COGNITIVE UNDERPINNINGS OF GREAT APES’ TOOL-USE 1

Supplemental Materials

The Cognitive Underpinnings of Flexible Tool-Use in Great Apes

by C. J. Völter & J. Call, 2014, Journal of Experimental Psychology: Animal Behavior Processes

http://dx.doi.org/10.1037/xan0000025

Experiment 1: Number of trials per configuration

Analysis

If subjects failed the 16 trials for a given configuration, we scored 17 trials for this configuration (i.e. the minimum number in which they could have solved the configuration). A second coder scored 20% of the trials to assess interobserver reliability, which was excellent

(clear condition: number of trials per configuration rS=1.0, N=113, p<0.001; tool condition: number of trials per configuration rS=1.0, N=105, p<0.001).

Subjects who performed above chance with regard to number of trials per configuration but not for the other dependent variables were indicative of significant post- error corrections by avoiding previously made errors.

We used Pearson's correlations to test the relation of age with the dependent variables. All p-values reported here are exact and two-tailed. The assumption of normality was met for the current data (Kolmogorov–Smirnov test: p>0.05). At the individual level, we used the Wilcoxon signed rank test for the count variable number of trials per configuration to test against the chance level.

To test whether the dependent variables number of trials per configuration were influenced by the factors condition (tool, no-tool), level of planning (LoP), changes in direction (CiD), repetition of configurations, and the age of the subjects, we used a generalized linear mixed model that included these five predictors as fixed effects and subject as well as configuration identity as random effects. The models were fitted in R using the function lmer of the R-package lme4 . The significance of the full model as compared to the null model was established using a likelihood ratio test . Therefore, we used the R function anova with argument test set to "Chisq." All models reported here were found to be significant (p<0.01). COGNITIVE UNDERPINNINGS OF GREAT APES’ TOOL-USE 2

As the dependent variable ‘number of trials’ was a count response, we used a poisson error distribution. We z-transformed all predictors to a mean of zero and a standard deviation of one to get comparable estimates. The intercepts of the models represented the sample mean assumed by the models. In the case of the poisson model, the fitted mean is revealed by the inverse log-function (exp(intercept)). As the chance value of the count variable “number of trials per configuration” was 4, we subtracted the log-transformed chance value from the estimate of the intercept and calculated the z- and P-value based on the adjusted estimate. However, since the dependent variables were not based on a simple linear function of the given predictor variables in the models, there was a minimal deviation of the sample mean assumed by the model from the actual sample mean. We corrected for this small deviation by adjusting the scaled variables by adding a constant value chosen such that the absolute difference between the actual sample mean and the fitted mean was minimized (the corresponding function was written by Roger Mundry and is available on request). Doing so did not affect any terms of the model except for the intercept. Thereby, the intercept in these models became a reliable test of subjects’ performance against chance while controlling for the covariates and random effects.

Results

Overall, the full model was clearly significant as compared to the null model (likelihood ratio test: χ²=129.67, df=7, p<0.001). The GLMM indicated a significant effect of age and repetition (see Table S1): younger subjects performed better than older ones, and subjects performed better in the second round compared to the first one. In contrast, there were no significant main effects of tool-use, LoP, CiD or interactions between tool-use and LoP / CiD.

Neither tool-users nor non-tool-users performed different from the chance level of 4 trials (no-tool: 3.54 ± 0.42 trials, z=0.60, p=0.55; tool: 4.67 ± 0.50 trials, z=0.25, p=0.80). At the individual level, six non-tool-users (five chimpanzees and one bonobo) and two tool- users (two chimpanzees) solved the maze task in significantly less than 4 trials (all p<0.05).

Table S1

Experiment 1: Output of GLMMs for the Number of Trials per Configuration

Number of trials Model terms Est 95% CI p Tool-use 0.26 [-0.03, 0.54] 0.08 LoP 0.02 [-0.06, 0.11] 0.54 CiD 0.09 [-0.05, 0.23] 0.21 Age 0.16 [0.02, 0.29] 0.03 COGNITIVE UNDERPINNINGS OF GREAT APES’ TOOL-USE 3

Repetition -0.17 [-0.20, -0.14] <0.01 Tool-use x LoP 0.03 [-0.03, 0.09] 0.38 Tool-use x CiD -0.08 [-0.25, 0.10] 0.39

Experiment 2: Number of trials per configuration

Analysis

Same as in experiment 1.

Results

Overall, the full model was clearly significant as compared to the null model (likelihood ratio test: χ²=15.81, df=7, p=0.027). The GLMM indicated no significant effects of tool-use, LoP, CiD, and age or an interaction between LoP / CiD and tool-use (see Table S2). Overall, subjects performed above chance in the tool (1.38 ± 0.08 trials, z=16.1, p<0.001) as well as in no-tool condition (1.22 ± 0.04 trials, z=16.6, p<0.001). This was also true across the different levels of planning and changes in direction. At the individual level, all subjects performed overall above chance in the no-tool and tool condition (all p<0.001).

Table S2

Experiment 2: Output of GLMMs for the Different Dependent Variables

Number of trials Model terms Est 95% CI p Tool-use 0.10 [-0.09, 0.30] 0.30 LoP 0.10 [-0.04, 0.24] 0.15 CiD 0.07 [-0.06, 0.21] 0.30 Age 0.07 [-0.02, 0.17] 0.13 Order of condition 0.03 [-0.06, 0.13] 0.50 Tool-use x LoP -0.01 [-0.20, 0.18] 0.90 Tool-use x CiD 0.09 [-0.10, 0.28] 0.34

Experiment 3: Number of trials per configuration

Analysis

We scored and analyzed the data in the same way as in previous experiments. Additionally, we compared the data for the tool and no-tool condition of Experiment 1 with the new data obtained with the opaque-cued condition. A second coder scored 20% of the opaque-cued trials to assess interobserver reliability, which was excellent (number of trials per configuration rS=0.99, N=118, p<0.001). COGNITIVE UNDERPINNINGS OF GREAT APES’ TOOL-USE 4

Results

Overall, the full model was clearly significant as compared to the null model (likelihood ratio test: χ²=169.23, df=10, p<0.001). The GLMM indicated a significant effect of repetition and the opaque-cued condition (see Table S3): subjects performed better in the second round compared to the first one and subjects in the opaque-cued condition needed more trials than those in the no-tool condition and tool condition. In contrast, there were no significant main effects of age, LoP, CiD, or interactions between condition and LoP / CiD.

Overall, subjects in the opaque-cued condition did not perform significantly different from chance (6.23 ± 0.40 trials, z=0.3, p=0.764) like the subjects in the no-tool and tool condition (see Experiment 1). At the individual level, there was no subject in the opaque- cued condition that required fewer than four trials per configurations. In contrast, five subjects in the opaque-cued condition needed significantly more than four trials per configuration (four chimpanzees and one orangutan), the remaining six subjects performed at chance level.

Table S3

Experiment 3: Output of GLMMs for the Different Dependent Variables

Number of trials Model terms Est 95% CI p Opaque-cued vs. No-tool -0.62 [-0.87, -0.36] <0.01 Opaque-cued vs. Tool -0.35 [-0.61, -0.09] 0.01 LoP -0.02 [-0.12, 0.08] 0.75 CiD 0.12 [-0.01, 0.26] 0.07 Age 0.09 [-0.003, 0.18] 0.06 Repetition -0.14 [-0.16,-0.11] <0.01 Opaque-cued vs. No-tool x LoP 0.03 [-0.05, 0.11] 0.45 Opaque-cued vs. Tool x LoP 0.06 [-0.02, 0.14] 0.13 Opaque-cued vs. No-tool x CiD -0.03 [-0.19, 0.12] 0.68 Opaque-cued vs. Tool x CiD -0.11 [-0.27, 0.04] 0.16

Experiment 4: Number of trials per configuration

Analysis

Same as in experiment 1.

Results

Overall, the full model was clearly significant as compared to the null model (likelihood ratio test: χ²=235.05, df=10, p<0.001). The GLMM indicates a significant effect of condition (see Table S4): subjects performed better in the no-tool condition compared to the opaque- COGNITIVE UNDERPINNINGS OF GREAT APES’ TOOL-USE 5 cued and opaque condition. Moreover, subjects performed better in the opaque-cued condition compared to the opaque condition. Additionally, we found a differential effect of LoP between the no-tool and opaque condition. LoP had only a significant effect on performance in the no-tool condition (z=2.12, p=0.034) but not in the opaque (z=-0.39, p=0.697) or opaque-cued condition (z=1.24, p=0.215). For LoP 0 configurations, subjects performed significantly better in the no-tool condition compared to the opaque (z=8.10, p<0.001) and opaque-cued condition (z=4.93, p<0.001) and subjects performed better in the opaque-cued than in the opaque condition (z=3.82, p<0.001). Similarly, for LoP 1 configurations, subjects performed significantly better in the no-tool condition compared to the opaque (z=8.18, p<0.001) and opaque-cued condition (z=3.97, p<0.001) and subjects performed better in the opaque-cued than in the opaque condition (z=4.77, p<0.001). For LoP 2 configurations, subjects performed significantly better in the no-tool condition compared to the opaque (z=6.86, p<0.001) and opaque-cued condition (z=5.62, p<0.001) but there was no difference between the opaque-cued than in the opaque condition (z=1.50, p=0.133). Finally, we found a differential effect of CiD between the opaque-cued and opaque condition. However, across CiD, subjects performed better in the no-tool condition compared to the opaque (CiD 0: z=9.71, p<0.001; CiD 1: z=9.37, p<0.001) and opaque-cued condition (CiD 0: z=4.32, p<0.001; CiD 1: z=7.57, p<0.001) and they also performed better in the opaque-cued than in the opaque condition (CiD 0: z=6.03, p<0.00; CiD 1: z=2.19, p=0.029). Besides, there were no significant main effects of age, LoP, or CiD.

Overall, subjects performed significantly better than chance in the no-tool (1.55 ± 0.16 trials, z=-11.1, p<0.001) and opaque-cued condition (3.00 ± 0.28 trials, z=-2.35, p=0.019) but not in the opaque condition (4.20 ± 0.33 trials, z=0.1, p=0.941). At the individual level, all seven subjects required significantly less than 4 trials in the no-tool condition, 6 out of 7 in the opaque-cued condition, and 2 out of 7 in the opaque condition.

Table S4

Experiment 4: Output of GLMMs for the Different Dependent Variables

Number of trials Model terms Est 95% CI p

No-tool vs. Opaque 1.00 [0.85, 1.15] <0.01

No-tool vs. Opaque-cued 0.64 [0.49, 0.79] <0.01

Opaque-cued vs. Opaque 0.36 [0.24, 0.47] <0.01

LoP 0.13 [-0.01, 0.26] 0.08

CiD 0.05 [-0.12, 0.23] 0.55 COGNITIVE UNDERPINNINGS OF GREAT APES’ TOOL-USE 6

Age 0.08 [-0.04, 0.19] 0.20

Order of condition 0.02 [-0.03, 0.08] 0.42

No-tool vs. Opaque x LoP -0.15 [-0.29, -0.01] 0.04

No-tool vs. Opaque-cued x LoP -0.04 [-0.19, 0.11] 0.64

Opaque-cued vs. Opaque x LoP -0.11 [-0.23, 0.001] 0.05

No-tool vs. Opaque x CiD -0.05 [-0.19, 0.09] 0.50

No-tool vs. Opaque-cued x CiD 0.13 [-0.02, 0.28] 0.09

Opaque-cued vs. Opaque x CiD -0.18 [-0.30, -0.07] <0.01

References