The effect of sample size on estimates of covariation
While the sample sizes presented in this study are modest, they are
within the ranges suggested as sufficient in the literature. For
example, Cheverud’s analysis (1988) suggests that 40 individuals is a
large enough sample size for estimating genetic variance from phenotypic
variance. More recent simulation-based studies suggest that adequate
sample sizes likely vary depending on the statistics being used and the
properties of one’s dataset, but show that samples sizes of 16-32 are
large enough to accurately estimate most common statistics associated
with variance-covariance datasets (Grabowski and Porto 2017; Watanabe
2022). There are also several previously published empirical studies
examining the properties of variation-covariation with samples sizes
similar to our own (Polly 2005; Goswami and Polly 2010; Blankers et al.
2015). To further ensure that our sample size was adequate for detecting
differences in means between groups we performed power analyses to
calculate the minimum sample size required across a range of
correlations and effect sizes with 80% power. We also performed Monte
Carlo simulations across a range of hypothetical trait values and
standard deviations to determine at which sample sizes estimates of
variation stabilized. Full methodologies and results can be found in the
appendix B, but overall, we found that our samples sizes were sufficient
for detecting differences between groups even at small effect sizes
(η2 >0.02) and that estimates of trait
variation stabilized at 20 individuals.