Standardizing results PreviousNext
Mplus Discussion > Multilevel Data/Complex Sample >
 Emilie.franck posted on Saturday, August 17, 2019 - 4:53 am
Dear Dr. Muthen,

I am performing a three-level model (random) in MPlus. At the moment, my analyses are just explanatory.
I have predictors on each level and would like to have an idea of their explanatory power. However, Mplus does not perform standardization in a three-level model (when using weights). I was thinking if just using the simplified version of standardization (although I realize that in a later stage I will need to adapt my method as it is not 100% correct to use it in a multilevel model) with the following formula: stdcoefficient = Betax * stdx/stdy.

I am not sure if this approach is even allowed as exploratory method, but if I try to apply this formula, I am not sure with stdy I should use. The one I obtain for my Y00 is very small and thus I get values larger than 1 for almost all my variables. Should I use another stdy?
Or can I really not apply this formula to my results?

Many thanks in advance!

Kind Regards,

 Bengt O. Muthen posted on Saturday, August 17, 2019 - 11:35 am
Do you have random slopes?
 Emilie.franck posted on Saturday, August 17, 2019 - 11:57 am
Dear dr. Muthen,

Yes I included them. I know there is a lot of discussion within the literature on how to do it if random slopes are included though, and I will get into that topic. I just thought I could get a bit of an idea of what predictors could be more important than other. maybe this is not the case at all after all!

Many thanks,

Emilie Franck
 Bengt O. Muthen posted on Saturday, August 17, 2019 - 5:54 pm
Standardization with random slopes is only available in Mplus for twolevel and time series modeling I believe. It follows the approach of that Schuurman et al Psych Methods article.
 Tihomir Asparouhov posted on Monday, August 19, 2019 - 9:07 am
If you don't have random slopes output:stdyx should work. If you have random slopes the situation is a bit more complicated because R2 varies across clusters as the random slopes do. There are a couple of alternatives you can use as approximations.

1. Standardize the variables before the analysis - this way the the total variance would be 1 and the R2 would be 1 - the residual variance on the within level. This would give you all three level combined though. If you want the three levels separate use approach 2.

2. Estimate the variance for each of the levels from univariate runs for each variable separately. You can then use those in the original run to compute standardized coefficients.
 Emilie.franck posted on Monday, August 19, 2019 - 11:36 am
Dear Dr. Muthen, Dear Dr. Asparouhov.

Thank you very much for this information. I will try to do the first option and run 1 time the unstandardized model and one time the standardized model. I guess that all three levels combined is enough in this stage, as it is more exploratory to draw some conclusions.

Many thanks in advance for the quick answer!

Kind Regards,

Emilie Franck
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