Standardized output for DSEM models PreviousNext
Mplus Discussion > Multilevel Data/Complex Sample >
 Niyantri Ravindran posted on Monday, September 25, 2017 - 8:37 pm
I tested a set of models assessing lagged effects using DSEM. I find that my fixed effects in the unstandardized output is non-significant (but approaching significance), whereas the the standardized estimates (specifically, the Within-Level Standardized Estimates Averaged Over Clusters) is flagged as significant. When I run the same models in MLM, the fixed effects are significant and parallel the standardized output in DSEM. Why am I seeing these differences, and should I trust the unstandardized or the standardized output in DSEM? Thanks!
 Tihomir Asparouhov posted on Tuesday, September 26, 2017 - 9:48 am
I would recommend looking at the significance of the cluster specific effects (i.e. the random effect) instead of the fixed effect.

The "Within-Level Standardized Estimates Averaged Over Clusters" standard error does not have the same meaning as the model estimate standard error. That is because it is conditional on the sample. That SE can be used to determine if the average effect for the individuals in the sample is >0. The model estimate is regarding the average effect for the population where the sample comes from.

MLM also underestimates SE due to not accounting for the dependence of the variables due to proximity of times of observations.
 Niyantri Ravindran posted on Tuesday, September 26, 2017 - 3:50 pm
Thank you so much, that was really helpful. I had another question. I noticed that sometimes, the estimates for fixed effects/cluster-specific effects are significant in both MLM and DSEM, but they appear to flip in direction. In MLM, the estimate is negative, but in DSEM, the estimate is positive. Do you know why this happens?
 Tihomir Asparouhov posted on Wednesday, September 27, 2017 - 3:47 pm
These coefficients are not the same - you can estimate both of these coefficients in the same model - see top two rows of Table 11 in

This might also be an indication that the model needs AR(2) modeling rather than AR(1) or that you should use the lagged command also for the covariate and add the lagged covariate as a predictor as well.

I recommend reading Sections 2.6 and 5.4 in
 Niyantri Ravindran posted on Saturday, October 14, 2017 - 2:04 pm
Thank you for your response. The readings were really helpful too. I just want to clarify your response to my first question. When you said look at the significance of 'cluster-specific effects,' you meant the random effects in the unstandardized output, right? Or did you mean that we should look at the significance of the "Within-Level Standardized Estimates Averaged Over Clusters?"
Back to top
Add Your Message Here
Username: Posting Information:
This is a private posting area. Only registered users and moderators may post messages here.
Options: Enable HTML code in message
Automatically activate URLs in message