Facial expressivity in dominant macaques is linked to group cohesion

Abstract

Social living affords primates (including humans) many benefits. Communication has been proposed to be the key mechanism used to bond social connections, which could explain why primates have evolved such expressive faces. We assessed whether the facial expressivity of the dominant male (quantified from the coding of anatomically based facial movement) was related to social network properties (based on social proximity and grooming) in nine groups of captive rhesus macaques (Macaca mulatta) housed in uniform physical and social environments. More facially expressive dominant male macaques were more socially connected and had more cohesive social groups. These findings show that inter-individual differences in facial expressivity are related to differential social outcomes at both an individual and group level. More expressive individuals occupy more beneficial social positions, which could help explain the selection for complex facial communication in primates.

Keywords: facial behaviour; facial expressivity; individual differences; macaques; social networks.

Figure 1.

Sociograms of social groups. (a) Group BTJ, (b) group VNT, (c) group YOK, (d) group ZODI, (e) group STAR, (f) group ASAM, (g) group BAG, (h) group PLUM and (i) group WILL. Nodes represent individuals, Edge thickness represents social connections as calculated through DSI. Orange nodes represent the single adult male of each group, off-white nodes represent the adult females. All sociograms were built using igraph for R [35].

Figure 2.

Notable (significant) relationships. (a–c) The relationship between expressivity measures and individual-level social network attributes. (d–f) The relationship between expressivity measures on group-level attributes. Plots with data coloured in orange represent relationships with AU diversity, and plots with data colours in blue represent relationships with AU rate. Plots c and d are based on a reversed proximity score to depict stronger relationships as a higher value. All expressivity measures were z-scored so they can be directly compared visually. Error bars represent standard errors.

Figure 3.

Comparison of model fit: individual AUs. Each model contained AU proportion as the predictor variable, and eigenvector centrality as the outcome variable. Models ranked by their predictive value (R ²). Bottom right panel: example of AUs acting on the face of a macaque; coloured outlines show area on the face affected by the movement.