Random Versus Nonrandom Coefficient Models for Multilevel Analysis

It has been argued recently that adequate representation of the multilevel processes of interest in many educational settings will often require models with coefficients that are random rather than nonrandom. Computer simulation results presented here demonstrate the extent to which the two most common multilevel or contextual analysis procedures may mislead when data are, in fact, generated with a random coefficients process. In contrast, correct results are obtained with a third method, which has been recently proposed but not used yet in substantive research.