest situation, but in other contexts, for example, escape from predators, whereas the disadvantages of excessive persistence would arise from waste of time and energy in the contest itself. Third, the disadvantage of excessive persistence would be felt to the same extent by the ultimate winner as by the ultimate loser (E.O. Wilson, 1975).

To return to Wilson's question, “Why do animals prefer pacifism and bluff to escalated fighting?” Maynard Smith & Price (1973) showed that for a species capable of both ritualized and escalated fighting - the latter likely to lead to serious injury to one opponent - one ESS was to adopt the ritualized level, only escalating of the opponent did so first.
From these assumptions of the model, we can calculate the utility (the balance of benefits and costs) of playing a particular strategy against another.
if E[Iv(J)] is the utility of playing tactic I in a population playing J, the rules defining that I is ESS are as follows:
Either animals playing the ESS do better against others playing the same strategy than an animal adopting any other strategy can do against one adopting the ESS:

(A)    E[Iv(I)] = E[Jv(I)]

or if the previous utilities are equal, as they will be in a mixed ESS (see below), the return to an animal playing ESS against a mutant strategy must be greater than the return to an animal playing the mutant strategy against another which also adopts the mutant strategy

(B)    E[Iv(J) > E[Jv(J)]

Theorists have developed models to account for both symmetrical and asymmetrical contests. In the former, the opponents are equally matched, and equivalent in every other way; in the latter there in some asymmetry, either related to their ability to win the fight, or else some other difference, such as first arrival and second arrival at a food source, which distinguishes them (Caryl, 1981).

Several lines of evidence suggest that non-lethal patterns of settling disputes, such as agonistic displays that end short of fighting, ritualized combat, and submission and appeasement signals, have most likely evolved because such behaviors benefit the individual actors engaging in these behaviors, not because species preservation calls for such beneficial patterns of behavior. In other words, individual animals generally do not kill or seriously wound members of their own species because usually it is not in their own genetic self-interest to do so (Fry, 1980).
If one views the various types of ritualized aggressive behavior witnessed in many species as reflections of individuals generally pursuing evolutionary stable strategies, then overly pugnacious animals (and pacifists also) would appear to be penalized in terms of reproductive success and fitness. In many circumstances, so-called 'hawks' - or actors that fight more frequently or more forcefully than the majority of their conspecifics normally do - would stand a higher chance of serious injury than their less pugnacious peers. If for instance an overly aggressive fighter continues a struggle with an already submissive partner, the gentler opponent, acting out of self-defense, may in turn escalate its response and seriously injure the imprudent 'hawk' (Fry, 1980).
A submissive rhesus monkey (Macaca mulatta), for example, will accept incisor bites from the winner of a fight, but if the victor should use its deep cutting canines, the loser abruptly launches a defensive counter-attack (Bernstein & Gordon, 1974).
Another important reason why it may be more advantageous for the individual to engage in limited aggression rather than follow a 'hawk' strategy involves the concept of 'inclusive fitness'. Hamilton (1964) sets forth the concept as follows: “The social behaviour of a species evolves in such a way that in each behaviour-evoking situation the individual will seem to value his neighbours' fitness against his own according to the coefficients of relationship appropriate to that situation”. If some of the individuals with which the 'hawk' fights are his relatives - a likely event in many social species - they will share some of his genes. So by killing or injuring them, the 'hawk' would actually be lowering his own inclusive fitness to the degree that copies of its own genes are also carried by its relative opponents (Fry, 1980).
In vertebrate species, since half of an individual's genes come from its mother and half from Its father, the relatedness between parents and offspring is exactly 50 %. On the average, full siblings will also share 1/2 of their genes, half-siblings 1/4, first cousins 1/8, and so on. Therefore, natural selection should favor individuals who, in interacting with others, act as if they consider this degree of relatedness. This is not to suggest that animals consciously calculate coefficients of relatedness before behaving either altruistically or selfishly in relation to others (Fry, 1980). As Barash (1976) puts it, natural selection performs this cost-benefit on behaviors such as aggression. As previously indicated, a primary reason that group selection analyses were proposed in the first place was because certain behaviors observed frequently in various animal species were thought to lower the fitness of the individuals engaging in them. Wynne-Edwards (1962) and others suggested that, since agonistic encounters of limited severity appear to reflect acts of altruism on the part of the aggressors, ritualized aggression and other forms of limited combat must have evolved via group selection. They could not have possibly have evolved through Darwinian individual selection. As many scholars have recently stressed, this conclusion should be questioned (See Alexander, 1974; 1975; 1980; Dawkins, 1976; Clutton-Brock & Harvey, 1976; Maynard Smith, 1974; 1976; Symons, 1978a,b; and G.C. Williams, 1966).
The idea of evolutionarily stable strategies and the concept of inclusive fitness provide a theoretical framework for interpreting ritualized or limited aggression and other so-called altruistic patterns in terms of updated Darwinian natural selection. Behaviors formerly assumed to aid the group or species, to the detriment of the actor, may not be altruistic after all and in fact may even increase the inclusive fitness of the actor (Fry, 1980).

The Four Levels of Behavioral Explanation
In the explanation of behavior, the distinction between 'proximate causes' and 'ultimate causes' is of paramount importance. 'Proximate causation' concerns the direct mechanisms that bring something about. It is, in a sense, an account of the structure of the individual organism, of how the animal is organized so that different environmental factors and events influence its behavior in particular ways. 'Ultimate causation' concerns adaptive significance - that is, the value of a certain behavior, its selective consequences, which must ultimately entail reproductive consequences. It is causation on a generational time scale. The claim that it is causation at all rests on the assumption that these adaptive results of behavior have been instrumental in its evolving and being maintained by natural selection. A third level of explanation concerns development. Development within an individual life span is called 'ontogeny', in contrast to the evolutionary change that takes place over many generations and is called 'phylogeny', the fourth explanatory level. Explanations in terms of ultimate causation and phylogeny are intimately related, but these levels can be differentiated. The former is concerned with the functional significance of behavior with what it is for and with the consequences that are relevant for its maintenance by natural selection. Phylogeny is concerned with the raw material out of which the behavior evolved - the preexisting behavioral organization from which natural selection has been able to sculpt the behavior in question (Daly & Wilson, 1978). Proximate causation, ultimate causation, ontogeny and phylogeny correspond to Tinbergen's (1968) cause, function, development, and evolution. We decide which explanatory level is appropriate whenever we attempt to answer the question 'why'. Yet defenders of an explanation at one level often imagine that their pet theory obviates explanation at all others. Most of the acrimonious debates of behavioral science derive in large measure from this sort of narrow advocacy. Take the controversial subject of human aggression, for example.
There has been much fruitless debate among three groups: supporters of ontogenetic explanations of aggressiveness, such as the theory that attaches great importance to the behavior of parents as models for children; those who explain aggression as the proximate result of frustration; and still others who see it as an evolved aspect of behavior selected for its value in attaining useful goals. Lorenz (1966) correctly stressed that aggressive behavior has evolved because of its adaptive value to the aggressor. He repeatedly implied, however, that aggression must therefore be impervious to variations in early experience, which neither follows logically nor accords with actual evidence. Some critics, such as Montagu (1968), have replied with a similar non sequitur: Since aggressive behavior is learned ontogenetically, there is no validity to evolutionary explanation. Both arguments overlook the necessity for explanation at multiple levels. Evolution occurs because natural selection operates on animals that are the interactive products of their genes and their individual ontogenetic histories. If aggression serves its perpetrator, it will be selected for. Natural selection does not depend on whether or not the aggressive behavior is in part the result of learning. It is quite wrong to imagine that behavior that has been learned has not also evolved. But it is equally wrong to suggest that behavior that has evolved is therefore somehow 'contained in the genes', as if ontogenetic factors had no influence at all (Daly & Wilson, 1978).

Ritualization and communication
Strictly speaking, a ritualized response is specialized through evolution for communication. Ethologists prove ritualization of a response by tracing its phylogenetic relationship to a less showy and less stereotyped response, in the same or a related species, which might serve another purpose altogether. A ritualized threat might be derived from a feeding response. When the origins of a ritual are traced, its ceremonial features for signaling can be detected. The concept of ritual becomes abused, however, when a response with some of these features is explained as ritualized even though no ancestral behavior has been identified and no evidence of communication has been shown. An animal can exhibit ritual-like postures and movements when in a conflict between antagonistic behaviors or when in extreme fear, without sending any message to conspecifics. One can determine whether a possible display is communicative (and therefore ritualized) by calculating the degree to which the display of one animal alters the behavior in another (Brown 1975). The duet-like behavior of many threat displays between competitors is evidence of a high degree of ritualization. Measurement would give meaning to claims that the behavior of animal X is 'highly ritualized'; instead, such statements often are based on subjective impressions of beauty or ceremony. The requirement of communication would also remove many kinds of fighting from the category of ritual. The adjective stylized (Ewer, 1968) seems more appropriate for fighting that was shaped mostly by the advantages of noninjury rather than for communication (Schuster, 1978).
The misleading use of ritual can be appreciated in the lek aggression of lechwe. Leks have long been famous for a high degree of both aggression and ritualization (Lack, 1968), ever since Darwin (1871) relied heavily on lek ritualization to support his theory of sexual selection among competing males. In lechwe, the relationship between aggression and ritualization is more complex than any simple dichotomy between fighting and ritual. There is also no obvious trend toward ritualization if quantitative measures are taken.
If the total repertoire of lechwe threat behaviors is counted, the number in about eight ('about' because some never occurred alone or their status as displays in not proven). In this respect, lechwe are not unusual. Wildebeest have thirteen displays classed as 'agonistic' by Estes (1969), plus many others that appear regularly in territorial confrontations. Most lechwe threat behaviors - including a simple threat-chase, in which a male walks deliberately or rushes toward another with head at shoulder height and muzzle pointing forward - are shared by nonlekking antelope.
The most distinctive feature of lechwe aggression is the high frequency of the threat-chase, a behavior lacking the classic features of ritual and used to displace another male from its location when leks are organizing. In most males, the threat-chose is often executed with a slow, deliberate walk. The threat-chase of master males in usually a rushing charge, after which the defender races back to the females. The speed is functional. Challengers are denied the chance to enter the mating center by exploiting aggressive involvements of the master male. And master males often must interrupt a courtship sequence to expel a challenger. A briefly executed threat-chase allows the male to quickly resume the courtship in an almost uninterrupted sequence leading to copulation.
Ritualization in also absent when a challenger is not intimidated by the threat-chase of another. A continuous fight ensues, during which the horns remain interlocked for the duration - the animals vigorously pushing, twisting, and turning until one turns and flees. By means of such fights, interchanges occur on territories.
The only ritualized encounters occur during face-offs between territorial neighbors, sometimes known as 'challenge rituals' (Estes, 1969). These are dominated by sequences of threats and appeasements executed in a slow, hesitant manner and with a duet quality of simultaneity that fully justifies the description of ritualized. One feature in worth noting because of its rarity: Threats are frequently accompanied by penis erection, a physiological sign of sexual arousal in an aggressive context. Significantly, it is also reported in the lekking kob but is absent in most territorial antelope. The erection appears instantly with the onset of threatening and disappears just an quickly when the threat ceases, or during appeasement. It also disappears during actual fighting. Whether the erection can function alone as a threat in not known. But it is a classic example of a response undergoing a change in function and motivation, a feature of ritualized responses sometimes known an 'emancipation'.
Unlike encounters with threat-chases or fights, challenge rituals have no clear winner or loser, The purpose seems more a social interaction than a confrontation, signaling territorial presence and boundaries.
After a period of threatening, lasting as long as twenty minutes, the males slowly separate and return to their areas. Despite all the evidence of ritualization, the majority of challenge rituals include at least one brief period of fighting.
The pattern of lechwe aggression defies any attempt at a division between fighting and ritual. The most frequent behavior, the threat-chase, is an example of neither; the most ritualized confrontations include fighting. A more important distinction is between two different goals: to displace a rival or to interact with a territorial neighbor. These goals also seem to account for the degree of ritualization.
Displacement aggression, whether by threat or by fight, in a nonritualized contest between rivals for the same territory. When no territory is at stake, interactions are full of ritual. Perhaps communication is more important between neighbors signaling their status. Also, their social status in more nearly equal, explaining the hesitation and frequent appeasement that can develop into ritual. Master males, judging by their behavior, occupy a dominant position over all others and are able to dispose of challengers with a brief charge.
An accurate description of aggression in any species should avoid simple dichotomies and measure the overall frequency of aggression; the frequencies of individual behaviors; and the degree of fighting, of ritualization, of injury, and of death - all as separate dimensions that can combine in all possible combinations (Schuster, 1978).


A Simple Model: The War of Attrition
The War of Attrition is one of the simplest models that have been considered. It represents a contest which is settled by display alone. In the model, individuals are imagined to show their threat display at constant intensity until one gives up, leaving the other, which was prepared to go on at this point, as the winner. In this game Maynard Smith (1974) showed that the ESS is to choose the duration of the display, X, according to the negative exponential distribution

        P(X) = (1/V) exp(-X/V)

A result which is important for subsequent arguments is that the average cost of the contest (owing to the time wasted in the display) under this model is equal to V/2, where V is the gain from victory (Maynard Smith, 1974).

Fig. 5: The war of attrition (from Caryl, 1981).

In his analysis, Maynard Smith tacitly assumed that the cost of display would be a linear function of its duration. But this assumption may not be valid, even where the display is of constant intensity. To display, an animal must neglect other activities, and the cost per unit time of neglecting feeding, for instance, would be expected to rise sharply as the animal's food deficit becomes larger. In another context, there is evidence of cost function which are quadratic (Review by McCleery, 1978).
The assumption that the cost of display is linearly related to its duration is not necessary. Norman, Taylor & Robertson (1977) and Bishop & Cannings (1978) have shown that the model can be extended very easily to deal with cases in which the cost per unit time changes during the course of the display - the equation above is rephrased in terms of a cost function that relates the cost of the display to its duration (Caryl, 1981). In the simplest model of the War of Attrition the value of reward to either contestant is the same and equal to V. Maynard Smith (1974), Bishop & Cannings (1978), and Hines (1977) have argued that in such an environment the ESS is for each contestant to adopt the mixed strategy of bidding X or more with probability e-x/V.
However, there are two serious objections which can be leveled at the model and the inferences drawn from it (Riley, 1980). First of all, while adoption of the exponential mixed strategy yields, on average, a gain to the winner, there is no social gain from contests: the full value of the reward is, on average, just offset by the value of the resources committed to the contest. The second objection to the simple model is that the evolutionary stability of the exponential mixed strategy is suspect. In Riley (1979) it is established that only for infinite populations are the equilibrium conditions proposed by Maynard Smith sufficient for a strategy to be evolutionary stable.
Riley (1980) provides a strong stability condition which is sensitive to the population size. He then demonstrates that in the War of Attrition with uncertain rewards there is a unique 'strong evolutionary equilibrium' strategy. Strategy V is a strong evolutionary equilibrium strategy if it yields an expected return of 'fitness' exceeding that of any feasible alternative μ when the population adopting μ is small. As the population becomes large this is shown to approach the solution strategy proposed by Bishop, Cannings & Maynard Smith (1978).
Hammerstein & Parker (1982) provide a game theoretical analysis of the asymmetric War of Attrition with incomplete information. This is a contest where animals adopt different roles like 'owner' and 'intruder' in a territorial conflict, and where the winner is the individual prepared to persist longer. The term incomplete information refers to mistakes in the identification of roles. The idea by Parker & Rubenstein (1981) is mathematically worked out and it is confirmed that there exists only a single ESS for the model with a continuum of possible levels of persistence and no discontinuities in the increase of cost during attrition. The ESS prescribes to settle the conflict according to 'who has more to gain or less to pay for persistence'. The only evolutionary stable convention is thus to give the player access to the resources who has the role which is favored with respect to payoffs.
By contrast, it was shown earlier (Hammerstein, 1981) for various asymmetric versions of the 'Hawks-Doves' model (Vide infra) that an ESS can exist which appears paradoxical with respect to payoffs. The nature of this contrast is further analyzed by introducing elements of discreteness in the asymmetric War of Attrition. It turns out that some conditions must be satisfied in order to have the possibility of an alternative ESS which is not of the above simple commonsense type. First, a decision to persist (or escalate) further in a contest must typically commit a contestant to go on fighting for a full 'round', before he can give up without danger. Second, such a 'discontinuity' must occur at a level of persistence where the contest is still cheap, and, finally, errors in the identification of roles must be rare.

The War of Attrition is an example of an important class of models that Caryl (1981) calls 'continuous models'; the cost of a contest, dependent on its duration, is continuously variable. An alternative type of model is what Caryl calls the 'discrete model', and is exemplified by the model used by Maynard Smith & Price (1973). (In this model, the contest could be fought at two distinct levels of escalation, and the most important factor in bringing it to an end was serious injury, which produced a large, discrete, increment in the cost of the contest and caused the injured animal to cease fighting).
Game theorists have used discrete models to model escalation in animal contests, assuming that the escalation involves a series of steplike changes in the intensity and potential danger of the interactions. However, in his discussion of the ethological evidence, Caryl (1981) emphasized the progressive nature of the escalation in many animal contests, a quality which suggests that a continuous model might be more appropriate. The effects of changes in the vigor or intensity of display, and perhaps also changes in the risk of serious injury, might in principle be accounted for in terms of an appropriate cost function. But continuous models depend critically on assumptions about the function relating the costs to the duration of the contest. As yet, we have no information about the shape of these cost functions, although we can make guesses from the distribution of contest lengths (Caryl, 1981).

Unlabelled Ordinal Conflicts
In the War of Attrition model of animal conflict a reward is obtained by whichever of two opponents displays longer, each individual incurring a cost associated with the length of the contest. This model is generalized by Bishop & Cannings (1978) to allow more general reward and cost functions, and restrictions on the length of contest permitted. This permits unification of the War of Attrition model and the Graduated Risks model (Vide infra), and also the extension to models in which contests may end either due to injury, or to retreat. Bishop & Cannings (1978) consider a particular class of conflicts which are a generalization of several suggested by Maynard Smith. This class they term 'unlabelled ordinal conflicts'. They are unlabelled in the sense that there are no features or labels which distinguish the animals either in a way which is important in a physical sense, e.g., size or strength, or indeed any other label which might be used to develop a cooperative strategy. The conflicts are ordinal because the outcome is determined as a win for whichever contestant plays the larger value from some one-dimensional ordered set, M*, say. M* is termed the value space. In this case, the values and the pure strategies are identical.
Associated with any conflict is a pay-off, going to the winner, and a loss by both winner and loser representing the energy expenditure during the contest. Specifically, suppose that values x and y are chosen by two animals, with x>y, x,y ∈ M* then if X is 'play x always' and Y is 'play y always', Bishop & Cannigns take
        (i) pay-off to animal playing X = f(y) - g(y),
        (ii) pay-off to animal playing Y = - g(y)
In the event that x = y each animal gets pay-off [f(y)/2] - g(y). The rationale behind these pay-offs is simply that the conflict is assumed to progress, or escalate through the values [0,y) and end at y. The pay-offs are thus related to y rather than x. the f function is called the reward function, the g function is the energy function.
Two particular examples are the War of Attrition (Maynard Smith, 1974) and the Graduated Risks conflict (Maynard Smith & Parker, 1976). The details of these are as follows:
(1) War of Attrition: f(y) = V, V>0, g(y) = y. V is envisioned as the reward for winning some item such as a piece of food or a territory and y simply corresponds to the time invested in a competition. Here M* would normally be [0,∞] or [0,s].
(2) Graduated Risks: f(y) = V(1 - y), g(y) = y(D - V)/2. V is the reward for winning, D is the loss if injured and here y is a measure of the degree of aggression to which an animal is willing to raise the conflict, where y = probability of the fight ending in injury. Here M* could be [0,1] or some sequence {y0,. y1,.. yn,...} either terminating or not, perhaps corresponding to distinct behavioral patterns rather than a steady gradation of aggression.
Bishop, Cannings & Maynard Smith' (1978) essential conclusion is that the effort an animal is prepared to make to win a contest (in time or energy expended, or in risk run) will increase as the pay-off for victory increases.
Bishop & Cannings (1978) also include a model which is a mixture of the previous two.
(3) f(y) = V e-2λy, g(y) = μ/2 - (V - D)(1 - e-2λy)/2 where V is the reward for winning, D is the loss if injured, μ is the rate of using up energy during the conflict and λ is a measure of the level of aggression at which an animal is fighting. In fact λ is the rate of injuring your opponent, i.e., λδt = probability of injuring your opponent in time interval (t, t + δt) (δt, small) given that there was no previous injury. Usually, M* = [0,∞]. The contest is as per the War of Attrition model except that the possibility is allowed of either contestant being injured. If there is an injury, the contest finishes immediately with the uninjured animal taking the reward. The War of Attrition results can be obtained from this model by letting λ tend to zero and putting μ = 1, and the Graduated Risks model by putting e-2λy = 1 - y and μ = 0. In each case it is demonstrated that either there is no ESS or there is a unique ESS, which is fully specified.
In essence, the War of Attrition supposes that two animals are competing for an indivisible prize of value V, and that their only choice of strategy is of the length of time for which they will continue, and hence of the price they are prepared to pay. The ESS is to continue for a variable length of time t, with a probability density p(t) = 1/V e-t/V. In such contests, all individuals are assumed to be identical. If there is some external asymmetry associated with each contest - for example between the owner and the interloper in a territory, or the larger and smaller of two contestants - then the ESS is for the contest to be settled immediately in accord with the asymmetry (Maynard Smith & Parker, 1976). This of course depends on the asymmetry being perceived by both contestants.

Bishop, Cannings & Maynard Smith (1978) consider contests in which an animal may be in different states on different occasions, but in which the state is known only to the individual and not to his opponent. Suppose for example that an animal is sometimes hungry and sometimes less so. It is common sense that it should be willing to compete more strongly for food when hungry. But how much more strongly? It is important to remark that a contestant who is not hungry will not reveal the fact to his opponent, because it is always possible that his opponent is not hungry either, in which case the food may be won cheaply. The authors derive the ESS for contests with variable rewards, when the value is known only to the contestant and not to his opponent. They make a number of simplifying assumptions. In particular, the states of the two contestants are assumed to be independent, the pay-off to an individual is independent of his opponent's state, and the 'plays' open to an individual do not depend on his state, and extend from 0 to ∞.
They also note that the state of an individual might well depend on his success or failure in previous conflicts (he might become more hungry if defeated, less so if victorious). The extension of the analysis to non-violent Wars of Attrition also finds support. Thus, to take the example of the Siamese fighting fish, Betta splendens, Maynard Smith (1972) writes: “Ritual conflicts between males are usually followed by escalated fights, in which one or both rivals may be seriously injured. Conflicts between females however often end (typically after 5 -15 minutes) with the surrender of one fish, without escalated fighting. Simpson followed such conflicts in detail, measuring the frequency and timing of particular components of the ritual. He found no significant difference between the frequencies with which eventual winners and eventual losers performed particular acts, except during the last two minutes of a contest, when the eventual winner could be recognized from the fact that her gill covers were erected for a larger proportion of the time. The fact that the winner could not be distinguished from the loser until close to the end of a contest fits well with the prediction from game theory”.

Hawks and Doves
The simplest discrete model is the celebrated Hawks and Doves game (Maynard Smith, 1976; Maynard Smith & Parker, 1976). The pay-off matrix is shown in the following table.

  Hawk   Dove  
Hawk   V-D/2   V  
Dove   0   V/2  

Fig. 6: The Hawks and Doves game. The payoffs in the table are to the tactic in the row when played against the column. V is the gain from victory, D the cost of injury. For the matrix shown, which ignores the cost of threat, the ESS is to escalate with probability P = V/D (when V < D). If threat imposes a cost T on each opponent, the probability of escalation becomes P = (V + 2T)/(D + 2T) (Caryl, 1981).

Consider a simple model: a species that in contests between two individuals has only two possible tactics, a 'hawk' tactic and a 'dove' one. A hawk fights without regard to any convention and escalates the fighting until it either wins (that is, until its opponent runs away or is seriously injured) or is seriously injured. A dove never escalates; it fights conventionally, and then if its opponent escalates, it runs away before it is injured. At the end of a contest each contestant receives a payoff. The expected payoff to individual X in a contest with individual Y is written E(X,Y). The payoff is a measure of a change in the fitness of X as a result of the contest, and so it is determined by three factors: the advantage of winning, the disadvantage of being seriously injured and the disadvantage of wasting time and energy in a long contest. For the hawk-dove game suppose the effect on individual fitness is +10 for winning a contest and -20 for suffering serious injury. Suppose further two doves can eventually settle a contest but only after a long time and at a cost of -3. (The exact values of the payoffs do not affect the results of the model as long as the absolute, or unsigned numerical, value of injury is greater than that of victory).
The game can be analyzed as follows. If the two individuals in a contest both adopt dove tactics, then since doves do not escalate, there is no possibility of injury and the contest will be a long one. Each contestant has an equal chance of winning, and so the expected payoff to one of the doves D equals the probability of D winning the contest (p - 1/2) times the value of victory plus the cost of a long battle, that is, E(D,D) equals (1/2)(+10) + (-3), or +2.
Similarly, a hawk fighting another hawk has equal chances of winning or being injured but in any case the contest will be settled fairly quickly. Hence the expected payoff E(H,H) is equal to (1/2)(+10) + (1/2)(-20) or -5. A dove fighting a hawk will flee when the hawk escalates, so that the dove's expected payoff is 0 and the victorious hawk's payoff is +10. Now suppose the members of a population engage in contests in the hawk-dove game in random pairs and subsequently each individual reproduces its kind (individuals employing the same strategy) in proportion to the payoff it has accumulated. If there is an ESS for the game, the population will evolve toward it. The question, then, is: Is there an ESS for the hawk-dove game?
It is evident that consistently playing hawk is not an ESS: a population of hawks would not be safe against all mutant strategies. Remember that in a hawk population the expected payoff per contest to a hawk E(H,H) is -5 but the expected payoff to a dove mutant E(D,H) is 0. Hence dove mutants would reproduce more often than hawks. A similar argument shows that consistently playing dove is also not an ESS.

Serious injury = -20            E(H,H) - 1/2(+10) + 1/2(-20) = -5
Victory = +10                E(H,D) = +10
Long contest = -3            E(D,H) = 0
                    E(D,D) = 1/2(+10) + (-3) = +2
  Hawk (H)   Dove (D)  
Hawk (H)   -5   +10  
Dove (D)   0   +2  

There is, however, a mixed strategy that fulfills the requirements of an ESS. A mixed strategy is one that prescribes different tactics to be followed in a game according to a specified probability distribution. The mixed strategy that is evolutionarily stable for the hawk-dove game is play hawk with probability 8/13 and play dove with probability 5/13. The hawk-dove model predicts that mixed strategies will be found in real animal contests, either in the form of different animals adopting different tactics (such as hawk and dove) or in the form of individuals varying their tactics (Maynard Smith, 1978).
If the cost of injury D is so great that it exceeds the value of the prize, V, then hawks cannot exclude doves from the population: the ESS is a mixed equilibrium with p = V/D where p is the proportion of hawks. If D < or equal V then all animals are hawks (Caryl, 1981).

Prior Fitness
Treisman & Collins (1980) demonstrated that in addition to the value of the prize and the possible damage inflicted, the animal's fitness prior to a contest (the parameter Fo) may affect the ESS for that contest in the hawk-dove game. When Fo is low the expected probability of attack may be higher, for D large, than would otherwise be predicted. Computer simulations of selection over 1000 generations indicate that the effect may be less marked when multiple contests may ensue than for a single contest. An animal may face a contest which will occur once only but which is not the sole determinant of its fitness; for example, it may have the chance of taking over its sire's harem, or of being expelled and having to obtain mates by recruitment or to rely on insemination as a hanger-on. But if all its chances of future reproduction will occur once only, then necessarily, so Treisman & Collins predict, it must always attack. Something close to this situation may arise when the established male, in sole possession of a harem (and who, if displaced, will find no other) faces a younger challenger. Certainly the older male should always attack as fiercely as he can.

The Prudent Hawk Gambit

Fig. 7: The prudent hawk gambit (from Caryl, 1981).

It could be argued that display only rarely involves a high cost, whereas this is required in every escalated contest. But suppose a new gambit arose which involved escalating to the same level as hawks, but withdrawing after a suitable period of time even if no injury had occurred. The principle involved could be the same as that which allows animals to decide when to terminate display, and the period could be adjusted so that the occasional serious injury produced an average cost of V/2. Caryl (1981) calls this gambit the Prudent Hawk. The table shows the payoff matrix for this game.

  Hawk   Prudent Hawk   Dove  
Hawk   V-D/2   -(V-D/2)V/D   V  
Prudent Hawk   V-D/2.V/D   0   V  
Dove   0   0   V/2  

To give the doves a chance, Caryl reverts to the convention that threat carries no cost. The new gambit would always win over doves; it would also sometimes win over hawks, although it would sometimes be injured in these contests. When the probability of injury in contest between two prudent hawks is V/D (so that the cost of these contests is equal to their average payoff, V/2), the ratios of types are:

         Hawks      Prudent Hawks    Doves
1/(α-1): 1 : 1/α

where α = D/V, the 'riskiness' of escalating. When α = 2, hawks and prudent hawks each form 40 % of the population, but when α = 8, prudent hawks have risen to 79 %. Thus by escalating, but stopping when prudent, an individual can do very well, and under this model most combats should be escalated. However, few should lead to serious injury - most would stop before this occurs. Intuitively, this seems to fit the biological facts better than the original model in which all escalated contests ended in serious injury. Geist's (1971) review of data on moose and other species showed that it involved risk of injury or death (respectively) of about 10 % and 4 % per year, not per contest, while data for mule deer give an estimate of 10 % per year as the chance of injury (Geist, 1974), and data for musk oxen give values of 5 % to 10 % per year for the chance of death (Wilkinson & Shank, 1977).

The War of Nerves
It will be apparent that there is nothing magical about the choice of V/2 for the cost of an escalated combat. Animals could be imagined to vary the cost in a continuous fashion by fighting for a longer or shorter period at the escalated level. More important, Bishop & Cannings (1978) have described an extension of the War of Attrition in which the possibility of injury to one opponent is considered. The injured animal is assumed to stop fighting at once, an assumption which could be justified in terms of Parker's (1974) argument that an injury would decrease the fighting ability of the animal, but which may not always be biologically relevant. Bishop & Cannings showed that the ESS for this model was a negative exponential distribution of the form

        P(X) = (1/M)exp(-X/M)
with mean
        M = V[[μ - λ \(V - D)]

where μ is the cost per unit time of display, and λ is the chance of injury. Although Bishop & Cannings place little emphasis on the model, it seems important enough to warrant a name of its own, and so Caryl (1981) called it the War of Nerves. Clutton-Brock et al. (1979) have described the fights of red deer stags as Wars of Attrition in which the rate of payment of costs mounts with the duration of the contest, with the additional possibility that a false move may lead to injury to either contestant at any stage after the two animals have locked antlers. This description is that of a War of Nerves in which the cost per time unit, μ is a function of time, μ(t), rather than a constant (Caryl, 1981).

Dying Gasp Retaliation
It has been widely accepted that one functional difference between escalated combat and threat is the cost of the behavior. Escalated combat involving physical fighting is often clearly more energetic than the displays that may begin the contest; the individuals may also be less responsive to events outside of the combat, and hence more susceptible to predation; and the combat may produce injuries which carry a cost in terms of a period in which the efficiency of some or all types of behavior is decreased, even if they do not result in death. Recent work has corroborated the earlier analysis of Geist (1971) in showing that serious injury and death do occur as consequences of fighting, and provides some support for the idea that the risk from escalated combat may be considerable.
In some models (e.g., the Hawks and Doves model) it is argued that this fact is sufficient to explain the persistence of some form of low-cost non-escalated combat within the population: although an animal which is prepared to escalate further than its rival may win the contest, the escalator will pay a heavy cost when it meets another individual like itself, and as this is likely in a population composed mainly of escalators, pacific individuals will be at an advantage (Caryl, 1981). Imagine two alternative fighting strategies in a highly venomous species such as a scorpion. 'Hawk' scorpions go all out for the kill, and use their lethal sting. 'Retaliators' wrestle with their claws, and never sting except in retaliation. A retaliator who has been stung attempts, with his dying gasp, to sting his murderer. If he succeeds, both die. Calculation shows that the outcome of this simple game depends upon the probability, p, that a mortally wounded scorpion, whether hawk or retaliator, will succeed in stinging his murderer with his dying gasp. Hawk is an ESS only if retaliation is totally ineffective (p = 0). If p is nonzero but small, a stable mixture of hawks and retaliators evolves. If p is large, retaliator, on its own, is the ESS. Let us make the plausible assumption that p is large, and that the population is therefore dominated by retaliators.
Now we come to the interesting point. As far as his survival or genetic success is concerned, retaliation is pointless for the individual retaliator. Once he has been stung he is doomed. Stinging back does him no good at all. Yet retaliation is the dominant strategy in this model population, because it is the ESS. The point is that, even if an act does not benefit the inclusive fitness of the animal doing it, it can still predominate in a population because it is a manifestation of an ESS (Dawkins, 1980).
Maynard Smith & Price (1973) have devised a more subtle model where one had five types: Hawk, Dove (which they call 'Mouse'); Bully who comes on as a Hawk, but quickly changes to a Dove or flees if the opponent is also Hawkish; Retaliator who comes on as a Dove, but who turns Hawkish if the opponent is Hawkish; and Prober-Retaliator, who plays Retaliator most of the time, but every now and then tests out the mettle of the opponent by turning Hawkish. By assigning certain plausible values the authors show that probably the population would evolve to mainly Retaliator and Prober-Retaliator, with just a small number of Mice (who will never fight back); but as the authors point out, in any real population, one is almost bound to get some Mice: the young, the old, the sick, and so on (Ruse, 1979).
Dawkins (1980) warns that the classic Retaliator in the original model of Maynard Smith & Price (1973) is now known not to be, strictly, an ESS at all. It is an equilibrium, since in a population of retaliators no other postulated strategy would do better. But it is not stable, for, if the population composition drifted a little from the equilibrium, selection would not tend to restore it. Instead it would push it further away, presumably to the other known equilibrium, a stable mixture of 'hawks' and 'bullies' (Gale & Eaves, 1975). It is not difficult to rephrase the model slightly so that retaliator becomes a true ESS. Dawkins & Krebs (1978) incorporate one such improvement in their summary of the model.

Pure, Mixed, and Conditional Strategies
A mixed strategy is the opposite of a pure strategy. Hawk, dove and retaliator are all pure strategies. So is 'wait for 5 1/2 minutes' a pure strategy. But 'play hawk with probability p' is a mixed strategy, and so is 'wait for a time t where t is drawn at random from a probability density function'. The diagnostic feature of a mixed strategy is that its specification contains at least one probabilistic statement. 'Stochastic' might be a better label than 'mixed'. To anticipate a later section, a mixed strategy should not be confused with a 'conditional' strategy such as -retaliate IF your opponent attacks you'. The mathematical equivalent of a mixed strategy can be achieved if each individual plays a pure strategy, the population as a whole containing a mixture of pure strategists. We can thus think of the hawk-dove game as ending in a stable polymorphism, a mixture of pure hawks and pure doves in critical proportion, p. But equivalently the ESS could consist in each individual being a stochastic 'dawk', choosing to play dove or hawk at random, with a built-in bias corresponding to the critical proportion, p. Any combination of these two extremes would be stable, provided that in the population as a whole the strategy hawk was played p of the time and dove 1 - p of the time. Similarly, the War of Attrition could lead to a stable polymorphism, the frequencies of the morphs following the predicted negative exponential distribution; alternatively each individual could produce a random sequence of waiting times of different length, drawing them from a negative exponential probability density function (Dawkins, 1980).
A conditional strategy is like a computer program with an 'IF' statement. Whether pure or mixed, most of the strategies we have so far met have been unconditional. Even 'dig with probability p' is unconditional, since what the animal does on a given occasion is determined by a random process and not by a recognizable event.
Retaliator is an example of a conditional strategy, as is 'fish if the weather is fine, steal if it is wet'. Maynard Smith & Parker (1976) have considered the often surprising consequences of postulating strategies conditional upon asymmetries in aggressive contest between two individuals, for instance, 'attack if larger, retreat if smaller' (Vide infra).
Selten (1980) has proved that mixed strategies cannot be stable in such asymmetric games. He has shown that in game models of asymmetric animal conflicts where the opponents assume different roles like 'owner' and 'intruder', evolutionary stable strategies must be pure strategies if a condition of information asymmetry holds. The condition is satisfied if two opponents always have different roles. Information about the opponent's role may be incomplete. Great attention is being paid at present to the possible existence of alternative male strategies in vertebrate populations. For instance, in mammalian species where dominant males hold harems of females, subordinate males sometimes adopt a policy known as kleptogamy (Clutton-Brock, Albon & Guinness, 1979; Cox & LeBoeuf, 1977). Kleptogamists sneak briefly into harems and steal hurried copulations before being chased away by the harem master. It is just possible that in some species kleptogamy and harem-holding genuinely represent two strategies in a stable mix. In this case the average benefit of the two strategies will be equal. But in most cases it is much more likely that harem masters fare consistently better than kleptogamists, and that the ESS is the pure conditional strategy: 'if possible hold a harem; if you can't, be a kleptogamist'. Then in the stable state all males will be playing this one strategy, and the behavior that an individual actually shows will be conditional on factors like his size or skill in combat (Dawkins, 1980).

Asymmetric Contests
It is obvious that real animals can adopt strategies that are more complex than 'Always escalate', 'Always display' or some mixture of the two. For example, some animals make probes, or trial escalations. Other employ conventional tactics but will escalate in retaliation for an opponent's escalation. There is, however, another important way in which many real animal contests do not conform to the hawk-dove model. Most real contests are asymmetric in that, unlike hawks and doves, the contestants differ from each other in some area besides strategy.
Three basic types of asymmetries are encountered in animal contests. First, there are asymmetries in the fighting ability (the size, strength or weapons) of the contestants: differences of this kind are likely to affect the outcome of an escalated fight. Second there are asymmetries in the value to the contestants of the resource being competed for (as in a contest over food between a hungry individual and a well-fed one): differences of this kind are likely to affect the payoffs of a contest. Third, there are asymmetries that are called uncorrelated because they affect neither the outcome of escalation nor the payoffs of a contest. The uncorrelated asymmetries are of special interest because they often serve to settle contests conventionally (Maynard Smith, 1978; Cf. Maynard Smith, 1974; 1976; Parker, 1974; Maynard Smith & Parker, 1976; Parker & Rubenstein, 1981).
Perhaps the best example of an uncorrelated asymmetry is found in a contest over a resource between the 'owner' of the resource and an interloper. In calling this an uncorrelated asymmetry it is not meant that ownership never alters the outcome of escalation or the payoffs of contests; it simply means that ownership will serve to settle contests even when it does not alter those factors. To demonstrate the effect of such an uncorrelated asymmetry Maynard Smith (1978) returns to the hawk-dove game and adds to it a third strategy called bourgeois: if the individual is the owner of the resource in question, it adopts the hawk tactic; otherwise it adopts the dove tactic.

The Hawk-Dove-Bourgeois Game
    E(H,B) = 1/2E(H,H) + 1/2E(H,D) = -5/2 + 10/2 = +2.5
Serious Injury = -20
    E(D,B) = 1/2E(D,H) + 1/2E(D,D) = +0 + 2/2 = +1
Victory = +10
    E(B,H) = 1/2E(H,H) + 1/2E(D,H) = -5/2 + 0 = -2.5
Long Contest = -3
    E(B,D) = 1/2E(H,D) + 1/2E(D,D) = +10/2 + 2/2 = +6
    E(B,B) = 1/2E(H,D) + 1/2E(D,H) = +10/2 + 0 = +5
  Hawk (H)   Dove (D)   Bourgeois (B)  
Hawk (H)   -5   +10   +2.5  
Dove (D)   0   +2   +1  
Bourgeois (B)   -2.5   +6   +5  

In the hawk-dove-bourgeois game it is assumed that each contest is between an owner and an interloper, that each individual is equally likely to be in either role and that each individual knows which role it is playing. The payoffs for contests involving hawks and doves are unchanged by the addition of the new strategy, but additional payoffs must be calculated for contests that involve bourgeois contestants. For example, in a contest between a bourgeois and a hawk there is an equal chance that the bourgeois will be the owner (and so playing hawk) or the interloper (and so playing dove); hence E(B,H) equals 1/2E(H,H) + 1/2E(D,H) or -2.5. The remaining payoffs are calculated in a similar manner. The main point, however, is that there can never be an escalated contest between two opponents playing bourgeois, because if one is the owner and playing hawk, then the other is the interloper and playing dove. Therefore the payoff E(B,B) is equal to 1/2E(H,D) + 1/2E(D,H), or 5. When this figure is compared with the other payoffs, it is not difficult to see that consistently playing bourgeois is the only ESS for this game. Thus ownership is taken as a conventional cue for settling contests (Maynard Smith, 1978). In biological terms, a method for settling contests by taking into account some asymmetric feature, such as first arrival, which could not by itself influence the outcome, can be evolutionary stable. The commonplace adoption of 'owner wins' conventions or 'bourgeois strategies' is often taken as evidence that uncorrelated asymmetries exist in nature. However, it is not clear how such conventions can be maintained by selection, especially when there exist asymmetries in RHP (Vide infra) or resource value (Parker & Rubenstein, 1981).

Conventional Fighting as Assessment of Resource Holding Power (RHP)
Parker (1974) examined the view that the adaptive value of conventional aspects of fighting behavior is for assessment of relative resource holding power (RHP) of the combatants. According to this view, outcome of aggressive disputes should be decided by each individual's fitness budget available for expenditure during a fight (determined by the fitness difference between adoption of alternative strategies, escalation or withdrawal without escalation) and on the rate of expenditure of the fitness budget if escalation occurs (determined by the RHPs of the combatants). Thus response thresholds for alternative strategies ('assessments') will be determined by natural selection on a basis of which opponent is likely to expend its fitness budget first, should escalation occur. This 'loser' should retreat (before escalation) and the winner should stay in possession of the resource. Many aggressive decisions depend on whether one is a resource holder, or an attacker. Assuming the RHP of the combatants to be equal, there are many instances of fitness payoff imbalances between holder and attacker which should weight the dispute outcome in favor of one or other opponent by allowing it a greater expendable fitness budget. Usually the weighting favors the holder; the attacker therefore needs a correspondingly higher RHP before it may be expected to win.
The rationale behind this is explained by Parker (1974) as follows: Once 'retaliator' has stabilized as an ESS, any mutant individual able to assess from the conventional fighting stage how its own RHP compares with that of its opponent would have a selective advantage, since it could withdraw without damage when the RHP of its opponent exceeds its own by a sufficiently large amount. It is assumed that RHP is a measure of the absolute fighting ability of a given individual. If this character spreads, we may end up with a 'total peace' strategy, where all disputes are settled conventionally.
In this case, provided that the characteristic of retaliation is not lost, a mutant deficient in responding to the signals of RHP during conventional fighting will not spread - it will be disadvantageous since it will not gain any extra resources and will be beaten in encounters with individuals of higher RHP. Thus, our 'conventional assessor/retaliator' becomes the ESS. It has certain problems to face, however. Firstly, there is the obvious difficulty that selection will immediately favor exaggeration of those cues used to assess RHP. The selective advantage of this form of 'evolutionary cheating' is simple: if (for example) size is used as the cue for RHP, then where for other reasons it is disadvantageous to increase absolute size (and RHP), what will be favored are mechanisms to increase apparent size (and therefore apparent RPH). That this has happened often seems very likely. The canid threat posture involves raising the neck hair and standing erect (Darwin, 1872), so does that of many other groups including rodents (Eibl-Eibesfeldt, 1970). Lions have manes, fish often raise fins, birds fluff out feathers. Certain species have inflatable pouches. Examples are legion. Another cue very commonly used could be weaponry. Much of threat display involves exaggeration and display of teeth, antlers, claws or even hind legs, which are the main defensive weapons in locusts. It seems quite likely that these features might initially have given good indices of RHP. Where there is this type of drive for 'evolutionary cheating', a counter-selective compensatory adjustment of the assessment mechanism would continually follow in its wake.
More reliable measures of RHP might be provided by direct trials of strength between combatants. Pushing and pulling contests, or head and/or tail beating clashes abound in all groups from fish to ungulates. Very often conventional fighting consists of combinations of 'unreliable' display and 'reliable' contests of strength, implying that many cues may be used to assess relative RHP. Very often the odds appear heavily weighted in favor of the resource holder, and the absolute RHP (as judged by human eyes) of the attacker apparently has to exceed that of the holder very considerably before a take-over occurs. There are several possible reasons why this effect operates:
(1) Suppose that the 'resource structuring' is near perfect; i.e., by the assorting action of disputes, the population is perfectly truncated with the highest RHP individuals occupying all the resources. In this case, restructuring can only occur by inputs and outputs of competitors and resources or by changes in RHP status of individuals in one or both of the two groups (holders or non-holders). Here most of the observed disputes would obviously be won by the holders.
(2) The tenure of the resource may well itself increase RHP, especially where the resource is a food source. Also the outcome of a fight may involve experience of the local environment, hence tenure of the resource may increase RHP in this way. Position of the holder in guarding the resource may be very important, for instance, in the female-guarding behavior where the male clings to the female because the attacker must prise the holder off before the take-over can occur.
(3) Pay-offs may be different for the holder and attacker. Supposing that the holder will lose more than the attacker will gain, it might be expected that the holder could afford to sacrifice more units of fitness in the fights than the attacker could afford to expend. Hence where the combatants are of equal RHP, the attacker should withdraw because it will run out of expendable fighting units before the holder. Hence an attacker must be of higher RHP before it can win.
According to Parker (1974), there seems little doubt from the literature that assessment of RHP is occurring in most cases of animal combat. To avoid any implications of teleology, it must be stated that 'assessment' in this context means only that the individual responds differentially to opponents on a basis of their RHP relative to its own; the only assessment of what is the appropriate response is the unconscious one performed by selection.
Size, strength, weaponry, and experience all seem involved in RHP. Males are usually dominant over females. This often relates to RHP disparity because males are bigger; in some instances however secondary sexual characters are used as signals, e.g., comb size is a determinant of dominance in chickens (Collias, 1943). It seems possible that because of sexual selection male fitness may be increased by adopting a more dangerous strategy if this gives an overall increased insemination rate. Thus males of the same RHP as females may have a higher fitness budget for fighting over, say, food - because being in peak condition may affect male fitness more because of intra-sexual competition.
It is interesting in this context that females with young often (but not always) increase markedly in rank. They may have a higher fitness budget in such circumstances.

The Assumption of Unlimited Credit
Caryl (1981) concluded that discrete models of the type so far discussed give little help in understanding escalation, in the ethological sense of the term. What other factors may determine the choice of a level of escalation?
In these models, the mean payoffs to individuals adopting different tactics are the same, but the variance differs. The greater the variance, the greater the chance of doing very badly over a series of contests. Could this be a factor which was important in determining the choice of level of escalation? Do animals adopt nonescalated tactics to reduce the riskiness of combat?
Theorists have assumed that it is the mean payoff over a series of encounters which is important, and have not discussed the variance. This is equivalent to the assumption of infinite credit. If we were considering, for instance, the gains and losses incurred in contests over food in a small Passerine this would imply that all the losses in December, no matter how severe, could be compensated by a spell of good luck in January. In many real situations it would presumably not be possible to compensate in this way; a Passerine that did very badly over the course of a single December day might perish during the following night.
The assumption of 'limited credit' leads us to ask how an individual should choose its tactics over a series of contests to avoid 'going bankrupt'. If credit was particularly limited for certain categories of individual, e.g., the young or sick, we might expect them to choose a Dove-like, low variance strategy. In species in which an unusually high success in contests could be reflected in an equivalently large number of offspring sired by the successful contestant we might expect escalation to occur more readily than it would in those species in which there was a lower variance in the number of offspring per contestant.
Arguments such as these presuppose that success is measured in some intermediate currency, such as energy, and that the balance between profit and loss in this currency is mapped onto the ultimate scale of fitness. Previously, games theorists have dealt directly with measures of fitness, and these should (by definition) eliminate problems such as these; for instance, the cost (in units of ultimate fitness) of a given type of injury or period of display would not be constant, but would depend on the time and circumstances in which it was incurred. But it is probably realistic to look at the primary effects of most fighting in terms of a currency such as energy; for most of the reproductive cycle, the chance of reproducing immediately is zero, and the contribution of behavior to the probability of reproduction must take the form of an investment in energy, status, etc., which will later affect the chance of reproduction (Caryl, 1981).

Contests With Small Injuries
In the War of Nerves, an important assumption is that the injury (which is the only component of cost that is not partitioned equally between the opponents) is so serious that the injured animal stops fighting. The increments in costs will not always be equally partitioned; in contests in which many of the injuries are slight and do not reduce the individual's fighting abilities (e.g., Betta contests, perhaps), equally matched opponents may receive injuries in a random sequence, each involving an increment of cost to one opponent. Alternatively, rather than receiving an injury, one opponent might choose to perform a more energetic display, involving a greater cost than its rival's less energetic behavior. How can continuous models be extended to cope with this type of contest?
Caryl (1981) sketches the outlines of a possible answer to this question, as follows: It is useful to think of the cumulative cost of the fight to each opponent as being plotted along separate axes (Fig. 8). If the axes are at right angles, and if the costs are equally partitioned and the opponents equally matched, the progress of the contest will be represented by movement along the 45 degree line (line a in Fig. 8) (if the cost function was nonlinear, the speed of movement along this line would not be constant; for instance, it would increase with distance from the origin if the cost function was increasing).

Fig. 8: Probability density functions (from Caryl, 1981).

In the right-hand panel of Fig. 8 the probability density functions P(X) = (1/V)exp(-X/V) representing the probability that each opponent will give up (under the assumptions of the War of Attrition) have been plotted separately along the two axes. The function along axis 1 represents both the probability that each opponent will give up (under the assumptions of the War of Attrition) have been plotted separately along the two axes. The function along axis 1 represents both the probability that opponent 1 will retreat and also (because this will give victory to opponent 2) the reward function to opponent 2. If opponents differed in ability (for instance, if one was smaller and had to exert itself more to match its rival's displays) the asymmetry would lead to movement along a line away from the 45 degree axis (e.g., line b). Once the asymmetry became apparent it would be worthwhile for the weaker animal to give up. Even in a contest in which it was equally matched, the expected cost would be V/2, i.e., on average to win it would have to invest as much as it could expect to gain from winning or would require to find an equivalent to the contested item (Maynard Smith, 1974). When moving along line b, it must invest more than this to have the same probability of winning. In this situation, it would require a lower investment to look for an alternative item. Conversely, it would be worthwhile for the stronger animal to carry on fighting when the asymmetry became apparent.
Line c in Fig. 8 shows the course of a contest in which the cost involves a series of small injuries, and is therefore unequally partitioned. If the opponents are equally matched, the overall gradient of the trajectory will be 45 degrees, but by chance the line may depart considerably from the 45 degree line in some places.
Are the principles governing an animal's choice of tactics in this situation similar to those in situations in which the cost is equally partitioned? The only theoretical analysis which appears to be relevant here is Maynard Smith & Parker's (1976) discussion of a contest involving rounds, each ending with an injury to one opponent, which might or might not give up at this point. They examined tactics which involved giving up if injured in the first, second, etc. round. It is difficult to generalize from the authors' results, since they used a cost function for the successive rounds which increased exponentially, and thus much more steeply than those usually considered for a War of Attrition, and also because they assumed that the decision of whether or not to stop fighting was based on the result of the immediately preceding round alone, and independent of the earlier rounds, which would also affect the location on the cost plane.
Nevertheless, it is clear that the model involves an important new property, which is a feature of the payoff matrix rather than the ESS that may arise from it. In a standard continuous War of Attrition model (or in the discrete models discussed by Bishop & Cannings (1978) which are direct analogues of it) in which costs are equally partitioned, tactics which involve 'playing high' do well in contests with those involving 'playing low'. For instance, if V = 10, the tactics 'give up after 10 units of cost' and 'give up after 20 units of cost' both do well when pitted against an opponent which plays 'give up after 1 unit', and both do equally well against this opponent. This is also true for models with unequally partitioned costs in which the cost function is linear (i.e., all rounds have the same cost for defeat).
But when the successive rounds have progressively greater costs, the tactic 'give up after a defeat costing 20 units' does less well than 'give up after a defeat costing 10 units', even when both are pitted against opponents playing 'give up after 1 unit', because the individual playing the more hawkish tactics may suffer by chance a long run of defeats, even against a very dove-like opponent. This disadvantage to individuals making plays which are greater than V, the value of the reward, is in addition to the disadvantage to such individuals of the losses in combat against others like themselves, and this increases the pressure which tends to eliminate from the population tactics which involve playing greater than V.
The effect is very clear when the cost function is steep, and for symmetrical contests, Maynard Smith & Parker (1976) found a bimodal ESS which contained only two tactics. Caryl (1981) has confirmed this result in his own simulations using their very steep cost function; but when a more gradual cost function is used, the dynamics become much more complex, and it is still not clear whether one or more true ESSs exist, and from what range of initial populations they could evolve.

Contests as Random Walks
Suppose that instead of defining the costs of each round, we were to relax the rules and allow the contestants to choose the size of the next step at each point. Is there a rule governing the size of step that should be chosen at a particular stage in this contest?
Simpson (1966) drew an analogy between Fighting Fish and bidders at an auction to explain the pattern of escalation and matching seen in the contests. Caryl (1981) speculates that it might be fruitful to consider the contest as a random walk in the plane defined by the two cost axes in Fig. 8, and to look for analogies with the rules for successful gambling. It is striking that in real contests it is not necessarily the winner-to-be that escalates. Thus in the Fighting Fish, the eventual loser is as likely as the eventual winner to bite first (Simpson, 1966).
Does this readiness to escalate in the animal which (at least in the conventional view) is ultimately proved the weaker reveal a parallel to the gambler's best strategy in a game biased against him - if it is worth continuing, play boldly? Or is the conventional view incorrect? Perhaps, instead of providing a 'true' reflection or relative endurance, the results of such contests (in which the opponents are not grossly mismatched) merely demonstrate the effects of chance in determining which of the two opponents, each playing a distribution similar to those found in the War of Attrition, decides to give up first.
Caryl's (1981) argument that the weaker animal should play boldly if it is worthwhile for it to continue fighting presupposes some asymmetry in payoff, which compensates for the difference in fighting ability. This suggestion has received some support from the analysis of rutting behavior in red deer, in which it seems likely that solitary stags are in general inferior to harem-holders in fighting ability (Clutton-Brock et al., 1979). They noted that solitary stags which initiated combat with harem-holders stood to gain hinds, but could obviously not lose any, while the converse was true of holders which initiated fights with solitaries (since their hinds could be stolen by kleptogamists). They suggested that this difference in payoff could explain the observation that solitary stags initiated fights with holders more frequently than expected by chance, and holders initiated fights with solitaries less frequently than expected, but they did not discuss the relationship of events within the fight to this asymmetry in payoff (Caryl, 1981).

A Benefit of Destructive Combat?
Treisman (1977) was primarily concerned with investigating what would happen if he introduced one or two new types of mutant individuals into a Hawks-Doves population. He considered a new type that fought even more dangerously than existing Hawks, and won over them without damage, and showed that rather than a complex balance between the Doves, the existing Hawks, and these new 'Super-Hawks', the population would be a simple mixture of Doves and the new mutant. Hawks, if they were present, would be forced to display only, and thus be indistinguishable from Doves. Treisman noted that the average return to an individual in such a population increases as the danger of the escalated combat increases. The more dangerous a type of Hawk that has arisen, the greater the overall payoff. He clearly expected this result might be applicable to real animals, since he wrote: “The existence of a few members of the species who are potentially highly destructive results in an overall population strategy which increases the expected return for all... Animals... possessing a choice and which can discriminate will do better to join heterogeneous (i.e., where D>V) rather than homogeneous populations (i.e., where D<=V) assuming they adopt the appropriate behavioral strategy in each case” (Treisman, 1977).

Fig. 9: The return E to members of a Hawks and Doves population as a function of D, the cost of serious injury. The return is minimum when D = V, the gain from victory (From Caryl, 1981).

According to Caryl (1981), it is a mistake to extend the predictions of the model in this way. Although the model incorporates three mechanisms for settling disputes, only one of them involves an explicit cost which may represent what occurs in nature. Disputes between Doves are settled by a mechanism which carries no cost and is not discussed in detail.
Consider a population in which D>V: in Treisman's terms, one that is 'policed' by a few potentially highly destructive animals. In this population, most disputes are between Doves, and by assumption carry no cost.
When a large proportion of the disputes are decided without any cost in this way, it is hardly surprising that the average return is greater than when D=V, for only at this point is the cost of disputes fully accounted for by the assumptions about cost that are made in the model. (At the point where V=D, all the population are Hawks, and the assumption about the cost of injury is sufficient by itself to explain why an injured animal stops fighting: If the injured animal persisted, its chances of sustaining the next injury would be 1/2, and so the expected cost of another round, D/2, would be equal to V/2, the expected return from victory. There would be no incentive for it to continue to fight, and whatever the effect of the injury on its fighting ability, its occurrence could decide the contest.)
Caryl suggests that one factor leading to Treisman's result might be the neglect of the cost of display in the standard Hawks and Doves model. What happens when we take this cost into account? Maynard Smith (1976) has provided a formula which includes the cost of threat. In this case we should expect the cost of threat to be V/2, by analogy with the War of Attrition. Substituting this value we find that the expected return to members of the population is independent of the proportion of Hawks (i.e., of the value of D) when D>V.
Similar arguments can be applied when D<V. For instance when D is very small, why should an individual stop fighting for a valuable prize as the result of a very slight injury? Thus, Treisman's prediction is one which depends on a pair of decision rules which are unrealistic, because they are not explicitly justified on biological grounds, and it is a mistake to extend it to real contest situations (Caryl, 1981).

Relatedness and Aggression
Game theory has been used by some authors to analyze evolutionary limits to the expression of aggression in theoretical haploid parthenogenetic species. Others have examined frequency dependent selection, of which aggression may be a case, by applying population genetic models to diploid species. Treisman (1981) presents a model which attempts to combine these two approaches. Game theory is used to determine evolutionary stable strategies and corresponding stable polymorphisms for a two-strategy game played by members of a diploid sexual species, when choice of strategy is determined by two alleles at a single locus. Results are given for dominant, co-dominant and recessive determination of choice of the more aggressive of two strategies, for two levels of relationship: unrelated players and sibs. It is found that for a range of models of single locus inheritance the ESS determined for haploid species remains the stable population strategy for diploid sexual species, when players are unrelated. In sibling contestants aggression is reduced.
The mixed strategy haploid ESS underestimates, but the pure strategy haploid ESS provides a good indication of the degree to which relatedness lessens aggression in diploid species. For both haploid and diploid species there may be a considerable advantage to confining conflicts to kin.


In most competitive situations asymmetries (which may be asymmetries in fighting ability or resource holding potential, asymmetries in resource reward values, or 'uncorrelated asymmetries) play a major role in determining the outcome of contests (Vide supra). Maynard Smith & Parker (1976) showed that if contests are characterized by the asymmetric roles A and B, then either of the conventions: 'when in A, be prepared to escalate to cost Z; when in B, retreat without contesting' or 'when in B, be prepared to play Z; when in A, retreat' can be an ESS if Z>V, where V is the value of the resource. Examining contests in which either RHP or resource values were asymmetric, they defined a 'commonsense' ESS as one where the individual with lower RHP or with least to gain retreats, and a 'paradoxical' strategy as the reverse (the bigger animal, or the one with more to gain, is the one that retreats). In general, commonsense ESSs were much more likely to be found in nature, though the possibility of paradoxical ESSs could not be entirely eliminated.
In an extension of Maynard Smith & Parker's analysis, Hammerstein (1981) has stressed the formal possibility of paradoxical strategies. An important assumption in those models is that opponents have perfect information about the asymmetries - they always 'know' which role they occupy. The stability of conventional strategies of the above type depends critically on Z, the level of damage an animal is prepared to play to when it is challenged by an individual that breaks the convention. Parker & Rubenstein (1981) call this (Z) the 'reserve strategy'; they argue that correct deductions about reserve strategy are the key to modeling asymmetric contests in which role assessment is almost perfect. In a population in which all individuals observe a given convention and never make mistakes about roles, then the reserve strategy is never shown. If Z>V, a mutant that fails to retreat and breaks the convention always experiences a fitness less than mean population fitness, and hence cannot invade. But so, of course, do the individual he fights against. Thus recurrent mutation for convention-breaking can exert selection to reduce reserve play. Perhaps even more important, Parker & Rubenstein show that rare mistakes about roles will also exert selection on the reserve strategy. They argue that the result of such selection, in animal contests that obey the rules of the War of Attrition, is generally to adjust reserve strategy to such a level that only commonsense conventions are possible. They find that paradoxical conventions can exist only temporarily as a result of drift to a maladaptively high reserve strategy level. Rare mistakes about role, and recurrent mutation for convention-breaking, will often erode the reserve strategy by selection to a level where the paradoxical convention is invadable.

Interactions between Asymmetries: Can Assessment Evolve?
The central idea of assessment strategy in animal contests is that individuals are expected to monitor relative RHP and relative resource value. As a result it was suggested that disputes will generally be settled without escalation by a convention set by an interaction between the asymmetries in RHP and V. The interaction proposed (Parker, 1974) was simply that the individual with the lesser score for
                resource value V
            ---------------------------------------------- (1)
            rate of cost accrual if both escalate

should be prepared to retreat, but not his opponent. Parker & Rubenstein (1981) attempt to show that a convention based on (1) is not only an ESS: it approximates to the only ESS for the case where opponents can assess their roles accurately before beginning an escalated contest, and where the costs of an escalated contest would rise continuously at fixed rates. They call a strategy based on (1) an 'assessor' strategy because it defines which individual would have the expectation of being the first to enter a range where payoff from the particular interactions becomes negative if both were to continue to fight indefinitely. Parker & Rubenstein define interactions between RHP and V asymmetry as two types:
(i) Non-contradictory. The opponent with higher RHP also has most to gain (higher V) from the contest. Alternatively, either RHP or payoff may be symmetric, with opponents differing in only one respect. Non-contradictory interactions of the latter type were studied by Maynard Smith & Parker (1976), and apart from providing further evidence for the implausibility of paradoxical strategies, Parker & Rubenstein (1981) add little to their conclusions for cases where information about roles is perfect.
(ii) Contradictory. Here the opponent with higher RHP has least to gain from the contest, and vice versa. Such forms of interaction were not investigated by Maynard Smith & Parker (1976) and they generally present more problems for the development of assessor strategies than non-contradictory asymmetries (Parker & Rubenstein, 1981).
Parker & Rubenstein's extensive analyses of asymmetric conflicts suggest that where animal contests follow War of Attrition principles (victory goes to the individual that is prepared to persist longest), selection will favor abilities to acquire information about roles. Contests will obey rules set by an interaction between asymmetries in fighting ability (RHP) and resource value (V). When role assessment approaches perfection, the ESS is likely to be 'commonsense' and will approximate to the form of an optimal assessor strategy.
But just because assessment can evolve does not mean that escalated contests will not occur. Contests will be settled entirely peacefully only if reliable estimates of RHP and V can be obtained quickly and cheaply (e.g., by a short assessment phase, or display). This is the assumption of Parker & Rubenstein's first model. Because phylogenetic constraints will doubtless limit assessing abilities, animals will occasionally make mistakes about their roles, especially where 'role asymmetries' are weak. The authors define a role as being set by the interaction between asymmetries in RHP and in V. Hence a weak role asymmetry can arise either because asymmetries in both aspects (RHP and V) are weak, or because asymmetries in both aspects are strong but contradictory, so that their effects tend to be compensatory. Weak role asymmetry, undetectable by short assessment, will lead to escalation. However, if opponents do escalate into a true contest, role assessment need not have ended. The assumption of Parker & Rubenstein's second model is precisely that animals will use the contest itself to increase information.
Despite a vast literature on animal fighting, good evidence for assessment strategy is not easy to obtain. What evidence there is is often compatible with the prediction that fights tend to escalate or increase in length when opponents are closely matched (e.g. mirror experiments) or where asymmetries in RHP and V are contradictory (Vide infra). Parker & Rubenstein (1981) see assessment displays primarily as adaptations to obtain information about an opponent; this necessarily produces selection for bluff, or even concealment of cues (Parker, 1974; Maynard Smith & Parker, 1976). During a contest, an individual may itself convey information by making a probe to obtain information. In a sense, this is a form of (inadvertent) communication.
In complete contrast, there can exist contest situations in which selection may have actively favored specialized signals. For instance, Clutton-Brock & Albon (1979) argue that red deer stags monitor roaring rate as a means of RHP assessment and that this is uncheatable because roaring contests are so exhausting. Escalated contests are very costly.
Parker & Rubenstein's (1981) models for information acquired during a contest are intended to develop the proposition that role assessment may have costs. Good information about RHP may be obtainable only during some form of trial of strength or weaponry; role assessment may improve as contest costs increase. Behavior like push-pulling contests are plausible candidates for assessment tournaments, but though 'trials of strength' (or weaponry) were commonly discussed in ethological literature, quantitative evidence is not abundant. Much, as Parker & Rubenstein have shown, will depend on how much information can be gained for a given amount of risk or energy expenditure; and much will depend on the current 'tactical convention' of the population.
Hammerstein (1981) has made an extensive analysis of games with perfect information that obey hawks-doves rules and in which asymmetries relate to RHP, V, and ownership. Reserve strategy in his models is fixed by the pure hawk strategy, with injury at a given cost - D. This generates an 'ambiguous zone' in which bourgeois (or, incidentally, 'anti-bourgeois') can be an ESS against the RHP asymmetry. This gives the fundamental result that unless 'payoff-relevant aspects' (RHP and V) are relatively strongly asymmetric, they may be ignored in favor of bourgeois, even when this appears paradoxical. There is some impressive evidence that bourgeois conventions do exist in nature. For instance, Davies (1978) found that even very short tenure of a territory was enough to count as 'ownership' in speckled wood butterflies. Austad et al. (1979) have argued that tenure of the territory may here correlate with RHP.
Three studies (Kummer et al. 1974 for hamadryas baboons; Hazlett et al. 1975 for crayfish; and Reichert 1978 for spiders) appear to conform rather closely with Hammerstein's predictions in that ownership is respected unless size disparity is fairly large. This need not always be so, however. Klingel (1967) narcotized dominant male zebra; the harem was quickly taken by an interloper. But as the faculties of the narcotized male became restored, he immediately displaced the new owner without significant dispute. Perhaps for animals that meet repeatedly and can recognize each other, ownership is respected on a less transient basis than in, say, an insect like the speckled wood butterfly. However, it is difficult to see reasons why this might not also apply in baboons; and indeed, the most recent work (Kummer et al., 1978) suggests that prior knowledge of opponents does exert a significant influence on whether ownership is respected (Parker & Rubenstein, 1981; See also Maynard Smith, 1982 and references therein).

Ownership in Asymmetric Games (Maynard Smith, 1982)
Kummer, Götz & Angst (1974) studied contests between male Hamadryas baboons over females. In the wild, a male Hamadryas forms a long-lasting association with several females. It was shown that if male A was permitted to form a bond with a strange female, then a second male, B, who has watched the interaction will not subsequently challenge A for ownership. If, on a later occasion, male B forms a bond with a female, he will not subsequently be challenged by A. Escalated fights do occur between two males if each perceives himself as the owner of the same female. It seems clear that ownership, and not any perceived difference in size or strength, is decisive in settling contests. Bachmann & Kummer (1980), however, have shown that female choice can have some influence on the outcome. In an experimental situation, low- and middle-ranking males showed greater respect for an owning male if the female preferred the owner in choice tests; dominant non-owing males did not alter their behavior in response to female preferences. This makes sense, because if a female prefers a male she is more likely to stay with him, and is therefore more valuable to him. There is some evidence that female choice is relevant in the wild. Abegglen (1976) observed a troop in which male fighting had resulted in extensive redistribution of females. Several mother-daughter pairs were separated by the fighting, but were found. to be reunited months later, indicating that their preferences had influenced the course of events.
To summarize on Hamadryas baboons, there are escalated fights between adult males over females, and there is evidence that female choice can affect both the initiation and outcome of such fights, Nevertheless, most potential contests are settled by prior ownership, and need not depend on perception by the contestants of differences in fighting ability.
Davies (1978) studied territorial behavior in the speckled wood butterfly Pararge aegeria. Males defend patches of sunlight on the floor of the woodland, moving as the sun moves. Davies was able to show that it is ownership which decides contests, the 'owner, being a male which has settled in a territory, if only for a few seconds.
Similar results were obtained by Gilbert (pers. comm. in Maynard Smith, 1982) on male swallowtails, Papilio zelicaon. which hold hilltop territories. Owners always win, and there was an escalated contest if two males perceived themselves as owners (they had been permitted to own the same hilltop on alternate days). Baker (1972) found a more complex situation in the peacock, Inachis io. Males hold patches of nettles as territories; these are oviposition sites for females. Typically, contests are won by the owner after a spiral flight. Sometimes, however, a second male may settle in a territory, perhaps because the owner is away courting a female. A longer spiral flight then ensues; this is usually but not always won by the original owner, probably because the intruder cannot find its way back to the territory when the spiral flight is broken off.
If both males do return to the territory, a succession of spiral flights takes place, the eventual winner being the stronger flier, which is able to keep above and behind its opponent.
One final example of the significance of ownership in settling contests is taken from the work of Pusey and Packer on lions (pers. comm. in Maynard Smith, 1982). Groups of males cooperate to take over female prides, but once in control of a pride, males compete for estrous females. A male forms an exclusive consortship with an estrous female and prevents other males from coming into close proximity to the female. As long as the consorting male has clear ownership of the female a rival male will not seriously challenge the owner for the female. One male may be owner of the female during one estrous period but be a rival for the female during the next. A male can consort continuously with a female throughout her estrus, which lasts for several days. There is, however, competition between males to be the first to establish ownership; in particular, a male may guard a female for several days before she shows signs of receptivity.
Particular interest attaches to those situations in which two males in the same cooperating group fight over possession of females. These fights occur in two situations, which have in common that the asymmetry between owner and non-owner has broken down. The first, and more obvious, case arises when the owner wanders too far from the female, enabling an intruder to come closer to her; ownership is then unclear. The second case arises when two consort pairs come into close proximity. There is then no longer an asymmetry, and a fight may ensue; one male may try to acquire the other's female and thus may come to control two females simultaneously, but in some cases no such attempt is made, and the fight seems to result merely from the intolerance felt by an owner of the presence of a second male. This letter case affords a dramatic (because counter-intuitive) example of the importance of asymmetries in settling contests. In thinking about it, it is important to remember that the cost of a contest between male lions is high. Not only is there risk of injury in the contest itself; even an uninjured male would pay a price if its opponent was injured, because a group of males in which some are injured is less likely to be able to defend the female pride against other groups. Because the price is high, dependence on the asymmetry will be strong, and the risk of escalation on the relatively rare occasions when the asymmetry breaks down correspondingly great.
To summarize, asymmetric games of the Hawk-Dove type, with a finite set of discrete possible pure strategies, can have both common-sense and paradoxical ESS's, but only a mutant adopting the former strategy can invade a population whose members ignore the asymmetry and adopt the appropriate mixed strategy. Asymmetric games of the war of attrition type, with a continuously distributed set of possible actions, can be analyzed if one assumes that errors in role identification occur. If so, and if payoffs in the two roles are unequal, only the common-sense ESS exists; that is, the contestant to whom the value of winning is greater wins, and the other contestant gives in cheaply (Maynard Smith, 1982).

Are Assessment Strategies observed in Nature?
A number of authors have pointed out that organs used in fighting are displayed prior to fighting, and that such displays may settle contests without escalation. Geist (1966) has shown that, in Stone's sheep (Ovis dalli stonei), horn size is more variable than body size; that horns are displayed in agonistic encounters between the males; and that, when a ram enters a new band, the great majority of its interactions are with sheep of the same degree of horn development. It seems likely that assessment of horn size is used in determining, without escalation, the position of the new ram in the dominance hierarchy.
In similar vein, Packer (1977) has pointed out that fighting ability of baboons (Papio anubis) declines with wear and injury to canines, and that the canines are shown in yawning displays during agonistic encounters. Morton (1977), expanding on an earlier remark by Collias (1960), pointed out that, in both birds and mammals, low-pitched sound are usually associated with aggression and high-pitched ones with fear and appeasement. He suggests that this association has evolved because low pitch is usually associated with large size. In the toad, Bufo bufo, Davies & Halliday (1978) have shown that depth of croak is indeed used in assessment in intermale fights.
The role of roaring in the assessment of fighting ability in red deer (Cervus elaphus) has been studied by Clutton-Brock & Albon (1979) and Clutton-Brock et al. (1979). The authors suggest that roaring rate may be a better predictor of fighting ability than, for example, body or antler size, because the ability to roar declines with age after 11 years (as does fighting ability), and declines in an individual which is exhausted after holding a harem for a long time.
At the end of September or early October, hinds congregate in particular areas, where they are joined by stags which have spent the rest of the year in bachelor groups. Stags compete with one another for the possession of groups of hinds, or 'harems'. Individual stags between the ages of 7 and 11 years are most successful in holding a harem. In this age range, a stag can hold a harem for 2-4 weeks; during this period he may have to fight another stag on average once in five days. Fighting success is positively correlated with reproductive success and, in turn, with antler size. There are large differences, within and between age classes, in the success of stags in holding harems. Stags lose up to 20% of their body weight during the rut. Fighting is potentially dangerous: 6% of stags were injured per year, indicating a chance of about 25% of serious injury per lifetime. Fighting is also costly for a harem holder because, during a protracted fight, his harem will be dispersed by younger stags.
As a seriously-wounded stag has little chance of survival, it is apparent that the optimum strategy is to avoid contests with superior rivals or those in which there is a high risk of injury. It is therefore to be expected that assessment will be used to settle contests. There are good grounds for thinking that roaring contests and parallel walks are concerned with the assessment of fighting ability.
During a roaring contest, stags usually roar in alternating bouts, and direct their roars towards their opponent. Clutton-Brock et al. found that stags roar most in conditions in which they are likely to be challenged and that the pitch of roaring is related to the size of the animal, larger stags having lower-pitched roars. For ten mature harem-holding stags, estimates were made of fighting ability (based on success in actual fights against opponents whose success was also known) and roaring rate (average number of roars per minute during a contest). There was a significant correlation of +0.80 between these measures. Fights were most frequent between stags which were approximately equal in roaring ability. The authors suggest that roaring rate may be a better predictor of fighting ability than, for example, body or antler size, because the ability to roar declines with age after 11 years (as does fighting ability), and declines in an individual which is exhausted after holding a harem for a long time.
Roaring appears to function for the two stags as a means of assessment. If the protagonist remains after the roaring contest, the contestants generally commence parallel walking during which two stags walk up and down at a distance of a few meters from one another. Such walks are most frequent between equally matched opponents. The fact that there is a strong correlation between the duration of parallel walking (a test of stamina) and that of subsequent fighting again indicates a trial of strength.
This suggests that if there is a substantial difference in fighting ability it will be detected by a long parallel walk, but that if a long walk fails to reveal such a difference the ensuing fight will be a long one, because the contestants are equally matched (Maynard Smith, 1982).

Figs. 10-12: Escalation of assessment of fighting ability in red deer.

Should the other male continue to challenge, however, the antlers are used and a fight develops. Fighting stags lower the head and ram the opponent with their antlers. As the rival faces his attacker, their antlers lock together and a pushing match ensues. Each stag attempts to get uphill from his opponent and to push him backwards. The loser finally withdraws and takes to flight. Pursuit is rare because attacking involves lowering the head which reduces the male's speed. If, however, the loser slips or falls, the winner will take the opportunity to jab the flank of its fallen rival, often wounding it severely.
Clutton-Brock et al. were able to show that roaring, parallel walking, and antler-pushing are three separate - ritualized - stages of escalation in a process of mutual assessment of fighting potential. This method of assessment (of size discrepancy, vigor of display, stamina and strength) is probably relatively immune to cheating.
These data support the view that roaring and parallel walks provide a means whereby rival stags can assess one another's fighting potential. Stags seldom challenge older, larger individuals, so that roaring contests rarely occur if there is an obvious visible discrepancy in size; very powerful stags of approximately equal strength, however, generally do not fight after long parallel walks, thus avoiding serious injury.
In general, when smaller males withdraw upon perceiving an honest signal from a larger male, both parties gain; smaller males do not waste time and energy in a battle they are unlikely to win, and larger males save time and energy that would otherwise have to spend struggling with annoying smaller competitors (e.g., Alcock, 1998).
“To conclude, assessment of resource-holding power (RHP) is taking place in animal contests. In the examples discussed, the signal is correlated with fighting ability, and does affect the behavior of animals receiving it. Horn size in sheep, canines in baboons, roaring rate in red deer, and depth of croak in toads are all indicators of RHP which would be expensive or impossible to fake” (Maynard Smith, 1982).
An analysis of games in which more than one asymmetry exists has been undertaken by Hammerstein (1981). If two asymmetries exist, it might seem that the asymmetry with the larger payoff difference would be used in conventional settlement. This, however, turns out not to be the case. Hammerstein shows that it is possible for an aspect which does not affect payoffs to settle contests, even if a second aspect may have such a strong effect on payoffs also exists. In other cases, one aspect may have such a strong effect on payoffs that it is necessarily used to decide contests.
One of the most extensive studies of contests in the field concerns male fiddler crabs (Uca pugilator) over burrows (Hyatt & Salmon, 1978). In 403 contests observed, the owner won in 349, and the intruder in 54, cases. However, in the latter cases, the intruder was larger in 50 contests and smaller only in one. Clearly, differences both in size and ownership are relevant to the outcome. Typically, ownership is taken as the arbiter, but a sufficiently large size difference can override this. It is difficult to analyze this further because little is known of the payoffs involved. Males have claws powerful enough to crush an opponent. Hyatt & Salmon, however, observed no injuries resulting from fights, so the main costs are in time and energy. In contrast, Jones (1980) reports that in a related species, Uca burgersi, 25 % of males had damage to their major chelae of the kind to be expected if it had occurred during fights. It is also hard to measure the value of the burrows (which are mating stations) over which the fights take place. Nevertheless, the basic conclusion, that both size differences and ownership influence outcomes, is well established and will probably prove to be typical (Maynard Smith, 1982).
The best illustration of the way in which asymmetries of size and ownership, and variations in payoffs, can influence contest behavior is Riechert's (1978, 1979, 1981) study of the funnel-web spider Agelenopsis aperta, and Sigurjonsdottir & Parker's (1981) analysis of dung fly struggles (for details see Maynard Smith,1982).

Dynamics of the Evolution of Animal Conflicts
Zeeman (1981) introduced a dynamic into the game-theoretical models of agonistic behavior. The dynamic is based on the hypothesis that the growth rate of those playing each strategy is proportional to the advantage of that strategy, and turns out to be a system of cubic differential equations on the population space, which is an n-simplex, where n + 1 is the number of strategies.
Solving the differential equations gives the evolutionary flow on the simplex, in other words the paths along which the population will evolve from any given initial distribution of strategies. In particular, if there is an ESS then this will be an 'attractor' of the flow. However, the converse is not true because not every attractor is an ESS.
Zeeman illustrates this notion by reanalyzing the original HDBR-game of Maynard Smith & Price (1973 et seq.). Suppose that individuals in a population are competing for something that affects their reproductive success. Suppose that during each contest an individual can either display, or escalate the fight, or run away. Suppose that each individual pursues one of the four strategies:

   Strategy   Initial Tactic    Tactic if opponent escalates
H hawk escalate escalate further
D dove display run away
B bully escalate run away
R retaliator display escalate

In order to compute the payoff matrix the following scores are introduced: Win 6; Lose 0; Injury -10; Waste time -1. Quantitatively the numbers themselves are not important, because changing them does not alter the qualitative nature of the game, provided we do not alter their signs or alter the order of their magnitudes.
The interpretation is as follows. If a hawk meets a dove then the hawk wins the contest with payoff 6, and the dove loses with payoff 0, implying that the hawk is more likely to produce offspring. If two hawks meet then they both escalate the fight until one is injured, which carries a heavy penalty since it may also adversely affect subsequent contests; therefore there is a 50 % chance of winning and a 50 % chance of getting injured, and so the payoff = expected gain = 1/2(6 - 10) = -2. If two doves meet then they both waste a lot of time displaying, which incurs a small penalty because the time could be more fruitfully employed mating or feeding, etc. There is a 50 % chance of winning, and so the payoff = (1/2 x 6) -1 = 2. A retaliator behaves like a hawk towards a hawk, and like a dove towards a dove or another retaliator, and so scores accordingly. A bully wins against a dove, but loses against both hawks and retaliators. If two bullies meet the first one to escalate causes the other one to run away, with no waste of time, so the payoff 1/2 x 6 = 3. Therefore Zeeman has calculated the payoff matrix A (aij), where aij denotes the payoff to strategy i played against strategy j.

    H    D    B     R  
H    -2    6    6    -2  
D    0    2    0    2  
B    0    6    3    0  
R    -2    2    6    2  

It is a non-zero sum game because A is not skew symmetric. No single strategy is immediately recognizable as the best, because no single row has all its terms greater than those of all the other rows.
Before he introduces the dynamic, Zeeman illustrates the idea by a graphical treatment of a simple example. Suppose the population consisted of only hawks and doves; Zeeman calls this the HD-subgame. Describe the population by x - (x0, x1) where x0 is the proportion of hawks, x1 the proportion of doves, and x0 + x1 = 1. Assuming the population is large, then the probability of meeting a hawk is x0, and of meeting a dove x1. Therefore if an individual always plays hawk his payoff E0, or expected gain, is given by E0 -2x0 + 6x1. Similarly if he always plays dove his payoff is E1 0x0 + 2x1. The set of populations is represented by the 1-simplex joining the points H = (1,0) and D = (0,1). The graphs of the two payoffs are shown in Fig. 13

Fig. 13: Graphs of pay-off in the hawk-dove subgame (from Caryl, 1981).

Let P denote the population where the two graphs cross; then P is given by E0 = E1. Therefore, -2x0 = 6x1 = 2x1. Therefore P = (2/3, 1/3). If x is to the left of P then E1>E0, so the dove strategy pays off better. Therefore the doves will tend to have more offspring per head than the hawks, and so the proportion of doves in the population will increase, and x will move to the right. Conversely is x is to the right of P, then the proportion of hawks will increase, and x will move to the left. Therefore there is a flow on the 1-simplex, towards P. Hence P is an attractor, and the population will reach a stable equilibrium there.
The reason that the two-strategy game is easy to solve is that the population space is only one-dimensional, and so the flow has to flow along it, one way or the other. If there are three or more strategies then the population space is two or more dimensional, and so to obtain the flow lines it is necessary to solve a differential equation.
It should be noticed that Zeeman's main hypothesis depends implicitly upon three assumptions: (1) Each individual plays a fixed pure strategy. (2) Individual breed true, in the sense that offspring play the same strategy. (3) Payoff is related to fitness, in the sense that the more payoff means the more offspring. In a later section of the analysis assumption (1) is generalized to allow individuals to play mixed strategies. The main results of Zeeman's analysis are as follows. The retaliator strategy is a weak attractor, but this is only a transient property because the game is structurally unstable. When the game is stabilized the retaliator becomes an evolutionary stable strategy. At the same time another ESS appears comprising a mixture of hawks and bullies, and if individuals are allowed to play mixed strategies then this tends to produce a pecking order. Thus the stabilized game offers an explanation for the evolution of hierarchical societies in terms of natural selection acting on individuals.

Honesty versus Cheating in Information Transfer
Van Rhijn & Vodegel (1980) studied the possible consequences of individual recognition for the settlement of conflicts by means of simulation. They considered four strategies: (1) Retaliator (based on Maynard Smith's models and used as a control condition), (2) Threat-right (threatens towards a submissive and will follow with attack if the submissive does not retreat; retreats from a threatening or attacking dominant), (3) Attack-right (as 2 but without threatening), and (4) Threat-dominance (as 2 but with a low probability of threatening and attacking a dominant). Their main conclusions:
If the knowledge about strength or dominance of the other individuals is perfect, the Threat-right strategy (thus a warning before a real attack) turns out to be the most successful under a wide variety of conditions. If that knowledge is not perfect (during the learning phase), other strategies can yield better results.
One of the predictions of the War of Attrition model (Maynard Smith, 1974) was that information about the probability to attack (or information about intentions) should not be conveyed. This model refers to asymmetric conflicts so that its predictions may be relevant for conflicts between individuals which are unknown to each other. Maynard Smith & Parker (1976) also considered the possibility of bluff in asymmetric conflicts. They reasoned that “if contests are settled by asymmetric cues, the evolution of features which exaggerate apparent size (or whatever feature is used as a cue)” may be important. Van Rhijn & Vodegel (1980), however, stress that if individual recognition plays a role in the evaluation of that asymmetric cue, bluff (about intentions and RHP) can hardly evolve, because bluffers shall mostly be recognized.
Rose (1978) analyzed the Scotch Auction evolutionary game. The Scotch Auction game is effectively a sealed bid auction. Organisms contest a single indivisible payoff in pairs, each 'playing a bid' with a cost directly proportional to the size of the bid. The player making the largest bid wins the payoff item, paying the full cost of its bid, irrespective of the size of the opponent's bid. The loser pays its full bid too. Rose showed that the Scotch Auction is a pathological game, in the sense that there is no ESS and no asymptotically stable polymorphism among discrete or continuous probability rule-obeying strategies. Furthermore he showed that cheating mutants can take over populations of Scotch Auction rule-obeying strategists, and that such cheating strategists will become more efficient, wholly undermining the Scotch Auction's sealed-bid rule. The controversy is, by now, clear: should or should not an organism convey information about its intentions or RHP. Should or shouldn't it cheat as much as possible?

In considering the evolution of information transfer, there is a basic distinction between two types of information; (1) Information about RHP; i.e., about size, weapons, etc. which might influence the outcome of an escalated fight. (2) Information about motivation; i.e., about what an animal will do next. The reason why these two must be distinguished is as follows. A genetic change causing an animal to behave differently in a given situation (i.e., a change in its motivational system) can occur with little selective cost to the animal, except in so far as the change in behavior itself has selective consequences (e.g. escalation may lead to injury). A genetic change increasing an animal's RHP will be costly outside the contest situation. No difficulty arises in accounting for the transmission of information about RHP; this is what happens in assessment. The difficulties arise in accounting for information about motivation, essentially because there is nothing to prevent animals 'lying' about what they will do; more formally, it is hard to see how selection could maintain a consistent relationship between signal and subsequent action (Maynard Smith, 1982). The same basic point has been made by Caryl (1979) and Zahavi (1981).
An extensive literature exists to show that, in some sense, information is transferred during animal contests (e.g. Stokes, 1962a,b, Dunham, 1966, Andersson, 1980, on birds; Hazlett, 1966, 1972, Hazlett & Bossert, 1965, 1965, Dingle, 1969, on crustaceans; Simpson, 1968, Dow, Ewing & Sutherland, 1976. Jakobsson, Radesäter & Järvi, 1979, on fish; Rand & Rand, 1976, on lizards; Reichert, 1978, on spiders).
The data on information transfer during contests (reviewed by Caryl, 1979; and Maynard Smith, 1979; 1982) tend to show the following:
(1) It is common for an animal to use a range of actions during a contest; these actions can plausibly be arranged on a scale of increasing aggressiveness.
(2) Information is present in these acts, in the sense that there is a correlation between the act now performed by one individual, and the next act by the same individual.
(3) Information is received, in the sense that there is a correlation between the acts now performed by one individual, and the next act performed by its opponent.
(4) A common pattern is for the contest to start with acts at a low level on the scale of aggression, and gradually escalate, as each animal matches any increase in aggression by its opponent. Nevertheless, it is difficult or impossible to predict from the acts performed during a contest which animal will be the ultimate winner.
These statements refer to ritualized display movements, and not to unritualized movements. Caryl (1979) and Maynard Smith (1979, 1982) have both argued, from point (4), that the 'function' of the various displays is not to convey information about how long, or to what level, an animal is prepared to persist. To that extent, the evidence agrees with the theoretical prediction. There remains the problem of why there is often a variety of displays and why some information is encoded in them. Some displays are concerned with indicating RHP, but that is not the whole explanation. Maynard Smith (1982) presents two suggestions which might contribute towards a solution: (1) If the resource being contested is divisible - e.g., space for territories - 'bargaining' becomes a possibility, and bargaining requires variable signals (Maynard Smith, 1979; 1982). (2) If repeated contests take place between the same two opponents, new possibilities arise for honest communication (Van Rhijn & Vodegel, 1980).

An important question is how far the acts performed by individuals carry information about long-term intentions, and in particular about which will be the ultimate winner. Caryl (1979) has reanalyzed the data on tits (Stokes, 1962a,b), grosbeaks (Dunham, 1966) and skuas (Andersson, 1980) with these questions in mind.
His conclusions are as follows:
(1) Displays are poor predictors of physical attack. No one display is followed by attack with high probability. A particular display may be correlated with attack at one time of year but not another. In 2 out of the 3 species, the most aggressive display by one bird did not correlate with retreat by the other.
(2) Some displays are good predictors of immediate retreat.
(3) Two different displays may have the same effect on the receiver but predict different things about the actor.
The existence of an 'I surrender' signal is easy to understand; similar signals are found in other groups (e.g., Dow, Ewing & Sutherland, 1976, for the fish Aphysemion striatum). Apart from this, there is little reason to think that the displays are conveying information about motivation, or about the level to which a bird will escalate. A rather similar conclusion emerges from the studies of fish: Simpson (1968), Betta splendens; Dow, Ewing & Sutherland (1976), Aphysemion striatum; Jakobsson, Radesäter & Järvi (1979), Nannacara anomala.
Maynard Smith (1982) concludes that the apparent examples of 'announcement' in animals may depend either on some special circumstance which makes lying unprofitable, or on a physiological constraint making it impossible. Andersson (1980) has offered the following explanation of why many species have a variety of different threat displays:

The Variety of Threat Displays
In his review of the behavior of gulls, Tinbergen (1959) points out that there is a remarkable diversity of threat displays in the signal repertoire of each species, and he adds “it is of course necessary to ask why this should be so”. The high number of threat postures, movements, and calls contrasts with the low number of displays used in non-agonistic encounters, for example between parent and offspring, and between sexually motivated mates. In such situations, one or a few displays seem to suffice. Tinbergen suggested that each of the different threat displays are adapted to one of a limited number of situations. He referred to the threat display of gulls as 'distance-increasing' signals, which are used to drive away competitors. He suggested that among the threat displays some are more offensive, indicating a high probability of a subsequent attack, whereas others are less likely to be followed by attack. He also hypothesized that different kinds of opponents are best treated by different kinds of threat displays: accidental or casual territory intruders by mild threat, and would-be settlers by the strong language of offensive threat.
There seems to be at least one problem with this explanation, as Andersson (1980) points out: what survival value is there in performing the less offensive threat display? Obviously, if the task is to get rid of a casual intruder, the offensive threat should be at least as efficient as the milder threat. Then why not always use the stronger threat signal? Tinbergen did not explain this. Perhaps his hypothesis is correct, but al least it seems to need some supplementary arguments.
The second hypothesis suggests that different threat signals are adapted for communication with different categories of receivers, in particular with receivers at different distances from the sender. Tinbergen suggested that the 'long call' of gulls is adapted for long-distance threat. Because of its good carrying power, the long call would seem to be the gull display best suited for function at a distance, and there is much evidence that it is indeed used particularly against distant opponents. Not only calls, but also postures may be adapted for dealing with intruders at different distances.
As Andersson (1980) points out, it should be important to the sender that the correct, intended receiver of a signal does get the message, and this might sometimes explain part of the diversity of threat signals in a repertoire, particularly in gregarious species.
In addition to the two previous functional hypotheses, the 'conflict theory of displays' (e.g., Tinbergen, 1952; Baerends, 1975) is often mentioned in connection with display repertoires. Cullen (1972) emphasized that it offers an explanation for why there are more than one display in certain situations. The different displays may reflect a difference in the relative strength of tendencies to attack or retreat or, perhaps better, to move towards or away from an opponent. Associations between displays and subsequent actions in well-defined situations seem to support the idea that at least some displays, which are lumped together under the 'agonistic' label, do in fact represent different messages about subsequent actions. If it is advantageous for senders to inform receivers about such differences in motivation, this might explain much of the diversity of threat displays. However, the conflict theory primarily aims to answer questions about the phylogeny and causation of different displays, not about the functional significance of the diversity of threat signals, which is here at issue (Andersson, 1980).
The previous hypotheses were based on, or compatible with, the traditional view that displays serve to transmit accurate information about the sender's probable subsequent behavior (e.g., W.J. Smith, 1977). A different approach was taken by Maynard Smith & Price (1973) in their game-theoretical analysis of animal conflicts. The analysis suggested that bluff might be an important element in animal communication ('bluff' can be said to occur for example when a behavior pattern, which has been used in communication as a reliable predictor of a certain action, begins to be used without that action necessarily following). Several important consequences of this possibility have been stressed by Maynard Smith (1979, 1982), Maynard Smith & Parker (1976), Parker (1974), Zahavi (1977), Dawkins & Krebs (1978) and Caryl (1979). In a slightly different context similar ideas were expressed by B. Wallace (1973) and Otte (1974). Barnard & Burk (1979) suggested that bluff or 'cheating' of status cues might be important in dominance hierarchies (Vide infra).
Possibly, bluff has also been important for the evolution of a diversity of threat signals in a species' repertoire. Ethologists seem to agree that many threat displays, perhaps most of them, evolved from intention movements for attack, or from other patterns occurring prior to fighting (e.g., Tinbergen, 1952; W.J. Smith, 1977).
Imagine what might happen to an unritualized attack pattern when it gains use as a threat signal. Originally it forms part of actual attack. Therefore in its early evolution as a display, the pattern is used by few individuals as a signal, de-coupled from attack. On average the pattern therefore is a fairly reliable predictor of attack, which will usually occur together with it. In this early stage, the pattern should therefore be an efficient display, which convinces receivers , that the sender is very likely to attack. Opponents should take this message and quit, unless they have good reasons to do otherwise.
However, because the pattern is now available for use separately from attack, victory in many contests will go to cheaters, which bluff and display the pattern even when not particularly likely to attack. Cheaters, because they can win contests by bluff even when their motivation to attack is low, will be at an advantage and spread to the population. Consequently, the pattern will occur more frequently without attack following. The display then becomes a less reliable predictor of attack. This in turn selects for a reduced tendency to be impressed by the display, which therefore becomes less efficient in chasing away opponents. However, the pattern never becomes total bluff, because it will usually occur each time there is a real attack.
Suppose now there exists a second intention movement or other pattern, consistently associated with attack. If that pattern can also become decoupled from attack, which it predicts more reliably than the first, 'bluff-infested' pattern, then the previous chain of arguments can be applied to this second pattern as well. Individuals which use it as a threat signal should be favored over those using the older one, until bluff has also permeated the second pattern and reduced its reliability as a predictor of attack. Unless other kinds of selective pressures operate on the displays, both should finally have similar efficiency in deterring opponents.
In essence, there should be competition among threat displays, and a state of balance in their use might arise due to frequency-dependent selection, where each display becomes less efficient for chasing away opponents as the display becomes more frequently used as bluff. In this respect, threat displays differ from many other signals, for example synchronizing precopulatory displays, and calls used by parents to attract young to food, when there is usually nothing to gain by using the signal as 'bluff'. Such signals therefore will remain reliable, and there is no advantage in developing several alternative signals for the same message, contrary to the case with threat signals (Andersson, 1980).
Rohwer (1982) argued that, when a population may be characterized by interference competition for resources, variation in fighting ability among individuals, and repeated confrontations between individuals, together with difficulty of individual recognition, badges of status should invade as recognition marks that render good fighters memorable. Reliability of such badges can be maintained by negative frequency-dependent selection when individuals of different appearance (and status) either play mutually beneficial roles or employ alternate competitive tactics. In territorial social systems intraspecific mimicry of recognition badges should evolve, because, in contrast to group-living situations, the cost to a cheat of being discovered is low when individuals are dispersed. The general result of such mimicry is that good and poor fighters become similar in appearance.

Sexual Selection and the Advantage of Seeming Fitter
M.B. Williams (1978) argues that there is a specifiable type of fitness indicating trait which is particularly likely to be exaggerated because of fitness-faking modifier genes. The most firmly founded explanation for instances of sexual selection (Mayr, 1963) suggests that if the female has to choose between phenotypically similar males of different species, then natural selection should fix a female preference for traits which best distinguish males of her own species from other males; sexual selection may then lead to the fixation, and possibly the exaggeration, of these traits.
Another explanation (Trivers, 1972) points out that if a male courts more vigorously when his sperm level is high, then natural selection should fix a female preference for more vigorous courtship; subsequently, if secondary structures used in display heighten the appearance of vigorousness, sexual selection may exaggerate these structures.
A third explanation (Selander, 1972) points out that if position in a dominance hierarchy is determined by both age and physiological condition, then natural selection should fix a female preference for males of higher rank, since such females mate preferentially with males all aspects of whose fitness have been 'demonstrated'; this would lead to the accentuation of traits, e.g., aggressiveness, which raise a male's position in the hierarchy.
The above three suggestions are about intersexual selection; similar considerations apply to intrasexual selection: e.g., if fighting between males leads to permanent damage of the weaker male and males gain strength with age, then natural selection could fix in males a deference to stronger males; subsequently if secondary structures used in agonistic display heighten the appearance of strength, intrasexual selection may exaggerate these structures. All of these explanations consider the traits selected on to be fitness indicating traits whose disadvantageous properties are merely side effects, more than counterbalanced by the advantage of choosing better fathers. Williams proposed that sexual selection should, by favoring age-faking and sperm count faking modifiers, cause an escalation in the size, intensity, precocity, etc., of those traits.

Inconsistent Cues and Skeptical Recipients
Moynihan (1982) has argued that in agonistic interactions signals of intention should be perceived readily but analyzed skeptically by the recipient. Caryl (1982) points out that this hypothesis is consistent with predictions from games theory, rather than incompatible with these predictions as Moynihan believes. Applied to cues about asymmetries such as those in RHP, rather than to signals of intention, the hypothesis of the 'skeptical recipient' appears to make sense of some striking observations in the literature. In asymmetric contests, transfer of information is to the advantage of both opponents. Games theorists have shown that animals would be expected to transfer information about differences in strength, etc, which affected the probability of winning a contest (e.g. Maynard Smith, 1979; Parker & Rubenstein, 1981), while at an empirical level, these differences are known to influence the results of contests in many species. Hence animals are expected to evolve signals which make conspicuous such differences in RHP, and recipients are expected to utilize whatever cues to RHP they can extract. In some species, several cues carry information about RHP, and the recipient would be expected to be sensitive to all of these. But how should the recipient respond if the different cues provide information which is inconsistent?
Caryl (1982) suggests that analogy with the model discussed by Maynard Smith (1974) leads to the following rule of thumb: “An opponent may try to cheat by exaggerating cues, but it is unlikely to gain any advantage by underemphasizing cues to RHP. Hence, where cues to RHP are inconsistent, it is likely to be better to believe the least impressive information that the opponent provides”.
Some of the most striking observations on the response to inconsistent cues of RHP have been made by Rohwer (1977) and Rohwer & Rohwer (1978). In Harris sparrows (Zonatrichia quereula) variations in dominance are marked by the variation in pigmentation of the feathers of the throat, breast and crown. Dominant birds have dark feathers, while subordinate birds have light feathers. Rohwer found that subordinates whose throats and crowns were experimentally darkened to look like dominants (producing dominants who were 'cheats') were persecuted by the legitimate dominants. Apparently their behavior lead them to be attacked, because when it was modified by testosterone injections given in conjunction with the experimental change in plumage, the darkened birds were not attacked. In this context, where plumage and behavioral cues provided inconsistent information the birds responded to the behavioral cues. Rohwer found that when he bleached the plumage of legitimately dominant birds, their conspecifics now responded to the morphological rather than to the behavioral cues. The bleached birds could not get their opponent to retreat by threat alone, but only after a real fight, which was sometimes very lengthy (The results of this fight confirmed that their natural dominant markings were associated with real fighting ability.)
Rohwer noted that the morphological and behavioral cues were taken at their face value when they were consistent, but not when there was incongruity between them. In its original form, his 'incongruence hypothesis' does not explain why behavioral cues should be disbelieved in one context and morphological cues in another. But if animals are using the rule of thumb 'When in doubt, pay attention to the least impressive cue - that way you will not get cheated', it becomes clear why we should expect the results that Rohwer obtained. The new hypothesis proposed by Caryl (1982) also makes it unnecessary to postulate a tendency to 'punish cheats' (a tendency which is difficult to account for in terms of advantage to the individual, although it would clearly be good for the species; Maynard Smith, 1979).
The data on information transfer during contests (reviewed by Caryl, 1979; Maynard Smith. 1979) tend to show the following:
(i) A species often has an array of signals, which can plausibly be arranged in an ascending scale of aggressiveness.
(ii) An act contains information about the future behavior of the actor, since it is correlated with his next act.
(iii) Some information is transmitted, since the acts of the receiver are influenced by the actor.
(iv) Nevertheless, it is difficult or impossible to predict from the acts performed during a contest which animal will be the ultimate winner. These statements refer to ritualized display movements, and not to unritualized movements. Caryl (1979) and Maynard Smith (1979, 1982) have both argued, from point (iv), that the 'function' of the various displays is not to convey information about how long, or to what level, an animal is prepared to persist, To that extent, the evidence agrees with the theoretical prediction. There remains the problem of why there is often a variety of displays and why some information is encoded in them. Some displays are concerned with indicating RHP, but that is not the whole explanation.
Maynard Smith (1982) presents two suggestions which may contribute towards a solution:
(i) if the resource being contested is divisible - e.g., space for territories - 'bargaining' becomes a possibility, and bargaining requires variable signals (Maynard Smith, 1979; 1982);
(ii) if repeated contests take place between the same two opponents, new possibilities arise for honest communication (Van Rhijn & Vodegel, 1980).

Hinde (1981) has argued that the 'meaning' of a given display depends both on the situation and on the context of other displays in which it is given (e.g., Tinbergen, 1959; W.J. Smith, 1965; Beer, 1975, 1976; Hayward, Gillett & Stout, 1977; Amlaner & Stout, 1978). But even though the sequelae of a given display are not constant, it may still serve as a signal of probable subsequent behavior. Such an interactional hypothesis, that the behavior which follows a threat display is determined in part by the behavior of the reactor, is compatible with data on a wide range of species, from fiddler crabs (Hyatt & Salmon, 1979) to mammals (Lehman & Adams, 1977). Hinde (1981) reviews the evidence and concludes that comparative data suggest that the small evolutionary steps leading to ritualization could be described as involving 'manipulation'; that studies of the behavior following threat displays show that they provide approximate but not accurate indications of what the actor will do next; and that an interactional approach suggests that this may be because the actor is genuinely undecided.
However, the games theory approach does prompt further questions. Would an individual do better to be more or less vague about his current state? Would he do better to accentuate or play down his tendency to attack or to flee? In each case, of course, the responsiveness of the reactor is assumed (temporarily) constant: if actors changed their signaling thresholds, and did better thereby, reactors would no doubt in due course evolve changed responsiveness (partially) to compensate (e.g. Dawkins & Krebs, 1978). The issue is whether the actors would temporarily be at an advantage.
A clue here lies in the fact that the social releasers used in threat elicit both attack and flight from other individuals. Thus exaggeration of the probability of subsequent attack or minimization of escape might augment the probability or viciousness of an attack on the displaying organism; while minimization of attack or exaggeration of the probability of escape might permit attack from a rival who would otherwise be deterred (See also W.J. Smith, 1977). And if there are situations in which it is better to signal that you might attack (e.g., if the opponent might thereby be provoked to leave), or that you might leave (if you thereby prevent a damaging attack), and if it is also often better to conceal your precise state, then evolution towards optimal ambiguity seems inevitable (Hinde, 1981; Cf. Maynard Smith, 1979).

Threat Displays
The complexities of display behavior caught the attention of ethologists early on (e.g., Tinbergen, 1939, 1948). It was suggested that many displays (though not all - see e.g., Marler 1959; Andrew, 1961, 1972) depend on conflicting tendencies (defined as in Hinde (1970)) to behave in incompatible ways (e.g., Tinbergen 1951, 1952). In particular it was suggested that many threat displays depend in part on incompatible tendencies to attack or approach and to move away from or flee from the rival.
The evidence for this view, which has recently been reviewed by Baerends (1975; see also Dabelstein, 1978) comes from a number of different sources, including the situation in which the display occurs, the behavior which accompanies it, the behavior which precedes and succeeds the display, and the nature of the movements used in the display. In addition there is some limited experimental evidence involving independent manipulation of the supposed 'tendencies' (e.g., Blurton Jones, 1968). It is agreed that the conflict may involve a number of different levels of integration (Andrew, 1956, 1972; Blurton Jones, 1968).
Several authors have argued that tendencies in addition to attacking (or approaching) and fleeing (or moving away) must be involved - for instance to 'stay put' or 'feed' (Baerends & van der Cingel 1962; Stokes 1962a; Blurton Jones, 1968; Baerends, 1975), and there is indeed evidence that the frequency of displays may change with hunger - a phenomenon which could be understood on this basis (Williams, Kirkawa & Morris, 1972).
The evidence that threat movements are given when conflicting tendencies are present comes in part from the fact that they may be followed by more than one type of behavior. Qualitative data showed early on that a given threat display might be followed by either attack or escape (e.g., Tinbergen, 1942; Hinds 1952), and quantitative data confirmed it (e.g., Moynihan, 1955). While the frequencies with which different types of behavior (attack, escape, etc.) follow a given display have been used as evidence for the general nature of the conflicting tendencies underlying a display, the postulation of conflict depends an the fact that threat displays do not forecast precisely what a bird will do next. This
does not mean that displays carry no information about what the actor is likely to do next (Hinde 1981).
Species usually have several threat postures, or postures which vary in some degree, and the variants differ in their sequelae. Some have a higher probability of being followed by attack or escape than others (Moynihan, 1955). But predictions from displays are always probabilistic: on the basis of studies of blue tit (Parus caeruleus) threat displays, Stokes (1962a) wrote “It seems therefore that we should use the occurrence of postures only as general indicators of the underlying tendencies to act. What a bird actually does following a specific posture appears to depend upon a combination of internal and external stimuli which are only partially reflected in the bird's postures”.
On the basis of these data (Linde, 1972b, 1975) suggested that displays were given when the bird was uncertain what to do, and that what it did next depended on the behavior of the rival. It seemed reasonable to suppose that an animal that was definitely going to attack or to flee would do best to do so immediately, without giving warning by displaying first. On this view, threat displays would be useful only in moments of 'indecision': if what one individual would do depended in part on the probable behavior of the other, threatening by the former might elicit a response from the letter which would precipitate a decision by the initial actor (Hinde, 1981).
Although the 'conflict hypothesis' has recently been used most in discussions of the immediate causation of displays, it was developed in part to explain their phylogeny. Comparative evidence suggests that most threat displays evolved from intention movements of displacement activities. It further suggests that these display movements, and the structures on which they depend, have been elaborated in evolution for a signal function - that is, they have become more effective in eliciting responses from other individuals (Tinbergen, 1948, 1952). While such an evolutionary course must be constrained by the mechanisms determining responsiveness in the reactors (Tinbergen, 1964). it is implicit in Tinbergen's writing that natural selection acts as the actor to make the signal more effective. He writes of displacement activities as having “the function of releasing responses in other individuals” and social releasers as being “adapted to the function of sending out stimuli to which other individuals react” usually innately (Tinbergen, 1952). Many of the evolutionary changes can be understood in terms of natural selection acting on the actor, making his signal more effective by exploiting the reactor's responsiveness to supernormal stimuli (e.g., Tinberqen, 1951).
It has been further emphasized that these processes of 'ritualization' often involve some degree of 'emancipation' of the movement from its motivational basis. This may involve the development of 'typical intensity', the movement has the same form over a wide range of motivational states (Morris, 1957). It has been suggested that this makes the movement more readily recognizable by another individual, and thus that it thereby becomes more effective in eliciting a response. In other cases the signal movement seems to have become largely (Weidmann, 1958), though probably never completely (Baerends, 1975) divorced from the motivational factors that, earlier in evolution, controlled it (Hinde,1981).
Finally, there in increasing evidence that the 'meaning' of a given display depends both on the situation and an the context of other displays in which it is given (e.g., Tinbergen, 1959; W.J. Smith, 1965; Beer, 1975, 1976; Hayward, Gillett & Stout, 1977; Amlaner & Stout, 1978). But even though the sequelae of a given display are not constant, it may still serve as a signal of probable subsequent behavior (Hinde, 1981).
Caryl (1979) rejects an interactional hypothesis of the type proposed here in part because it can not easily be tested; however this is only a matter of practical difficulty which could be solved (Hinde, 1981). While it would be unwise to generalize the discussion too far, the view that the behavior which follows a threat display is determined in part by the behavior of the reactor is at least compatible with data on a wide range of species, from fiddler crabs (Hyatt & Selman, 1979) to mammals (Lehman & Adams, 1977).
We thus see that comparative data suggest that the small evolutionary steps leading to ritualization could be described as involving 'manipulation'; that studies of the behavior following threat displays show that they provide approximate but not accurate indications of what the actor will do next; and that an interactional approach suggests that this may be because the actor in genuinely undecided. However the games-theory approach does prompt further questions. Would an individual do better to be more or less vague about his current state? Would he do better to accentuate or play down his tendency to attack or to flee? In each case, of course, the responsiveness of the reactor is assumed (temporarily: constant: if actors changed their signalling thresholds, and did better thereby, reactors would no doubt in due course evolve changed responsiveness (partially) to compensate (e.g., Dawkins & Krebs, 1978). The issue is whether the actors would temporarily be at an advantage (Hinde, 1981).
A clue here lies in the fact that the social releasers used in threat elicit both attack and flight from other individuals. Lack (1939) showed that a robin's (Erithacus rubecula) red breast feathers elicit attack from a territory owner, and Tinbergen (1948) demonstrated the importance of the color markings displayed by a stickleback (Gastarosteus aculeatus) in threat by assessing how often models were attacked (see also e.g., Stout & Brass, 1969), if the amount of black in the plumage of Harris sparrows (Zonotrichia querulea) is artificially increased, they are attacked more, even though the darker birds are usually dominant (Rohwer, 1977). Thus exaggeration of the probability of subsequent attack or minimization of escape might augment the probability or viciousness of an attack on the displaying bird; while minimization of attack or exaggeration of the probability of escape might permit attack from a rival who would otherwise be deterred (see also WJ. Smith, 1977). And if there are situations in which it is better to signal that you might attack (e.g., if the opponent might thereby be provoked to leave), or that you might leave (if you thereby prevent a damaging attack), and if it in also often better to conceal your precise state, then evolution towards optimal ambiguity seems inevitable. Maynard Smith (1979) has already made a somewhat similar suggestion about the value of signalling escape, but his tentative suggestion that individuals are selected to punish lying is, as he points out, difficult to account for in terms of individual selection, and less economical than that suggested by Hinde (1981).

The suggestion that so-called 'deceit' is, for one reason or another, not advantageous may have considerable generality. For example, Halliday & Houston (1978) have argued that it may well be to the advantage of a male newt who has only one spermatophore not to behave as if he had more. Clutton-Brock & Albon (1979), on the basis of their own data on red deer and a review of relevant data on other species, conclude “we should expect selection to have favored individuals which honestly advertized their real potential by the cheapest possible means”.
Fagen (1980) concluded on the basis of his models that assortative encounters favor. the evolution of conventional fighting tactics, and that a 'conspiracy of doves' can be an evolutionary stable population under these conditions. Perhaps in evolutionary terms too, honesty is the best policy.


'Home range' is the area that an animal learns thoroughly and habitually patrols (Seton, 1909; Burt, 1943), while the 'core area' is the area of heaviest regular use within the home range (Kaufmann, 1962; Jennrich & Turner, 1969). 'Territory' is an area occupied more or less exclusively by an animal or group of animals by means of repulsion through overt defense or advertisement (Noble, 1939; J.L. Brown, 1964, 1975; E.O. Wilson, 1971, 1975; Cf. Dyson-Hudson & Smith, 1978; Barash, 1982). The territory need not be a fixed piece of geography. It can be 'floating' or 'spatiotemporal' in nature, meaning that the animal defends only the area it happens to be in at the moment, or during a certain time of day or season, or both (Cf. Leyhausen, 1965). Territoriality, like other forms of aggression, has taken protean shapes in different evolutionary lines to serve a variety of functions. And like general aggression, it has proved difficult to define in a way that comfortably embraces all of its manifestations. Nevertheless, there is some consensus among evolutionary biologists that overt defense is the diagnostic feature of territoriality.
According to E.O. Wilson (1975) the exclusive use of terrain must be due to one of the following five phenomena: (1) overt defense, (2) repulsion by advertisement, (3) the selection of different kinds of living quarters by different life forms or genetic morphs, (4) the sufficiently diffuse scattering of individuals through random effects of dispersal, or (5) some combination of these effects. Where interaction among animals occurs, specifically in the first two listed conditions, we can say that the occupied area is a territory.
Territorial behavior is widespread in animals and serves to defend any of several kinds of resources (food supply, access to females, shelter, space for sexual display, nesting, etc.). The functions served by territorial defense are idiosyncratic and difficult to classify. E.O.Wilson (1975) is nevertheless able to distinguish several major categories in which the known or probable function matches the size and location of the defended space. The following classification is an extension of one developed for birds in sequential contributions by Mayr (1935), Nice (1941), Armstrong (1947), and Hinde (1956).
Type A: a large defended area within which sheltering, nesting, and most food gathering occur.
Type B: a large defended area within which all breeding activities occur but which is not the primary source of food.
Type C: a small defended area around the nest.
Type D: pairing and/or mating territories (leks).
Type E: roosting positions and shelters.
The home range of an individual, whether defended as a territory or not, must be large enough to yield an adequate supply of energy. At the same time it should ideally be not much greater than this lower limit, because the animal will unnecessarily expose itself to predators by traversing excess terrain. This optimal area or optimal yield hypothesis seems to be borne out by what little data we have bearing directly on the energy yield of home ranges (C.C. Smith, 1968; Altmann & Altmann, 1970).
Among terrestrial vertebrates there is a surprisingly consistent correlation between the size of the animal and the size of the home range it occupies (McNab, 1963). It is also approximately true that the rate of energy utilization is a linear function of the metabolic rate, while the metabolic rate increases as a logarithmic function of the animal's weight. It follows that the area of the home range is a logarithmic function of energy needs (E.O. Wilson, 1975).
Why should animals bother to defend any part of their home range? MacArthur (1972) proved that pure contest competition for food is energetically less efficient than pure scramble competition. This is a paradox easily resolved. Territoriality is very special form of contest competition, in which the animal need win only once or a relatively few times. Consequently, the resident expends far less energy than would be the case if it were forced into a confrontation each time it attempted to eat in the presence of a conspecific. Its energetic balance sheet is improved still more it if comes to recognize and to ignore neighboring territorial holders - the 'dear enemy' phenomenon (E.O. Wilson, 1975).
Clearly, then, a territory can be made energetically more efficient than a home range in which competition is of the pure contest or the pure scramble form. But if this is the case, why are not all species with fixed home ranges also strictly territorial? The answer lies in what J.L. Brown (1964, 1975) has called economic defendability. Natural selection theory predicts that an animal should protect only the amount of terrain for which the defense gains more energy than it expends. In other words, if an animal occupies a much larger territory than it can monitor in one quick survey, it may itself trotting from one end of its domain to the next just to oust intruders, an energetically wasteful activity. Furthermore, territorial defense is curtailed if it exposes animals too much to predation. There is also the phenomenon of aggressive neglect: defense of a territory results in less time devoted to courtship, fewer copulations, and neglected and less fit offspring. In short, the territorial strategy evolved is the one that maximizes the increment of fitness due to extraction of energy from the defended area as compared with the loss of fitness due to the effort and perils of defense (E.O. Wilson, 1975; J.L. Brown, 1964, 1975; Schoener, 1971; Crook, 1972).
If there is less than enough for all of some requisite for reproduction - food, cover, mates, or nest sites - some individuals will probably receive less than others of the resource is short supply. The 'haves' would then leave more offspring than the 'have nots', other things being equal. The rewards of aggression depend on the stakes. If there is little to be gained by aggression and much to be lost by it, territorial behavior will be selected against. If there is much to be gained or guaranteed by aggression and little to be lost by it, territorial behavior will be selected for. Under steady-state conditions of competition, a norm for intensity of territorial behavior will most likely be established, with extremes in both directions selected against (J.L. Brown, 1975).
A second, competing hypothesis is that territoriality evolved by group selection, particularly interpopulation selection. This model also assumes that food, or less probably some other resource, is the ultimate limiting factor. Territoriality is a device evolved by the entire population, including the unfortunate floaters (non-territory holders who have no 'home' of their own, who are frequently driven out to suboptimal feeding areas, are exposed more to predation, which may be facilitated by the prey's undernourishment and lack of familiarity with safe places for escaping from predators), to hold population densities at or below the carrying capacity of the environment. The regulation is achieved at least in part by altruistic restraint and even self-sacrifice, especially on the part of the floaters.
Existing evidence strongly favors the first, individual-selection hypothesis. From the large amounts of data on birds, we can obtain a rough idea of the magnitude of individual mortality associated with territoriality as opposed to the rates of extinction in territorial populations (reviews by Lack, 1966; J.L. Brown, 1969; and Krebs, 1971). These data seem decisive. The differential mortality associated with territorial exclusion is heavy, on the order of 10 % or more per generation, enough to drive the evolution of territorial behavior with even a small amount of heritability in innate components of the behavior. The rates of population extinction, by contrast, must be very low, even if we assume restricted genetic neighborhoods and small effective population sizes. The group-selection hypothesis therefore appears to be excluded in at least the bulk of the better-analyzed cases (E.O. Wilson, 1975).
The economic defendability model of territoriality was elaborated by Dyson-Hudson & Smith (1978). Economic defendability has several components that interact to produce a cost-benefit ratio. The costs of territoriality include (1) the time, energy, and/or risk associated with defending an area; (2) the possible diversion of time and energy from other necessary activities; and (3) the possible negative consequences of relying on a spatially limited area for resources. The benefits of territoriality are simply those that result from exclusive access to critical resources; however, this benefit is conditioned by factor 3 (above) and is relative to alternative (nonterritorial) modes of resource utilization.
For any case of territoriality, the ratio of benefits to costs should exceed 1.0 (and probably by a comfortable margin). It can also be argued that adaptive processes in the long run will tend to produce optimal results and, thus, that the benefit/cost ratio for a territorial system should have an average value greater than the nonterritorial alternative available for the individual or group. However, this last expectation involves the assessment of a broad range of opportunity costs, and the economic defendability model is not sufficient for this purpose (Dyson-Hudson & Smith, 1978).
The cost/benefit ratio of a territorial strategy is highly dependent on the pattern of resource distribution. For a general model of economic defendability, as presented by Dyson-Hudson & Smith, the important parameters of resource distribution are predictability and abundance. Predictability has both a spatial component (predictability of location) and a temporal one (predictability in time). Abundance or density of a resource can be measured in several ways: in terms of average density over a broad area (the average for the territory or home range), as an average value within a particular type or microhabitat (within-patch density), and in terms of the fluctuation in density over time (the range of variability). Resources that are predictable in their spatiotemporal distribution have greater economic defendability than unpredictable resources. A habitat where critical resources are predictable will be most efficiently exploited by a territorial system (holding other resource distribution parameters constant). Geometrical models of foraging indicate that it is more efficient for individuals to disperse to mutually exclusive foraging areas when food resources tend toward a uniform distribution and are predictable (Horn, 1968; C.C. Smith, 1968). Unpredictability of resources results in lowered benefits of territorial defense (in terms of resources controlled), and, below a certain threshold, territoriality will be uneconomical or even unviable (J.L. Brown, 1964).
With a sufficient degree of resource predictability, clumping of individuals (coloniality) is expected to occur. Under these situations, efficient resource utilization may depend on the pooling of information about the location of ephemeral resource concentrations. In general, increased average density of critical resources makes a territorial system more economically defendable, simply by reducing the area that needs to be defended and thus reducing defense costs. However, density of resources within a patch combined with a high degree of unpredictability reduces the economic advantage of territoriality. That is, with sufficient within-patch density and patch unpredictability, localized and ephemeral superabundances result, where the temporary glut of resources is more than can be consumed and thus is best shared (either actively or passively) rather than defended.

Table X (after Dyson-Hudson & Smith, 1978)

Relationship between resource distribution and foraging strategy
Resource Economic Resource Degree of
Distribution Defendability Utilization Nomadism
A. Unpredict. Dense Low Info-sharing High
B. Unpredict. Scarce Low Dispersion Very High
C. Predict. Dense High Territoriality Low
D. Predict. Scarce Fairly low Home ranges Low-medium

Evolution of Territoriality and Spite
Most models for the evolution of territorial behavior have assumed that natural selection will favor the size X of territory which maximizes individual reproductive success f(X) (See Davies, 1978 for review).
The most serious weakness of simple cost-benefit models is their inability to deal with frequency-dependent selection (Parker & Knowlton, 1980). Hamilton's (1970; 1971) papers on the evolution of spite point to another flaw in the simple cost-benefit approach; the strategy of harming oneself to harm others more (by defending resources in excess of what can be used, for example) can never be that which maximizes net benefits, f(X). In Hamilton's models spite directed at individuals of less than average relatedness is favored because inclusive fitness is considered, rather than simply individual reproductive success.
Verner (1977) has also stressed that territorial individuals may benefit by excluding others from resources because this increases their relative contribution to the gene pool of the next generation. An essential difference between the two models is that in Hamilton's the spite is discriminate (directed towards individuals of less than average relatedness), whereas in Verner's it is indiscriminate. Alleles coding for indiscriminate spite can sometimes spread, but their evolutionary stability has been questioned (Rothstein, 1979; Tullock, 1979). Knowlton & Parker (1979) showed that some degree of indiscriminate spite will always be an ESS, but that the spite component of territoriality will be substantial only in small populations (See also Rothstein, 1979).
Parker & Knowlton (1980) present two classes of models for the evolution of territory size. Both are frequency-dependent (employing the concept of ESS), and are based on finite population sizes and continuous strategy sets. A territorial strategy is defined in the first class by the size of territory an individual defends if it is one of the individuals successful in obtaining a territory in a patch of resources when others are excluded. The ESS for territory size may be spitefully large when the potential fecundity of individuals is low, but as fecundity (or population numbers) increases, the ESS decreases towards the size which maximizes individual reproductive success. When the costs of defending a territory are a function not only of area but also of the number of competitors (excludeds plus holders) against which the area is defended, then the individuals excluded from resources (exludeds) are likely to lower the ESS more markedly than the individuals who have acquired territories (holders).
In the second class of models, all individuals gain access to resources, but the amount an individual acquires depends on its defensive effort strategy relative to the defensive effort of other competitors. Here, in contrast, the ESS for defensive effort increases with increasing potential fecundity. When the second class of model is extended to include phenotypic variation in individual success (at a given defensive effort), variance in phenotypic ESSs diminishes as the number of competitors increases. If the term spite implies that an individual invests more in territorial defense than the level that maximizes individual reproductive success, then a spiteful ESS can be found only in the exclusion models (Parker & Knowlton, 1980).
Rothstein (1979) has given a detailed survey of the sort of conditions that will favor the evolution of spite. It now seems clear that the amount of spite must be rather less than many authors have envisaged (e.g., for territoriality: Chitty, 1967; Gill, 1974; J.L. Brown, 1974; Verner, 1976; for sexual interference: Arnold, 1976).
Rothstein (1979) and Knowlton & Parker (1979) independently reached similar conclusions concerning the likely magnitude of spite in natural populations (See also Tullock, 1979; Pleasants & Pleasants, 1979). Rothstein made the further point, however, that even small degrees of spite would be evolutionary unstable against 'resistors' (individuals who are not in themselves spiteful but who thwart the efforts of neighbors attempting to take spitefully large territories). In Knowlton & Parker's competitor exclusion models, successful individuals obtain the number of resource units specified by the strategy they are playing by virtue of some convention associated with an asymmetry independent of the resource use strategy (such as prior ownership or RHP). A resistor mutant would suffer the considerable costs of escalation against the convention in addition to the energetic costs of resistance itself.
Although Rothstein did not formally analyze resistance in his models, the strategy resembles 'retaliator' in Maynard Smith's (1976) hawk-dove game (Getty, 1979). If the two are formally similar, then resistance may be an important component of models based on discrete strategy sets (spiteful vs. non-spiteful behavior); it is not likely to affect the conclusions of Parker & Knowlton's (or other similar models) which seek an ESS from a continuous strategy set.
Considering the two strategy game (spiteful vs. non-spiteful), Colgan (1979) suggested that in a mixed population the non-spiteful territory holders benefit from the action of spiteful individuals with superterritories. He therefore suggests that “spitefulness has become altruism”. This semantic paradox arises through considering specific individuals rather than aggregate effects on specific genes. In estimating the fitness of the non-spiteful strategists, one must include both territory holders and non-holders, since they both carry genes for acting non-spitefully. When the spiteful strategy is rare, there is never likely to be more than one superterritorial individual per patch, and any extra exclusion must therefore fall upon non-spiteful strategists. So superterritoriality can spread, essentially because it effects can be non-random with respect to alternative alleles for spiteful and non-spiteful. Thus although superterritoriality may be termed 'indiscriminate' spite (to imply no recognition of non-relatives), it can spread only because its effects can act discriminately (Parker & Knowlton, 1980).
According to Rothstein's (1979) analyses, inhibitory (i.e. spiteful) traits have only limited evolutionary importance. Selection nearly always favors channeling time and energy into improving an individual's own performance rather than harming others. Hamilton (1970) has argued that spiteful behavior is unlikely to evolve for three reasons: that all spiteful actions incur appreciable costs; that animals will not be able to differentiate between individuals which are of less than average relatedness to them and those which are more closely related, and may therefore damage their own kin; and that the trait is only likely to spread in small populations which it may help to extinguish. But these conditions are met in higher mammals. In species where marked dominance hierarchies exist, disruptive actions directed by dominants at subordinates may cost the dominant little. Also there is evidence that individuals may be able to distinguish their degree of relatedness to other animals quite nicely. Finally, if one supposes that spiteful aggression is only shown in certain contexts (for example at high population densities) there is no reason to think that it would extinguish the lineage in which it arose. While it therefore seems possible that spite may evolve, it will be extremely difficult to distinguish from selfish disruption. For example, even where apparently unsolicited aggression occurs, it is possible that it increases the initiator's inclusive fitness (Clutton-Brock & Harvey, 1976).
One possible example of spite might be called 'punishment'. Where a dominant individual's access to resources is jeopardized by the behavior of s subordinate, it may be to the former's advantage to punish the latter, reducing its fitness enough to make further attempts unprofitable. Hyenas (Crocuta crocuta) that intrude into their neighbors' ranges may be attacked and killed even after they have submitted and are attempting to escape (Kruuk, 1972). By occasionally killing such intruders, clans must reduce the potential advantages of poaching as well as removing competitors. Punishment is only likely to occur where the punishment is experienced or observed by several individuals, thus increasing the benefits of punishing (Clutton-Brock & Harvey, 1976).


Popp & DeVore (1974, 1979) proposed the following theory of aggressive competition and social dominance (with special reference to primates):
Consider a simple cost-benefit analysis of aggressive competition in a model of optimal behavior, where costs and benefits are measured in units of reproductive success, and aggression is viewed as a strategy that, under a given set of environmental and genetic constraints, may maximize an individual's contribution of genes to succeeding generations. In discussing the principles of natural selection that determine the optimal rate and form of aggressive competition, it is logical to ask first what costs of competition can be adaptively incurred by an aggressive actor. In calculating the maximum costs that an actor should be willing to accept in an aggressive encounter (Ca), there are four groups of variables of primary significance: (1) the probabilities of access to the disputed resource through aggression or through alternative strategies, (2) the benefit of access to the object of competition, (3) the effects of the competition on relatives of the actor, and (4) the intrinsic competitive abilities of the actor and his competitor.
In determining an actor's maximum adaptive expenditure in a competitive encounter, the following algorithm can be used. Multiply the probability of the actor's access to the disputed object if he competes aggressively (pa1) times the benefit of access to the disputed object to the actor if he competes aggressively (Ba1); subtract from this quantity the probability of the actor's access to the disputed object if he does not compete aggressively (pa2) times the benefit of access to the disputed object to the actor if he does not compete aggressively (Ba2). Hence Ca < pa1 Ba1 - pa2 Ba2.
The above inequality takes into consideration only the potential costs and benefits of aggressive competition to an actor, and assumes that the competitor is unrelated. Following Hamilton (1964, 1972), Maynard Smith (1964), and others, however, it is clear that the actor must also take into account the potential costs and benefits of the competition to his competitor if that competitor is a relative - a not unlikely condition in a social group of primates. Thus, the probability that a related competitor will gain access to a resource if the actor is or is not aggressive (pc1 and pc2, respectively) and the potential benefits to the competitor under both of these conditions (Bc1, Bc2), as well as the costs of the aggressive competition to the competitor (Cc), must be multiplied by Hamilton's (1972) regression coefficient or relatedness (bac), where bac is defined in the absence of inbreeding as the probability that a gene is present in the genotype of the actor that is identical by descent with a gene in a randomly chosen gamete of the competitor. We would expect, therefore, that natural selection will favor an individual that competes aggressively with a conspecific when the following inequality is true:

    Inequality 1: Ca < pa1 Ba1 - pa2 Ba2 + bac (pc1 Bc1 - pc2 Bc2 - Cc)

The actor's maximum adaptive expenditure for an aggressive encounter is represented by Ca. This includes such components as the caloric expenditure during a fight, the cost of repairing injuries, the risk of permanent injury or death, and so forth, all expressed in terms of their effects on reproductive success. Likewise, Ba1 represents the total benefit of access to the object of competition. For example, in competition over a food item, Ba1 includes components such as the net nutritional value of the item relative to other food items in the environment, measured by its effects on reproduction, as well other consequences of the competition (such as a higher rate of access to similar resources in the future, if such action means that the same competitor is more likely to avoid future competition for those resources).
According to Inequality 1, aggressive competition is most adaptive when: (1) the cost of aggressive competition to the actor is small; (2) the benefit of the disputed object in terms of the reproductive success it will confer upon the actor is large; (3) aggressive competition confers a probability of access to the disputed resource by the actor that is large compared to the probability of access if aggressive competition does not occur; (4) the competitor (i.e., the recipient of the aggressive act) is genetically unrelated; and (5) the cost of aggressive competition to the competitor is small if the competitor is genetically related (bac > 0). In fact, with some exceptions (vide infra), if the competitor is unrelated, the actor will be selected to discount the costs of aggressive competition to the competitor (Cc) entirely.
When Inequality 1 is true for one or more actors, aggressive competition is expected to result and continue over time until the inequality is no longer true for any of the actors. Individual α is predicted to win in an encounter with individual β when the following inequality is true:

    Inequality 2: Kα Cα > Kβ Cβ

where, Kα = α's intrinsic competitive ability, i.e. α's ability to inflict costs of competition on β per unit time;
Kβ = β's intrinsic competitive ability;
Cα = the maximum adaptive expenditure by α for aggressive competition, as determined by Inequality 1;
Cβ = the maximum adaptive expenditure by β (considering β as the aggressive actor) for aggressive competition, as determined by substitution in Inequality 1.
More generally, if a winner in a dyadic aggressive encounter is defined as the individual that takes possession of the disputed resource when the aggressive interaction is terminated, then the winner can be predicted to be the actor with the higher value for the quantity Ka Ca.
In theoretical terms, aggressive competition may be viewed as an attempt by an individual to modify the cost-benefit function of its competitor in such a way that continued attempts by the competitor to gain access to a resource are unprofitable and maladaptive. Injury, or the threat of probable injury, to a competitor can elevate the costs of continued competition until these are no longer offset by the potential benefits of access to the disputed resource. When such a point is reached (i.e., when Inequality 1 becomes false), natural selection will favor the competitor that terminates the aggressive interaction; in a dyadic encounter with β, the amount of β's losses are predicted by the value of Cβ.
The last individual to terminate its aggressive behavior will normally gain access to the resource, and therefore be defined as the winner. If α wins in an encounter with β, its cost of competition equals the quantity KβCβ/Kα , which is less than or equal to its maximum adaptive expenditure (Cα) for that encounter.
It follows that in aggressive encounters an individual will be selected to act in ways that will increase the rate of expenditure of an opponent's competitive effort while reducing its own rate. In populations where this competitive strategy includes attempts to inflict physical damage on opponents, natural selection will favor those individuals who develop defense strategies (assuming the defense strategy is less costly than the injuries it prevents). As a result, an aggressive actor need not necessarily inflict physical injury on an opponent for the actor's behavior to be adaptive; it is sufficient that the actor force his opponent to adopt a strategy of defense that is costly in time and energy. As Maynard Smith (1972) noted: even 'ritualized fights' (which may involve little chance of injury to the participants) can thus be settled.
Asymmetry in the cost-benefit functions of competitors is the key to the termination of aggressive interactions. If two competitors knew the precise value of all variables in Inequalities 1 and 2 prior to an aggressive encounter, we would not expect such an encounter to occur; the losing competitor would be known in advance, and would be selected to avoid the competition. Even when the values for the relevant variables are not precisely known, the disparity between the values of KaCa for two potential competitors may be so large that the inferior competitor actively avoids the interaction; thus, some potential conflicts may be selected with little or no costs of competition. In other cases, competitors may be so closely matched in value of KaCa that aggressive conflict will occur. In this circumstance even the winner will be forced to expend most of his Ca before the outcome of the encounter is determined.
If follows from Inequalities 1 and 2 that when two individuals are equal in other respects, the competitor that can derive a greater benefit from the object of competition is expected to win the encounter. Similarly, a competitor with the lower costs of competition per unit time of interaction has a competitive advantage when other variables are equal. The cost of aggressive competition per unit time relates directly to the concept of intrinsic competitive ability that was introduced in Inequality 2; it is not the individual with the greater Ca that wins the encounter, but rather the individual that expends the smaller relative portion of Ca per unit time. Hence morphological structures used in fighting evolve in populations, where such variations arise by mutation, because the costs of competition per unit time are higher to individuals who lack such structures than to individuals who possess them (Popp & DeVore, 1974; 1979).
Several other pertinent predictions can be derived from Popp & DeVore's model:

(1) Threat displays. It is easy to appreciate that agonistic displays may benefit the recipient of such displays, since advance knowledge of an impending attack enables the individual to prepare to flight or flee. It is much less obvious, however, why an individual would be selected to provide unambiguous signals of aggressive intent for a potential victim; it would seem that greater damage could be inflicted on the opponent through surprise attacks. It is important to recognize that individuals are not selected to maximize the losses of their opponents, but to maximize their own gains in reproductive success; these two operations are not necessarily equivalent. Inflicting costs on an opponent beyond those required to win a given encounter does not usually result in added benefits for the aggressor. In fact, when the cost of seriously injuring the opponent exceeds the cost of winning the encounter, or when bac is greater than 0, such behavior will be disadvantageous to the actor.
Agonistic displays will be favored by natural selection when they alleviate the need for forms of aggressive competition that would be more costly to the individual. Thus, if agonistic display alone is sufficient to drive a competitor away from a disputed resource, and if display is on the average less costly to the individual than aggressive behavior, and if the benefit of agonistic display in preventing needless aggressive escalation is greater than the cost of putting one's potential opponent on guard, then agonistic display will be an adaptive strategy.

(2) Submissive gestures. The ability to predict the outcome of an aggressive encounter before it occurs is also advantageous to a potential aggressor. In Inequality 1, it was shown that as the probability of winning an encounter approaches 0, the maximum costs that an individual is willing to suffer also approaches 0. Predicting the outcome of a conflict permits an individual to avoid the conflict completely or terminate it early when it is apparent that he would lose. As a result, individuals who are sensitive to cues that are useful in predicting the outcome of an aggressive encounter can be favored by natural selection. Individuals are expected to modify their behavior according to such characteristics of their opponents as body mass, horn size, and stage of dental eruption when these characters relate to their opponent's intrinsic competitive ability, that is, when they potentially affect the outcome of an aggressive encounter.

(3) Exaggerations. In populations in which individuals use the behavioral and/or morphological characters of potential competitors to determine whether or not aggression will be adaptive, it is widely recognized that those characters on which intrinsic competitive ability is estimated will tend to be emphasized or exaggerated by natural selection. For example, in species in which body size is a determinant of the outcome of aggressive encounters and individuals modify their behavior as a function of their opponents' body size, then individuals that exaggerate their body size during display will have a competitive advantage. Wallace (1973), in a discussion of the evolution of misinformation in animal communication, argues convincingly that, where exaggerations evolve, large exaggerations produce a disproportionately large increase in fitness of the actor compared to small exaggerations; 'small lies' have about a 50:50 chance of improving a competitor's understanding of a situation, whereas 'big lies' deceive competitors with near certainty. Similarly, it is Popp & DeVore's belief that selection for increasing exaggeration of characters relating to intrinsic competitive ability (Ka) is expected to proceed until the specific character is so exaggerated that a positive correlation between its appearance and the intrinsic competitive ability of the individuals within the population no longer exists (or until such characters reach the stage of development where counterbalancing selection through other pathways inhibits further change). one such counterbalancing selection force is the evolution of compensatory adjustment capabilities in the conspecifics that observe these displays, i.e., the ability of competitors to devalue the exaggeration by the appropriate amount. Trends in the evolution of exaggerated body size and fighting structures have been generally appreciated since Darwin; however, some additional principles follow from the model of aggressive competition presented here. Inequality I includes a number of variables relevant to the outcome of aggressive encounters. Exaggerations by individuals need not be limited to morphological structures that are related to intrinsic competitive ability (Ka), but may theoretically include any of the variables in the cost-benefit functions for aggressive competition. Thus, an actor who can exaggerate the value of the disputed resource (Ba1), so that the opponent overestimates the benefit that the actor can derive from it, will have a competitive advantage. In functional terms, such exaggerations might be made by an individual who mimicked higher levels of motivation, to gain access to a disputed resource, than he in fact felt. Or an individual may exaggerate the value of winning an encounter (Ba1) by misrepresenting the circumstances under which the encounter occurs. For example, 'displacement activities' such as sham incubation behavior may have straightforward adaptive significance; the presence of a nest or eggs would usually imply higher benefits from winning the encounter for their owner than for the intruder, and individuals who advertise such asymmetry when it exists, or delude an opponent into believing that such asymmetry exists through sham behavior, will have a competitive advantage.
Another consequence suggested by this model is that actors who exaggerate their own probability of winning the encounter (pa1) have an advantage; an effect that can be achieved by displays that project a confident, self-assured manner. Likewise, concealing the costs of competition (Ca) that one is suffering (at least as long as there is some probability greater than 0 of winning the encounter), would lead an opponent to underestimate the rate of cost he is inflicting upon you, and to overestimate the length of time you can compete adaptively - thereby increasing the probability that the opponent will terminate the encounter early.
An individual who can convince its opponent that an unrealistically high regression coefficient of relatedness exists between them would also have an advantage.
Finally, an individual can gain a competitive advantage in an aggressive encounter with a competitor that is known to be a relative by exaggerating the costs of competition to himself, i.e., the actor gains an advantage by suggesting to the related opponent that the competition is inflicting very high costs upon the actor. Thus, under some circumstances related individuals may be expected to feign serious injury, or higher rates of costs of conflicts to themselves, during a fight.
There are yet subtler strategies that probably find their greatest expression among humans, e.g., convincing the opponent that the disputed resource is valueless to him.

(4) Degree of ritualization. If an individual in an aggressive interaction terminates its aggressive behavior at or shortly after the time that its opponent gestures submissively, it will gain access to the disputed resource and on the average gain a net benefit for the entire interaction. If the winner continues to act aggressively toward the already submissive opponent, however, the situation changes substantially. If we assume that the victor continues the aggression with the intent to kill or seriously injure his submissive opponent, a new set of cost-benefit functions rapidly develop. Since no cost of competition could ordinarily exceed the costs of a fatal injury or, alternatively, since the benefit of saving one's life is considerably higher than the benefit that could be derived from a disputed resource, the individual whose submissive behavior has failed to terminate its opponent's aggression, will under most circumstances (i.e., where high bac is not involved), fight desperately in an all-out self-defense. By contrast, the only benefit for the potential assassin would be the elimination of just one of many competitors. The costs that a potential assassin is willing to incur, therefore, would usually be lower than the costs the victim is willing to suffer in self-defense, If we assume that there is no great disparity in the value of Ka for the two opponents, the aggressor will find it adaptive to terminate the aggressive encounter prior to killing an opponent. We can conclude that winning an aggressive encounter over a disputed resource that is of trivial value, compared to the value of continued survival, is a poor indicator to the victor that he could kill an opponent in continued combat. In addition, under natural conditions the submissive animal often has the opportunity to escape, and this further reduces the mortality directly attributable to aggressive competition (although it does not necessarily reduce the number of competitors who are eliminated because they have been excluded from essential resources).
Note that the preceding argument is not at all equivalent to the frequent assertion that organisms possess an innate inhibition against killing conspecifics: whenever differences between two competitors in intrinsic competitive ability times the maximum adaptive expenditure for aggressive competition, as represented in Inequality 2, are sufficiently large, murderers can be favored by natural selection. Although there is a number of noteworthy examples of strategies favored by natural selection that lead to the killing of conspecifics, the cost-benefit functions do not often meet such criteria.
Exceptions include some primate species (e.g,, Hrdy, 1977) in which infants are killed by presumably unrelated or distantly related adult males, who are competing for access to a female's reproductive effort. Additional data on mortality resulting from aggressive competition are provided by Geist (1971) on ruminants, Bergerud (1974) on caribou, Schaller (1972) on lions, LeBoeuf, Whiting & Grant (1972) on elephant seals, Mykytowycz & Dudzinski (1972) on rabbits, and Mayburg (1974) on eagles. Among eagles, like some species of hymenoptera, natural selection favors fratricide i.e., patterns of aggressive competition that lead to the killing of full-sibs. In numerous species mortality arises from cannibalism; such behavior has been well documented and is not discussed here.

Dominance Theory and Aggressive Competition
In paired encounters, when individuals are acting optimally in their own evolutionary interests, the individual with the relatively higher value for KaCa will win the encounter and be defined as dominant; the individual with the lower value for KaCa will lose and be defined as subordinate. When all possible pairings between an individual and all other conspecifics with which it interacts are made, its rank in a hierarchy is defined as the number of individuals over which it is dominant. Popp & DeVore have already discussed how the value of a resource is relevant in determining the amount of aggressive competition that is adaptive in attempts to attain access to it. Further, one should not assume a priori that: (1) two different resources are of equal value to an individual, or (2) a particular resource is of equal value to two different individuals, or (3) a particular resource is of equal value at all times to an individual. Perhaps the single most important conclusion from a consideration of social dominance in terms of aggressive competition theory is that dominance hierarchies are expected to be time- and resource-specific. (Territoriality may be viewed as the consequence of aggressive interactions that are location-specific.)
This conclusion suggests that a number of widely accepted concepts and methods associated with earlier paradigms of social dominance are invalid. The notion that hierarchies produced by measuring rates of access to a single resource may be generalized to speak of social behavior in other contexts may frequently lead to error.
Popp & DeVore do not imply that an individual's dominance status with respect to access to a particular resource will never be correlated with its dominance status in competition for access to other resources; indeed such a correlation may occur frequently. Their point is that such similarities between resource-specific dominance hierarchies are, in fact, correlations - a particular dominance status with respect to access to a food item does not necessarily imply that an individual will have the same dominance status with respect to other disputed resources. Positive correlations between the dominance rankings of an individual in competition for two different resources are expected to arise from similarities in the value KaCa in competition for both resources, relative to other members of the social group. Obviously, under some circumstances differences in the intrinsic competitive ability (Ka) among the members of a social group may be so large that short-term fluctuations in dominance status with respect to commonly disputed resources are rare or nonexistent. Such a consistency in dominance status, however, must be demonstrated, not assumed.
In short, the concept of general dominance, as it is frequently applied in primate studies, leads to methodological error when it is employed to determine dominance rankings and individual dominance status. More seriously, it obscures the issues that underlie the outcomes of single encounters, and precludes interpretations of social dominance and aggressive competition in terms of life history strategies, reproductive success, and inclusive fitness.

As previously discussed, it is adaptive for an individual to be able to predict the outcome of an aggressive encounter; such an ability permits him to reduce costs by avoiding conflicts that will be lost and to increase benefits by competing to the end in encounters that he can win. In species that form long-term associations among a small set of individuals, one method of predicting the outcome of a competitive encounter is by the recollection of past encounters with a specific opponent. Past competitive experience with a known opponent under circumstances similar to the present competitive interaction can be useful in estimating the cost-benefit function for the opponent in aggressive competition. For example, if individual β has lost in all previous aggressive encounters over food items with individual α, it will be to β's advantage to avoid such competition in the future until the variables in Inequality 2 change in his favor. The best strategy for a subordinate individual who knows from past experience that it cannot win an encounter (i.e., pa1 is near 0) is to avoid the competition. It is this principle that is responsible for the often observed decline in the frequency of aggressive behavior when the members of a social group have had sufficient time to form dominance hierarchies.
Considering the same example from α's viewpoint, it may be to its advantage to seek an increase in the rate of competitive interactions with β, since it can dominate β with relative certainty. In the extreme form, it will be to β's advantage to show submissive behavior immediately upon being challenged by α, since any expenditure in competitive effort always represents a net loss.
Appropriate behavior by subordinates that are acting in their own best interest may, therefore, involve active avoidance of the dominant individual or active avoidance of situations with the potential for conflict. e.g., the avoidance of foraging activity near the dominant individual. Behavior such as this has led to a great deal of confusion in the social dominance literature: to the unwary observer, the illusion is created that dominant individuals are passively dominant in a role that has been conferred upon them by subordinates for reasons unspecified. In contrast, Popp & DeVore emphasize that both dominant and subordinate individuals must be viewed as actors that have been selected to display behaviors appropriate to the natural social environment for the maximization of their reproductive success.
Dominance hierarchies do not exist because they bring harmony and stability to the social group, but as the consequence of self-interested actions, in the evolutionary sense, by each group member.

Among the great apes, male chimpanzees employ several reproductive strategies, some of which are alternatives to direct aggressive competition. Both Goodall (1968) and McGinnis (1979) report male-male aggression in competition for estrous females, and consistent with these findings Nishida (1979) reports that a single male was involved in 46 % of all copulations observed during his study. Thus, under some conditions dominant male chimpanzees may be able to monopolize sexually receptive females.
Alternatively, Goodall (1971) and McGinnis (1979) report highly promiscuous breeding patterns in some circumstances, where males that are unable to exclude male competitors from access to estrous females form large groups attending a single female and copulate successively, or opportunistically steal copulations when more dominant males are temporarily distracted. Under such conditions, one expects that sexual selection will favor those males that deposit the largest quantity of sperm in the female reproductive tract - hence, the extremely high relative testicular weight among male chimpanzees as compared to the other great apes and man.
Thus, in addition to aggressive competition, male chimpanzees employ sperm competition as a reproductive strategy.
'Prostitution behavior' by subordinates, in the attempt to attain access to limiting resources, is another alternative to aggressive competition. The fact that estrous female chimpanzees have differentially high access to food items in the presence of sexually mature males is well known in both captive and wild populations (Yerkes, 1940; Teleki, 1973). Since proximity to a sexually receptive female is conducive to sexual access, driving away an estrous female in competition over food items confers fewer net benefits than driving away a nonreceptive female. Hence, males show high rates of sharing with estrous females, and females are expected to recognize such tendencies and respond in a way that is most efficient in exploiting them, e.g., by unhesitatingly approaching males that are ordinarily dominant to them with respect to food or approaching with sexual presentation.
A similar analysis can be made of grooming behavior shown by subordinates in their attempts to gain access to a desired object that is in the possession of a more dominant individual.
Subordinates that act as though the status of dominant individuals is immutable, however, will not necessarily be acting optimally. Subordinates that recognize and capitalize upon opportunities to reverse the dominance status between themselves and higher ranking individuals will be favored by natural selection. Such behavior is analogous to the ESS of prober-retaliator discussed by Maynard Smith & Price (1973).
Since conspecifics will estimate the rank of an individual and its ability to maintain that rank by its behavior and appearance, dominant individuals are selected to provide unambiguous signals of their high status and ability to maintain it. Strutting behavior, tail-erect postures, and open-mouth threats are well-known gestures in primates that are displayed primarily by dominant individuals. High-ranking individuals past their prime are expected to attempt to conceal their declining competitive abilities (Popp & DeVore, 1974; 1979).

Dominance and Cheating
Barnard & Burk (1979) distinguish three basic types of hierarchy: those based exclusively on asymmetries of fighting ability (statistical hierarchies), those involving decisions based on internal state ('confidence' hierarchies) and those depending on cues received from an opponent (assessment hierarchies).

Statistical hierarchies. Consider a groups of animals competing for a limited resource. Each individual differs from the others in competitive ability (i.e., the ability to win in competitive encounters) and fights every time it encounters a competitor. Fights last until one individual has clearly lost at which point it retreats and the winner claims the resource. An observer watching the group over a period of time and monitoring a number of such fights would be able to construct a rank order of fighting success within the group based solely on asymmetries in fighting ability. If there were no changes in the physical or behavioral characteristics of group members, the stability of the hierarchy would not change with time. Barnard & Burk call such a hierarchy a 'statistical' hierarchy. Its important features are a lack of any form of assessment or memory of past encounters.
The apparent rank ordering of captive sea-anemones (Brace & Pavey, 1978) may be an example. However, statistical hierarchies are unlikely to be widespread in occurrence. A series of mathematical models of hierarchy formation (Landau, 1968; Chase, 1974 et seq.) have shown that stable statistical hierarchies are unlikely in groups larger than five unless individual differ very markedly in competitive ability. Nevertheless, for reasons which will become clear later, statistical hierarchies form the foundation for more complex organizations.

'Confidence' Hierarchies.
For reasons of energy conservation and risk of injury, it will usually pay individuals within a competing group to avoid becoming involved in escalated fights. To do this, they must develop a means of predicting the outcome of potential fights. If an individual has been involved in a succession of fights it can follow a simple rule with respect to future fights: 'If I won in the past, be aggressive; if I lost, be cautious'. Each individual can be thought of as updating his estimate of his own competitive ability to that of the average individual in the group. If all members of the group fight according to the rule, individuals who won their first few fights (by beating a lower ability competitor or by some chance arbitrary advantage) would become more aggressive (Scott & Fredericson, 1951) and more likely to win the next few fights. Similarly, individual who lost would tend to carry on losing.
An individual's estimate of its competitive ability can be called 'confidence' because it reflects an expectation of the outcome of future fights. The confidence of any given individual may depend on its initial competitive ability and the frequency with which it encounters strong winners and losers. Such effects of previous wins and losses on fighting behavior have been shown in a number of groups (Ginsburg, 1942; Alexander, 1961) and computer simulations have shown that a relatively long memory of past outcomes is sufficient in itself to account for stable dominance hierarchies. In contrast to statistical hierarchies, confidence hierarchies may frequently change in stability with time.

Assessment Hierarchies. The models discussed so far suggest how stable hierarchies may develop using only the physical differences between individuals or differences in their fighting experience. However, for most populations, such methods of settling contests will be highly uneconomical. For any given individual within the population, there is likely to be a number of other individuals by which it will always be beaten, irrespective of its internal state. Selection will normally favor any individual which avoids contests with opponents of higher competitive ability, an individual that is whose fights no longer occur on the basis of random encounters but are instead calculated risks. In order to avoid such contests, individuals must be able to distinguish between opponents.
In a population whose individuals vary in competitive ability, there may be range of characteristics which correlate with competitive ability. For example, if competitive ability depends on body weight, then body size may signify an individual's status. Similarly the length of canine teeth or the size of horns may indicate competitive ability. These characteristics all happen to pertain to weaponry used in fighting but selection will favor individuals which take into account any cue that correlates with competitive ability. High status individuals may be those with, for example, large crests or darker pelage. If a single cue (C1) which correlates with competitive ability is used to settle contests, a stable hierarchy will result.
Assessment using a single cue has been recorded in various groups (Ginsberg & Allee, 1942; Greenberg, 1947; Bovbjerg, 1953; Tyler, 1972). However, as long as the relationship between C1 and competitive ability is no more than a statistical correlation, it is open to cheating. If C1 can be faked or if mutant low-ability individuals arise bearing high ability characteristics, cheating will be rapidly selected for since the advantages gained from access to a limited resource are appreciable. Clearly assessment cannot be by C1 alone; assessors should test the information they receive from potential opponents by probing (Maynard Smith & Price, 1973). The frequency of probing will be negatively influenced by the density of cheats in the population. This is because if the density of cheats is low, probing is likely to result in a genuinely high-ability individual being challenged and a high risk of injury to the challenger.
It is important to remember that individuals are simultaneously assessors and cheats and that selection acts on both characters. Individuals are selected both to emulate high status and to detect the true status of opponents. The cost of assessment in terms of time and energy spent probing rises with the density of cheats in the population. When the cost of assessment exceeds the benefits gained from distinguishing between opponents, assessors should develop an alternative means of detecting cheats. Barnard & Burk (1979) suggest that this may be achieved by elaborating C1 to C1 max at which point a second cue C2 will be added to the assessment mechanism. C2 is any cue other than C1 which aids in distinguishing the relative status of competitors. It may be a pre-existing feature or arise de novo in the population. As long as, in conjunction with C1, it is in some way correlated with competitive ability, it may be used in assessment.
Opponents now consist of two variables in the eyes of the assessor; firstly a statement of competitive ability (C1) which may be true or false, and secondly a reference cue (C2) which is genuine. Possible examples of recognition at this level are the use of song type and position on the territory boundary of competitors in territorial white-throated sparrows (Zonotrichia albicollis) (Falls, 1969) and the use of croak pitch and body size in toads (Bufo bufo) (Davies & Halliday, 1978). At least two cues also appear to be involved in assessment in Harris's sparrows (Zonotrichia querula) (Rohwer, 1975) and chaffinches (Fringilla coelebs) (Marler, 1955). It has been shown experimentally that cues are assessed additively; manipulation of one cue usually results in the altered individual being recognized as a cheat (Rohwer, 1977). However, reference cues (C2,3,...,n) are potentially as open to cheating as C1, except that two or more cues now have to be faked if the cheat is to be convincing. If cheats arise, however, the process described above will be repeated and a further cue C3 will be added to the assessment mechanism.
The essential point is that assessors should add reference cues to the assessment mechanism because taking a new cue into account is likely to be energetically cheaper than maintaining relationships by probing.
Furthermore Barnard & Burk suggest that only cues correlating with high status will be added in because the cost of probing (Pc) is twofold. There is a time cost Pt which is incurred simply by being involved in a probe, and an injury cost Pi which is incurred if a higher status individual is challenged. Selection will favor assessors who recognize and avoid high status individuals (Pc = Pt + Pi) rather than low (Pc - Pt) and simultaneously, high status individuals will be at a selective advantage through priority of access to limited resources.
Initially high status individuals should elaborate the current cue Cn; if assessment is for example by horn size, selection will favor high status individuals with larger and larger horns, if it is based on darkness of the pelage, selection will favor darker and darker pelage. Eventually, however, a point is reached when high status individuals can elaborate Cn no further. This may be due to physiological cost (e.g., of increasing the size of bodily structures) or to selective cost (e.g., of increased conspicuousness to predators). At this point Cn reaches Cn max and high status individuals can no longer stay ahead in the 'arms race'.
Since the density of cheats in the population is likely to rise rapidly once Cn max is reached, probing and hence the cost of expressing status also rises. It is at this point that selection will favor the introduction of a new cue. Although initially assessment cues may be only slightly correlated with competitive ability, selection is likely to result in a stronger and stronger correlation because high status individuals bearing a recognizable cue will be at an advantage through a reduced frequency of challenges and increased priority of access to limited resources. Eventually alleles for assessment cues and those for characteristics determining competitive ability may be linked through selection for chromosomal segregations maintaining the alleles together. Furthermore there may be simultaneous elaboration of the neural machinery required to deal with the assessment of progressively more complex cue patterns.
Barnard & Burk (1979) propose to call the array of cues used in assessment the 'assessment unit'; opponents are recognized by whatever sized unit is required for reliability. However, there is nothing especially significant about the physical individual as an assessment unit. Below the level of the individual, the unit will comprise as many cues and whatever degree of cue elaboration it takes to make the assessment mechanism (at least temporarily) uncheatable. Nevertheless there may be constraints on unit size such that, even though the mechanism is cheatable, current cues cannot be elaborated further and new cues do not arise. In this case relationships may be maintained by probing. High frequencies of probing appear to be a common feature of hierarchies in which there is no clear-cut assessment mechanism (Barnard, 1978).
Nevertheless it may be possible for the assessment unit to extend beyond the individual. In some primate species, it is common for individuals to solicit the help of others in tackling dominants. In this way two subordinates can supplant an individual higher in rank than either of them (Rowell, 1974; Kawai, 1958; de Waal, 1982). Such 'dependent rank' (Kawai, 1958) is also found in birds (Boyd, 1953) and appears to operate on the basis of additive rank rather than simply on the physical superiority of more than one individual.
It is clear that there is no distinct division between assessment involving particular cues and so-called individual recognition. An individual is after all simply an array of cues which are additively assessed and recognized as a pattern. Furthermore, an assessment unit need not be based on visual or tactile experience of individuals. For example, it may comprise a song type and a position on a territory boundary (Falls, 1969) or it may comprise olfactory cues and time (Leyhausen, 1956). Opponents in the eyes of an assessor will consist of whatever cues can be reliably used to distinguish between them.
'Individual recognition' is not a prerequisite for dominance hierarchies but dominance hierarchies may be a driving force of the evolution of 'individual recognition'. Theoretically, the addition of cues can go on indefinitely. In practice, however, several factors will militate to halt it. This is because (a) with each addition, cheating becomes more difficult because it is statistically unlikely that a multiplicity of correctly faked cues will arise; (b) it is possible that there will be no pre-existing or newly arising cue which correlates with competitive ability; (c) a cue already in the assessment mechanism may be elaborated to a degree which is energetically or selectively limiting and is proportional in degree of elaboration to the status of the individual and (d) a new cue may arise which is similarly expensive and which can only be assessed by high status individuals.

Status-limited Cues. In (c) and (d) above Barnard & Burk (1979) suggested that the arms race between assessors and cheats would be brought to a halt if energetically or selectively expensive cues evolved. The authors call these 'status-limited' cues because only individuals able to benefit appreciably from limited resources could 'afford' to develop them. The idea that evolutionary expensive cues may be selected for under certain conditions is, of course, not new. Zahavi (1975; 1977) has frequently suggested that 'handicaps' may evolve to increase reliability in communication between competing individuals. To avoid being cheated, high status individuals develop expensive cues and unequivocally state their superiority.
Several objections can be raised to the 'handicap' principle (Dawkins, 1976). In particular it is difficult to rationalize the possession of 'handicaps' if these render the bearer no fitter (because of physiological or selective cost) than a low status individual. For example, although perhaps starting with a higher energy reserve, the development of the 'handicap' may require a high status individual to use up its energetic advantage until its energy quota falls to minimum needed to sustain the animal. Assuming this minimum level is the same for all individuals in the population, it follows that all could be equally 'strong' despite overt differences in their competitive ability. The assessment mechanism is therefore valueless and will collapse.
This does not mean, however, that selection will not favor status-limited cues under any circumstances. Energetic reserves may be distributed through a population such that cues can be elaborated asymmetrically but still leave high status individuals with an energetic advantage. This would result in graded cues which would be indistinguishable at any given time from graded series which were temporary expressions of an arms race and may provide a functional explanation for widespread allometry in assessment cues. Such graded cues would correspond to (c) above.
Alternatively, it is possible for a population of competing individuals to be divided into those which possess an expensive cue and those which do not. A good example may be the territorial system found in many animals. In the Uganda kob (Adenota kob thomasi) for example, males compete for possession of small breeding territories (TGs) which can only be held at considerable expense (Leuthold, 1966). TGs provide little or no grazing so holder males are unable to feed efficiently during their tenure and competition for ownership is fierce. However, females can select mates from among the TG holders so such males are at a considerable selective advantage. Possession of a TG therefore appears to be a status-limited cue (only strong males can afford to defend a TG) which females can use to choose reliably between potential mates.
Other forms of territoriality such as the system of feeding territories in grouse (Lagopus I. scoticus) (Watson, 1967) may also be status-limited. In this case non-holders do not usually challenge holders, perhaps because ownership is likely to have been secured through superior competitive ability. However, the possession of a feeding territory may itself further increase the asymmetry in favor of the holder (Barnard & Burk, 1979). For an analysis of the evolution of reliable and unreliable badges of fighting see also Rohwer (1982).

Female Dominance Relations in Primates
The control of aggression (e.g. Carpenter, 1954; Kummer, 1971; Porier, 1974) and the regulation of population density (Wynne-Edwards, 1962) are frequently suggested as functions of dominance hierarchies. These evolutionary explanations based on group selection have been challenged by Maynard Smith (1964), Lack (1966), G.C. Williams (1966) and others. Clutton-Brock & Harvey (1976) and Deag (1977) have advocated evaluation of individual costs and benefits in dominance interactions. West-Eberhard (1975) suggested that subordinance may be understood as a form of altruism enhancing inclusive fitness and he drew attention to the importance of kin selection in the evolution of dominance. Several investigations of the relationship between male dominance and reproductive success in primates have been conducted (for a review see Bernstein, 1976). However, only three studies have investigated the association between female rank and reproductive success. Two of these studies were done on rhesus monkeys (Macaca mulatta).
Drickamer (1974) analyzed ten years of life history data from the island colony of rhesus at La Parguera, Puerto Rico, and found a significant correlation between the total number of offspring produced by a female and her dominance rank. Sade et al. (1977) found a high correlation between the intrinsic rate of increase of genealogies and their dominance rank for the island colony of rhesus at Cayo Santiago, Puerto Rico. The third study (Dunbar & Dunbar, 1977) was conducted on free-living gelada baboons (Theropithecus gelada) in Ethiopia. Again, a positive correlation was found between the dominance rank of females and the number of living offspring per female.
Hrdy & Hrdy (1976) proposed a functional model of female dominance hierarchies in Presbytis entellus. They termed the female langur hierarchy 'age-graded' and contrasted it with the so-called 'nepotistic' rank system found in rhesus and Japanese macaques. Hrdy & Hrdy suggested that female langurs are ranked according to their reproductive value, “individual reproductive value declines with age, and older female langurs defer to younger ones. Since such a system depends to a large extent on the compliance of low ranking animals, it is only expected to occur in groups composed of close relatives” (Hrdy & Hrdy, 1976).
Chapais & Schulman (1980) demonstrated that a genealogy-based rank system and a reproductive value-based rank system can be integrated into one explanatory model. Moore (1978) described a 'partially age-graded' rank system for Papio anubis in which some females outrank their mothers. Such a system seems intermediate between the langur and macaque hierarchies. Schulman & Chapais (1980) presented a functional explanation of rank relations among macaque sisters. Using demographic data they demonstrated that during the life history of macaque sisters, their relative reproductive values are reversed, and that the characteristic and permanent reversal in dominance rank among sisters, may result in part from mothers favoring daughters with the highest reproductive value.
Hausfater (1978) presented a unitary explanation of the various types of female dominance hierarchies found in non-human primates. His thesis is that the capacity of a female to maintain her position in dominance hierarchy varies across species primarily as a function of longevity and the rate of senescence. However, the factors affecting the capacity to maintain rank are not made explicit. Presbytis entellus would be a candidate species in which aging exerts a strong influence on dominance. Female langurs would thus be generally outranked by younger females. Since it is assumed that senescence in rhesus occurs at a more gradual rate, it would be expected that rhesus females would be capable of maintaining rank at a more advanced age. Baboons would be intermediate to langurs and rhesus, and only some females would be outranked by younger females. Thus, according to Hausfater's model, different life history traits give rise to different dominance systems.
This insightful model ignores one fundamental aspect of dominance, the phenomenon of coalitions and the whole pattern of agonistic support and aid found in many primates. Thus, the model does not take advantage of the powerful insights afforded by kin selection theory. An old female could retain her dominance position if she is aided by relatives who have an advantage in this relationship. The fact that dominance relations are more than dyadic power contests is well documented (Kawai, 1958; Kurland, 1977; de Waal, 1982). Rank of adult females is independent of body weight in langurs (Hrdy & Hrdy, 1976) and rhesus monkeys (Sade, 1967; Missakian, 1972). Recent studies (Kaplan, 1977, 1978; Kurland, 1977; Massey, 1977) have shown that macaques aid close relatives more than distant relatives, and distant relatives more than non-relatives in agonistic conflicts.
Furthermore, grandmothers aid granddaughters but not vice versa. Thus it appears that patterns of agonistic aid are in accord with the hypothesis that inclusive fitness is being optimized (Chapais & Schulman, 1980).
Numerous advantages and disadvantages of group living and sociality have been proposed (e.g., Alexander, 1974; Geist, 1974; Jarman, 1974; Treisman, 1975a,b; Hamilton, 1971; Rubenstein, 1978), and experimentally confirmed (Hoogland & Sherman, 1976; Andersson & Wiklund, 1978).
Chapais & Schulman (1980) assume that social rank differentiation is a consequence of selection at the individual level to meet, primarily, the disadvantageous contingencies of group living such as increased intraspecific interference competition (R.S. Miller, 1967). Summarizing 10 years of life history data from the La Parguera rhesus colony Drickamer (1974) states: “The most important feature which emerges from this investigation is the pervading effect of the social rank of the female. Higher ranking females had daughters that matured earlier, they had a higher percentage reproducing each year and their infants had a higher probability of surviving to the age of one year. In contrast, low-ranking females had daughters who matured at a later age, fewer reproduced each year and their infants had a significantly lower probability of survival. Also, females of low rank tended to have their infants later in the birth season than those of higher rank. If this fact is coupled with the lowered survival rates for infants born very late in the birth season, yet another factor supporting the importance of high rank emerges” (Drickamer, 1974).
Sade et al. (1977) partitioned the life table for the Cayo Santiago rhesus colony and found that high ranking matrilines were significantly more productive than low ranking matrilines. The factors responsible for these correlations in rhesus monkeys have yet to be determined. However in the case of Theropithecus gelada, Dunbar & Dunbar (1977) have shown that the correlation could be attributed to lower ranking females experiencing more infertile cycles than did higher ranking females. In the population they studied, dominant females were seen harassing lower ranking females during the estrus part of the latters' menstrual cycle: this could have disrupted their ovulatory cycles and so delayed conception.
The effect of dominance on the fitness of females could also be expressed through differences in mortality rates according to rank. This is suggested by observations made on Amboseli vervet monkeys by Wrangham (unpubl. data cited in Chapais & Schulman, 1980). Macaques are semi-terrestrial primates which live in multi-male groups. Males reach sexual maturity at 3 1/2 years of age, and generally migrate from their natal group by the time they attain 4 or 5 years. Adult males of a social group can be arranged in a linear dominance hierarchy based on the outcomes of agonistic interactions (Sade, 1967; Kaufmann, 1967). In dyadic agonistic interactions, adult males usually rank above adult females. The male hierarchy is unstable over time in part due to male migration during the mating season (Missakian, 1972). Unlike males, macaque females remain with their social group for life. The female dominance hierarchy is stable, and changes which occur follow a highly predictable sequence (Sade, 1972; Missakian, 1972).
Three rules are sufficient to describe the dominance system of female rhesus, M. mulatta (Sade, 1967; Missakian, 1972), and Japanese macaques, M. fuscata (Kawai, 1958; Kawamura, 1958; Koyama, 1967): (1) Daughters rank above all females to whom their mother is dominant, and rank below all females to whom their mother is subordinate. (2) Daughters do not outrank their mother. And (3), by the time a female reaches 4 years of age she rises in rank above older sisters. Thus as adults, rank is inversely correlated with age. The rank relation of any female dyad can be derived from these three principles. This pattern occurs with extraordinary empirical regularity (Sade, 1972; Missakian, 1972; Schulman & Chapais, 1980).
Chizakawa et al. (1979) describe a characteristic pattern of rank reversal in female rhesus. The daughter (not yet fully sized) would begin attacking her mother and the mother would fight back “by biting back or knocking down their daughters, but they never pressed the advantage, and the encounters ended quickly”. Later, the daughters would solicit aid from higher ranking females but the mother would not. The reversal in rank was thus probably influenced by apparent inhibitions of the mothers in fighting their daughters. This illustrates well a prediction made by Chapais & Schulman (1980): the mother, even if she has an intrinsic advantage in dominance, should accept subordinance instead of risking escalated conflict.

Evolutionary Analysis of Parent-Offspring Conflict
Trivers (1974) presented a model of parent and offspring interactions that suggested that antagonism between them might be a result of natural selection. His discussion is analogous to that of Hamilton (1972) on the spread of altruistic traits via kin selection. Trivers argued that parents and offspring have large areas of common interest in terms of individual fitness because they are highly related to each other (i.e., share substantial portions of their genomes). However, in any sexually reproducing species which is not completely inbred, parents and offspring are not genetically identical. Therefore the optimal strategies for maximizing the fitness of parent and offspring, respectively, may not be identical.
More specifically, Trivers argued that a parent is related equally to each of its offspring (r = 1/2, where r is the probability of an allele in the parent being identical by descent). However, an offspring is related to itself by r = 1 and to a full sibling by r = 1/2. In terms of fitness, therefore, an offspring values itself twice as much as a full sibling. Thus, Trivers suggested that selection on an offspring should favor the spread of alleles that skew parental investment (PI) toward itself at the expense of the parent's other offspring. From this analysis, selfish alleles would be expected to spread as long as the ratio of cost for full siblings to benefit for the selfish offspring was less than 2 to 1 (Stamps & Metcalf, 1980).
Conflict between parent and offspring has been viewed as a special case of kin selection (Trivers, 1974; Dawkins, 1976; Trivers & Hare, 1976; Fagen, 1976). Alexander (1974) raised serious objections to the concept of parent-offspring conflict. He argued that alleles causing selfish behavior of the type described by Trivers cannot spread because they lower the reproductive success of the selfish individual carrying them. The main point is contained in the following quotation from Alexander (1974): “Suppose that a juvenile... cause(s) an uneven distribution of parental benefits in its own favor, thereby reducing the mother's own overall reproduction. A gene which in this fashion improves an individual's fitness when it is a juvenile cannot fail to lower its fitness more when it is an adult, for such mutant genes will be present in an increased proportion of the mutant individual's offspring”. To paraphrase Alexander, selfish genes cannot spread because a selfish individual would have a lower reproductive success than would a nonselfish individual. The gain in extra parental resources going to a selfish animal as a juvenile would be insufficient to outweigh the net loss in reproductive success when it became a parent and had selfish offspring.
No less than three models have been independently developed to resolve the question raised by Alexander as to whether selfish genes could ever spread (Blick, 1977; Parker & MacNair, 1978; and Stamps, Metcalf & Krishnan, 1978). All three analyses are in agreement that alleles responsible for selfish behavior can spread to fixation. Blick and Stamps et al. make the further point that selfish alleles can spread despite the fact that they decrease the reproductive success of the selfish individual. This is because in mixed families of selfish and nonselfish individuals, the nonselfish offspring is always hurt more than the selfish offspring (Stamps & Metcalf, 1980).

Traditionally brood reduction has been viewed as an adaptive response of the parents to adjust their brood size to the food levels prevailing during the nesting period (Lack, 1954). When food is plentiful all of the offspring are raised; when food is scarce one nestling is selectively starved and dies, so that sufficient food is available for the survival of the remaining young. Hence brood reduction may be viewed as a strategy which increases parental reproductive success during periods of food shortage.
However, O'Connor (1978) suggested that brood reduction may be result of parent-offspring conflict, and that selfish offspring may cause brood reduction to the detriment of their parent's reproductive success. O'Connor's analysis assumes that the sacrifice of one offspring will reduce the food-related mortality of the surviving offspring. Since the parent is related equally to all of its offspring it will selectively starve one sibling only if the total number of surviving offspring is increased by this strategy. An offspring will value itself more than it will value a full sibling. Hence, in general, it will be more apt than the parent to sacrifice its sibling in times of moderate food shortage. In particular, O'Connor assumes that an offspring will sacrifice a sibling whenever its total inclusive fitness is increased, where its inclusive fitness is measured by the increase in the survivorship of the other surviving siblings multiplied by their coefficient of relationship to the selfish sibling, minus the loss in survivorship to the sacrificed sibling multiplied by its coefficient of relationship to the selfish sibling (Stamps & Metcalf, 1980).
A number of species lay two eggs but raise only one offspring. Brood reduction occurs by fratricide, with one offspring killing the other, usually within a few days of hatching (e.g., hawks (Meyburg, 1974; Skutch, 1976; Steyn, 1977), skuas (Doward, 1962; Young, 1963), cranes (Miller, 1973)). O'Connor interprets this as an example of parent-offspring conflict: the fratricide has a higher Inclusive fitness by eliminating its sibling than if it allowed it to survive, whereas the parent would have preferred to attempt to raise both offspring to adulthood.
Another possibility is that parents continue to lay an extra egg as insurance against inviability, loss, or damage. In any event, it is clear that conflict theory alone cannot explain why parents continue to lay two eggs when one offspring always seems to kill the other (Stamps & Metcalf, 1980).
On the other hand, several workers have suggested that the insurance hypothesis can explain this phenomenon without invoking parent-offspring conflict (Doward, 1962; Miller, 1973, Meyburg, 1974; Skutch, 1976). They have proposed that brood reduction by fratricide actually works to increase parental fitness and reproductive success. The idea is that parents can only raise one offspring, but lay two eggs as insurance. If both eggs hatch then the parent leaves it up to the most vigorous offspring to kill its sibling. Skutch (1976) noted that fratricide is most prevalent among predatory species, the offspring of which possess both the weapons and the behavior patterns necessary to make a quick and efficient kill.
While O'Connor (1978) emphasized conflict in avian species, a similar approach might prove rewarding in mammals. Simultaneous sibling aggression and rivalry have been reported in several species, the most notable example being pigs, which fight for preferred teats and wound each other with their sharp teeth (McBride, 1973; Fraser, 1975). The production of a 'runt of the litter' is a well-known phenomenon in mammals; Dawkins (1976) suggested that it might represent a parental strategy to channel offspring competition onto one, less viable victim.
Stamps, Metcalf & Krishnan (1978) have, in general, confirmed Trivers' analysis for cases in which offspring are produced sequentially, i.e., when each offspring is independent of its parents before the next is conceived.
Implicit in the analysis by Trivers is the assumption that costs are absorbed by the selfish offspring's sibling(s) randomly, without regard for their phenotypes. Metcalf, Stamps & Krishnan (1979) argued that under certain circumstances the costs of a selfish offspring's behavior will be absorbed unequally by selfish and non-selfish offspring without regard for their coefficient of relationship (r). Specifically, their analysis suggests that when multiple offspring are produced simultaneously, alleles causing selfish behavior may spread at cost/benefit (c/b) rations greater than those predicted by kin selection theory, i.e., alleles causing selfishness will spread when c/b > 1/r (where both b and c are measured in terms of offspring survivorship).
Hines & Maynard Smith (1979) extended the idea of an ESS to games between related individuals. There are two possible approaches. The 'personal fitness' approach, proposed by Grafen (1979), modifies the fitness of an individual by allowing for the fact that an individual is more likely than other members of the population to meet an opponent adopting the same strategy as himself. The 'inclusive fitness' approach adds to the payoff of an individual r times the payoff to his opponent, where r is the coefficient of relationship between them.
Hines & Maynard Smith regard the former method as correct, but show that if a strategy p meets the inclusive fitness criterion this is necessary but not sufficient to ensure that it meets the personal fitness criterion. If the vector p gives the probability distribution of policies in a mixed ESS, and p* the frequency distribution of pure strategists in a stable population playing the same game, p and p* are in general not the same if opponents are related, as they usually are if opponents are unrelated.


Evolution may be defined as any change in the organization of a system that persists for more than one generation. Those systems whose organization persists for more than one generation include the gene pool of a species, social systems, and ecological systems (ecosystems). Thus there are three general classes of evolutionary change: genetic change, which is the primary phenomenon in biological evolution; social change, the basic phenomenon in cultural evolution; and ecological change. Scott's (1981) concern is with the processes of evolutionary change at all levels, as applied to agonistic behavior.
Agonistic behavior, in turn, is defined as a behavioral system including those patterns of behavior that are primarily adaptive in situations involving conflict between members of the same species. It includes both aggressive and defensive fighting, escape, avoidance and freezing, as well as threats and other patterns of behavior used as signals in such interactions. It excludes predation, which involves interaction between two different species and is generally organized into a separate behavioral system in those species that show it.
The polysystemic theory of agonistic behavior has its origin in multi-factorial theory which states that causal factors affecting agonistic behavior could be demonstrated on every level of organization from the genetic to the ecological. Abundant evidence has been accumulated to support this assumption. The multi-factorial theory, however, is deficient in that it does not provide for interaction between factors at different levels. Polysystemic theory, on the other hand, assumes that living phenomena are organized in nested systems and subsystems, each corresponding to a level of organization, all related to each other and showing interaction between systems at different levels. Such interactions are in many cases reciprocal (Scott, 1975, 1981).
According to Scott, evidence has been gradually accumulating that destructively violent agonistic behavior is largely the result of disaggregation of systems at any level of organization. Thus he considers destructively violent agonistic behavior to be a more or less pathological phenomenon. There is a hierarchy of stability in systems, with genetic and physiological systems being the most stable, and social and ecological systems being less so. Consequently, disaggregation is most probable, and hence most important, in social and ecological systems.

The genetic theory of evolution in the context of polysystemic theory.
The Hardy-Weinberg principle states that, in an infinitely large population in which the individuals mate at random (panmictic population), the proportions in which any variable gene exists will remain constant, resulting from an intrinsic property of the chromosomal process of the transmission of heredity. This principle may be termed the constancy of gene frequency; thus from a genetic viewpoint the problem of evolution is to determine how changes in gene frequency are produced. The major classes of change mechanisms that act on variation to shift gene frequencies and thus produce differences between species are as follows: mutation, inbreeding and genetic drift, sexual reproduction, and selection.
Essentially, selection is any process that results in the differential survival of potential parents or, from the genetic viewpoint, differential survival of genes transmitted by such parents (Hamilton, 1964). Such differentials can be produced in two general ways: differential mortality and by unequal rates of reproduction. Such processes are generally conceived of as pressures that shift gene frequencies in one direction or another.
According to Scott, differential reproduction (usually called selection) can be produced by factors originating from any level of organization, including the genetic level itself and extending up to ecosystems and the physical environment. Conversely, it is theoretically possible for selection to act on any level of organization, even that of ecosystems (Lewontin, 1970). The relative importance of selection at these different levels is an arguable point, depending on the particular kind of character that is affected, but there is no logical reason why such multi-level effects cannot occur.
Of particular importance to the evolution of agonistic behavior, Scott insists, is selection at the social level. Social systems themselves may have different capacities for survival. Campbell (1975) has suggested that this is true of culturally based systems as well as biological ones; that selective survival of varying human cultural systems takes place. One obvious variable in such systems is the uses of and emphasis on agonistic behavior. Correlated with this, each social system exerts pressures on its component individuals that we may call social selection (Cf. the concept of 'social spoiling': van der Dennen, 1980).
Systems theory thus implies that selection pressures are very complex, not only because of their numerous sources, but because the sources themselves interact. Further, complexity arises because cultural evolution and ecosystem change occur independently of genetic change. Consequently, selection pressures originating from these system levels are constantly being altered.
In conclusion, the effect of polysystemic theory is to render the genetic theory of evolution much more complex as it transforms an essentially reductionistic theory that implies one-way causation from lower to higher levels into an interactive theory that assumes two-way causation between all levels. Thus social systems, often called mating systems by biologists such as Orians (1969), influence genetic systems by modifying both inbreeding and selection, as well as the reverse.
Likewise the phenomenon of differential survival, or selection, becomes enormously more complex, with interactive effects possible between all sources of selection differentials, and reciprocal effects between these sources and the systems affected. It follows that any attempt to explain evolution on the basis of some simple principle is bound to be inadequate.
According to S. Wright (1969), the most favorable condition for rapid evolution is one in which a species is divided into small semi-isolated populations which will tend to be moved apart both by inbreeding, thus producing random changes, and by differential selection, producing changes that are adaptive in nature. The reality of these processes has been confirmed by the field studies of Selander & Johnson (1973). These two processes are not mutually exclusive, and so can act simultaneously.
One of the general properties of living systems is that they become increasingly organized and as they do become more stable. Genetic systems are no exception. Subjected to selective pressures, they tend to develop resistance to them, a phenomenon which Lerner (1958) called genetic homeostasis. Consequently, a further factor contributing to rapid evolution is a situation in which the population is subjected to selection pressures against which it has not been previously buffered. This should occur when a population moves into a new environment and could account for the rapid changes seen in cases of adaptive radiation.
Behavior has two important characteristics that affect its interaction with evolution processes. First, behavior is actively adaptive; i.e., it varies according to the stimuli and the situations in which the animal finds itself. Therefore, behavior is a secondary source of variation which is largely independent of genetic variation. Second, behavior is learned; i.e. it is organized in response to experience. Again such organization is largely independent of genetic organization and further increases the possibilities for variation.
With respect to mutation and function of variation, adaptive behavior magnifies variation. Learned behavior further magnifies variation among individuals but tends to limit behavioral variation within an individual. Learned behavior can thus produce consistent variation among individuals that mimics genetic variation. Sexual reproduction is modified by sexual behavior. Finally, behavior interacts with the selection process. The behavior of a species can evolve in two directions. One is the evolution of fixed patterns of behavior, each of which is adaptive in one particular situation. The other is in the direction of a general capacity for adaptive behavior than can be used in a variety of situations. Most species among the higher animals actually follow both of these evolutionary courses. In most cases, the latter direction is superior in adaptive value, and one would expect that species that have evolved in this fashion would survive at the expense of others that were evolved in a simpler fashion. Learned behavior, on the other hand, has a tendency to interfere with selective survival. The animal that has learned an effective behavioral adaptation is far more likely to survive than a younger one which has not learned the response, even though the younger might have potentially greater capacities. We would expect then, that among animals that have learning capacities, the greatest mortality would take place among young animals, as is indeed the case.
Differential survival that has an evolutionary effect therefore should take place principally between older individuals that have developed their full capacities. Even here the survival of an individual may depend as much upon its previous opportunities to learn as upon its actual capacity for learning. The general effect of behavior is to buffer the individual against the forces of selection and to render their action much more indirect.
Social organization always involves two or more individuals and the behavioral interaction between them. This means that the expressed behavior is always affected by genes in at least two individuals. In social behavior, independent expression of gene effects is impossible, and this complicates the problem of genetic analysis (Scott, 1977).
Social organization has an effect upon behavior, and so does behavior have an effect upon social organization. This is one of the fundamental characteristics of a living system: mutual causation or feedback. A major consequence of this fact is that the resulting social system will determine which kinds of behavior are adaptive within that system and so act as a selective force. The effects of a social system on the four major processes of evolution are as follows. A social system has an effect similar to that of organized behavior on variation. It limits variation within itself but increases variation between systems. Within a social system, effects on variation are expressed through the creation of social roles, which standardize and limit the variation of the individuals that take on these roles. There may be a sharp differentiation between individuals in different roles and great similarities between those who have the same role. The simplest kind of role organization is that which assigns different roles to the two sexes.
With respect to sexual reproduction, a social system organizes sexual behavior into a mating system, again decreasing the random nature of sexual reproduction. Finally, the major adaptive function of social organization is to modify the environment in ways that are favorable to the survival of the individuals included within the system. Thus social systems modify and change the sources of selection pressure and incidentally introduce new ones favoring the survival of individuals who fit the system.
Scott derives some important inferences from this polysystemic theory of evolution. First, there are no 'genes for' particular characteristics. There is no gene for aggression, for peace, for altruism, or any other characteristic. If there were, it would be consistent with the theory of preformation, a theory which has been discredited in embryology for some centuries. Instead, a gene can modify the expression of behavior in various ways, including its frequency, intensity, and so on. Second, gene action is always expressed as part of a genetic system, and the action of a particular gene varies with the system involved. A gene never acts as an independent individual, although this sort of assumption has been made by some recent authors.
Furthermore, Scott asserts, the polysystemic theory rules out any hypothesis that assumes individual selection only. Such a theory, for example, might assume that an aggressive individual would be able to accumulate all of the resources available to a group and hence be more likely to survive. Attractive as this theory is on the face of it, no individual in a highly social species can survive long by itself, and such behavior is always tempered by its effects on a group. In any situation involving social behavior, selective survival always involves more than one individual; at this level the theoretical problem of individual versus group selection become purely academic.
Finally, Scott contends, agonistic behavior is primarily functional with respect to social systems and hence it should be most strongly affected by social selection pressures. For example, at the time when small, isolated human cultural groups were rapidly diverging from one another, different cultures should have selected different aspects of agonistic behavior, but today such differential selection is decreasing rather than increasing. In any case, effective selection always emphasizes survival and reproduction. There should always be, in either individuals or groups, selection against reckless bravery and death and in favor of cautious survival. Opposed to this tendency, a social group as a whole may benefit from possessing individuals who vary in their agonistic behavior. Extremely timid individuals may serve to sound a note of alarm, and bolder individual may stop to take a second look, both contributing something to the total system. Those who flee may contribute to the survival of a group, but so also may those who stand and fight. On balance, most evolutionary processes in man contribute to the increase of individual variation in agonistic behavior, although cultural uniformity should have some effect in the opposite direction.

Mutual deterrence
Most primitive peoples may be said to live in a condition of mutual deterrence, backed up by metaphysical (magic) and intermittent physical (war) means. Zagare (1985) examined the prototypical structure of the relationship of mutual deterrence and found that there is a natural congruence between this structure and that of Prisoners' Dilemma, a notorious game generally believed to impose suboptimal outcomes on those who play it. Embedding the theory of deterrence within a dynamic framework developed by Brams & Wittman (1981), he shows that a strategy supporting a Pareto-optimal status quo outcome is both long-term stable and farsightedly rational, provided that this outcome is the initial outcome and that both players have the ability to punish a defection from it by the other.
When one player is seen to prefer not to punish the other's unilateral defection from the status quo, deterrence is not stable. Curiously, when this preference is symmetrical and neither player is seen to prefer to punish the other's unilateral defection from the status quo, stable mutual deterrence is reestablished, although the stability of deterrence in this case depends on the ability of both players to pass through or endure mutual punishment, as in a limited war. Deterrence rests upon each player's fear of the other's destructive capability. Yet, too much fear is destabilizing.

Altruism Reexamined
It has been proposed that the avoidance of serious injury in most lion conflicts may not indicate the existence of groups adaptations or reflect altruism at all. Predator defense strategies employed by savanna-living baboons (Papio cynocephalus) also illustrate how a behavioral pattern once thought to be altruistic can be reevaluated by focusing on individual fitness rather than species survival. The defense of the troop has been described by DeVore & Washburn (1963): “Because they are in front of the troop by twenty to forty yards, the peripheral adult males are usually the first troop members to encounter a predator and give alarm calls. If a predator is sighted, all the adult males actively defend the troop. On one occasion we saw two dogs run up behind the troop, barking. The females and juveniles hurried ahead, but the males continued walking slowly. After a moment an irregular group of some twenty adult males was between the dogs and the rest of the troop. When a male turned on the dogs, they ran off. On another day we saw three cheetahs approach a troop of baboons. A single adult male stepped toward the cheetahs, gave a loud, defiant bark, and displayed his canine teeth; the cheetahs trotted away”.
Similarly, Stoltz & Saayman (1969) report: “When packs of dogs threatened troops... the large adult males immediately interposed between the troop and the attacking animals and it was not uncommon for a single dominant male to maim of kill three or four large dogs before retreating in the direction taken by the troop”.
The first researchers to study baboons extensively in their natural habitat explained the protective aggression of adult male baboons as group level adaptations existing to contribute to the survival of the troop and species (See DeVore & Washburn, 1963; DeVore, 1964; Hall, 1968). For instance: “The adult male baboon, with powerful muscles, large canine teeth, and a body weight twice that of the female, is specialized for defense of the troop... The most important function of the adult males... is protection of the other troop members from danger which originates outside the troop” (DeVore, 1964).
Possible fitness benefits to the protecting males themselves were not considered. Benefits for the troop as a unit, and ultimately for the species, were assumed to be the primary adaptive functions for which protective behaviors had evolved. And with the attention of early field researchers centered at levels above the individual, the altruistic nature of the male baboon's protective actions seems to have been largely taken for granted (Fry, 1980).
However, group selection and adaptation need not be evoked as evolutionary processes for explaining the function of behaviors like male baboon defense of the social group. As an alternative explanation, individual selection and inclusive fitness theory also account for male protection; natural selection simply may have favored males that protect their own offspring and other relatives within the troop. The inclusive fitness model holds that for a particular type of behavior to evolve, the benefits of the behavior (i.e., keeping close relatives alive) to the adult male's inclusive fitness must be greater than the costs (i.e., possible injury or death) to his inclusive fitness. Perhaps the risks of attacking wild dogs are not as great for a male baboon as is sometimes assumed, especially if other baboons join in the assault. This assertion gains support from the descriptions of dog-baboon and cheetah-baboon interaction quoted above (e.g. DeVore & Washburn, 1963; Stoltz & Saayman, 1969). Baboons are described dispersing, maiming, and even killing their adversaries. And from their own observations and the accounts in the literature, Altmann & Altmann (1970) report that only unsuccessful or incipient attacks by dogs and cheetahs have ever been witnessed on baboons.
Lions, however, with their large body size, powerful jaws, sharp teeth, and claws pose a more serious threat to baboons than do smaller dogs and cheetahs. “Lions... will put a baboon troop to flight” write DeVore & Washburn (1963). Fighting with a lion would probably not enhance a male baboon's inclusive fitness; fleeing appears to be the better option (Fry, 1980).
Additional support for the contention that male defense of other troop members is not truly altruistic but to the contrary may increase the protector's fitness can be gained by observing the effects of habitat on male responses. On the open savanna where baboons can usually see the nature of the predatory danger, the males flee from serious threats (e.g., lions) but stand their ground against less ferocious animals (e.g., dogs, cheetahs). A forested environment offers predators better opportunities to catch prey by surprise. When an alarm is first given, troop members may not know the nature of the danger. Under such circumstances, flight would seem to be the safest response even for dominant males. In her study of forest-living baboons, Rowell (1966) only rarely saw adult males protect the troop. “This deployment - fleeing troops forming a bodyguard of adult males between the source of danger and the rest of the troop - was seen on occasion, but only when the cause of alarm was so slight that the more confident adult males did not respond to something that set the juveniles running; a stronger stimulus produced precipitate flight, with the big males well to the front and the last animals usually the females carrying heavier babies”.
In the face of real or uncertain danger, male baboons become frightened and protect their own skin; however, if the threat is slight, they may simultaneously protect relatives and others within the troop.
This foregoing reevaluation of baboon behavior as beneficial to the animals actually engaging in the protecting parallels other reinterpretations of presumed altruism such as the regulation of clutch or litter size supposedly for 'population control', warning calls, predator distraction displays, and helping behavior (Dawkins, 1976; Lack, 1958; Sherman, 1977; Stern, 1970; Trivers, 1971, 1974; Wiens, 1966; Woolfenden, 1975). Indeed, among nonhuman species, behaviors that benefit individuals other than the actor or benefit the group as a whole and cannot be shown also to benefit the actor are either very few and far between or entirely unknown (Dawkins, 1976; Lewontin, 1970; G.C. Williams, 1966; E.O. Wilson, 1975). Thus at least among nonhuman species, most 'altruism' upon closer examination might more fittingly be called 'pseudo-altruism' (Fry, 1980).

If behaviors can be shown to be advantageous to the individual, as well as to the group, we are faced with at least two types of competing hypotheses for explaining the evolution and ultimate function of the behavior: Individual-kin selection on the one hand and various types of group selection on the other.
While remembering that selection at different levels may not be totally mutually exclusive, it also should be recognized that individuals are the units of reproduction. Groups, populations, species, and ecological communities are not. The mere fact that a population or species survives through many successive generations is not sufficient evidence that population level adaptations (or species level adaptations) actually exist. Group or species survival may be merely an incidental reflection of the survival of well-adapted individuals; the distinction must be made between well-adapted populations and populations of well-adapted individuals (G.C.Williams, 1966).
Williams argues that when an adaptation can be accounted for at both the level of the individual and at the group level, it is not simply a matter of taste as to what level of explanation to apply. He argues that Occam's razor must be employed; that is, adaptations must not be recognized at higher levels than are really necessary to explain the behavior.
E.O. Wilson (1975) criticizes what he sees as Williams' overemphasis on parsimony as a criterion for choosing individual level accounts over group level explanations. “The goal of investigation should not be to advocate the simplest explanation, but rather to enumerate all of the possible explanations, improbable as well as likely, and then to devise tests to eliminate some of them”. Although in the practice of science it is impossible to list 'all possible explanations', in principle Wilson's point is well taken. We should examine instances of animal aggression and attempt to determine if individual animals by and large are behaving for the good of their social group or species or to enhance personal inclusive fitness (Fry, 1980).


Rapid evolution is impossible unless selection is constantly 'presented' with a broad array of variation form which to 'select'. S. Wright (1931 et seq.) has described a 'variation-generation' system which meets these requirements. By 'genetic drift' he means more than the narrow conception of the term described in textbooks (and sometimes called the 'Sewall Wright effect') which in this narrow sense refers only to random changes in gene frequency due to small population size. For Wright himself genetic drift has a larger meaning: any and all factors, including inbreeding and selection, which promote genetic divergence between small component groups within a larger population. Any species which is characteristically subdivided and persistently re-divided into small, relatively isolated groups of a few hundred individuals, will become very diverse genetically - even in an otherwise stable environment. Such a species will carry great potential for rapid evolution, and if subjected to a very strong selective force, this potential will be realized. Anthropological and ethological evidence suggests that such a population structure was characteristic of australopithecines, Homo erectus and Homo sapiens during the Pleistocene (Bigelow, 1971).
Mutation is the ultimate source of genetic variation, but gene mutation alone cannot, in higher organisms, provide sufficient variation for really rapid evolution. Sexual recombination enormously increases variation, by producing, at every generation, a new array of unique gene combinations. When a species is divided into many small, relatively isolated groups, inbreeding and random changes in gene frequency increase variation in yet another dimension. Inbreeding leads to genetic uniformity within a given group, but when separately applied to many different groups within a species, the variation of the species as a whole is greatly increased.
Certain levels of gene flow (migration) between groups ensure enough outbreeding to counteract the harmful effects of excessive inbreeding, and may enhance rather than eliminate the intergroup genetic variation. Prehistoric men were probably extremely variable genetically and selection has always had plenty from which to select (Bigelow, 1971).
S. Wright (1931) studied how the speed of evolution is related to the ratio between the mutation rate and the size of the interactive population. When the mutation rate is low and the size of population is large, the interbreeding between mutants and 'normal' individuals, or between mutants of opposite directions, tend to cancel out the mutations, and the speed of evolution is very slow.
When the mutation rate is very high and the size of population is small, there is much inbreeding effect which tends to amplify the mutations rapidly, and evolution may take place so fast that the newly evolved species has no time to adapt to the environment or to seek an appropriate new environment, no time to work out new types of relationships between individuals, and the whole species may become extinct.
When the mutation rate is moderate and the population size is neither too large nor too small, a mixture of stabilization and change takes place, and the phenomenon of random drift occurs; a change may be amplified for a while, stabilization takes place, another change in a new direction takes place, etc. A faster rate of evolution occurs when the total population is subdivided into semi-isolated colonies, connected with occasional interbreeding (Bigelow, 1971).

Genetic Tracking and Culture
It is true that we do not inherit a particular language or a set of customs and attitudes. In the same sense a mammal does not inherit knowledge of its parent's home range. But it does not automatically follow that dominant traits of cultures are never genetically fixed through selection of genotypes conforming to the prevailing culture. The process whereby a trait ('phenotype') is induced first by environmental influences and later fixed by selection of the genes most prone to exhibit the trait is called genetic assimilation (E.O. Wilson, 1970, 1975). The key to the genetic assimilation of given cultural traits lies in their heritability. The speed with which a trait is evolving in a population increases as the product of its heritability and the intensity of the selection process. In other words the susceptibility of human behavioral characteristics to evolution depends on the amount of genetic material there is to work on in a given population and the intensity with which certain genetic types are favored in the process of differential reproduction. There exists abundant evidence that many of the most important measurable behavioral characteristics do possess a moderately high heritability.
E.O. Wilson (1970) has provided impressive evidence that strong selection at the level of single genes can occur under moderate selection pressure in ten generations or less. Such evolution is not only well accommodated by genetic theory but it is a commonplace in experimental animal populations. And second, that moderate behavioral evolution can occur at rates at least comparable to the process of ordinary species formation. There is thus every justification from both genetic theory and experiments on animal species to suppose that rapid behavioral evolution is at least a possibility in man. By rapid Wilson means significant alteration in say, emotional and intellectual traits within no more than ten generations - or about three hundred years. The changes could affect genes shared by a large proportion of the population.
Moreover, the necessary conditions are present for such swift microevolution. It has already been mentioned that many measurable behavioral characteristics in man have been demonstrated to have moderate degrees of heritability. There is also overwhelming evidence for the existence of the second prerequisite for continuing evolution, namely varying reproductive performance among different families and family groups within the same society. This is especially characteristic of primitive and impoverished societies (Spuhler, 1959; Stern, 1960; Dunn, 1962; Mayr, 1963).
Finally, some cultures evolve slowly enough so that their dominant features remain essentially unchanged for hundreds of years, long enough, it would seem, to permit the occurrence of some amount of genetic assimilation. Man therefore has the genetic capacity to track some of the dominant features of particular cultures. Whether he does so - and to what degree - remains an open question.
What relevance does this 'genetic tracking' have to human aggression? For one thing: “Suppose it were true that fifteen million or more years ago out australopithecine ancestors evolved into carnivores who hunted in packs, that these creatures were highly aggressive and territorial, and that their habits persisted to the dawn of agricultural societies around five thousand years B.C. Even if Dart's (1953) most sanguinary imaginings were thus momentarily accepted for the sake of argument, would it follow that australopithecine traits persisted as a genetic legacy into civilized times? The answer is no. If there were selective pressures in other directions, time enough has elapsed since the appearance of the first agricultural and urban societies, and even since the fall of Rome, for the evolution of some behavioral traits to have reversed itself many times over. Given even small amounts of heritability and selection pressure, both of which are indicated by our rudimentary understanding of behavioral genetics in man, it seems inevitable that man makes himself genetically as he goes along. If aggressive behavior under stress has a genetic basis, it is likely to be due less to Pleistocene genetic inertia than to the fact that such behavior has continued to be adaptive into modern times” (E.O. Wilson, 1970).
Furthermore, the assumption of homogeneity in gene pool composition (i.e., the assumption that all mankind has exactly the same balance of heredity abilities and drives) is neither necessary nor probable (Cf. Freedman, 1964). And finally: “it is not enough to point to the absence of a behavioral trait in one or a few societies as conclusive evidence that the trait is environmentally induced and has no genetic disposition in man. The difference could still be genetic and, in fact, only recently evolved” (E.O. Wilson, 1970).

Culture and Fitness
Unfortunately, most biological theories and models to date have been designed exclusively for the selective retention of genetically programmed characteristics. Since phenotypes are also very much a product of learning in many species (particularly in Homo sapiens), there is an urgent need to expand our models to accommodate the retention of traits whether based on chemical instructions genetically inherited at conception, or an accumulated 'wisdom' passed along sometimes continuously and from many 'parents'.
The case for an integrated theory of biological and cultural evolution is even stronger if the benefits of culture can often be measured in the same terms and at the same level as the usual benefits of biology. Accordingly, we must ask whether the adaptive significance of most cultural practices, too, lies in their contribution to the inclusive fitness of individual human beings. In theory, there are good reasons why this may be the case. If it is true that culture is a 'superorganic extension' of human adaptation, as has so often been claimed, we might well expect the selective retention of learned characteristics to complement genetic selection. Unless it can be shown that cultural evolution has somehow run consistently counter to the trends of genetic selection (an unlikely proposition, see Alexander, 1971), it is not likely that individual humans generally act in opposition to their own inclusive fitness (Durham, 1976a,b; 1979).
On the contrary, it is reasonable to suggest that people tend to select and retain from competing variants those cultural practices whose net phenotypic effect most enhances their individual inclusive fitness. This kind of selective retention ('cultural selection') would be operationally independent of natural selection; that is, widespread phenotypic changes in a population could occur within each generation and prior to appreciable changes in gene frequency. At the same time, it would be functionally complementary to natural selection. Indeed, the human 'capacity for culture' is commonly thought to be a biological attribute and therefore a product of organic evolution (Spuhler, 1959; Dobzhansky, 1961; Caspari, 1963).
If so, then this capacity most probably contributed, like other biological traits, to the ability of individuals to survive and reproduce in a given environment. Over the course of human evolution, natural selection would have favored the capacity for the continuing selection of biologically adaptive cultural traits even when the reproductive consequences were not consciously recognized. If it be assumed, then, that this capability is not a very recent product of organic evolution, the result of past genetic selection could be a bias for genotypically selfish selectivity in the retention of cultural attributes (Durham, 1976a,b; 1979).
By this hypothetical mechanism, an important source of culture change would be the selective retention by individuals of traits that maximize their ability to survive and reproduce in a given environment. Accordingly, the rate of cultural evolution would depend upon the collective results of individual discrimination, in turn a function of the rate of dissemination, the expected magnitude of impact on fitness, and the variance in relative advantage among the alternatives. It is necessary, of course, that there be a source of variation, but for the purposes of this argument we need not distinguish diffusion, invention, or accident. In this way the cultural adaptation characterizing a particular social group would come to consist of those traits, including behavior, that past experience had shown to be the most advantageous for the survival and reproduction of individuals. This selective retention would be specific to features of the natural and social environment, including properties of available resources and characteristics of any neighboring or sympatric social groups. Existing cultural practices would also be a part of the environmental influences on the further evolution of culture. In predictable environments, fitness-enhancing traits would tend to be retained in a cumulative fashion from generation to generation. There would be a significant advantage to carefully educating and socializing offspring. Traditions and rituals would become, by analogy, the 'supergenes' of cultural evolution - elaborate behavioral sequences linked together and inherited as a package. Competing variations that tend to reduce the relative reproductive success of individuals would eventually disappear.
In this manner, the accumulated wisdom of past generations would allow a closer fit of behavior and environmental conditions. In such a system, genotypically selfish memory of past experience is not limited to nucleotide sequences (Durham, 1976a,b; 1979).
The process outlined by Durham is similar to other theories of cultural evolution by selective retention (e.g., Murdock, 1956; Sahlins & Service, 1960; Campbell, 1965; 1975; Flannery, 1972; Ruyle, 1973; Cloak, 1975; Dawkins, 1976; Richerson & Boyd, 1978) except for the present emphasis on the role of individuals within a society in the process of change and on the role of inclusive fitness as the criterion of selection (See also LeVine, 1973).
This argument is not meant to deny or discredit the evidence for the spread of certain cultural traits by 'intersocietal selection' (for example, see Davie, 1929; Carneiro, 1970a,b; or Harris, 1971). As with organic evolution, the important question concerns not the possibility of selection at higher levels, but the relative effectiveness and direction of selection at those levels. Group-level selection through warfare, for example, could easily reinforce and accelerate the trends of the individual-level process described above, particularly if there is a tendency for individuals to cooperate best in order to fight hardest and longest when their own personal interests are at stake. (This hypothesis has now been advanced by a number of authors: Keith, 1949; Alexander & Tinkle, 1968; Bigelow, 1969; 1972; Alexander, 1974; E.O. Wilson, 1973a; 1975).
On the other hand, there are circumstances where a high rate of extinction among small and independent groups of closely related individuals could conceivably counteract trends in individual selection (Darlington, 1972; E.O. Wilson, 1975).
In general, though, considering both the rapidity with which individual-level cultural selection may proceed, and the fact that an interindividual process must actually generate the intergroup variability prerequisite for group selection, there is reason to believe that cultural forms of group-selected genotypic altruism may be rare, though perhaps important in special cases.
In short, Durham proposes the fundamental hypothesis that humans generally behave in ways that maximize the propagation of their genes. For convenience, this can be called the hypothesis of genotypic selfishness.
Cultural practices produced in this way by individual-level selection (whether or not reinforced at the group level) would complement biological adaptations produced by natural selection. Indeed, these cultural and biological influences could often be confounded in the phenotype. Consequently, a general theory for human evolution would do well to allow for both cultural and genetic processes of selective retention, particularly in reference to human social behavior. Durham therefore proposes the term 'Selection' (with a capital S) to refer to selective retention by the individual fitness criterion, regardless of whether the predominant process is a variety of cultural or natural selection (Durham, 1976a; 1979).

Please send your reactions, suggestions and personal URL to