Detecting Measurement Bias Using Regularized Structural Equation Modeling

• In Event: Advances and Accessibility in Statistics

Abstract

Measurement bias could result in the comparison of psychological constructs across groups not meaningful. This simulation study aims to compare the performance of six regularization methods for detecting uniform and non-uniform bias, and for estimating penalized parameters in restrictive factor models. The six methods implemented with maximum likelihood estimation are: ridge, lasso, adaptive lasso, elastic net, smoothly clipped absolute deviation, and minimax concave penalty. Design factors include the types of bias, sizes of bias, and sample sizes. Preliminary results show that ML and ridge have the highest false positive rates, while other regularization procedures have good balance of true positive and false positive rates with large sample sizes.

Author

• Xinya Liang, University of Arkansas