2. Criteria for animal personality

What observational criteria could be used to infer personality in nonhuman animal species? As Table 1 already shows, quite a variety of approaches have been used, generally requiring substantial interpretation by human observers. The settings vary from the laboratory to zoos to naturalistic environments and immediately raise two critical questions. First, how should behaviors be classified or coded? Second, how much observational data do we need to do this? These two challenges are crucial to identify ethologically meaningful and generalizable patterns in behaviors whose regularity across time is the basis for inferring personality traits. These challenges have been discussed at length by Uher (Reference Uher2008). Building on this discussion, we would emphasize three critical ingredients: collect as complete a dataset over as extended a time as possible; code the data by expert consensus in light of the best knowledge that we have of the animal’s ethology; and collect the data in rich, naturalistic settings. The last point raises a further issue: context dependency.

Table 1. A nonexhaustive list of research on animal personality after 2001. See Gosling (Reference Gosling2001) for a list of research on animal personality before 2001. See Kralj-Fiser and Schuett (Reference Kralj-Fišer and Schuett2014) for a list of additional invertebrate studies

The issue of context dependency offers some valuable criteria for animal personality. Personality constructs need to refer to individual differences that generalize to some extent across time, that is, they should be temporally stable. On the other hand, we would argue that the behaviors from which we can infer the personality constructs need to be context dependent (context sensitive) to some degree (again, contrary to some views in Kaiser & Müller’s Reference Kaiser and Müller(2021) work that have instead stressed contextual consistency). In humans, the context sensitivity of how personality is expressed in behavior has been a well-known feature ever since Walter Mischel’s seminal work (Mischel, Reference Mischel1996).

The inferred personality constructs from the behaviors must have consistency across contexts, otherwise the criterion of temporal stability would be violated as context changes over time. To operationalize this, the consistency merely needs to apply at the level of the inferred personality constructs from recurring patterns of behaviors – not the precise bodily movements themselves. Indeed, behaviors that are too rigid across very different contexts may be inappropriate indicators of personality as we are defining it – they may instead be like sea anemone traits or Parkinsonian tremor in humans: biological traits that are not psychological. Some flexibility and plasticity seem essential (Dall et al., Reference Dall, Houston and McNamara2004). We would speculate that more open-ended contexts, such as complex social situations, might reveal behavioral differences that can best be explained by personality (Gosling, Reference Gosling2001; Uher, Reference Uher2008). An important methodological point would thus be to study personality with an emphasis on those situations/contexts where personality actually has a measurable role in explaining and predicting behavior.

The behavioral patterns from which we infer personality in animals, as in humans, should exhibit broad, situation-dependent (context sensitive) variability. This context- or situation dependence is precisely the evidence that personality processes (together with other psychological processes) are interposed between the situational stimuli and the behaviors that they elicit. The other psychological processes, such as affect and cognition, together with environment/context, constitute an important aspect of personality according to many views (Revelle, Reference Revelle, Robins, Fraley and Krueger2007; Wilt & Revelle, Reference Wilt and Revelle2015; Arden et al., Reference Arden, Bensky and Adams2016; Baumert et al., Reference Baumert, Schmitt, Perugini, Johnson, Blum, Borkenau and Wrzus2017; Renner et al., Reference Renner, Klee and von Oertzen2020). For example, a person being an extrovert (as referring to a temporally stable personality trait) will behave differently depending on the situation, and on what mood they are in, whether they are tired or attentive, etc. Such a distribution of the varied behaviors could fruitfully be thought of along the lines of Fleeson’s whole trait model (Fleeson & Jayawickreme, Reference Fleeson and Jayawickreme2015): we can think of a personality trait as the parameters describing the shape of the distribution of a measurable state/behavior that is expressed transiently in different contexts. How personality expresses itself in behavior, through layers of cognitive and affective processing, accounts for the width of the distribution. Density distributions that are very narrow across very different contexts would argue against an effect of personality because the behavior is insufficiently sensitive to context. On the other hand, too broad density distributions that show extreme variability over context would make it difficult to discover a stable pattern. Animal and human personality should show distributions intermediate between these extremes: temporally stable personality traits are inferred from behaviors that are nonetheless sensitive to different contexts. Of course, it is not merely the density distribution per se, but consistent patterns, and interpretable changes in patterns over time, that are the hallmark of context-sensitive personality (Revelle & Condon, Reference Revelle and Condon2015; Revelle & Wilt, Reference Revelle, Wilt and Rauthmann2021).

In addition to measuring behavior across a range of different contexts in which personality can manifest, other major practical challenges for the study of animal personality include accessibility, sample size, and ecological validity. In humans, we derive personality traits across a typically large and representative population, but many animal studies will be limited in both sample size and representativeness (cf. Table 1). For instance, two mice might have somewhat different behavioral patterns, but we would need to study many more before concluding that these are indeed behavioral patterns explained by personality, or merely individual differences due primarily to external environmental factors that do not generalize to other mice. In addition, personality in laboratory mice will depend on the strain and differ from wild animals (Broadhurst, Reference Broadhurst1975; Blanchard et al., Reference Blanchard, Hebert, Ferrari, Palanza, Figueira, Blanchard and Parmigiani1998; Crabbe et al., Reference Crabbe, Wahlsten and Dudek1999; Augustsson & Meyerson, Reference Augustsson and Meyerson2004). It will be helpful for the field to have a common set of approaches (perhaps including common tasks and observational methods (Kaiser & Müller, Reference Kaiser and Müller2021) as well as common algorithms for analysis, such as tests of reliability and coherence), together with shared databases. This will help accumulate knowledge across larger and larger numbers of animals, and across species. As with other such approaches, funding initiatives aimed at consortia to make this possible would be important. Box 1 summarizes some of the challenges that distinguish the study of animal personality.

3. Current personality frameworks

One approach to animal personality applies extant (human) personality theories (Table 2) to animals. There are a number of theories on personality dimensions in humans, from Cattell’s 16 (Cattell, Reference Cattell1945; Heather et al., Reference Heather, Cattell, Mead, Boyle, Matthews and Saklofske2008) to the Big Five (McCrae & Costa, Reference McCrae, Costa, Boyle, Matthews and Saklofske2008) to Eysenck’s three (Eysenck, Reference Eysenck1991), as examples. While there are other criteria that might decide in favor of one of these over another (e.g., biological evidence for the reality of a particular dimension), from a statistical point of view they could be compatible with one another: a full trait space would have as many dimensions as there are items or measures, but any one of a number of dimensionality reductions (using factor analysis or principal component analysis, for instance) could compress this to an arbitrary number of fewer dimensions depending on the desired amount of represented variation. In this view, there is no unique personality space, and translation between different spaces is generally straightforward as long as each of them is based on a sufficiently complete set of measures at the outset (Markon et al., Reference Markon, Krueger and Watson2005; Ludeke et al., Reference Ludeke, Bainbridge, Liu, Zhao, Smillie and Zettler2019; Bainbridge et al., Reference Bainbridge, Ludeke and Smillie2022).

Table 2. Theories of human personality, brief description, and rating of applicability to animal studies (0 = not applicable, 1 = applicable with substantial modification, 2 = applicable with minor modification)

Given the challenges of collecting sufficient observational data from animals, our overarching recommendation for applying extant theories of human personality to behavioral data from animals is to begin small. One might characterize a particular facet/dimension of personality, such as timidness vs. boldness, and then gradually build out from there to a more complete personality once sufficient data are available. In general, we would recommend sticking with one of the most widely accepted theories on personality dimensions, such as the Big-Five, and then to see if perhaps one or more of the dimensions drop out because they have too sparse observations or they seem to be absorbed by another dimension. As can be seen in Table 1, most studies of animal personality offer a relatively small number of dimensions that are often a subset of those in the Big Five, or an amalgamation of some of its factors. Once a study has undertaken such a characterization of a nonhuman species’ personality, however limited, it would then be important to see how well it predicts behaviors to a future situation and how well it generalizes across time and individuals. Using existing human personality theories as the approach could observe animal behavior and score it along the dimensions of the theory as well as design specific situations or tasks intended to assess a particular dimension. Testing across multiple contexts would also be important, as noted above, and emphasizes the need for comprehensive and well-curated data sharing among studies. In general, these approaches have confirmed that at least some dimensions commonly used to describe human personality also apply to nonhuman animals; for example, studies of chimpanzees and bears (Table 1) have used the Big Five (Table 2).

Another approach comes from the study of animal behavior itself, which also offers theories to explain the behavior, but from the perspective of ethology rather than derivation from theories of human personality. Some examples come from careful studies of animal behavior by trained human observers. When studying boldness and exploration, Fox et al. (Reference Fox, Ladage, Roth and Pravosudov2009) discovered that mountain chickadees’ behaviors in their exploration of a novel environment and a novel object are uncorrelated, suggesting two independent personality dimensions and contradicting other studies (Carter et al., Reference Carter, Feeney, Marshall, Cowlishaw and Heinsohn2013). In personality studies of chimpanzees, social dominance and dominance-related activities play a pervasive role, conceptualized as a sixth factor distinct from the Big Five in humans (King & Figueredo, Reference King and Figueredo1997). As can be seen from just these brief examples, the ethological factors typically used to characterize variability in some aspects of behavior tend to differ between species and can differ from those popular in humans.

Extant theories of both human and animal personality as described above all have dimensions that are readily interpretable. But all these approaches raise a nagging worry: they are ultimately very much the design and interpretation of humans. An advantage is that we can understand the dimensions used to characterize animal personality. The worry is that we are anthropomorphizing. This brings us to data-driven approaches that can aim to minimize the risk of anthropomorphizing and that could suggest entirely new personality dimensions we might never have thought of intuitively.

4. Data-driven approaches to animal personality

The behavioral data collected for a data-driven approach should be as complete as possible over a long time in varied environments to ensure that all variation in behavior is sampled (Uher, Reference Uher2008; Uher et al., Reference Uher, Asendorpf and Call2008). By applying a data-driven algorithm on such a large high-dimensional dataset, one can extract factors/dimensions or categories. If based on sufficiently diverse and ecologically valid behaviors, this approach can generate personality dimensions from nonhuman animal data that may be quite distinct from those featured in human personality theories. In principle, this could allow us to obtain species-specific personality dimensions suited to a particular nonhuman species’ behaviors, avoiding anthropomorphism (Uher, Reference Uher2008).

A recent example of such a data-driven approach comes from a study by Forkosh et al. (Reference Forkosh, Karamihalev, Roeh, Alon, Anpilov, Touma and Chen2019) in mice (Figure 1). Mouse behaviors were recorded over multiple days in a rich environment (Figure 1B). From video recordings, 60 behavioral features were collected based on location tracking, including social behaviors (Figure 1C). Linear discriminant analysis (LDA) (Figure 1A) was applied to reduce the 60 behavioral-feature dimensions into 4 stable so-called “identity domains” (ID, analogous to personality dimensions) by maximizing the ratio of between-mice to within-mouse variability. These 4 identity domains (Figure 1C) were shown to be replicable and stable over developmental stages and social contexts using separate validation data, and the 4 IDs were shown to capture transcriptomic variance in the brain and variance in genetically driven behavioral differences.

Figure 1. Data-driven inference of personality in mice. A. Linear discriminant analysis (LDA) was used for dimensionality reduction by maximizing the ratio of between-subject to within-subject variability. B. Schematic of the enriched group-housing environment. C. Running LDA on 60 behavioral-feature dimensions (showing 13 representative dimensions) collected from video recordings resulted in four validated identity domains (ID1 - ID4), corresponding to personality dimensions. The IDs themselves are uncorrelated. The width of blue and red connecting lines indicates the strength of the correlation between the four IDs and the 60 behavioral-feature dimensions. For instance, the first personality factor, ID1, is positively correlated with “Chase” and negatively correlated with “Escape”.

Adapted with permission from “Identity domains capture individual differences from across the behavioral repertoire” by Forkosh et al., 2009. Copyright 2019 by The Author(s), under exclusive license to Springer Nature America, Inc.

Some of the tools developed for data-driven personality studies in animals could fruitfully be applied to humans as well. For instance, densely sampled data are now available from social media, such as smartphone usage. While studies have generally simply attempted to map such data onto extant personality theories such as the Big-Five (Stachl et al., Reference Stachl, Au, Schoedel, Gosling, Harari, Buschek and Bühner2020), one could imagine using LDA instead and comparing the results to the four identity domains discovered in the mouse study above.

There may be no clear answer whether to use more “top-down” approaches that start with theorized personality dimensions (as suggested in Section 3 above) or more “bottom-up” approaches that depend on a relatively complete sampling of neural or behavioral data, as suggested in the current section. Uher (Reference Uher2008) has provided a useful taxonomy of the different approaches to animal personality that illustrates the diversity available. Our own preference would be to use a mixture of “top-down” and “bottom-up” approaches, with each being used to inform and revise the other. As we mentioned, a main challenge is the number of samples of observational data available. While semi-automated methods are now making very large datasets from quite diverse situations possible (as in the Forkosh et al., Reference Forkosh, Karamihalev, Roeh, Alon, Anpilov, Touma and Chen2019 study) and thus enabling the discovery of strongly data-driven personality dimensions, ecologically based approaches have also been successful in characterizing animal personality. For instance, studies of fear and anxiety have been inventive in constructing situations in the lab that aim to mimic relevant parameters encountered in the wild (e.g., Blanchard & Blanchard, Reference Blanchard, Blanchard, Blanchard, Griebel and Nutt2008; Kumar et al., Reference Kumar, Bhat and Kumar2013; Mobbs & Kim, Reference Mobbs and Kim2015).

5. Concluding comments

The motivations for studying personality in nonhuman animals go well beyond our curiosity and interest in understanding animal behavior per se. Animals offer powerful models for investigating the biological and genetic basis of personality (Gosling, Reference Gosling2001), for understanding how it may have evolved (Dall et al., Reference Dall, Houston and McNamara2004, Reference Dall, Bell, Bolnick and Ratnieks2012), and for explaining pathology, topics we have not discussed here for reasons of space. However, we want to stress that neuroscience (and biology more broadly) certainly should figure both in the creation of hypotheses and in the interpretation of results. In this respect, we see biologically inspired theories of personality, such as the line of work from Pavlov through Eysenck to Gray (see Corr & Perkins, Reference Corr and Perkins2006, for an overview), as excellent candidates also for animal personality.

Complex behavior in both human and nonhuman animals requires psychology for explanation. We have stressed that animal personality ought to be considered as one component to explain complex behavior. This means that there must be at least some other psychological processes that are also influenced by personality and contribute to explanation of complex behavior, analogous to the “thinking” and “feeling” that figure in theories of human personality. We think that these other psychological processes will include processes such as affect and cognition, but it is important to stress that we simply do not yet have a full theory of animal psychology available. Many more studies will be needed to thoroughly characterize animal behavior in rich environments. Building a mature science of animal personality will go hand in hand with building a mature science of animal psychology.