Individual humans, and members of diverse other species, show consistent differences in aggressiveness, shyness, sociability and activity. Such intraspecific differences in behaviour have been widely assumed to be non-adaptive variation surrounding (possibly) adaptive population-average behaviour. Nevertheless, in keeping with recent calls to apply Darwinian reasoning to ever-finer scales of biological variation, we sketch the fundamentals of an adaptive theory of consistent individual differences in behaviour. Our thesis is based on the notion that such ‘personality differences’ can be selected for if fitness payoffs are dependent on both the frequencies with which competing strategies are played and an individual's behavioural history. To this end, we review existing models that illustrate this and propose a game theoretic approach to analyzing personality differences that is both dynamic and state-dependent. Our motivation is to provide insights into the evolution and maintenance of an apparently common animal trait: personality, which has far reaching ecological and evolutionary implications.
Ever since Darwin, biologists have used natural (and sexual) selection to explain biological diversity. Initially, adaptive arguments were restricted to explaining differences between genera and higher level taxa. By the 1960s, however, it was no longer controversial to discuss differences between closely related species and sub-species in such terms (e.g. Lack 1961). Moreover, by the 1980s, the concept of adaptation to local conditions by different populations of the same species had been accepted widely (e.g. Endler 1986). More recently there have been calls to take this progression a step further, for evolutionary biologists to ‘adopt a general set of expectations about adaptive individual differences within single populations that can be applied to all traits’ (Wilson 1998, p. 199). However, while adaptive individual differences have been recognized in particular instances (e.g. mimicry polymorphisms: Gilbert 1983; alternative male-mating tactics: Gross 1996; social foraging roles: Giraldeau & Caraco 2000; brood size differences: Morris 1998), intraspecific differences in behaviour are still widely assumed to be non-adaptive variation surrounding (possibly) adaptive averages. Nevertheless, interindividual variation in behaviour is often distributed in a non-random manner, along particular axes (Gosling & John 1999; Gosling 2001), suggesting that it is likely to have consistent ecological and evolutionary consequences and hence be a focus for selection.
From an adaptive perspective, it makes sense for individuals to adjust their behaviour according to current conditions (including their internal state), and this can result in individual differences in behaviour if there is between-individual variance in local conditions. More problematic, from this perspective, is how to make sense of individual differences in behaviour that are consistent over time (within or across generations) and/or different contexts. In some cases, fundamental biological factors, such as age or sex, are responsible for generating such correlations in behaviour. However, a complete theory of consistent individual differences requires an understanding of the evolution of behavioural correlations over time and context in general, including cases where such obvious biological constraints are not apparent (Sih et al. 2004). In fact, people of all ages and sexes differ consistently in their tendencies to be neurotic, agreeable, extravert, open and conscientious (Gosling & John 1999). Moreover, psychologists and behavioural biologists are becoming increasingly aware that consistent individual differences in behaviour are not restricted to humans. Indeed, it is clear that such differences exist in a wide range of contexts and species, from primates to insects and cephalopod molluscs (Gosling 2001; Sih et al. 2004). For example, individuals of a number of species often differ in their aggressiveness (Huntingford 1982; Riechert & Hedrick 1993); although individuals will alter their aggression levels in a context-dependent manner, some are consistently more aggressive than others. Similar consistencies have been observed for activity (Henderson 1986; Sih et al. 2003), exploration (Dingemanse et al. 2002), risk-taking (Wilson et al. 1994; Fraser et al. 2001), fearfulness (Boissy 1995) and reactivity (Koolhaas et al. 1997). In humans, these differences are referred to as personality types (Pervin & John 1999), while in non-human animals terms such as coping styles, behavioural tendencies, strategies, syndromes, axes, or constructs are used (Gosling 2001; Sih et al. 2004); we will use the term personality differences henceforth when referring to consistent individual differences in behaviour, in time and/or across contexts, for both human and non-human animals.
Regardless of how they are maintained, personality differences have significant evolutionary and ecological consequences. Indeed, as a number of models now indicate, maintenance of different personality types may favour the evolution of cooperation (e.g. Crowley & Sargent 1996; Lotem et al. 1999; Fishman et al. 2001; McNamara et al. 2004). Moreover, accounting for the maintenance of genetic variability in the face of selection is an outstanding issue in evolutionary biology, and understanding the maintenance of heritable personality differences will shed light on this general problem. In addition, the presence of personality differences may also affect how animal populations respond to change. If personality differences evolve jointly with morphology and other traits, they can facilitate speciation and the conservation of adaptive change (Wcislo 1989; Wilson 1998). Furthermore, for many species, environmental change as a result of anthropogenic influences is likely to be a major factor in shaping their future. Therefore, a key issue in conservation biology is how well species can respond to such change, and the presence of different personality types in populations will determine species’ ability to persist. For instance, bold, neophillic individuals may be able to locate new resources if the traditional resource base fails, or aggressive individuals may be better at competing for resources as they become more limiting, thereby enabling local populations to persist under anthropogenic challenges (see Sih et al. 2004 for a detailed discussion).
Despite their apparent ubiquity in a wide range of species and their potential evolutionary and ecological importance, the evolution of consistent individual differences in behaviour is poorly understood. Here we focus our discussion on the conditions under which personality differences will be favoured by selection and sketch the fundamentals of an adaptive theory of personality. To this end, we review existing theory that illustrates our thesis. We elaborate the notion that personality differences can be selected for if the fitness payoffs of the actions available to individuals are dependent on both the frequencies with which they are performed, and the behavioural history of individuals. The former suggests the application of evolutionary game theory (Maynard Smith 1982), which can often predict outcomes that run counter to simple intuition (e.g. Clark & Mangel 1984). In addition, such logic suggests that predicting what an animal should do in any given instance means keeping track of a wide range of variables (e.g. an individual's state, as well as those of its competitors). Given the complexity of this task, solving for optimal (evolutionarily stable) behaviour is unlikely to be possible analytically, and will involve numerical techniques such as dynamic programming (Houston & McNamara 1999; Clark & Mangel 2000).
Frequency dependent selection and personality differences
Since the advent of evolutionary game theory (Maynard Smith 1982), it is well known that making fitness payoffs dependent on the frequencies that competing strategies are played (frequency dependent selection) can lead to the stable coexistence of different behavioural ‘types’ in populations. For example, in the basic Hawk–Dove game of competition over resources of value v, given that getting into an escalated fight costs c > v, the evolutionarily stable strategy (ESS) is for a proportion v/c of individuals to play Hawk (always escalate if challenged) while 1 − (v/c) play Dove (always capitulate without fighting) at any given time (Maynard Smith 1982). There are, however, two ways in which the ESS mixture of tactics in a population can be maintained by frequency dependent selection: each individual can perform actions randomly with fixed probabilities and thus generate the predicted mix of strategies in large populations, or fixed proportions of individuals can play each strategy consistently. Only the latter scenario would generate personality differences. The conditions favouring the evolution of one or other of these forms of mixed evolutionarily stable strategies have yet to be elucidated in general, although investigating the evolutionary dynamics of biological games (Nowak & Sigmund 2004) may generate insights. For instance, adding the possibility of ‘eavesdropping’ (basing tactics on the outcome of an opponent's last interaction: Johnstone 2001) to the Hawk–Dove game with replicator dynamics selects for consistent individual differences in aggression in monomorphic populations in which all individuals play Hawk and Dove randomly at ESS probabilities. This is because, with eavesdropping (only escalate against losers) in the population, more consistent aggressiveness (high or low) is favoured since, by being more predicable, individuals can avoid getting into extended (costly) fights. Moreover, with increased interindividual variation in aggressiveness, increased levels of eavesdropping will be favoured to minimize the chance of fighting with the more aggressive individuals (who are more likely to have won their last fight), and so on. This dynamic feedback will eventually result in polymorphic populations that are composed of extreme types at ESS frequencies, in which individuals are always either Hawks, Doves or Eavesdroppers (R.A. Johnstone & S.R.X. Dall, unpublished data). Thus, consistency can be selected for when being predictable gets competitors to respond in the future so as to improve focal individuals’ payoffs. Further work is needed to establish whether this is common whenever outcomes of social interactions can be observed by others and inform future interactions; for instance, when animals signal in communication networks (McGregor 1993) or if individual ‘public image’ or ‘reputation’ is constantly on the line in games involving a choice between cooperation and defection (Nowak & Sigmund 1998; Leimar & Hammerstein 2001). Indeed, Fishman (2003) demonstrates that when individuals cannot always afford to help others, cooperation may involve a stable mixture of two types of individual: indiscriminate (consistent) altruists that always try to help and discriminating altruists that only try to help individuals that helped in their preceding interaction. This is because the long-term stability of cooperation requires the occasional forgiving (albeit incidental) of inadvertent failures to donate help, which would otherwise elicit mutually detrimental tit-for-tat-style dynamics and the evolution of universal defection.
However, although there is evidence, from both the human (Bouchard & Loehlin 2001) and non-human literature (e.g. Koolhaas et al. 1997; Drent et al. 2003), that personality traits often have significant heritable components, selection for a mix of fixed pure strategists over monomorphic mixed strategists is not the only potential adaptive route to animal personality differences. Indeed, it has been suggested that monomorphic but probabilistic mixed strategies are rare in nature when individuals can employ more than one tactic over their lifetimes (e.g. Gross 1996). Instead, individuals can switch tactically between alternative strategies according to the behaviour of others, their own condition and current ecological circumstances. This offers another, more commonly observed route to maintaining a mix of types in a population (see Gross 1996 for a discussion). Furthermore, as we now describe, such conditional, state-dependent behaviour can generate consistent differences in behaviour between individuals that are genetically identical.
State-dependency and behavioural specialization
At any given instant, it is possible to characterize an animal by its state; the variables that make up this quantity may represent features of an organism such as its size or energy reserves, or aspects of the environment, like its territory size or the recent behaviour of other animals. Crucially, an organism's state limits the actions it can perform, and influences the costs and benefits of those options that are available to it. For instance, when an animal's energy stores are low, it may not be able to invest in developing its reproductive organs, and hence the option to reproduce is unavailable. Moreover, if its reserves are so low that an animal is unlikely to survive an expected lean spell, the costs of exposing itself to predators and other dangers can be outweighed by the potential benefits of obtaining food, which would not be the case if the animal were in better condition. In addition to the importance of state in isolation, the sequence in which an organism performs its actions over time is likely to be significant; the current action may influence future state and hence future decisions. These issues can be explored using dynamic state variable models, which have proven useful in understanding the tradeoffs involved with energy reserve management in general, including the influences of food supply, metabolic costs, predation risk and social interactions (see Houston & McNamara 1999; Clark & Mangel 2000 for reviews).
The state-dependent dynamic programming approach, with payoffs to actions also depending on the frequencies with which interacting conspecifics adopt actions, will be useful for characterizing conditions under which otherwise identical individuals can become locked into different subsets of state-space, and hence develop distinct behavioural tendencies. For instance, Rands et al. (2003) model a pair of foragers that must choose between remaining in a refuge where they are safe from predators but cannot feed, and emerging to look for food under the risk of predation. In such circumstances an individual should rest safely in the refuge until its reserves fall below a certain threshold level when it should emerge to forage. The higher the risk from predators the lower this threshold should be (Houston et al. 1993). Rands et al. (2003) consider two foragers with initial energy reserves that differ stochastically. They demonstrate that, although the animal with lower reserves will leave the refuge first, the other individual will soon follow because it is safer to forage as a pair than alone. Since the animals forage together, they will enjoy similar foraging success and so the individual with higher reserves will exceed its (revised) threshold first and return to the refuge. This makes foraging more dangerous for the other animal (reducing its threshold level of reserves), which will then return to the refuge. Consequently this animal still has the lower reserves and will emerge to forage earlier next time, and so on. This feedback between experience and optimal behaviour, whereby stochastic initial differences between otherwise identical foragers are maintained, can thus result in individuals that differ consistently in their tendency to take risks: in this case to leave a refuge first and return last (Fig. 1). Dominance hierarchies that emerge from repeated competitive interactions between the members of a stable group are another potential example of consistent differences between individuals that can arise as a result of history; individuals with a history of losing become increasing likely to get into escalated fights if they challenge and so it ceases to be worth their while attempting to do so (e.g. Van Doorn et al. 2003a,b).
Future work exploring the conditions under which the feedback discussed above will generate consistent differences among individuals should be focussed on contexts related to axes along which between-individual behaviour is typically observed to vary consistently: exploration, risk taking, aggressiveness and mating tactics (Gosling 2001). In addition, it will also be important to build models that allow cross-context consistencies to be explored since, for instance, highly aggressive individuals often take more risks (Sih et al. 2004). Furthermore, it will also be important to consider strategic alternatives that are not discrete (e.g. forage versus rest) since consistent behavioural differences between individuals can be distributed continuously within populations (e.g. Dingemanse et al. 2002, 2003; Drent et al. 2003). As well as exploring the conditions favouring personality differences systematically, such work will illustrate the importance of ‘de-atomizing’ behavioural ecology as a discipline (Sih et al. 2004); animal behaviour at any given time and in any given context must be considered functionally as a ‘snapshot’ of an entire life-history strategy.
Costs to flexibility
Finally, personality differences may also simply emerge if there is variation in genetically determined behaviour within populations and behavioural flexibility is not always beneficial. From an adaptive perspective, within-population genetic variation can persist for a variety of reasons, including frequency dependent selection (Roff 1998), a balance between mutation and (weak) selection (Santiago 1998) and behavioural optima that vary in space or time (e.g. Mangel 1991). Under such conditions a fundamental question arises for the evolution of heritable adaptive personality differences, as well as for the evolution and maintenance of phenotypic plasticity in general (see DeWitt et al. 1998 for a discussion): when should unconditional pure strategies be selected for over those that are conditional or state dependent? The key issue here may be that flexibility is likely to be costly in a world that changes continuously regardless of individual actions and behaviour. This is because responding adaptively (behaviourally, physiologically or morphologically) to any portion of such an environment is likely to improve the accuracy of expectations about local conditions but increase uncertainty about the rest of the environment (i.e. information about alternative environmental states is constantly deteriorating while they are not being experienced: Dall & Cuthill 1997). This will increase the chances of responding inappropriately, or taking longer to respond appropriately, under alternative conditions that will not have been experienced as recently. Error-rate costs to flexibility can select for developmental canalization (Tufto 2000; Sih et al. 2004), and, along with potential response-time lags, may mean that flexibility should always be somewhat costly in systems that respond to their environments, such as biological systems with epigenetic development and learning. Formalizing this notion will allow the circumstances under which (inherited) unconditional strategies are selected for to be elucidated.
Furthermore, costs to flexibility in behavioural biology are likely to limit tactical responding to every eventuality even when conditional strategies are selected for. This may select for more strategic behaviour, or rules that perform well across subsets of the circumstances that individuals can expect to encounter. Indeed, various classes of outcome may be treated in the same way. For example, it is conceivable that an animal could respond differently to each threatening situation it faces. Instead, many animals seem to have a general state of fear that can be induced by many circumstances. In this sense, an emotional state is an internal state of an organism that is a generalized response to stimuli in a class of related stimuli (Leimar 1997; McNamara & Houston 2002). Rolls (1999) argues that emotional states make it easier for animals to perform appropriate behaviour in response to a stimulus. Depending on the rules governing the dynamics of such states and how they translate into actions, this strategic, rule-based behaviour sets the stage for the kind of frequency and state dependence of payoffs that can select for personality differences. For example, McNamara & Houston (2002) consider a situation in which an individual can either be trustworthy or untrustworthy. They show that if behaviour is completely flexible then all individuals should be untrustworthy. In contrast, if individuals are consistently either trustworthy or untrustworthy then these two types of personality may persist at a frequency-dependent equilibrium.
A behavioural ecology of personality?
Overall then, we contend that consistent individual differences in behaviour – personality differences – can be selected for when payoffs to actions depend both upon the frequency with which they are employed in a population and upon the behavioural histories of individuals. An animal's behavioural history, which can be subject to stochastic influences, determines its current state. In some cases the crucial component of state is an aspect of an individual's physiology (e.g. Rands et al. 2003). In others it is based on an individual's reputation; consistency can also be selected for when behaving predictably gets competitors to respond in the future to improve a focal individual's payoff, as illustrated when eavesdropping is possible in the Hawk–Dove game (R.A. Johnstone & S.R.X. Dall, unpublished data). Furthermore, costs to flexibility together with frequency-dependence can maintain consistent individual differences in behaviour. A challenge in behavioural ecology is to determine when the selection regimes discussed here occur, and thus establish the degree to which observed personality differences (Gosling 2001; Sih et al. 2004) are adaptive, and not merely ‘noise’ around population-average adaptive behaviour.
Thanks to Rufus Johnstone for extensive discussions. We are also grateful to Andy Sih and two anonymous referees who greatly improved the final version of this paper. SRXD was supported by grants to I. L. Boyd and T. H. Clutton-Brock (AFI1/03) and R. A. Johnstone (NER/A/S/2002/00898) by the Natural Environment Research Council (U.K.).