

Standardization in MLM using BAYES es... 

Message/Author 


Hello, I run a twolevel regression model for a continuous dependent variable with a random intercept and random slopes. I have level1 and level2 continuous and categorical predictors. I obtained the STDYX solution with the BAYES estimator in Mplus 8. I wanted to ask if you could please explain a little how these standardized effects are computed with the BAYES estimator. In the Mplus User's guide, for a time series example, it is stated "When a model has random effects, each parameter is standardized for each cluster. The standardized values reported are the average of the standardized values across clusters for each parameter". I wonder if this explanation also applies for a multilevel regression analysis with crosssectional data. Thanks, Paulina 


For crosssectional models the standardization is the regular kind. The new standardization for timeseries models has headings that make it clear it is not the usual standardization. 

S Napolitano posted on Wednesday, December 05, 2018  8:03 am



Dear Dr. Muthen, As a followup question to your answer to Paulina’s post, when you say that standardization in crosssectional multilevel models is “the regular kind”, do you mean that in a 2level model, level 1 SDs of predictor and outcome are used to standardize level 1 associations? I ask because the Mplus manual specifically refers to parameter standardization within cluster, following by their average across clusters for models with random effects (p. 801). If parameters are indeed first standardized within clusters, how does mplus treat standardization for instances with 0 withincluster variability? Are these clusters included in the denominator when averaging across clusters, or are they excluded? Thank you, Skye 


Page 801 talks about standardization with random slopes (so using the  statement). This is done only in the Bayes context and is only recently introduced. By standardization of the regular kind, I meant random intercept models where as you say you use e.g. the level 1 SDs to standardize level 1 estimates (and vice versa for level 2). For zero withincluster variance, you would declare the variable as Between and regular standardization applies. 

Back to top 

