Wild primates copy higher-ranked individuals in a social transmission experiment

The study investigates how social learning strategies shape the diffusion of novel behaviors in wild primates and how individuals integrate social with individual learning. Building on evidence that social network structure influences information flow, the authors test whether social transmission is option-specific (copying the observed technique) or generalizes across options, and whether model-based biases (e.g., rank, sex, age, kin) govern from whom individuals learn. Using open diffusion experiments in wild vervet monkeys and dynamic observation networks within Network-Based Diffusion Analysis (NBDA), the study evaluates the hypotheses that higher-ranked individuals exert greater social influence (directed social learning), that social transmission is option-specific, and that learning one option affects subsequent acquisition of the alternative option. The work aims to clarify mechanisms underlying the emergence of stable traditions and culture in wild primates.

Prior work shows that social transmission can drive diffusion of behaviors and traditions across animal groups, with social network centrality and proximity shaping spread. In primates, selective copying has been reported: captive chimpanzees show prestige and majority biases; wild vervets display multiple strategies including mother-offspring matching of food-processing techniques, female model bias in some contexts, and conformity among dispersing males. However, findings are inconsistent: some studies find no dominance-based learning bias in vervets. Other primates (e.g., capuchins) exhibit payoff-biased and age-related copying. Copying biases can arise from performance bias (models act more), attention bias (models observed more), and social information use bias (greater weight per observation). Many studies impose trained models and two-action tasks, potentially constraining ecological validity and confounding who first innovates with who gets copied. The authors address these issues using open diffusions and dynamic observation networks to isolate per-observation social effects and test multiple biases concurrently.

Subjects and site: Two habituated groups of wild vervet monkeys (Chlorocebus pygerythrus) at Inkawu Vervet Project, Mawana Game Reserve, South Africa: Noha (NH, N=28) and Kubu (KB, N=12). Adults vs juveniles defined by dispersal (males) and first birth (females). Individual identities confirmed via photographs. Dominance hierarchies were established from ad libitum agonistic interactions and food competition tests (May–Oct 2017), with significantly linear hierarchies (NH h′=0.29; KB h′=0.80). Ethics approvals noted. Experimental design: Open diffusion artificial-fruit task with eight boxes available simultaneously (two sites × four boxes; boxes spaced ~2 m; sites ~20–50 m apart). Each transparent box could be opened by either lifting a top lid (“lift”) or pulling a front drawer (“pull”), each yielding an apple slice. Monkeys freely interacted; sessions at sunrise; boxes rebaited between depletions; sessions ended after ~one apple consumed by an individual or when group left. Video recorded with identities of actors and attending observers announced. Attendance defined as oriented toward actor within 0–30 m with unobstructed view; multiple observers per event possible. Across May–Aug 2017, 17 sessions (NH) and 12 sessions (KB) were run (avg session duration NH 81.44 min; KB 62.01 min). Some naive observers (marked with *) were later opportunistically tested with a single box when frequent solvers were absent to assess learning from prior observations. Data coding: Videos analyzed in slow motion/frame-by-frame by two observers (20% double-coded; inter-observer reliability 0.87). Coded variables: date/time of each manipulation, actor identity, technique used (lift/pull and variants), and attending individuals. Dynamic observation network and NBDA: Constructed a dynamic observation network capturing, for each dyad, the cumulative number of times an observer saw a demonstrator solve the task with a given option prior to the observer’s acquisition. Used NBDA (order of acquisition diffusion analysis, OADA) to model the order in which individuals acquired each option. Two social transmission pathways modeled with separate s parameters: option-specific (OS: observation of lift increases rate of learning lift, and similarly for pull) and cross-option (CO: observation of one option increases rate of learning the other). Also modeled asocial learning rate, allowed for option differences in asocial learnability, and tested for learning generalization effects whereby solving one option changes the asocial learning rate of the other. Individual-level variables: Sex, age class (adult/non-adult), and rank included as covariates potentially affecting asocial and social learning (unconstrained model). Multi-model inference used AICc across model sets (OS-only; OS>CO; generalized OS=CO; CO-only; asocial-only). Transmission bias analyses: For rank, sex, age, kin biases, the dynamic observation network was partitioned into mutually exclusive sub-networks corresponding to directional pathways (e.g., higher-to-lower rank vs lower-to-higher). Separate s parameters estimated per pathway, compared via AICc (bias vs no bias vs transmission-only pathways). Additional GLMs/GLMMs tested performance and attention components: effects of rank/group on manipulation rate and success (quasi-Poisson GLMs with time offsets) and effects on observation rates (Poisson GLMM with observer/observed as random effects and time offset). Software: R 3.5.2; NBDA package v0.7.10; Gephi 0.9.2.

• Participation and first solvers: NH (N=28): all touched boxes; first solver was highest-ranked adult female (lift). 19/28 achieved at least one success. KB (N=12): all touched boxes; first solver was highest-ranked adult male (pull). 10/12 achieved at least one success.
• Evidence for social transmission and copying fidelity: Strongest support for exclusively option-specific (OS) social transmission (total model support 63.8%), followed by models with OS > CO transmission (22.2%). Little support for generalized transmission sOS=sCO (9.0%) or CO-only (2.5%). OS social transmission per observation relative to baseline asocial rate: sOS=0.237 (95% CI 0.086–2.00); cross-option sCO=0.019 (0–1.15). Estimated proportion of acquisitions by each pathway: OS social transmission 45.1% (95% CI 31.0–53.6); CO 5.6% (0–18.4).
• Learning generalization effect: After an individual solved one option, it was 31x more likely to asocially learn the other option than individuals naive to both (multiplicative effect x30.7; 95% CI x11.3–x110.6; model support 100%). Triangles in model predictions plotted at high probabilities indicate that second-option learning occurs readily regardless of observational experience.
• Sex effect on social transmission: Females tended to socially learn faster than males (multiplicative effect female/male x2.15; 95% CI x1.1–x9.3; support 57.1%).
• Rank transmission bias: Models with transmission only from higher- to lower-ranked individuals received most support (69.8%); additional support for higher-to-lower > lower-to-higher (18.8%); little support for no bias (10.8%) or lower-to-higher only (0.02%). Estimated sHL=0.23 (95% CI 0.020–1.09); sLH=0 (0–0.59).
• Performance, attention, and group effects: Higher rankers manipulated more (GLM estimate −1.55; SE 0.70; z=−2.21; p=0.035) and succeeded more (GLM estimate −1.43; SE 0.61; z=−2.35; p=0.027) than lower rankers. Rank did not affect being observed (GLMM estimate −0.054; SE 0.75; z=−0.72; p=0.47). KB (smaller group) showed more observations than NH (GLMM estimate −1.56; SE 0.52; z=−2.98; p=0.0029), manipulated more (GLM estimate −1.44; SE 0.35; z=−4.05; p=0.0004), and succeeded more (GLM estimate −1.43; SE 0.32; z=−4.52; p=0.0001), likely due to reduced competition.
• Option preference: Most individuals ultimately preferred the lift technique; in NH there was a significant group-level preference for lift (binomial test p=0.003).

The study demonstrates that novel foraging techniques in wild vervet monkeys spread primarily via option-specific social transmission, indicating that individuals copy the specific technique they observe rather than merely being attracted to the task. The pronounced increase in asocial acquisition of the second option after learning the first shows that initial learning facilitates subsequent exploration rather than locking individuals into a single method, aligning with theories that high-fidelity copying alone does not ensure cultural stability. The detected dominance (rank) bias—observations of higher-ranked individuals having greater impact per observation than those of lower-ranked—clarifies inconsistencies in prior vervet studies by isolating the social information use component from performance and attention. Females’ tendency to learn socially faster fits with their philopatry and presumed ecological expertise. Together, these results address the research question by identifying the transmission pathways and social biases underpinning the diffusion of innovations in a naturalistic setting, and they underscore the value of dynamic observation networks for disentangling multiple interacting social learning strategies.

This work combines ecologically valid open diffusion experiments with dynamic-network NBDA to show that: (i) social transmission in wild vervet monkeys is largely option-specific; (ii) learning one solution substantially increases the likelihood of subsequently asocially learning the alternative; and (iii) higher-ranked individuals exert greater social influence on observers than lower-ranked individuals. These findings advance understanding of how multiple social learning strategies interact to shape cultural transmission in primates and support the view that varying degrees of fidelity and model-based biases contribute to cultural dynamics. Future research should: analyze full behavioral sequences using approaches such as experience-weighted attraction (EWA) models to test for conformity and payoff biases alongside rank effects; expand to more groups and species to assess generality; manipulate model rank/payoff factorially; and examine developmental trajectories (e.g., infants vs older juveniles) of changing biases (maternal vs rank-based).

• Sample size and scope: Only two groups were tested, potentially limiting generalizability across populations and contexts.
• Methodological scope of NBDA: NBDA examines initial acquisition order and cannot assess how individuals settle on long-term preferences once both options are learned, precluding direct tests of conformity within this framework.
• Prior experience and group differences: NH had more prior exposure to box experiments than KB, and group size differences likely influenced competition, access, and observation opportunities.
• Age range: Only individuals older than one year participated; infant-specific mother biases reported elsewhere could not be evaluated here.
• Performance and access: Higher-ranked individuals manipulated and succeeded more, potentially affecting exposure patterns; although attention bias by rank was not detected, unequal access due to dominance could still shape opportunities in subtle ways.
• Opportunistic end-phase tests: Some naive observers were tested later without frequent solvers present, which may differ from the main open diffusion context.