Retention rates at F1, F2, and F3 were 87.04%, 81.50% and 73.61%, respectively. We conducted attrition analysis to test whether children at each time-point who did and did not have data differed. Out of 48 contrasts, only age of the child at BL was associated with the presence/absence of data on NEPSY at F3 (Mean[SD] with and without data, respectively: 4.18[.51] vs. 4.42[.39] years; F (1,214) =14.89; p<.001). Thus, data were considered to be missing at random (MAR). We employed a full information maximum likelihood (FIML) method, which may provide less biased estimates than list-wise or pair-wise deletion (Arbuckle, 1996; Enders & Bandalos, 2001). Additional data (23.61%) on NEPSY at F3 was missing as this measure was temporarily dropped from the study. A model-based stochastic regression imputation strategy was employed to assign values to these variables within the multi-group model framework. This form of imputation preserves the variability of the data and yields unbiased parameter estimates when the data are MAR (Little & Rubin, 2002; Enders, 2010).
