The Impact of Partial Measurement Invariance on Tests of Moderation for Single- and Multilevel Data

• In Event: Multilevel Modeling: Nonnested Multilevel Data, Measurement Invariance, and Nonlinear Models

Abstract

The measurement invariance literature was relatively silent about its impact on prediction and moderation than on group mean comparison. In simulation study 1, the relationship between measurement invariance on the power and Type I error of Wald test for moderation will be investigated for single-level data. Similar analyses will be conducted in simulation study 2 for multilevel data, with variables such as intraclass correlation and cluster size as additional design factors. Preliminary results showed that when metric (or weak) invariance did not hold and was not modelled correctly, test of moderation had very poor performance, with severely inflated Type I error rate and severely deflated power as sample size increases.

Authors

• Yu-Yu Hsiao, Texas A&M University - College Station

• Hok Chio Lai, Texas A&M University - College Station