The Effect of Within- and Cross-Level Multicollinearity on Parameter Estimates and Standard Errors in Multilevel Modeling With Different Centering Methods

• In Event: Roundtable Session 117
  In Roundtable Session: Complexities Encountered When Modeling Multilevel Data

Abstract

The prevalence of correlated predictors in different research disciplines and limited research regarding the effects of multicollinearity in multilevel models propelled the current Monte Carlo study. The study focuses on examining the extent of bias introduced into parameter estimates and their standard errors when multicollinearity is present in a multilevel model. Specific attention is paid to cross-level interactions in the presence of within- and cross-level multicollinearity. In addition, the use of group-mean centering as a viable solution to the negative impacts of multicollinearity is addressed. Controlled factors include the number of groups, cases per group, conditional interclass correlation coefficient, the correlation between two level-1 predictors, the correlation between a level-1 and level-2 predictors, and centering method.

Authors

• Thomas James Blaze, University of Pittsburgh

• Feifei Ye, University of Pittsburgh