The importance of first impression judgements in interspecies interactions

Close human-wildlife interactions are rapidly growing, particularly due to wildlife tourism popularity. Using both laboratory and ecological observation studies we explored potential interspecies communication signalling mechanisms underpinning human-animal approach behaviour, which to date have been unclear. First impression ratings (n = 227) of Barbary macaques’ social and health traits were related to the macaques’ facial morphology and their observed behaviour supporting a shared facial signalling system in primates. These ratings significantly predicted intended approach to the macaques during hypothetical interactions. Finally, real-world interspecies proximity was observed and found to be best predicted by the interaction between human first impression perception and animal behaviour. Specifically, perceived macaque health in interaction with actual macaque dominance drives close interactions despite human proclivity to avoid dominant animals, raising safety concerns in interspecies interactions.

animals and will prefer to look at and approach 'cute' animals whilst avoiding 'ugly' animals, possibly due to a perceived risk of contamination 22 . Therefore, perceived attractiveness, cuteness and healthiness judgments might be key to human's willingness to interact with animals.
Baby schema are cues of 'cuteness' 22 , defined as a set of infantile features consisting of a round face, large head, big eyes, high forehead, chubby cheeks, small nose and small mouth 23 . The features of a round face and large eyes have also been found to be more typical of a female face 24 . So, a 'cuter' face with a high baby schema should be typical of females and this is supported by the established 'cute is female' stereotype 25,26 . It has been found that humans are not only sensitive to human infant baby schema but also to baby schema as a rating of cuteness in adult faces 27 . In addition, humans are sensitive to animal baby schema 28,29 and have an instinctive ability to recognise cuteness in the form of baby schema in animals 30,31 . Therefore, baby schema has the potential to be a valid predictor of cuteness and sex in animals, and might influence human intention to approach/interact with animals.
The rapid expansion of human populations and wildlife tourism popularity have brought humans and wild animals into frequent proximity, which often results in close interspecies interactions. Such close interactions have been shown to increase potential risks to both human and animal welfare [32][33][34] . For example, primate bites are the second most common animal bite risk to travellers after dog bites, and account for up to 21% of all animal bites worldwide 35 . Despite the potential risks, humans often participate in interactions by either approaching and observing, feeding or taking a photograph with animals 36 . These behaviours are particularly common with primates, such as the Barbary macaque 34,37,38 , yet little is known about the interspecies communication mechanisms underlying human-animal interactions.
This research aims to test for the first time different aspects of first impression formation between species by exploring (1) whether human social trait judgements of Barbary macaques influence intended proximity for approaching, feeding or taking a photograph with the macaques; (2) whether these judgements are based on macaque facial morphological features; (3) if these judgements accurately reflect the real behaviour of the macaques. Finally, this research will explore (4) whether human first impression judgements and/or macaque behaviour best predict human-macaque proximity during real interactions. We hypothesise that human-made social, demographic and health trait judgements will be related to macaque facial morphology and will predict intended proximity and observed macaque behaviour. More specifically, human made dominance judgments will positively relate to fWHR and a high fWHR will negatively predict intended approach. Whilst cuteness, attractiveness and healthiness judgements will positively relate to baby schema with high baby schema positively predicting intended approach. In addition, we hypothesise that both human social trait judgements and observed animal behaviour will be important determinants of proximity in real-life interactions. We conducted an online questionnaire asking 227 participants to rate 17 neutral macaque faces on nine individual characteristics (dominance, trustworthiness, attractiveness, cuteness, healthiness, socialiness, activity, age and sex). We then asked how close participants would get to simply approach, feed and take a selfie with the macaques. In addition to the questionnaire, we also measured each macaque's fWHR and baby schema based on the pictures. Finally, we used behavioural and human-macaque proximity data previously collected for the 17 macaques shown in the questionnaire.

Results
Participants were more likely to report that they would choose to come in to greater proximity to feed the macaques than to take photographs with the macaques (N = 17, t = −2.871, P = 0.011) or simply approach them (N = 17, t = −9.904, P < 0.001, Fig. 1). In addition, participants reported they would be willing to get closer to take a photograph with the macaques than they would to simply approach them (N = 17, t = −11.293, P < 0.001).
Question 1: Do human social, demographic and health trait judgements of Barbary macaque faces predict intended human-macaque proximity?
From the online questionnaire, perceived social, demographic and health traits predicted how closely people would approach the macaques to interact with them (Table 1). People were more likely to approach in general, to feed or take a photograph with the macaques when they were perceived as trustworthy, subordinate, cute, social, young and female. In addition, people tended to be more likely to approach the macaques to feed them when they were perceived as healthy, but this did not reach significance.
Question 2: Are social, demographic and health judgements based on macaque facial morphology? Participants seemed to base their social and health judgements on cues found in macaque facial morphology. fWHR was positively correlated with perceived dominance, and negatively correlated with perceived Figure 1. Distribution of the average distance in meters participants reported being willing to approach each macaque for the three conditions: approach, feeding or taking a photograph.

Discussion
Close human-wildlife interactions increase risks to both human and animal welfare. Therefore, being able to make an accurate judgement of survival-related social, health and demographic characteristics of wild animals would lead to an effective interspecies communication, which could reduce the risks for both species. Overall our findings support our hypothesis that human first impression judgements related to intended approach to primates. Our results indicate that people are more likely to intend to approach, approach to feed or approach to take a photograph with a macaque when the macaque was perceived as trustworthy, subordinate, cute, social, young and female. Facial morphology measures were highly correlated to social characteristics and sex, which suggest that humans may use these morphological clues to determine these characteristics in macaques. In addition, perceived macaque social characteristics were highly correlated with observed macaque aggression rates, perceived health was negatively correlated with observed macaque self-scratching rates-an indicator of animal welfare. Finally, although people seem to be able to accurately assess survival-related characteristics in macaques based on pictures of their neutral facial expression, and the perceived characteristics predict their intended proximity to each macaque; this is not related to observed human-macaque proximity recorded in the real world. The interactions between perceived macaque health and actual macaque dominance were the best predictors of the observed distance between both species.  Table 3. Relationships between perceived macaque social traits and observed macaque behaviour. www.nature.com/scientificreports www.nature.com/scientificreports/ Our findings support the hypothesis that humans and other primate species share a social and health facial signalling system 39 , that is used by humans to determine intended approach to macaques. Humans intended to closely interact with macaques that they perceived as trustworthy, subordinate, cute, social, young and female. These perceived social and demographic traits influenced the distance they would be willing to approach each macaque independent of the type of interaction, observation, feeding or taking a photograph. These are the same characteristics preferred for human-human interactions. People are more likely to approach and interact with someone perceived as trustworthy, subordinate, attractive, social, young, and female 26,[40][41][42] . Interactions with a person presenting these characteristics is thought to be preferred because they are 'safer' with lower risk of conflict 5 . Humans might also prefer these characteristics to reduce possible health risks when closely interacting with another primate species, or more broadly another animal species.
As previously mentioned, humans have a preference to look at 'cuter' animals 22 . For instance, hikers were more likely to stop and approach and take photographs with animals they viewed as 'cute' 43 . The authors explain that this behaviour was due to the 'cute response' , which they described as 'the compulsion to nurture and not hurt the cute entity' . This so-called 'cute response' has also been linked to anthropomorphism by transposing human societal values onto animals 44,45 . The anthropomorphic views of wildlife as being 'cute' removes the concept of unpredictability in potentially dangerous wild animals 46 , meaning humans will not perceive approaching 'cuter' animals as dangerous or risky so they will be more inclined to do so.
Our results indicate that perceived social and health traits were significantly correlated to observed macaque behaviours, and thus suggest that humans can accurately assess primate's social traits and health status based on their faces. Trustworthiness, cuteness and socialness were negatively correlated to observed rates of macaque aggression, but perceived dominance was positively related to observed rates of macaque aggression. No perceived social, health or demographic traits were related to observed macaque dominance. Perceived health was negatively associated with observed rates of macaque self-scratching, a proxy of animal welfare 33 . However, humans were not able to assess the sex of the macaques. No actual record of macaque age was available to determine the accuracy of perceived age. Previous studies have found that humans are able to accurately assess dominance, trustworthiness, agreeableness (here: cuteness), healthiness, socialness, age but not sex in chimpanzees 9,10 . In addition, human ability to accurately assess these traits in faces in different primate species might explain the high correlation found between trait ratings given by familiar observers and behavioural coding methods (e.g., rhesus macaques 47 ; vervet monkeys, Chlorocebus pygerythrus 48 ; chimpanzees 49,50 ).
Our findings also demonstrate that facial morphology in macaques (fWHR and baby schema) can transmit similar social, health and demographic information as do human faces, and that humans can identify and use this information when deciding to approach macaques. Indeed, higher fWHR has been negatively related to trustworthiness and positively associated with dominance in many primate species including humans 4 , capuchins, Sapajus Cebus sp 14 , macaques, Macaca genus 18 . Recently, Japanese and rhesus macaques have been shown to display preferential attention toward human faces with lower fWHR, suggesting that macaques can form human like first impressions based on facial features 11 . Therefore, it appears that both macaques and humans may share the cognitive mechanisms required for processing social and health traits from facial morphology, supporting the idea of an honest interspecies communication.
Contrary to our predictions baby schema measurements were not correlated with perceived cuteness or age. Perceived cuteness is highly associated with age in animals, with younger individuals perceived as cuter than adults 31,51 . The age variability in our macaques was limited as all macaques were adults, which may explain why the baby schema measurements did not correlate with perceived age nor perceived cuteness. However, baby schema measurements were positively correlated with perceived health and sex. Previous research has shown that baby-faced adult humans were judged as less dominant, more honest, feminine and approachable 24,52,53 . Therefore, such facial morphological clues might be used by humans to assess health and sex in macaques, and such clues influence intended approach proximity towards the macaques. However, perceived sex was not related to sex in macaques, which can be explained by the low sexual dimorphism in Barbary macaque neutral faces.
Although our findings support our hypothesis that human first impression judgements related to intended approach, intended behaviour made by our participants did not reflect observed proximity between humans and macaques observed at the field site over an 11-months period. Actual macaque dominance in interaction with perceived health best predicted the observed distance between humans and macaques (over 60%). Given that our findings suggest that humans intend to approach subordinate animals more closely, this suggests that macaques play an active role in human-macaque interactions, with dominant individual macaques approaching more closely than humans intend or perceive to be safe. Previous studies have found that dominant primates have precedence over clumped food resources, and conflicts often occurs between macaques when food distribution is spatially limited 54,55 . As most human-macaque interactions at the field site were driven by food 38 , the greater proximity of dominant macaques compared to subordinate individuals is therefore expected. In addition, perceived health seemed to influence the proximity to which the humans interact with individual macaques. Although perceived health appears to accurately reflect overall health status of macaques, it appears that human-macaque interaction rates do not significantly change when macaques are ill as reported in 56 using the same macaque group as in this study. This highlights serious concerns for potential pathogen transmission between both species.
Although intended approach to each macaque was not related to observed human-macaque proximity recorded in the real-world it is important to consider some further points in relation to this finding. Firstly, in the real-world additional cues such as different and changing macaque facial expressions, body gesture and vocalisation, not only neutral faces can be used to guide macaque approachability. In addition, observed human approach attempt could have been more representative of the intended approach measured in the questionnaire than observed proximity, which is co-dependent of both human and animal behaviour. However, such measure was not available. Therefore, future research should investigate the relationships between human approach attempt and achieved proximity to better dissociate the influence of the human first impression judgements from (2020) 10:2218 | https://doi.org/10.1038/s41598-020-58867-x www.nature.com/scientificreports www.nature.com/scientificreports/ the macaque behaviour. Also, intended or observed interspecies proximity may be associated with human traits such as risk-taking, sensation seeking and neuroticism. However, to date, none of these factors have been studied in relation to wildlife approachability and further research should be conducted to establish the relative importance of these factors in interspecies interactions. Finally, it should be noted that intended approach in this analysis was the aggregate from 227 individuals and that observed proximity for real-life interactions was determined from a different group of individuals potentially masking a significant correlation between intended and observed proximity.
In conclusion, our results provide evidence of a shared facial signalling system in primates. Such honest facial signalling may be an evolutionary adaptation that confers advantages in intra-and interspecies interactions, e.g. preventing potentially detrimental conflicts or the spread of pathogens. Such information is used by humans to make accurate judgements of social and health traits of macaques, resulting in an adjustment of their intended behaviour towards the macaques. However, an inter-species interaction requires the involvement of at least two parties. Here our findings suggest that feeding wildlife increases the risks to human safety by bringing dominant animals in to greater proximity with humans, who would otherwise be less likely to intend to approach these animals based on their first impression judgement. We suggest that regulating the feeding of wildlife is an urgent requirement to reduce human and animal welfare risks.

Methods
Participants. Two hundred and twenty-seven participants took part (37 males, 189 females & 1 non-binary), aged between 18 and 76 years (M = 24.9 years, SD = 13.3 years). All aspects of the data collection and analysis reported were carried out in accordance with the guidelines approved by the University of Lincoln School of Psychology ethics committee (ethics code PSY1718540).
procedure. The study was conducted online via Qualtrics software (Qualtrics2018 v. 8). The link to the study was distributed to students at the University of Lincoln in exchange for course credits, and to the general public via social media. Participants then rated faces of 17 Barbary macaques on seven social traits (dominance, trustworthiness, attractiveness, cuteness, healthiness, socialness and activity) on a 7-point Likert scale from extremely low to extremely high. Demographic-like traits of age and sex were rated on a 7-point Likert scale from very young to very old and very feminine to very masculine. Participants saw each monkey face individually, and rated each face on all nine traits before moving on to the next face. Participants were then asked to select how close they would get to approach, approach to feed or approach to take a selfie with each monkey; from 11 options ranging from 0 m to 100 m, including an option to not approach.
Stimuli. The images used were all faces of different Barbary macaques from the same group, 8 males and 9 females taken by the corresponding author. All images showed Barbary macaques with neutral facial expressions, in a frontal pose, and in high resolution to allow facial morphological measurements.
Measurements. All facial measurements were taken using the image manipulation programme GIMP 2.10.4, including the measure tool and rectangle select tool (Fig. 3). Each macaque image was scaled to a width of 130 mm and measurements were collected in mm and recorded to two decimal places. All measurements were www.nature.com/scientificreports www.nature.com/scientificreports/ taken by two independent researchers. All measurements were within 10 mm of one another, so all were accepted, and the final measurement used for the following calculations was the average of both measurements.
Measuring the facial width to height ratio (fWHR). Facial width to height ratio (fWHR) was calculated from the maximum horizontal distance from the left to right zygion referred to as bizygomatic width and the upper face height, measured as the vertical distance from the upper lip to the highest point of the eyelids 14,18 (Fig. 1). Using the GIMP 2.10.4 rectangle select tool, a box was drawn out to these specifications and then measured accordingly. The final measurement was calculated by bizygomatic width divided by upper face height.
Measuring baby schema. Baby schema measurements were adapted from two papers 23,31 . The distance between selected facial landmarks were measured (Fig. 1): Forehead length (AO1), Left eye width (E1F1), Right eye width (E2F2), Average eye width (EF; calculated from E1F1 and E2F2), Face width (CD1), Face length (AB1), Mouth length (KL), Nose length (OH), Nose width (IJ). Using these measurements, five facial parameters were required to calculate baby schema: AO/AB1, EF/CD1, OH/AB1, IJ/CD1, KL/CD1. The results of these calculations were all then scaled appropriately so that AB1 equalled 100 mm, to control for the variance of facial height among each image. The mean and standard deviation for each facial parameter were then calculated. Comparing each scaled measurement to its mean, if AO/AB and EF/CD were greater than the mean it showed high baby schema. If OH/ AB, IJ/CD and KL/CD were less than the mean it showed high baby schema. If the facial feature indicated high baby schema it stayed a positive number. If it indicated low baby schema, it was changed to a negative number, for example from 0.5 to −0.5. Final baby schema scores were the average of each facial parameter z scores.
Observed macaque behavioural and human-macaque interaction data. This 56 was used to assess average rates of aggression and self-scratching, and scan sampling was used to measure the distance between humans and macaques such as 0 m, 0-1 m, 1-2 m, 2-5 m, and 5-10 m (for more details 38,57 ). The dominance rank of each macaque was calculated using corrected normalized David' scores based on the outcomes of all visible same-sex dyadic conflicts with no counter-aggression using ad libitum sampling 38 .
Data analysis. Prior to analysis, the perceived trait, age and sex data were ranked from lowest score (1) to highest score (7), and an index of distance was used ranging from touching (0) to would not approach (11). All categorical variables were then z-transformed to improve the interpretability of the variables 58 . Question 1: A series of GLMMs was used to explore whether participants' willingness to approach, approach to feed the macaques and approach to take a photograph with the macaques (dependent variables) were influenced by macaque characteristics (predictors: dominance, trustworthiness, attractiveness, cuteness, health, sociality, activity, perceived age and sex). All models were fitted using R software (R 3.5.3, R development core team 2018). All ranked variables such as index of distance and perceived traits were z-transformed to improve the interpretability of the variables 58 . The three models had as non-nested random factors the identities of the macaques and the participants. The models used the function lme of the R-package nlme for Gaussian linear mixed-effects models 59 . For each model, the significance of the full model was compared to the corresponding null model, i.e. model with all predictions replaced by '1' , using a likelihood ratio test (R function ANOVA). All models were checked to assess whether they violated any assumptions, which none did, including collinearity (VIF function, all VIF results <4, ranging 1.3 to 2.2), outliers (Cook's distance = 0.007, no outlier found), distribution and homogeneity of the residuals 60 .
Questions 2-3: Spearman's rho correlations (Spearman's R, using the cor function in R) were then used to investigate the relationships between trait judgement scores and macaques' facial morphological measurements. A Wilcoxon test (R function wilcox.test) was used to compare perceived sex ranked to macaques' facial measurements and Chi square test (R function chisq.test) was used to compare binary perceived sex (Female ≤3.5 average scores, Male >3.5 average score) to actual sex of the macaques. No Bonferroni correction was applied as analyses on each variable were hypothesis driven and tested fewer than ten times, as well as to ensure that the likelihood of the type I error was low 61 . These tests were conducted using R (R 3.5.3, R development core team 2018), significance at α = 0.05, two-tailed. All continuous variables were not normally distributed (Shapiro tests p < 0.05) or ranked, so non-parametric tests were used.
Question 4: First a spearman correlation was used to explore the relationships between intended behaviour and observed human-macaque proximity per macaque (N = 17). Then, a series of linear regressions (R function lm) was used to determine which human's first impression judgements (average scores rated by the 227 participants for each macaque) and/or macaques' behaviour best predict human-macaque proximity during real interactions. A series of simple linear regression were conducted to assess which macaque behaviours best predicted real human-macaque proximity, i.e. dominance, aggression rates and self-scratching rates. Then, as 50% of the variability in proximity was predicted by macaque actual dominance (see result section), interactions between judgment traits and actual dominance were then explored, i.e. cuteness * dominance, health * dominance, age * dominance. Due to the high collinearity between the predictors, several simple linear regressions were conducted rather than one multiple linear regression. Only interactions that increased human-macaque prediction were accepted (R 2 > 0.506). All regressions were checked to assess whether they violated any assumptions, including outliers, distribution and homogeneity of the residuals 60 .