Extended Data Fig. 8: Alternative calculations of correlations in fetal entropy measures.
From: Sex differences in prenatal development of neural complexity in the human brain
Because our fetal data contained multiple recordings from the same fetal subjects and, moreover, the random effect term significantly increased the fits of most models predicting entropy, we did not wish to rely on Pearson correlation coefficients (a) between entropy measures from each recording, as these correlation estimates may over-represent subjects with multiple recordings. For this reason, we instead utilized standardized model coefficients (betas) from linear mixed models that predicted entropy measures from each other while accounting for random effects (b). Differences between Pearson coefficients and averaged beta coefficients (c) are very small (β − r < 0.1 in all cases). However, standardized betas in LMMs are not generally symmetrical (that is, βi,j ≠βj,i), since they depend on the variance of the random effect, and the random effect may often contribute more to one variable or the other. To address this problem, we used the mean of βi,j and βj,i to represent the correlation between entropy measure i and j. Here, for transparency, we show the asymmetry in the standardized betas prior to averaging (d), which primarily affected correlation estimates for PermEn64.
