Within and between person correlation with mixed regression

[Morten Nordmo]

Hi all!

I am using mixed multi-level models with two leveled data. Every person has multiple measurements and I am interested in both changes between individuals and within changes over time.
To this end, I want to use STATA to produce standardized correlations with significance tests on both within person and between person levels. Is this possible?
I want to include within-between correlations in my descriptive statistics.

-Thanks

[Clyde Schechter]

So for definiteness, let's say you have a predictor variable called X. First calculate each person's mean value of X. Then calculate each observation's diference X - (personal mean X). Then instead of regressing on X, regress on the person mean variable and the difference variable. The coefficient of the former estimates the between-person effect of X, and the coefficient of the latter estimates the within-person effect of X. If you want to do this in a correlation metric, standardize everything.

[Morten Nordmo]

Thank you Clyde. That worked and produced the desired standardized correlations. However, I was hoping there was way to obtiain these in a simpler fashion, within the mixed model in STATA. In comparison, the multi level output from Mplus includes within-between correlations, as a standard.

Thank's for the help

[Morten Nordmo]

On second insepction, I believe this procedure may be less valid. Comparing the output to the multilevel correlations from Mplus reveals some key differences: Inflated between individual correlations. This makes me suspicions that the covariance structure is not handled in a correct manner, using this procedure. I am not sure exactly sure why this happens, but I suspect that valid between and within correlations are only possible in a multi-level analysis.