Bias and Efficiency for SEM With Missing Data and Auxiliary Variables: Two-Stage Robust Method Versus Two-Stage ML

摘要: This article compares parameter estimates by 2-stage ML (TSML) and a recently developed robust (TSR) method for structural equation modeling (SEM) with missing data. In the design, data are at random (MAR) after an auxiliary variable (AV) is included, they not (MNAR) otherwise. Results indicate that, when either substantive variables or AV nonnormally distributed, TSR most likely yields more accurate than TSML; TSML only slightly preferred to all normally distributed. Including distributed AVs reduces bias improves accuracy in estimates. However, distribution of has heavier tails that normal distribution, including them could result less When sample size N medium large, small, rate low, and...