|
|
 |
| Correlations for clustered variables |
 
|
| Message/Author |
|
|
|
|
Happy New Year!
I'm trying to estimate correlations for one set of variables with another for clustered data. I think I should be able to use the
v1 v2 WITH x1 x2 x3 ;
command and get standard errors for the correlations corrected for the dependent observations by asking for:
OUTPUT: STANDARDIZED(STDYX) ;
and using the CLUSTER var;
option.
However, my results aren't making sense, because although the CLUSTER option should correct standard errors, it shouldn't have a huge effect on the correlations, I'm thinking.
(When I run the analysis without CLUSTER in Mplus, the estimates are the same but the standard errors do differ.)
But, my thinking must be wrong, because the estimates are WAY different from correlations among the same variables in a program like (say) Stata without clustering.
Can you point me in the right direction?
Aren't those STANDARDIZED estimates correlations?
Thanks!
bac |
|
|
|
|
| Do you use Type=Complex or Type=Twolevel? Type=Complex doesn't change the point estimates, just the SEs - unless you have weights. |
|
|
|
|
Thank you very much, Bengt - Your question was very helpful! I was using Complex, and when I switched to using TwoLevel, I found that Basic was available (Duh!) and was able to get the correlations that way.
I also found that using Y1 WITH X1 X2 X3 tests the covariances adjusted for each other, so I did them one at a time and got tests for each pair.
Best,
BAC |
|
|
|
|
Saying
Y1 WITH X1 X2 X3;
does not test the covariances adjusted for each other, but the correlation for say Y1, X1 may change due to missingness, drawing on information from the X2, X3 variables. |
|
|
| Back to top |
|
|
