2017 Volume 56 Issue 2 Pages 142-152
This article proposes solutions to certain problems that are encountered when conducting multilevel analysis with dyadic data. First, problems encountered in Hierarchical Liner Modeling (HLM) that result from inputting mean scores of dyadic data as independent variables are discussed. Simulations have indicated that serious biases are encountered in Level 2 HLM estimations, when using mean scores of dyadic data, whereas estimates of Multilevel Structural Equation Modeling (ML-SEM) do not show these biases. Second, we focused on the problem of ML-SEM conducted with dyadic data often resulting in incomplete solutions. It has been indicated through examples that Bayesian estimations are more accurate and valid than maximum likelihood estimations. Finally, difficulties regarding the interpretation of individual level correlations are addressed.
