Personality links with lifespan in chimpanzees.

Life history strategies for optimizing individual fitness fall on a spectrum between maximizing reproductive efforts and maintaining physical health over time. Strategies across this spectrum are viable and different suites of personality traits evolved to support these strategies. Using data from 538 captive chimpanzees (Pan troglodytes) we tested whether any of the dimensions of chimpanzee personality - agreeableness, conscientiousness, dominance, extraversion, neuroticism, and openness - were associated with longevity, an attribute of slow life history strategies that is especially important in primates given their relatively long lives. We found that higher agreeableness was related to longevity in males, with weaker evidence suggesting that higher openness is related to longer life in females. Our results link the literature on human and nonhuman primate survival and suggest that, for males, evolution has favored the protective effects of low aggression and high quality social bonds.

Introduction

Life-history theory posits that strategies for increasing individual fitness lay on a continuum that describes an energetic trade-off between maximizing reproductive efforts and maintaining physical health as the organism ages (Stearns, 1976). At one end of this continuum are ‘r-selected’ populations. Individuals within these populations are characterized by early and frequent reproduction, the rapid onset of senescence, and a shorter lifespan. At the other end of this continuum are ‘K-selected’ populations. Individuals within these populations are characterized by later and less frequent reproduction, but delayed senescence, and a longer lifespan. Both ends of this continuum are viable fitness strategies, as are, depending upon ecological and social contingences, life history strategies between these extremes. These strategies are supported by behavioral adaptations (Stearns, 1976).

Differences in life history strategy have been advanced as one possible explanation for why individuals within populations exhibit stable differences in behavioral, affective, and cognitive dispositions, that is, personality traits (Dingemanse and Réale, 2005; Réale et al., 2010). A simulation study indicated that this theory is plausible (Wolf et al., 2007), and a meta-analysis on studies of boldness, exploration, and aggression in insects, fish, birds, and mammals offered mixed empirical support (Smith and Blumstein, 2008). This meta-analysis showed that bolder animals put themselves at greater risk and die at younger ages, but enjoy greater reproductive success than their shyer counterparts, which do not enjoy as many opportunities for copulation, but live longer, and so are able to invest more in their offspring (Smith and Blumstein, 2008). Boldness therefore is associated with a ‘faster’ (r-selected) life-history strategy. The findings of the meta-analysis for exploration and aggression were less clear: more aggressive individuals had greater reproductive success than less aggressive individuals, but this was not offset by reduced lifespan; individuals more prone to exploring their environment lived longer than neophobic individuals, but did not experience reduced reproductive success (Smith and Blumstein, 2008). Two concurrent reviews showed that, across a range of species, greater boldness, activity, and aggressiveness, and lower sociability and exploration, were associated with a faster life history strategy (Réale et al., 2010; Biro and Stamps, 2008).

Recent research found evidence that variation in the personality traits of humans and nonhuman primates are also associated with variables related to life history strategies. Studies of humans predominate this literature and, although there are exceptions (e.g., Alvergne et al., 2010; Gurven et al., 2014), this human literature grew out of personality psychology, health psychology, and epidemiology. Consequently, these studies did not set out to deliberately test whether personality variation reflected individual differences in life history.

The studies of human personality described above tended to focus on one or more of five traits - extraversion, agreeableness, openness, neuroticism, and conscientiousness - known collectively as the ‘Big Five’ or ‘Five-Factor Model’ (Digman, 1990). These five traits are operationalized as dimensions onto which several related lower-order traits cluster (Digman, 1990). Four of the five human traits correspond to personality traits studied by behavioral ecologists. Extraversion and agreeableness characterize how often and how well humans navigate their social world (Digman, 1990). Among other characteristics, extraversion features sociability and activity (Costa and McCrae, 1995), which are comparable to the same-named traits studied in behavioral ecology; agreeableness is the opposite of aggressiveness (Réale et al., 2007). Openness captures curiosity, originality, and a tendency to find novel ideas and situations appealing (Digman, 1990), and corresponds to exploration (Réale et al., 2007). Finally, neuroticism is related to fearfulness, vigilance, and emotional reactivity (Digman, 1990), and so appears to be the opposite of boldness, that is shyness or timidity (Réale et al., 2007). Conscientiousness describes individual differences in self-control, delay of gratification, and thoughtful planning (Digman, 1990). Animal analogues of conscientiousness have emerged in a few nonhuman primates, for example chimpanzees (King and Figueredo, 1997), and in Asian elephants (Seltmann et al., 2018). However, conscientiousness has only recently been operationalized in ways familiar to behavioral ecologists, that is as naturally occurring behaviors or responses to behavioral tests (Delgado and Sulloway, 2017; MacLean et al., 2014; Altschul et al., 2017). In this literature, conscientiousness is often termed ‘self-control’ (e.g., MacLean et al., 2014).

In addition to its focus on the Big Five traits, the life history variables most often examined in the human literature have been health outcomes, especially longevity. Meta-analyses of this extensive literature showed that people who enjoy better health and live longer tend to be higher in agreeableness, extraversion and conscientiousness, and lower in neuroticism (Strickhouser et al., 2017; Roberts et al., 2007). The explanatory theories emerging from this field posit that health-related behaviors, including diet, mediate relationships between personality and health (Turiano et al., 2015; Graham et al., 2017). The possibilities that agreeableness, extraversion and conscientiousness are related to a slower life history strategy, and that neuroticism is related to a faster life history strategy, are mostly not considered in this literature.

Studies of personality and life history in nonhuman primates are often narrower in scope than studies of humans. Specifically, they mostly test whether one or more personality traits related to social interactions are associated with health and/or mortality outcomes. This narrow focus is probably attributable to two characteristics of these species. First, nonhuman primates have relatively slow life-history strategies; lifespans are comparatively long and reproductive rates are comparatively low (Jones, 2011). Consequently, health and longevity are influential fitness measures in primates, including humans. Second, most primate species live in groups and are highly social (Napier and Napier, 1967). To date, whether they use rating and/or coding measures of personality, studies of personality and survival in nonhuman primates have shown that western lowland gorillas (Weiss et al., 2013), baboons (Silk et al., 2010; Archie et al., 2014; Seyfarth et al., 2012), and female rhesus macaques (Brent et al., 2017) that are higher in sociability live longer. However, a study of female blue monkeys found that the association between sociability and mortality was only true for individuals that had consistent bonds with groupmates (Thompson and Cords, 2018).

In addition to the fact that all but one of these studies focus on a narrow set of traits (Weiss et al., 2013), studies of primate personality and longevity have focused on a small number of species. In particular, New World monkeys are not represented and only one study was of a species of great ape (Weiss et al., 2013), the evolutionary line that includes humans. We wished to expand on what is known about the links between personality traits and life history strategy in nonhuman primates and in humans. To do so we examined these associations in chimpanzees, which are one of our closest living great ape relatives.

The present study was made possible by the existence of a database containing a large sample (n = 538) of captive chimpanzees living in zoological parks, research facilities, and sanctuaries located in the United States, the United Kingdom, the Netherlands, Australia, and Japan. Personality in this sample was assessed by ratings on two comparable questionnaires that assessed a wide range of traits. These ratings were made by keepers, researchers, and others who knew and worked with these chimpanzees for considerable lengths of time. Furthermore, the long follow-up times from when chimpanzees’ personalities were assessed to the present (7 to 24 years) meant that there were enough deaths to provide adequate statistical power for detecting associations between personality and mortality. The sample used in this study and the means of measuring personality deserve comment.

There is some disagreement as to whether chimpanzees or bonobos, which are as related to humans as chimpanzees, are the best model for ancestral humans (Stanford, 2012; Sayers et al., 2012). However, studies using similar personality measures in captive groups of chimpanzees and bonobos have found that the dimensions along which chimpanzee personality traits align themselves (King and Figueredo, 1997) are more similar to the human dimensions than are those of bonobos (Weiss et al., 2015). Specifically, in addition to a dominance dimension, which reflects competitive prowess, social competence, and fearlessness, that is not present in humans (King and Figueredo, 1997; Murray, 1998; Dutton et al., 1997; Freeman et al., 2013; Weiss et al., 2009; Weiss et al., 2007), chimpanzee personality is defined by five dimensions that resemble the human Big Five. These dimensions have been identified in many studies, including those that measured personality with different questionnaires (King and Figueredo, 1997; Murray, 1998; Dutton et al., 1997; Freeman et al., 2013; Weiss et al., 2009; Weiss et al., 2007; King et al., 2005; Martin, 2005; Buirski et al., 1978) and those that used coded behavioral observations instead of ratings (Freeman et al., 2013; Massen et al., 2013; Koski, 2011; Vazire et al., 2007; Pederson et al., 2005; van Hooff, 1970). In bonobos, questionnaire-based and coding-based methods revealed evidence for human- and chimpanzee-like agreeableness, conscientiousness, and openness dimensions, a dimension like the chimpanzee dominance dimension, and an additional dimension, attentiveness, which is distinct from conscientiousness (Weiss et al., 2015; Staes et al., 2016). However, these studies find next to no evidence for neuroticism and extraversion. Taken with findings from comparable studies of the other great apes (Weiss et al., 2006; Gold and Maple, 1994), one plausible scenario is that bonobo personality diverged from that of chimpanzees and the other great apes, including humans.

Some question the use of ratings to measure animal personality given the possibility of anthropomorphic projection (Uher, 2013). For studies of nonhuman primates, as noted in the previous paragraph, ratings and behavioral measures yield comparable personality traits. Moreover, a review and meta-analysis found evidence that different raters provide similar ratings, that these measures are heritable, and that they are repeatable (Freeman and Gosling, 2010), the latter being most recently demonstrated in ratings taken 35 years apart and made by two independent sets of raters on two different questionnaires (Weiss et al., 2017). In addition, the effects of anthropomorphic projection by raters, if present, are minimal (Weiss et al., 2012). These just-described findings are probably attributable to the fact that items on most questionnaires do not consist of a single word (typically an adjective), but include behavioral definitions, which limit the degree of subjectivity in interpreting the traits and making ratings (Uher and Asendorpf, 2008; Stevenson-Hinde and Zunz, 1978).

Another concern that some raise is the use of captive samples. Although they limit the conclusions that we can draw about ancestral humans, by using captive samples one is able to remove many extrinsic sources of mortality, for example predators and infectious diseases. Therefore, captive samples, such as that used in this study, control for potential confounds that might crop up in studies of wild samples. In addition, captive samples are uniquely suited to testing whether the associations between human personality and mortality risk reflect life history strategies followed by individuals apart from links between personality and health-related behaviors that are endemic to human personality studies.

We used these data to test six hypotheses, one for each chimpanzee personality trait. We will first describe the hypotheses for the chimpanzee personality traits of extraversion, agreeableness, openness, and neuroticism, which are closely related to traits studied by behavioral ecologists. We will then describe the hypotheses for conscientiousness and dominance, which were based on literature that we will discuss.

Because sociability and aggressiveness are associated with slower and faster life-history strategies, respectively (Réale et al., 2010; Brent et al., 2017), we expect that higher extraversion and agreeableness will be related to longer life. In nonhumans, lower boldness is related to a slower life-history strategy. In humans, although overall neuroticism is associated with poorer health and a shorter lifespan, aspects of neuroticism related to worry and vigilance, key characteristics related to lower boldness (Réale et al., 2007), are associated with better health and a longer lifespan (Gale et al., 2017; Weston and Jackson, 2018). We thus expect that neuroticism should be associated with a longer life-span. Exploration, in animals, is linked to some characteristics of a slower life history, and so we expect that openness in chimpanzees will be associated with longer life.

We expect that conscientiousness will be related to a slower life history, and so longer life. This expectation was based on the above-described finding that humans who are higher in conscientiousness enjoy better health and live longer. If we do not find such an association, it would suggest that the association between conscientiousness and better health in humans may be attributable to human-specific health behaviors, such as exercising, that are related to higher conscientiousness and lead to individuals being healthier (Turiano et al., 2015). Our basis for this interpretation of these results stems from the fact that captive chimpanzees do not have many (if any) opportunities to control their health, which is in fact maintained by humans.

Finally, among primates, social standing is related to physiological stress responses (Sapolsky, 2005) and high dominance is associated with higher stress, as well as faster, energetically intense growth in chimpanzees (Pusey et al., 1997). High-ranking individuals also mate more frequently and dominate resources to support their growth and reproductive efforts (Ellis, 1995). Higher rank in chimpanzees, therefore, is associated with a faster life history strategy. Because ratings on traits such as dominance in chimpanzees and other primates are related to rank, including in the wild (Buirski et al., 1978), we expected that dominance would be related to a shorter lifespan.

Results

Comparing captive and wild chimpanzee mortality

During the follow-up period, 187 chimpanzees died. A Kaplan-Meier plot (Figure 1) shows survival functions for our sample and a wild sample (Bronikowski et al., 2011). Unlike wild chimpanzee populations in which infant mortality is high, captive chimpanzee populations have strikingly reduced infant mortality, live longer, and display accelerated mortality in older ages. These results show that captive chimpanzees benefit from protection against extrinsic sources of mortality, for example shelter from elements and predators, good health care, and abundant food.

Figure 1.

Survival curves of captive and wild chimpanzees.

Lines indicate survival probability of each group over the lifespan. The solid lines represent the captive population used in this study and the dashed line corresponds to a wild group (Bronikowski et al., 2011). The shaded areas indicated the 95% confidence region for reach group.

Associations between personality and age

Inspection of the six chimpanzee personality dimensions (Figure 2), as well as prior studies (King et al., 2008) indicate that personalities change as individuals age, making it possible that an association between personality and longer life might be confounded. This is not necessarily undesirable, as it indicates that personality and lifespan are linked, but to be conservative, we modeled and therefore controlled for potential confounds between age and personality scores. We fitted generalized additive models (GAMs) for each personality dimension, regressing personality ratings on the age at which the individual was rated.

Figure 2.

Personality’s relationship with age and sex.

Each panel shows the personality scores of a specific dimension for all individuals in a scatterplot against age on the left, and on the right with bean plots showing the distribution of scores split by sex (females are on the left, males on the right). Relationships between age and each personality dimensions are illustrated in the figure supplements.

The carpet and scatter plot present the distribution of observations over which the best fit smoothed regression line is plotted with the 95% confidence region.

The carpet and scatter plot present the distribution of observations over which the best fit smoothed regression line is plotted with the 95% confidence region.

The carpet and scatter plot present the distribution of observations over which the best fit smoothed regression line is plotted with the 95% confidence region.

The carpet and scatter plot present the distribution of observations over which the best fit smoothed regression line is plotted with the 95% confidence region.

The carpet and scatter plot present the distribution of observations over which the best fit smoothed regression line is plotted with the 95% confidence region.

The carpet and scatter plot present the distribution of observations over which the best fit smoothed regression line is plotted with the 95% confidence region.

Generalized additive model of extraversion and age at personality rating.

Generalized additive model of conscientiousness and age at personality rating.

Generalized additive model of agreeableness and age at personality rating.

Generalized additive model of neuroticism and age at personality rating.

Generalized additive model of openness and age at personality rating.

The GAM regression lines for each model are plotted against the personality data in Figure 2—figure supplements 1 through 6. Curvilinear associations were presented between age and personality for all dimensions except neuroticism, where only a linear relationship was present.

Figure 2—figure supplement 1.

Generalized additive model of dominance and age at personality rating.

The carpet and scatter plot present the distribution of observations over which the best fit smoothed regression line is plotted with the 95% confidence region.

Because personality does change over time, some of the raw personality score variance could be attributed to rating age variance. Alternative, adjusted personality scores were therefore calculated as residuals from the regression function of each GAM. In the subsequent analyses, adjusted scores were fitted as predictors in separate survival models from the raw scores.

Decision tree survival models

We fit decision trees to test whether sex, origin (wild-born or other), or any personality dimensions were related to longevity. A conditional inference survival tree procedurally determined that among males, higher agreeableness was associated with longer survival (Figure 3). Specifically, males with agreeableness scores less than 0.063 standard deviations below the mean were at significantly higher risk than other males (p<0.027). These results held for the age-adjusted agreeableness scores as well.

Figure 3.

Conditional inference tree diagram indicating variables influencing survival.

Bottom panes indicate the survival curves of and number of chimpanzees in each sub-group. Sub-groups were split based on the growth of the tree and decision criteria are indicated below each node. Splits in numeric variables (e.g. agreeableness) are by standard deviations.

Weighted parametric hazard regression models

The association between agreeableness and survival in males was confirmed with parametric hazards modeling: in a AIC weighted model including all covariates and frailty effects, the hazard ratio for males was 0.66 (95% CI: 0.49 – 0.89) per standard deviation increase, and in a model where we adjusted personality scores to control for age, the hazard ratio associated with a standard deviation increase was 0.61 (95% CI: 0.42 – 0.89). In the models of only females, a positive association between openness and survival was also revealed with a hazard ratio of 0.77 (95% CI: 0.59 – 0.99) for unadjusted scores, but the association was not significant when we used the adjusted openness scores. Higher openness in males was not related to living longer nor was higher agreeableness in females (Table 1 presents a full description of the AIC weighted models). For a subset of the sample, more detailed rearing data were available, but survival analyses did not find any association between rearing conditions or origin and longevity (Table S1). A complete description of all survival analyses is available in the supporting information.

Table 1.

Weighted survival model estimates of personality and demographic variables related to longevity.

Values are model averaged parameter estimates and unconditional confidence intervals calculated from estimates shown in Supplementary Table 4.

	Unadjusted		Adjusted for age
Variable	Hazard Ratio	95% C.I.	Hazard Ratio	95% C.I.
Male (n = 216)
Wild-born	1.40	[0.68, 2.90]	1.35	[0.66, 2.74]
Agreeableness	0.66	[0.49, 0.89]	0.61	[0.42, 0.89]
Dominance	0.98	[0.74, 1.29]	0.99	[0.72, 1.37]
Extraversion	1.04	[0.71, 1.51]	1.01	[0.65, 1.57]
Conscientiousness	1.11	[0.78, 1.58]	1.19	[0.79, 1.81]
Neuroticism	0.91	[0.66, 1.25]	0.93	[0.66, 1.31]
Openness	1.09	[0.76, 1.55]	1.06	[0.78, 1.46]
Female (n = 322)
Wild-born	1.16	[0.72, 1.85]	1.17	[0.73, 1.87]
Agreeableness	1.12	[0.83, 1.50]	1.24	[0.84, 1.82]
Dominance	1.04	[0.83, 1.30]	1.05	[0.82, 1.35]
Extraversion	1.15	[0.80, 1.67]	1.02	[0.66, 1.57]
Conscientiousness	1.01	[0.76, 1.34]	0.98	[0.70, 1.38]
Neuroticism	0.93	[0.73, 1.17]	0.93	[0.72, 1.19]
Openness	0.77	[0.59, 0.99]	0.82	[0.66, 1.02]

Discussion

We found a clear pattern of relationships between personality and longevity in these data: among males, higher agreeableness was associated with longer life, even when agreeableness was adjusted for age. In other words, long-living captive male chimpanzees are those who engage in positive social interactions characterized by cooperation, geniality, and being protective. These findings match our prediction, although we did not necessarily expect to find the association only in males. However, this finding is consistent with the literature: in wild chimpanzees, male coalitionary aggression towards conspecifics is associated with greater chances of siring offspring (Gilby et al., 2013). Agreeableness, the opposite of aggression, ought to lie on the other end of the life-history spectrum, and be associated with longer life, as we found. More agreeable males may adopt a more cooperative dominance style (Foster et al., 2009), ultimately allowing for fewer, but more consistent reproductive opportunities over the course of a long life.

We were surprised to find no association between extraversion and longevity. Studies in monkeys (Silk et al., 2010; Seyfarth et al., 2012; Brent et al., 2017) have shown positive, protective relationships with extraversion. Of note, a positive association between extraversion and longevity was found in a study of gorillas that were also kept in captivity and assessed for personality by means of ratings (Weiss et al., 2013). Like their close chimpanzee cousins, captive gorillas show evidence for strong age-related declines in extraversion (Kuhar et al., 2006), yet extraversion was still associated with longevity. However, high sociability among primates does not support longevity in all circumstances (Thompson and Cords, 2018). The remaining difference between gorillas and chimpanzees that could explain our null findings for extraversion lies in the mating systems of these species. Specifically, gorillas have strict harems where one or two males have exclusive sexual access to multiple mature females (Harcourt et al., 1981). Chimpanzees, on the other hand, have a promiscuous mating system (Tutin, 1979).

There was no association between longevity and conscientiousness. It is possible that this finding reflects our captive sample in which the extrinsic benefits of being higher in conscientiousness have been removed. For instance, although chimpanzees are known to self-medicate using plants in the wild (Huffman and Wrangham, 1994), and while conscientious chimpanzees in captivity are more diligent (Altschul et al., 2017), individuals have no resources to use for self-medication in captivity. Our results thus suggest that the associations commonly found between conscientiousness and longevity in human is not related to intrinsic characteristics of the organism, but to the health-related behaviors associated with this trait (Turiano et al., 2015).

Females that were higher in openness lived longer, but the effect was not present when we corrected for confounding by age of rating. This is due to the strong curvilinear relationship between age and openness (Figure 2). Younger chimpanzees were much higher in openness and there was an association between lower openness and age, a limitation we might have missed had our sample been smaller. It is therefore impossible for us to conclude whether there is a protective association between openness and longevity in females or whether lower openness was a proxy for age.

Low boldness resembles one aspect of human neuroticism that is related to a longer lifespan, and so we predicted that neuroticism would be associated with greater longevity. However, we found no association in either direction. The absence of any effect of neuroticism in chimpanzees may be attributable to the fact that the health-harming and health-benefitting roles of neuroticism are, like conscientiousness, mediated by health behaviors, as well as the environment. For example, people who are higher in neuroticism tend to smoke, and this behavior explains some of the relationship between neuroticism and shorter lifespans (Graham et al., 2017). On the other hand, after the onset of certain diseases, some high neuroticism individuals are more likely to stop smoking (Weston and Jackson, 2018). Smoking does not explain the entire association in humans, however, as high neuroticism is also associated with greater reactivity to stressors (Chapman et al., 2011) and energetically expensive physiological responses (Réale et al., 2010), which could offset potential benefits of slow life-history benefits from neuroticism. Moreover, with the absence of predators in captivity benefits of vigilance would be reduced if not entirely eliminated, as danger and risks to health from agonistic social encounters remain.

Dominance, and the degree to which captive chimpanzees are characterized by their competitive prowess and fearlessness, and, consequently, the ability to enjoy the spoils of rank, had no bearing on how long individuals lived. In chimpanzees specifically, high-ranking individuals are generally less stressed (Goymann and Wingfield, 2004), but when the hierarchy is destabilized, high-ranking individuals become more stressed, and instability and reorganization can be common in wild chimpanzee groups (Muller and Mitani, 2005). Dominance may not play a major role in influencing longevity in captive populations because fission-fusion dynamics are not in play to the same extent as in the wild, thus group stability will be greater, and stressful disruption will be reduced. Moreover, in captivity there is less need for chimpanzees to compete with one another for resources, so traits such as dominance, that are related to rank, may not be related to mortality in this sort of environment.

This study had several limitations. Our data did not have measures of social variables like rank or social network, or psychological variables like intelligence. These chimpanzees lived exclusively in captive environments, which limits our ability to make evolutionary inferences regarding the associations between personality and survival. However, our captive sample was also a strength as it allowed us to identify extrinsic influencers that would be eliminated by captive environments and test novel hypotheses about the relationships between personality and life-history strategies in chimpanzees.

Our study also examined only a single species. More generally, future studies that incorporate multiple primate species could utilize phylogenetic approaches, which consider the importance of species differences in social organization and ecology (MacLean et al., 2012; Cornwell and Nakagawa, 2017). Phylogenetic analyses could allow researchers to identify which specific species differences moderate relationships between certain personality traits and measures of health and survival, as well as reproductive success and fitness more broadly.

The present study is a reminder of the complex, multifaceted nature of personality and sex, social relationships and the life course in chimpanzees. It also shows how studying the personality of our biological kin reveals that, as in humans, it is not the quantity of social relationships that matters, but the quality.

Materials and methods

Sample and experimental design

All research reported in this study was non-invasive. The research complied with the regulations and guidelines prescribed by The University of Edinburgh and the participating zoos, research institutes, and sanctuaries.

556 chimpanzees were assessed for personality between 1993 and 2010. Eighteen chimpanzees had to be removed from the sample due to incompatibilities with the study design, either because personality was assessed after death or because a veterinary staff member requested the individual not be analyzed and mortality data were thus withheld. Of the 538 remaining chimpanzees, 175 came from zoos in the United States, 164 came from the Yerkes National Primate Research Center (also in the United States), 156 came from zoos, a sanctuary, and two research centers in Japan, 21 came from the Taronga Zoo in Australia, 11 came from the Beekse Bergen Safaripark in the Netherlands, and 11 came from the Edinburgh Zoo in the United Kingdom.

Vital status was recorded throughout 2016 and 2017, yielding follow-up times ranging from 7 to 24 years, which is approximately equivalent to 10 to 36 human years (Napier and Napier, 1967). A total of 187 chimpanzees died during the follow-up period. As is standard in studies that seek to identify mortality risk factors, our analytic approach treated the remaining 353 chimpanzees as right-censored at the date that mortality data were gathered for that group. 336 individuals were known to be alive at the time of data collection, and 17 individuals were lost to follow-up and censored at the date of their last known record. All records were also left-truncated, beginning each record at the age at which the individual was assessed for personality.

Personality assessments

Fifty-four items comprising a trait name, for example ‘Fearful’ and a one to three sentence behavioral description, for example 'Subject reacts excessively to real or imagined threats by displaying behaviors such as screaming, grimacing, running away or other signs of anxiety or distress.’ were developed to assess the personalities of the chimpanzees (King and Figueredo, 1997; Weiss et al., 2009), Between 1993 and 2005, 43 of these items were used to assess the personalities of chimpanzees in the American zoos, the Taronga Zoo, and chimpanzees living at the Yerkes National Primate Research Center (King and Figueredo, 1997; Weiss et al., 2007). Starting in 2007, all 54 items were used to assess the personality of the chimpanzees living in Japan (Weiss et al., 2009), the Netherlands (Herrelko, 2011), and at the Edinburgh Zoo (Herrelko et al., 2012). The distributions of all six chimpanzee personality dimensions split by sex are shown in Figure 2.

The personalities of the chimpanzees in this study were assessed via ratings on these items by multiple keepers and researchers who knew the individual chimpanzees, sometimes for decades (King and Figueredo, 1997; Weiss et al., 2009; Weiss et al., 2007). In addition to showing that the interrater reliabilities are comparable to those found in human studies of personality, previous studies have shown that chimpanzee personality, measured this way, yields measures that are more reliable than behavioral codings (Vazire et al., 2007), that are heritable (Weiss et al., 2000; Wilson et al., 2017; Latzman et al., 2015a) and stable over time (King et al., 2008), and that generalize across samples (Weiss et al., 2009; Weiss et al., 2007; King et al., 2005), and are not adversely affected by anthropomorphic attributions on the part of raters (Weiss et al., 2012), Finally, these measures have been related to observed behaviors (Pederson et al., 2005), differences in brain morphology (Latzman et al., 2015b; Blatchley and Hopkins, 2010), and genetic polymorphisms (Wilson et al., 2017; Hong et al., 2011; Hopkins et al., 2012).

Generalized additive models

To adjust for confounding in the personality variables brought on by changes with age, we fit GAMs modeling the relationship between age at assessment and each personality variable (Wood, 2006). GAMs are an extension to linear models that allow the input data to ‘suggest’ non-linearities (Hastie, 2017) as opposed to requiring researchers to manually specify them, by, for example, adding a quadratic term to a model formula. To avoid overfitting, non-parametric transformations penalize roughness in the transformation function creating terms aptly called ‘smooths’ (Faraway, 2016). For our smooths, we used thin plate regression splines with a basis dimension (k) of 20. The basis dimension was verified as being acceptable using internal package functions; varying k did not alter any model fits. GAMs are difficult to interpret mathematically, but visually intuitive, so each GAM is described by its line of best fit, drawn in Figure 2—figure supplements 1 through 6. GAMs generate residuals like other regression models, thus, bivariate GAMs are a powerful method for identifying and controlling for the effects of confounders (Benedetti and Abrahamowicz, 2004).

Survival analyses

To be conservative, our survival models included all six personality scores. We also included sex and origin (whether the individual was born in the wild or not) as controls.

We used decision-tree analyses to identify associations between personality and longevity. Parametric and semi-parametric survival regression models force a specific link between variables and outcome, but decision trees do not impose any such assumptions; trees are able to automatically identify meaningful variables and even some interactions without prior specification (Bou-Hamad et al., 2011). Survival trees in particular have advantages over other techniques. In simulation studies of left-truncated right-censored decision trees with data much like ours, that is a large sample (N > 500) with many censored observations (>50%), conditional inference trees identified the correct predictors 94% and 93% of the time, respectively (Fu and Simonoff, 2016). This method can handle binary and continuous variables and is robust to the effects of time-dependent covariates, such as our chimpanzees’ personality dimensions, which could be confounded with age at rating.

We grew trees with both unadjusted and adjusted covariates. Adjusted covariates were residualized versions drawn from the GAMs used earlier to model the effects of age on personality. Using adjusted covariates had no meaningful effect on the conditional inference analysis; the tree grown was identical.

We validated our decision-tree analyses with fully parametric hazard regression models. We followed an information theoretical approach which allowed us to pool and average model estimates across a wide-range of possible choices of error distribution and variables to include (Burnham et al., 2011). We first built two sets of models, again, with unadjusted covariates and without adjusted covariates. Adjustment creates a different, alternative dataset which cannot be directly compared to the unadjusted data, so our evaluations of these models were necessarily kept separate. The linking distributions we used included the Weibull, log-logistic, Gompertz (Klein and Moeschberger, 2005), and semi-parametric splines survival functions (Goodman et al., 2011). There were no convergence issues and all splines were fit with 12 knots and κ = 10,000. The hazard models were fit with Gamma distributed frailty (random) effects to control for any influence that the different sample groups might have on survival, and estimated both jointly and separately by sex (Table S2 and Table 1, respectively). We also built models including and excluding the demographic covariates of sex and origin. No variation in specification affected our results (Tables S3 & S4).

In the interests of transparency, eLife includes the editorial decision letter and accompanying author responses. A lightly edited version of the letter sent to the authors after peer review is shown, indicating the most substantive concerns; minor comments are not usually included.

Thank you for submitting your article "Personality links with lifespan in chimpanzees" for consideration by eLife. Your article has been reviewed by three peer reviewers, one of whom is a member of our Board of Reviewing Editors, and the evaluation has been overseen by Ian Baldwin as the Senior Editor. The reviewers have opted to remain anonymous.

The reviewers have discussed the reviews with one another and the Reviewing Editor has drafted this decision to help you prepare a revised submission.

Summary:

The authors report a study that links personality with lifespan of a large sample of captive chimpanzees. They rely on a questionnaire method for assessing personality along six dimensions, and they use those data to find a relationship between personality and longevity. They find that for males longevity is related to agreeableness and for females to openness, and they discuss these results in terms of phylogenetic assumptions about human and great ape evolution.

Essential revisions:

Whereas the results are interesting, the reviewers and Reviewing Editor all have several major concerns with this study that cast doubt on the robustness of the theoretical framing, the methods, and as a consequence, the conclusions of the paper. A major revision would be required in order for publication in eLife, but as most of these involve incorporation of key discussion, literature, and modifications to the analytical approach, we feel these are feasible to accomplish within a two-month time frame. If you choose to revise and resubmit along the lines below, this will be a much stronger paper, and can be considered for publication in eLife. If these revisions are undertaken, it will have the potential to contribute not only an important dataset, but also to address some key theoretical problems in this area of research more broadly.

Below are the major issues/points that must be addressed in order to be considered in revised form. Large sections from the original reviews have been pasted into these comments, as they contain many helpful suggestions about theoretical framing, literature, and approach. Although each point is described at length, the substance of the revision should not simply be a longer version of the existing manuscript. Instead, it should be only a slightly longer version that revises much of the background and discussion in light of the valuable insights from the reviewers below.

1) Title. Given some of the uncertainties in the data and conclusions that are raised below, the title should be modified to ensure it faithfully reflects these issues.

2) Phylogenetic assumptions. The central framing of the problem with respect to human evolution (Introduction, fourth paragraph) makes a standard phylogenetic assumption: that behavioral attributes found in both chimpanzees and humans should also have been shared by the last common ancestor. There is much debate over the utility of this approach for behavioral attributes, or if it might suffer overmuch from homoplasy – especially when there appear to be so many differences between chimpanzees and bonobos, with both equally related to humans (work by Sayers is relevant here). The authors should include discussion and literature that explains why here they make the claim so strongly, or why they claim to pinpoint the evolution of specific male and female personality attributes to such a specific period of time (since the split with hominins). This is especially the case when their research question (as currently states) prioritizes similarities between humans and chimpanzees (rather than bonobos). The discussion that leads to this point also appears to somewhat answer the question in advance, because it is clear that the authors are already arguing that a link between personality and fitness is ancestral in primates (and therefore will also be linked in chimpanzees, humans, presumably bonobos, and their last common ancestor). This argument seeps back in later with the discussion of gorillas (Discussion, third paragraph), where they are referred to rather oddly (considering the phylogenetic argument up to this point and the actual genetic relatedness of gorillas and chimps versus chimps and bonobos or chimps and humans) as "close chimpanzee cousins". Thus, the phylogenetic argument should be very carefully constructed: what traits do they think are derived in chimpanzees, what traits ancestral to both chimpanzees and humans, what is the rationale, etc.?

3) Use of literature. The paper opens with a strong background discussion, but reads like a bank of examples rather than a coherent lead-up to a clear set of hypotheses (more on this below). Each cited study shows a suggestive tendency for proxies of fitness to link with some aspect of personality, but there is clear diversity within primates regarding how each of these relationships actually plays out (as well as in what fitness measures are used). This background captures some of the ambiguity and gaps in current knowledge, but it does not confront them head-on. The paper would increase its impact if it clearly describes the areas where there is more surety than others, and specifically where increased work is necessary and why. There is also a near-complete lack of citations to the non-primate animal personality literature. It is generally disconcerting that the primate personality and non-primate personality literatures don't often cite each other, but it is especially puzzling here since researchers working with short-lived, easily manipulated species like birds, fish, or insects, have much better means of studying links between behavioral differences and fitness outcomes, and can add a solid empirical basis to the theoretical framework needed in this study (see below).

4) Questionnaire method. There are some serious concerns about the questionnaire approach to assessing personality, but we recognize this is an approach that is used by some researchers and that the justification simply must be more robustly presented. Perhaps what is most problematic is that in this manuscript, the authors present this method as the method, thereby ignoring an enormous field of (ecological) studies that instead use an ethological approach and code actual behavior, or conduct experiments to test for consistency of behavior across time and context. Additionally, this field has proposed actual informed hypotheses about how personality influences life-history traits. The fact that the authors do not even refer to any of the work done by, to name only a few, Dingemanse, Réale and Sih, while talking about animal personality in light of evolution, is extremely problematic. Also, such studies do exist for chimpanzees; e.g. Uher and Asendorf 2008; Koski 2011; Massen et al., 2013, and also this should be acknowledged. Personality is defined as "inter-individual differences that are consistent over time and context" (something that the authors do not mention), and whereas they report here on inter-individual differences, they do not report anything about it being consistent or repeatable. It is common (and good) practice to use a test-retest design to check for such consistency, yet in primatology and especially when testing apes, this seems to be deemed unnecessary. Yet, as mentioned, this is how animal personality is defined and thus is very important. These authors (and others) tend to use the high inter-rater reliability they find as an argument against this. However, this is not the same as temporal consistency, and as mentioned before, this inter-rater reliability doesn't result from independent raters. Zoo keepers (and researchers alike) talk about their animals, and thus inadvertently but unavoidably, influence each other’s perception of, and consequently the ratings of, these animals. Further, all the references the authors use to validate the use of the rating method in the Materials and methods section (Weiss et al., 2009; King and Figueredo, 1997; Weiss, King and Hopkins, 2007; Herrelko, Vick and Buchanan-Smith, 2012; Vazine et al., 2007; Weiss, King and Figueredo, 2000; Wilson et al., 2017; Latzman et al., 2015; King, Weiss and Farmer, 2005; Weiss et al., 2012; Pederson, King and Landau, 2005; Latzman et al., 2015; Blatchley and Hopkins, 2010 –) are of people involved in this study, and thus not independent. In short, serious discussion needs to be undertaken to ensure the reader is aware that the questionnaire approach has its detractors, and the basis of those critiques. This has the potential to make this a much stronger paper because by providing a balanced view it can simultaneously present new data and pre-empt the problem of citation divergence (whereby some research groups cite only from select literature and others from a different set, and thus integration of these two literatures becomes compromised and thus detrimental to the overall scientific aims).

5) Hypothesis testing. The hypotheses are not clearly set up from the start, they are not embedded in any relevant theory (more on this below), and they are post hoc in nature. It appears to be a study where a large sample was input into some analyses to see what patterns emerged, and then those patterns were explained after the fact. A better approach would be to structure the lead-up so that it is clear what would be expected under what circumstances (phylogenetic, environmental, life history, etc.) and then test those hypotheses. In addition to modifying the setup to create a more rigorous set of well-supported expectations, the question be reworded to be more specific about chimpanzees and humans with respect to what traits should be linked with longevity. There is a good start to this discussion in the Introduction, and that could form the basis for a revised setup to the problem. There appear to be some expectations (Discussion, second paragraph), and these are discussed in an interesting way later, but the manuscript would be much stronger if these were clearly defined at the start and then systematically tested.

6) Theoretical framing. All predictions or interpretations are entirely based on previous empirical results, rather than derived from first principles. While this might be the norm in psychology, eLife is a biology journal, and evolutionary theory provides us with a framework from which to derive predictions about biological traits such as longevity and stable behavioral variation (that is presumably mediated by stable variation in neurobiology, metabolism, etc.). Thus, when examining links between consistent behavioral differences and longevity, an evolutionary biologist immediately thinks of life-history strategy as a possible underlying cause of both. Life history theory is especially relevant here as all sources of extrinsic mortality have been removed in this captive sample, and the chimpanzees are presumably dying because of intrinsic mortality; an individual's degree of investment in maintenance and repair, i.e. the things that reduce intrinsic mortality, is of course shaped by their life-history strategy, as are, arguably, consistent behavioral differences between individuals. Indeed, there is a large literature examining links between personality, longevity, and measures to reduce intrinsic mortality such as investment in immune function, in other animals called the 'pace-of-life syndrome' (e.g. Reale et al., 2010; Smith and Blumstein, 2008). This literature provides the kind of a priori predictions the current manuscript is lacking, such as certain personality dimensions being linked to longevity due to being part of a faster or slower life-history strategy. For example, achieving high dominance requires substantial investment in physical strength and muscle, associated with high testosterone levels and risk-taking behavior, which trade-off with investment in immune function etc. and are thus associated with a faster life-history strategy and higher extrinsic and intrinsic mortality (as is typical for most primate males compared to females); hence one would predict dominance to be negatively associated with longevity not positively, similar to the general sex difference in dominance and longevity (see e.g. Kruger and Nesse, 2005, Human Nature An evolutionary life history framework for understanding sex differences in human mortality rates). Conversely, an association between agreeableness and longevity is exactly what you would predict if agreeable individuals invest less in behavioral dominance and more in cooperation, which would be associated with a slower life-history strategy. Throughout the paper the authors speculate about possible causal links between personality traits and longevity (through 'controlling health', or 'health benefits conferred by intelligence'), which will need to be re-examined in light of theory that predicts both to be explained by a third variable (life-history strategy). For relevant arguments in humans, see e.g. several articles by Pepper and Nettle (2014 Human Nature, 2014 Applied evolutionary anthropology, 2017 Behavioral and Brain Sciences) that argue how a life history theory perspective can help explain variation in health behavior and thus SES-gradients in health. Of course, there are other evolutionary theories of personality (see e.g. Buss, 2009 How can evolutionary psychology successfully explain personality and individual differences?) but life-history theory provides the most direct link to longevity.

7) Use of a captive sample. The authors make strong claims about evolution and natural selection, yet test animals in a (non-natural) captive situation. As a consequence, selection pressures that have shaped evolution are being cancelled out and the effects of personality on longevity that the authors report are not informative for understanding the evolution of chimpanzees. For example, in this study there is no effect of extraversion (or boldness) on longevity, but it is obvious that such a trait may have an effect with actual predators around. Similarly, in the wild, were food is a limiting factor, dominance (which may not actually be a personality trait as it is not consistent if new opportunities arise) will have a major effect. As another example, the authors simultaneously argue that "observed effects in captive chimpanzees will be more comparable to effects found in similar human studies than would effects observed in wild chimpanzees". However, they go on to then offer an evolutionary explanation that seeks to describe their results in terms of ancestral behavior and the environment of selection (Abstract): "natural selection, after the divergence of hominins, favored the protective effects of high quality social bonds for males and exploratory behavior for females." The relationships they observe in fact seem equally explicable as factors that promote longevity specifically in captive situations. Although the authors do well to note this possibility, they appear to dismiss it in favor of their preferred alternative. Where they do find a lack of concordance with their expectations, the authors quickly engage in a useful discussion about the effects of captivity, but seem to discard this argument when they discuss their positive results. These alternative explanations must be carefully explored, and test implications set out (with substantive literature support) in order to seriously treat (and not just dismiss) the very real possibility that the observed pattern has no bearing at all on natural selection. One reviewer note that the captive sample can have its advantages, and these can be stressed. For example, the captive sample eliminates most extrinsic mortality, so that what remains is essentially how much individuals invest in maintenance and repair, which could well be related to their personality through life-history strategy (slow strategy = invest more = lower intrinsic mortality = 'nicer' personality). This still suffers from the problem that extrinsic mortality matters a lot in wild populations (and thus natural selection), but acknowledging these shortcomings, this study could be a good test of the idea that life-history strategy has consequences for both behavioral style and intrinsic mortality risk.

8) Analytical approach.a) It appears that the power analyses were conducted on the entire dataset (rather than pilot data), and thus constitute 'observed power'; this is unfortunately completely flawed and unnecessary. As demonstrated by Hoenig and Heisey (2001, Am Stat The abuse of power: The pervasive fallacy of power calculations for data analysis) there is nothing to be gained from such a retrospective power analysis, and indeed they may be entirely misleading. Power analysis only makes sense prospectively, using pilot data, and indeed eLife's transparent reporting form asks 'whether an appropriate sample size was computed when the study was being designed'. As this was not the case here, the power analyses should be removed.

b) I have to disagree with the dismissal of an age-confound on agreeableness based on a non-significant P-value of 0.077. P value thresholds are arbitrary conventions, and when there is an age pattern – the correlation of -0.08 is about as strong as the one for neuroticism at 0.09 – it should be controlled for, especially when one of the main findings is about an association between agreeableness and longevity. And while I appreciate that the authors fit several possible age models to the personality dimensions that did have significant correlations with age, I also disagree with selecting a single best model based on AICc (as the authors know, information criteria are better used to weight models and average predictions rather than select a single model [unless it receives all the weight]). Furthermore, polynomials are not ideal, and I would suggest using a spline term for age (using GAM) instead, which obviates the need to compare linear vs. non-linear fits. Incidentally, the fact that the best fit for the age effect on most personality dimensions was non-linear refutes the use of simple correlations. I would thus strongly suggest using GAM residuals for each personality dimension. As an aside, I was confused as to why date of birth rather than biological age was used?

9) Data accessibility. Two of the reviewers also expressed concern that the entire dataset may not be de-identified and available in published form (by assigning an ID to individual chimpanzees, and facilities). The editorial staff also had a discussion about the submission's compliance with the open-access policy of eLife. It is not clear to what extent the data can be precisely reproduced, given a lack of access to the full dataset that was used in the analysis. For example, how can personality links with lifespan be replicated without mortality data for the same individuals for which the personality attributes are known? Knowing social relationships, group size, etc. are also important because certain personality traits may be much more important in some specific settings than others.

1) Title. Given some of the uncertainties in the data and conclusions that are raised below, the title should be modified to ensure it faithfully reflects these issues.

Were-analyzed our data (discussed below), and these changes did not drastically change our findings. Since the results are unchanged, we do not see an obvious way to improve the title, but welcome the reviewers’ thoughts on this.

2) Phylogenetic assumptions. The central framing of the problem with respect to human evolution (Introduction, fourth paragraph) makes a standard phylogenetic assumption: that behavioral attributes found in both chimpanzees and humans should also have been shared by the last common ancestor. There is much debate over the utility of this approach for behavioral attributes, or if it might suffer overmuch from homoplasy – especially when there appear to be so many differences between chimpanzees and bonobos, with both equally related to humans (work by Sayers is relevant here). The authors should include discussion and literature that explains why here they make the claim so strongly, or why they claim to pinpoint the evolution of specific male and female personality attributes to such a specific period of time (since the split with hominins). This is especially the case when their research question (as currently states) prioritizes similarities between humans and chimpanzees (rather than bonobos).

We do not wish to understate how informative studies of bonobos are of these same issues; unfortunately, data from bonobos regarding longevity, are lacking. We have revised our text to acknowledge the importance of bonobos in this framework, though again, there is unfortunately little evidence from bonobos that we can bring to bear on these issues.

We have also addressed this concern in our addition of a phylogenetic perspective on great ape personality dimensions. To expand on our descriptions in the text, gorillas possess dominance, extraversion, and agreeableness (1), as well as dimensions like openness (2) and conscientiousness (3), but the evidence for these is weaker. Humans and chimpanzees possess the same Big 5 factors, but chimpanzees have the additional dominance factor (4-6). Bonobos differ from chimpanzees, humans, and possibly gorillas, as bonobos have a dominance-like factors (“assertiveness”), two factors describing different aspects of conscientiousness, an openness and agreeableness factor, and an extraversion factor that may also be termed “social withdrawal” (7).

From this, we can say that the most phylogenetically parsimonious tracing of personality in African apes is that dominance (or assertiveness), extraversion, agreeableness, neuroticism, openness, and possibly conscientiousness, are ancestral in African apes. The bonobo configuration regarding extraversion, conscientiousness, and neuroticism is derived, as is the lack of dominance in humans. A concise version of this explanation is now given in the manuscript (–Introduction, ninth paragraph).

The discussion that leads to this point also appears to somewhat answer the question in advance, because it is clear that the authors are already arguing that a link between personality and fitness is ancestral in primates (and therefore will also be linked in chimpanzees, humans, presumably bonobos, and their last common ancestor). This argument seeps back in later with the discussion of gorillas (Discussion, third paragraph), where they are referred to rather oddly (considering the phylogenetic argument up to this point and the actual genetic relatedness of gorillas and chimps versus chimps and bonobos or chimps and humans) as "close chimpanzee cousins". Thus, the phylogenetic argument should be very carefully constructed: what traits do they think are derived in chimpanzees, what traits ancestral to both chimpanzees and humans, what is the rationale, etc.?

Please see our previous response. We have revised our review of the literature, and in particular we have expanded our theory that explains our arguments as to which personality-longevity associations may be ancestral or derived. Please see our responses to query 5 concerning hypothesis testing, as well.

3) Use of literature. The paper opens with a strong background discussion, but reads like a bank of examples rather than a coherent lead-up to a clear set of hypotheses (more on this below). Each cited study shows a suggestive tendency for proxies of fitness to link with some aspect of personality, but there is clear diversity within primates regarding how each of these relationships actually plays out (as well as in what fitness measures are used). This background captures some of the ambiguity and gaps in current knowledge, but it does not confront them head-on. The paper would increase its impact if it clearly describes the areas where there is more surety than others, and specifically where increased work is necessary and why.

We have revised our Introduction and Discussion with an eye on confronting the strengths, weaknesses, and applicability of the related bodies of literature in primate behavior, psychology, and behavioural ecology. In doing so, we incorporated life-history theory, while also attempting to reconcile the nonhuman primate research with the different approaches and results in humans. We have attempted to address the main theoretical and empirical gaps in the existing literature, while maintaining a cohesive picture of the broader theory.

There is also a near-complete lack of citations to the non-primate animal personality literature. It is generally disconcerting that the primate personality and non-primate personality literatures don't often cite each other, but it is especially puzzling here since researchers working with short-lived, easily manipulated species like birds, fish, or insects, have much better means of studying links between behavioral differences and fitness outcomes, and can add a solid empirical basis to the theoretical framework needed in this study (see below).

We agree with the reviewers that better integration of the literatures should be a goal of such studies and regret that we did not set a good example in the previous version of this manuscript. To address this, we expanded our Introduction to recognize and incorporate relevant reviews and syntheses, which are largely based on non-primates. Moreover, where possible and appropriate, we have cited specific studies from non-primates.

4) Questionnaire method. There are some serious concerns about the questionnaire approach to assessing personality, but we recognize this is an approach that is used by some researchers and that the justification simply must be more robustly presented. Perhaps what is most problematic is that in this manuscript, the authors present this method as the method, thereby ignoring an enormous field of (ecological) studies that instead use an ethological approach and code actual behavior, or conduct experiments to test for consistency of behavior across time and context. Additionally, this field has proposed actual informed hypotheses about how personality influences life-history traits. The fact that the authors do not even refer to any of the work done by, to name only a few, Dingemanse, Réale and Sih, while talking about animal personality in light of evolution, is extremely problematic. Also, such studies do exist for chimpanzees; e.g Uher and Asendorf 2008; Koski 2011; Massen et al., 2013, and also this should be acknowledged.

Our original intention was to describe the history and validity of the method we used, but we apparently did so at the cost of discussing the broader issues. To address this weakness in the manuscript, we expanded the background of our methods to describe related methods including those that do not use questionnaires. We have done this in the current version, citing (to the best of our knowledge) all personality studies that have examined chimpanzees, using either subjective ratings or behavioral measures. This includes all the papers mentioned by the reviewers, as well as others. We also incorporated the foundational work of non-primatologist behavioural ecologists who study personality, Réale and Dingemanse in particular.

We believe this comprehensive take on the extant literature has improved the manuscript, as it demonstrates the convergence of personality measures across studies and approaches. We mention this in the paper (–Introduction, ninth and tenth paragraphs), but we wish to point it out here as well, that our questionnaire is more than just a set of adjectives – it includes both adjectives and a behavioural description for each adjective.

Personality is defined as "inter-individual differences that are consistent over time and context" (something that the authors do not mention), and whereas they report here on inter-individual differences, they do not report anything about it being consistent or repeatable. It is common (and good) practice to use a test-retest design to check for such consistency, yet in primatology and especially when testing apes, this seems to be deemed unnecessary. Yet, as mentioned, this is how animal personality is defined and thus is very important. These authors (and others) tend to use the high inter-rater reliability they find as an argument against this. However, this is not the same as temporal consistency, and as mentioned before, this inter-rater reliability doesn't result from independent raters. Zoo keepers (and researchers alike) talk about their animals, and thus inadvertently but unavoidably, influence each other’s perception of, and consequently the ratings of, these animals.

Before responding, we wish to correct a misperception. There is an understanding of the importance of repeatability as one way to establish the reliability of personality measures in the primate literature. We are unaware of anybody who argues that this is not important or who use interrater reliabilities as an argument against also establishing repeatabilities. In fact, repeatabilities or re-test reliabilities of personalities in primates have been examined since at least the 1970s, and this has also played a large role in studies of human personality research. However, given the origins of these subjects, it is not always possible to obtain these kinds of data (keepers often have limited time to participate).

Nevertheless, in our expanded Introduction, we discuss the repeatability of measures (tenth paragraph). Several studies (some involving a subset of our data and some independent studies) have examined temporal consistency in chimpanzee personality, occasionally over the course of decades (6, 8-10). In their review of the primate personality literature, Freeman and Gosling (11) note six studies, of rhesus macaques and the great apes, that presented test-retest reliability. Since that review, more papers have demonstrated repeatability (12-15).

Further, all the references the authors use to validate the use of the rating method in the Materials and methods section (Weiss et al., 2009; King and Figueredo, 1997; Weiss, King and Hopkins, 2007; Herrelko, Vick and Buchanan-Smith, 2012; Vazine et al., 2007; Weiss, King and Figueredo, 2000; Wilson et al., 2017; Latzman et al., 2015; King, Weiss and Farmer, 2005; Weiss et al., 2012; Pederson, King and Landau, 2005; Latzman et al., 2015; Blatchley and Hopkins, 2010) are of people involved in this study, and thus not independent. In short, serious discussion needs to be undertaken to ensure the reader is aware that the questionnaire approach has its detractors, and the basis of those critiques. This has the potential to make this a much stronger paper because by providing a balanced view it can simultaneously present new data and pre-empt the problem of citation divergence (whereby some research groups cite only from select literature and others from a different set, and thus integration of these two literatures becomes compromised and thus detrimental to the overall scientific aims).

Initially, we took this approach because other ratings studies were most comparable to our in terms of methods used, applications, etc. As mentioned above, we have now cited the chimpanzee (and other primate) literature more widely. In our Introduction we discuss the criticisms of rating techniques, and mention reliability and validity of our instrument in both the Introduction and methods. Ultimately, the revised manuscript gives what we hope is a balanced take on the strengths of the approach and the convergent validity of these methods for assessing personality in chimpanzees.

5) Hypothesis testing. The hypotheses are not clearly set up from the start, they are not embedded in any relevant theory (more on this below), and they are post hoc in nature. It appears to be a study where a large sample was input into some analyses to see what patterns emerged, and then those patterns were explained after the fact. A better approach would be to structure the lead-up so that it is clear what would be expected under what circumstances (phylogenetic, environmental, life history, etc.) and then test those hypotheses. In addition to modifying the setup to create a more rigorous set of well-supported expectations, the question be reworded to be more specific about chimpanzees and humans with respect to what traits should be linked with longevity. There is a good start to this discussion in the Introduction, and that could form the basis for a revised setup to the problem. There appear to be some expectations (Discussion, second paragraph), and these are discussed in an interesting way later, but the manuscript would be much stronger if these were clearly defined at the start and then systematically tested.

We have rewritten the entire Introduction to incorporate life history theory into our manuscript, as the central, guiding theory. From the outset, we develop the theory of personality in that context, before discussing specifics of primates and personality measurement, and finally situating our hypotheses in the life history framework.

6) Theoretical framing. All predictions or interpretations are entirely based on previous empirical results, rather than derived from first principles. While this might be the norm in psychology, eLife is a biology journal, and evolutionary theory provides us with a framework from which to derive predictions about biological traits such as longevity and stable behavioral variation (that is presumably mediated by stable variation in neurobiology, metabolism, etc.). […] Throughout the paper the authors speculate about possible causal links between personality traits and longevity (through 'controlling health', or 'health benefits conferred by intelligence'), which will need to be re-examined in light of theory that predicts both to be explained by a third variable (life-history strategy). For relevant arguments in humans, see e.g. several articles by Pepper and Nettle (2014 Human Nature, 2014 Applied evolutionary anthropology, 2017 Behavioral and Brain Sciences) that argue how a life history theory perspective can help explain variation in health behavior and thus SES-gradients in health. Of course, there are other evolutionary theories of personality (see e.g. Buss, 2009 How can evolutionary psychology successfully explain personality and individual differences?) but life-history theory provides the most direct link to longevity.

As mentioned in the previous response, we have rewritten the paper, from the Introduction onward, with life history theory at the front and centre. We believe that our adherence to the theory and the predictions we draw from it accurately reflect the extrinsic and intrinsic factors that are in play in this population.

7) Use of a captive sample. The authors make strong claims about evolution and natural selection, yet test animals in a (non-natural) captive situation. As a consequence, selection pressures that have shaped evolution are being cancelled out and the effects of personality on longevity that the authors report are not informative for understanding the evolution of chimpanzees. For example, in this study there is no effect of extraversion (or boldness) on longevity, but it is obvious that such a trait may have an effect with actual predators around. Similarly, in the wild, were food is a limiting factor, dominance (which may not actually be a personality trait as it is not consistent if new opportunities arise) will have a major effect. As another example, the authors simultaneously argue that "observed effects in captive chimpanzees will be more comparable to effects found in similar human studies than would effects observed in wild chimpanzees". However, they go on to then offer an evolutionary explanation that seeks to describe their results in terms of ancestral behavior and the environment of selection (Abstract): "natural selection, after the divergence of hominins, favored the protective effects of high quality social bonds for males and exploratory behavior for females." The relationships they observe in fact seem equally explicable as factors that promote longevity specifically in captive situations. Although the authors do well to note this possibility, they appear to dismiss it in favor of their preferred alternative. Where they do find a lack of concordance with their expectations, the authors quickly engage in a useful discussion about the effects of captivity, but seem to discard this argument when they discuss their positive results. These alternative explanations must be carefully explored, and test implications set out (with substantive literature support) in order to seriously treat (and not just dismiss) the very real possibility that the observed pattern has no bearing at all on natural selection. One reviewer note that the captive sample can have its advantages, and these can be stressed. For example, the captive sample eliminates most extrinsic mortality, so that what remains is essentially how much individuals invest in maintenance and repair, which could well be related to their personality through life-history strategy (slow strategy = invest more = lower intrinsic mortality = 'nicer' personality). This still suffers from the problem that extrinsic mortality matters a lot in wild populations (and thus natural selection), but acknowledging these shortcomings, this study could be a good test of the idea that life-history strategy has consequences for both behavioral style and intrinsic mortality risk.

We have scaled back our claims about the explanatory power of this sample, and have reoriented our findings in the context of the extrinsic and intrinsic factors, relevant to each personality dimension, that are or are not in play for captive chimpanzees. In our revised hypotheses, we consider all the major extrinsic and intrinsic differences between captive and wild environments, and how these would be related to strategic life-history differences. We also highlighted the drawbacks and advantages to working with a captive sample (–Introduction, eleventh paragraph; Discussion, seventh paragraph).

8) Analytical approach.a) It appears that the power analyses were conducted on the entire dataset (rather than pilot data), and thus constitute 'observed power'; this is unfortunately completely flawed and unnecessary. As demonstrated by Hoenig and Heisey (2001, Am Stat The abuse of power: The pervasive fallacy of power calculations for data analysis) there is nothing to be gained from such a retrospective power analysis, and indeed they may be entirely misleading. Power analysis only makes sense prospectively, using pilot data, and indeed eLife's transparent reporting form asks 'whether an appropriate sample size was computed when the study was being designed'. As this was not the case here, the power analyses should be removed.

We recognize the issues with power analysis in the context of our study. We hoped the analyses would be useful to a statistically literate audience, but since the reviewers point out that the analyses could be misleading or distracting, we have removed them.

b) I have to disagree with the dismissal of an age-confound on agreeableness based on a non-significant P-value of 0.077. P value thresholds are arbitrary conventions, and when there is an age pattern – the correlation of -0.08 is about as strong as the one for neuroticism at 0.09 – it should be controlled for, especially when one of the main findings is about an association between agreeableness and longevity. And while I appreciate that the authors fit several possible age models to the personality dimensions that did have significant correlations with age, I also disagree with selecting a single best model based on AICc (as the authors know, information criteria are better used to weight models and average predictions rather than select a single model [unless it receives all the weight]). Furthermore, polynomials are not ideal, and I would suggest using a spline term for age (using GAM) instead, which obviates the need to compare linear vs. non-linear fits. Incidentally, the fact that the best fit for the age effect on most personality dimensions was non-linear refutes the use of simple correlations. I would thus strongly suggest using GAM residuals for each personality dimension. As an aside, I was confused as to why date of birth rather than biological age was used?

We very much appreciate the suggestion that we used GAMs to control for age confounding in the personality data. We were not fully aware of the applicability of this method to the problem we faced. In response, we have now reanalysed the age and personality relationships for all traits in a GAM framework, and have used the GAM residuals in subsequent analyses instead of the simpler linear and polynomial model residuals. We have also changed the confounding variable from DoB to age at personality rating. These changes did not alter our results, except the association between longevity and residualized openness in females. The hazard ratio shifted by about 0.05, and the new 95% confidence interval now overlaps with 0. We have scaled back our conclusions accordingly, and our revised manuscript interprets openness and longevity in light of the ambiguity of this finding (Discussion, fourth paragraph).

9) Data accessibility. Two of the reviewers also expressed concern that the entire dataset may not be de-identified and available in published form (by assigning an ID to individual chimpanzees, and facilities). The editorial staff also had a

Since submitting the paper we have consulted with all authors and interested parties on the matter of data accessibility. We are pleased to say that all the data we used in our analyses can be published along with the manuscript, should it be accepted.

References

1) Gold KC, Maple TL. Personality assessment in the gorilla and its utility as a management tool. Zoo Biol. 1994;13(5):509-22.

2) Eckardt W, Steklis HD, Steklis NG, Fletcher AW, Stoinski TS, Weiss A. Personality dimensions and their behavioral correlates in wild Virunga mountain gorillas (Gorilla beringei beringei). Journal of Comparative Psychology. 2015;129(1):26-41.

3) Schaefer SA, Steklis HD. Personality and subjective well‐being in captive male western lowland gorillas living in bachelor groups. Am J Primatol. 2014;76:879-89.

4) King JE, Figueredo AJ. The Five-Factor Model plus Dominance in chimpanzee personality. Journal of Research in Personality. 1997;31:257-71.

5) Freeman HD, Brosnan SF, Hopper LM, Lambeth SP, Schapiro SJ, Gosling SD. Developing a comprehensive and comparative questionnaire for measuring personality in chimpanzees using a simultaneous top-down/bottom-up design. Am J Primatol. 2013;75(10):1042-53.

6) Dutton DM. Subjective assessment of chimpanzee (Pan troglodytes) personality: Reliability and stability of trait ratings. Primates. 2008;49:253-9.

7) Weiss A, Staes N, Pereboom JJM, Inoue-Murayama M, Stevens JMG, Eens M. Personality in bonobos. Psychol Sci. 2015;26(9):1430-9.

8) King JE, Weiss A, Sisco MM. Aping humans: Age and sex effects in chimpanzee (Pan troglodytes) and human (Homo sapiens) personality. Journal of Comparative Psychology. 2008;122:418-27.

9) Buirski P, Plutchik R, Kellerman H. Sex differences, dominance, and personality in the chimpanzee. Anim Behav. 1978;26:123-9.

10) Weiss A, Wilson ML, Collins DA, Mjungu D, Kamenya S, Foerster S, et al. Personality in the chimpanzees of Gombe National Park. Scientific data. 2017;4:170146.

11) Freeman HD, Gosling SD. Personality in nonhuman primates: A review and evaluation of past research. Am J Primatol. 2010;72:653-71.

12) Weiss A, Adams MJ, Widdig A, Gerald MS. Rhesus macaques (Macaca mulatta) as living fossils of hominoid personality and subjective well-being. Journal of Comparative Psychology. 2011;125:72-83.

13) von Borell C, Kulik L, Widdig A. Growing into the self: the development of personality in rhesus macaques. Anim Behav. 2016;122:183-95.

14) Neumann C, Agil M, Widdig A, Engelhardt A. Personality of wild male crested macaques (Macaca nigra). PLoS ONE. 2013;8:e69383.

15) Brent LJN, Semple S, MacLarnon A, Ruiz-Lambides A, Gonzalez-Martinez J, Platt MJ. Personality traits in rhesus macaques (Macaca mulatta) are heritable but do not predict reproductive output. Int J Primatol. 2014;35:188-209.