Effects of different intracranial volume correction methods on univariate sex differences in grey matter volume and multivariate sex prediction

Sex differences in 116 local gray matter volumes (GMVOL) were assessed in 444 males and 444 females without correcting for total intracranial volume (TIV) or after adjusting the data with the scaling, proportions, power-corrected proportions (PCP), and residuals methods. The results confirmed that only the residuals and PCP methods completely eliminate TIV-variation and result in sex-differences that are “small” (∣d∣ $$\approx $$ ≈ 93%) than scaling and proportions adjusted-data $$( \approx $$ ( ≈ 68%) or raw data ($$\approx $$ ≈ 45%). The replicated effects were meta-analyzed together and confirmed that, when TIV-variation is adequately controlled, volumetric sex differences become “small” (∣d∣ VOL features in predicting individuals’ sex with 12 different machine learning classifiers. Sex could be reliably predicted (> 80%) when using raw local GMVOL, but also when using scaling or proportions adjusted-data or TIV as a single predictor. Conversely, after properly controlling TIV variation with the PCP and residuals’ methods, prediction accuracy dropped to $$\approx $$ ≈ 60%. It is concluded that gross morphological differences account for most of the univariate and multivariate sex differences in GMVOL