Partitioning variance of predictor? PreviousNext
Mplus Discussion > Multilevel Data/Complex Sample >
 Shane posted on Monday, February 15, 2016 - 2:38 pm

I refer to example 9.1 in the Version 7 User Guide

I want to model the effect of socioeconomic factors between and within countries on life expectancy. I have 100,000 individuals evenly nested in 100 countries.

I have good reason to think that SES at the individual level and GDP at the country level influence life expectancy. I do not have reason to think that the group mean of SES for the 1000 individuals in each country influences life expectancy.

In this case I can use example 9.1a and put SES on WITHIN and GDP on BETWEEN and not include an observed group mean SES covariate in the model. I understand this as a classic ML modelling approach.

However, if I do this isnít the mean structure of SES still included in the within part of the model?

Should I instead be using the approach in 9.1b and decompose the variance of SES and only include the within part of SES in the within model and set the variance of SES to zero in the between model so that I am only modelling the effect of the within part of SES within and the GDP measure between?

Is this different to what us happening above?


 Bengt O. Muthen posted on Tuesday, February 16, 2016 - 6:56 pm
I think the first approach is ok if there is no variation across countries in average SES; otherwise take the second approach.
 Shane posted on Wednesday, February 17, 2016 - 1:25 am
Thank-you for your response.

A follow-up question if I may.

In the 9.1a example (classic ML model)I understand that the introduction of SES at the within group level can attenuate the between group residual variance and the effect of between group predictors.

Is this still the same in 9.1b if I fix the between group variance of (latent) SES to zero?
 Bengt O. Muthen posted on Wednesday, February 17, 2016 - 4:51 pm
I am not sure.
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