IQ heritability among humans is well-known state of research. Much less has been done on chimpanzees. Hopkins et al. (2014) published a recent study that touches the subject. 99 chimpanzees, aged 9 to 54 (mean = 24.55, SD = 10.67), were given the Primate Cognition Test Battery (13-tests).
This article is short, thus I may update the post if I find new studies in the future.
First, Hopkins et al. (2014) use PCA with varimax rotation, that revealed four components with eigenvalues >1.0, and these accounted for 54.20 percent of variance. Each of the 4 significant component scores was saved, and used in comparison of scores between sexes and rearing groups with MANOVA. No significant main effects or interactions were found between sex and rearing conditions (reared by humans or not) on the component scores. Thus, the performance is independent of rearing history and gender. One problem with virtually all applications of ANOVA-like analyses is that only the significance level is reported while what we should only care about is the effect size.
As has been done in previous studies in primates [15, 16], we used the program SOLAR (Sequential Oligogenic Linkage Analysis Routines) to estimate heritability . The overall ‘‘g’’ factor score as well the scores for each of the four components derived from the PCA served as the variables of interest in the heritability analyses. Age, sex and rearing history served as covariates.
SOLAR allows for analysis of genetic variance components, including linkage analysis, quantitative genetic analysis, SNP association analysis. They then attempt to determine the extent to which any two traits have the same set of genes that account for their variation. The genetic correlation between component score 1 and 3 is very high, rg=0.992 (SE=0.522, p<0.05). The number of factors to be retained is an important question. The authors rightly rejected the method of seletion based on eigenvalues>1.0 and the scree plot recommended by Costello & Osborne (2005). They prefer the parallel analysis, recommended by Ledesma & Valero-Mora (2007). This is a Monte Carlo simulation technique which evaluates what minimum eigenvalues are needed to reject the null hypothesis by subjecting a data set of random numbers to analysis (repeated many times, e.g., 1000 replications) when the data is adjusted for sample size and the number of variables. And it reveals that only component 1 had an eigenvalue that would be considered significant. Again, the problem of significance is related with sample size. They then correlate the Varimax-rotated factors with the unrotated (first) factor score.
Thus, individual differences in the derived ‘‘g’’ factor score correlated with scores for components 1 (r = 0.771, p < 0.001), 2 (r = 0.457, p < 0.001), 3 (r = 0.363, p < 0.001), and 4 (r = 0.256, p < 0.02), suggesting substantial overlap in the underlying or latent cognitive ability .
One problem seems to be that a one-factor solution is inconsistent with another study (Herrmann et al., 2007, 2010) that compares the factor structure of the PCTB between children and chimpanzees and orangutans. In children, the (correlated) 3-factor strucure yielded the best fit (spatial, physical and social factors). In chimpanzees, the best-fit model was a (correlated) 2-factor structure, spatial and physical-social, and the 1- and 3-factor models did not converge. The authors haven’t explained what happened. Problems of non-convergence can be due to identification problems because the model is ill-specified, too small sample size of multicollinearity (Lei & Wu, 2007). As I explained before, identification problems emerge when the number of freely estimated parameters exceeds the pieces of information (i.e., variances and covariances) available in the actual sample variance-covariance matrix. Or, convergence issue happens when we try to freely estimate a parameter that is not different from zero (Bentler & Chou, 1987, pp. 101-102). Sometimes, it is said that one can ease convergence by increasing the number of iterations, but the authors said nothing about it. The one-factor (i.e., g factor) were not preferred in each case. But a higher-order g factor model has not been tested, unfortunately.
Now, Hopkins et al. (2014) did not use CFA to assess model fit. Such comparison with Herrmann study is probably not justified.
Over a 2-year period, they re-tested 86 of the original 99 chimpanzees in this study on the 13 PCTB tasks. Performance was generally stable over time. For most tests they seem to show a retest effect, i.e., increased performance due to familiarity. PCA on this retest data shows pattern of loadings similar to what is displayed in table 1. The heritability for the g factor was h²=0.624 (SE=0.242, p<0.005).
Their discussion reads as follows :
Presumably, these attributes would have conferred advantages to some individuals, perhaps in terms of enhanced foraging skills or increased social skills, leading to increased opportunities for access to food or mating [24, 25]. These individuals would have then potentially had increased survival and fitness, traits that would have become increasingly selected upon during primate evolution, as has been postulated by a number of theorists, going all the way back to Darwin [26–30].
Herrmann, E., Call, J., Hernández-Lloreda, M.V., Hare, B., and Tomasello, M. (2007). Humans have evolved specialized skills of social cognition: The cultural intelligence hypothesis. Science 317, 1360–1366.
Herrmann, E., Hernández-Lloreda, M.V., Call, J., Hare, B., and Tomasello, M. (2010). The structure of individual differences in the cognitive abilities of children and chimpanzees. Psychol. Sci. 21, 102–110.
Hopkins, W. D., Russell, J. L., & Schaeffer, J. (2014). Chimpanzee Intelligence Is Heritable. Current Biology, 24:1-4.