Factored Regression Approach for a Continuous Non-Normal Incomplete Predictor

• In Event: AERA Roundtable Session Saturday 11:40 am Four Seasons Ballroom 4
  In Roundtable Session: Complexity in Data: Simulations and Applications in Complex, Multivariate, Multilevel, and Nonnormal Models (Table 2)

Abstract

The factored regression framework can easily accommodate models with nonlinear effects by factorizing the joint distribution into separate analysis and predictor models. Misspecification of the predictor model may occur due to the incomplete continuous predictor not following a parametric form. This study aims to examine the performance of the factored regression framework when an incomplete, continuous non-normal predictor is present in the model. A Monte Carlo simulation study compared two predictor models: a normal distribution and a Yeo-Johnson transformed normal distribution. Results revealed that the normal distribution was quite robust to violations of normality and that the Yeo-Johnson transformed distribution was more sensitive to the location of missingness and the order of factorization.

Authors

• Jiwon Kim, University of Texas at Austin

• Suyoung Kim, University of Texas at Austin