All analyses were run in Mplus Version 6 using bootstrap estimation with 5000 bootstrap resamples
and bias-corrected bootstrap estimates (Muthén & Muthén, 2010). Researchers argue that bootstrap
estimation is a more appropriate data analytic technique for smaller sample sizes because it does
not assume normality of the population distribution (Efron & Tibshirani, 1986; Preacher & Hayes,
2008). Given the non-normality of variables (see Table 1), significance of standardized estimates
and indirect effects were interpreted based on bias-corrected (BC) 95% confidence intervals (CIs). In
this sample, 11% of participating students had missing teacher data (due to teachers declining to complete
questionnaires) and 1% had missing WM data (due to examiner error in the administration of
tests). To handle missing data, maximum likelihood estimation was used (Muthén & Muthén,
2010). Given the association of gender with social functioning measures, the effect of gender was accounted
for in all analyses using the multiple indicators multiple causes model (MIMIC; Muthén,
1989); MIMIC allows for an estimation of the effect of gender on all other variables in the model.
The interaction of gender and WM also was originally considered and was not significant across models;
thus, the interaction term was dropped from reported analyses.