A Simulation Study of Parameter Estimates of Nonparametric Multilevel Latent Class Analysis With Covariates

• In Event: 30.074 - Issues in Parameter Estimation and Model Comparison

Abstract

The current research aims to advance our knowledge on how the quality of parameter estimates are affected by six study factors in conditional NP-MLCA: 1) the quality of indicators, 2) the number of indicators, 3) leve-1 covariate effects, 4) level-2 covariate effects, 5) number of groups-level units, and 6) group size. 384 conditions were examined. The findings are three: first, the covariate estimates are extremely biased when the quality of indicators is 0.6, especially in 6 indicators condition. Second, for level-2 covariate effect, samples size can compensate for the condition of quality of indicators 0.7. Third, more indicators and higher quality of indicators account for the total variability of the biases and coverage rates of CRP of latent classes.

Authors

• Chi Chang, University of New Mexico

• Xiaoran Tong, Michigan State University