Statistical analyses for this study were preformed using Mplus Version 5.21 (Muthén & Muthén, 2008). The following main idiosyncratic data issues were considered in the current analysis: (a) missing data, (b) nonnormality of some indicators, and (c) dependence of observations because of measurement error. Descriptive statistics showed that the percentage of missing data varied by indicator and time, ranging from 0% to 39.8% with a mean of 16% over all indicators and times. Missing data were analyzed as recommended by Shafer and Graham (2002) and were determined to be missing at random. Descriptive statistics also indicated that several of the indicators were skewed and slightly kurtotic. In order to account for data nonnormality, a robust maximum likelihood estimator (MLR) was utilized; MLR standard errors are computed by using a sandwich estimator based on a Huber–White algorithm (Muthén & Muthén, 2008). Maximum likelihood parameter estimates with standard errors and the chi-square test statistic are robust to nonnormality and nonindependence of observations. The MLR chi-square test statistic is asymptotically equivalent to the Yuan-Bentler (1997, 2000) T2 test statistic.
