Chimpanzees monopolize and children take turns in a limited resource problem

Competition over scarce resources is common across the animal kingdom. Here we investigate the strategies of chimpanzees and children in a limited resource problem. Both species were presented with a tug-of-war apparatus in which each individual in a dyad received a tool to access a reward, but tools could not be used simultaneously. We assessed the equality of tool use as well as the frequency of turn taking. Both species managed to overcome this conflict of interest but used different strategies to do so. While there was substantial variation in chimpanzee behaviour, monopolization was the common course of action: tool use was asymmetric with individual chimpanzees monopolizing the resource. In children, turn-taking emerged as the dominant strategy: tool use was symmetric and children alternated access to the tool at a high rate. These results suggest that while both species possess strategies for solving limited resource problems, humans might have evolved species unique motivations and socio-cognitive skills for dealing with such conflicts of interest.

3 apparatus. In these cases, trials had to be interrupted. They were repeated in the next session. On one occasion, subjects lost interest in the yogurt because it has previously been used as a vector for medicine. This trial was repeated in the next session.
To assess the influence of relationship quality, we derived friendship data from a questionnaire answered by the chimpanzee caregivers. The three most experienced caregivers of each housing group rated each chimpanzee test-dyad in terms of their relationship (scale: 'very good', 'relatively good', 'neutral', 'relatively bad', 'very bad'). Dyads were rated as friends if their relationship was rated on average as relatively good or very good. All dyads with worse ratings were rated as non-friends.
Children: All child dyads received five trials in a single test session. We reduced the trial number to ensure children's motivation remained as high as in the chimpanzee subjects. Children needed approximately 45 seconds to finish their stamping sheet in the individual pretest. Therefore, the trial length was set to 2.5 minutes to ensure both individuals had sufficient time to finish their task.
To assess friendship levels, before the experiment we asked participating dyads if they had ever played with each other before. Dyads were subsequently classified as friends if their answer was affirmative. In addition, half of the dyads were picked from the same kindergarten group while the other half of dyads was made up of children who had no or very limited previous interaction (i.e. they were recruited from different kindergarten groups). Chimpanzees: Individually, all chimpanzees had to successfully complete three steps of a familiarization phase before they proceeded to the test phase. In the first step, the subjects had access to just one room with a single yogurt box. In order to access the bait, they needed to pull out the stick already placed inside the box and use it for poking.
Step two resembled step one, but the stick was tied to the rope, introducing the tug of war-principle.
Step three resembled the experimental set up: Two rooms were each provided with a baited box and the connected stick-tools. But in contrast to the experiment, the door between both experimental rooms remained open to test the chimpanzees' understanding of the mirrored setup. The subjects had to feed from the reward box in each room. In each of the three familiarization steps, the chimpanzees had to gain access to and feed from the provided food in three consecutive trials to meet criterion. Then they moved on to the next step. In cases where individuals failed, they repeated the former, easier step. The subjects switched the room they started in after every session to avoid a room preference in other studies (starting position changed from left to right side).
A total of twelve chimpanzees (75%) passed all criteria of the familiarization phase. Four individuals (25%) showed no or too little interest in the apparatus and were excluded from further testing (see Table S1). Ten individuals succeeded in consecutively passing all criteria whereas two chimpanzees needed one back shift to the former condition (Kofi repeated step one and Robert step two).
Human children: Individually, the participants were presented with a single familiarization trial. The setup was the same as in the experimental phase but without a partner (each side provided with a stamping sheet, an ink-tube and a ropeconnected-stamp). First, the individuals forming the dyad were briefly introduced to 6 each other outside the experimental room. Then one of the children was assigned to one side of the apparatus, brought into the experimental room and taught how to stamp their sheet by an experimenter. Together, experimenter and subject practiced the technique. Here, all subjects learnt that they always need fresh ink for every stamp and are supposed to mark in each circle printed on the sheet. These two rules were repeated before every trial in the experiment. After one subject was familiar with the apparatus and the stamping technique, she left the room and the experimenter repeated the procedure with the second child. First, we fitted two GLMMs with binomial error structure and logit link function to investigate how the equality of tool use varied between the two species on a trial basis (addressing the level of inequality for each and every trial, GLMM 1) and on the dyadic level (addressing the overall inequality per dyad, across trials GLMM 2), because, potentially, the partners could alternate in monopolizing the tool and end up with an equal outcome. Therefore, we collapsed the equality of tool use of all trials (for each dyad separately) to investigate the overall equality for each dyad in GLMM 2.
The response in both models was a two-columns matrix comprising number of tool uses by the individual who showed more tool use and the individual who showed less tool use. Such a response means that the model fits the proportion of tool uses that was conducted by the subject who showed more tool use. Effectively this means to model the inequality in tool use of the two individuals, whereby total equality means this proportion being .5, and complete monopolization (i.e., maximum inequality) it being 1. Model 1 and 2 we fitted with binomial error structure and logit link function. The key test predictor in both models was 'species' which we included 8 as a fixed effect. Both models further controlled for the fixed effects of friendship, sex/gender, and same kindergarten group (two levels: 'yes' or 'no'; always 'yes' for chimpanzees; 50% of the children were recruited from the same and 50% from a different group within the same institution) and model 1 also for trial number. We included two random intercepts for the identities of the two subjects (grouped according to which individual showed more and which showed less tool use; when both subjects showed the same number of tool uses these were randomly assigned to the two random effects), a random intercept of dyad and in model 1 a random intercept of trial identity nested within dyad. To keep the type I error rate at the nominal level of .05 we included random slopes 4,5 of trial number within the random effects of the two subjects and the dyad in model 1, but we did not include the correlations among the random slopes and intercepts.
Model 3 tested for species differences in the total number of turn taking events (response) per dyad. The model was a GLMM, fitted with a Poisson error distribution and log link function 6 . To account for number of tool uses differing between dyads (limiting the possible number of turns taken) we included the number of possible turn taking events (total number of tool uses -1; log transformed) as an offset term into the model 6 . With regard to the fixed effects this model was identical to model 2, and we again considered species the key test predictor. As random intercepts, we included the identity of the two individuals per dyad (no random slope was identifiable since none of the fixed effects varied sufficiently within the two random effects; and a random intercept of dyad was not needed since each dyad provided only one value for the response).
As suggested, we ran additional analyses, which included the interaction of species*friendship keeping the structure of the original models (as described in the manuscript and above). We also added species*trial number to Additional  Table S4.
Summarising, the additional analyses replicated the original results. The R code (full and null models) is indicated below Table S4. The code needed for defining model terms (e.g. "resp") is identical with the R code of the original models (GLMM 1-3; R code indicated in Appendix C.) Note. Addl = Additional, b = Estimate, SE = standard error; NA = not available: data for the Likelihood ratio tests (χ 2 ) not meaningful for Intercepts and levels of the fixed effects (χ 2 always represents the factor and not the depicted level of it); : = interaction between factors Note. Red = Reduced, b = Estimate, SE = standard error; NA = not available: data for the Likelihood ratio tests (χ 2 ) not meaningful for Intercepts and levels of the fixed effects (χ 2 always represents the factor and not the depicted level of it); : = interaction between factors