Correlations with TYPE=COMPLEX PreviousNext
Mplus Discussion > Multilevel Data/Complex Sample >
Message/Author
 Lois Downey posted on Monday, January 12, 2009 - 9:24 am
My dataset includes responses collected over time from a group of respondents. I want to compute a correlation coefficient between two quasi-continuous measures, with standard errors corrected for clustering within respondents. However, when I ask for standardized parameter estimates, the standardized estimates match the unstandardized estimates, and both look like covariances, rather than correlations. I'm using the following syntax:

VARIABLE:
NAMES = ID wknum sncebase befordth
physhlth thnkrmbr recreate purpose
variety happy wnttodo PQOL
MSAStot MSASsom MSASpsyc Cncntrat
pain MSAS3 MSAS4 MSAS6 MSAS7 MSAS8
MSAS9 MSAS10 MSAS11 MSAS13 MSAS14
MSAS16 MSAS17 MSAS19 MSAS20 MSAS21
MSAS22 MSAS23 MSAS26 MSAS27 MSAS28
MSAS29 MSAS30 MSAS31 MSAS32
female age dxgrp cancer dropout died
nonwhite ed inc hospevr;

USEVARIABLES = PQOL physhlth;

MISSING = all (999);
CLUSTER = ID;

ANALYSIS: type=complex;

MODEL:
PQOL with physhlth;

output:
samp std;

What do I need to change in order to get correlations, rather than covariances?
 Linda K. Muthen posted on Monday, January 12, 2009 - 9:45 am
If you do TYPE=BASIC COMPLEX; without a MODEL command, you will get both covariances and correlations.
 Lois Downey posted on Monday, January 12, 2009 - 11:54 am
Is there a way to get p-values for the correlations, using the method you've indicated? My goal was to get standard errors corrected for clustering of responses within respondents, so I could evaluate the significance of the correlations.
 Linda K. Muthen posted on Monday, January 12, 2009 - 2:48 pm
No. If you want to use the MODEL command, you will need to use MODEL CONSTRAINT to define the correlation.
 Amanda Lemmon posted on Saturday, April 18, 2020 - 10:40 pm
Hello -

I have the same question as the topic starter. The syntax above (without MODEL CONSTRAINT) seems to work and produce correlations, standard errors, and p-values. I am not sure I understand why MODEL CONSTRAINT is needed?

I also wanted to ask a follow-up question. If I want to know correlations and their standard errors between a student variable (e.g., achievement) and a teacher variable (e.g., teaching style) where the teaching variable has the same values for all students taught by the same teacher -- would the method above with type = complex appropriate?

Thank you.
 Bengt O. Muthen posted on Sunday, April 19, 2020 - 12:24 pm
11 years have passed - now we give SEs with the standardized solution.

I would use Type=Twolevel and look at the standardized solution for the WITH statement on Between.
Back to top
Add Your Message Here
Post:
Username: Posting Information:
This is a private posting area. Only registered users and moderators may post messages here.
Password:
Options: Enable HTML code in message
Automatically activate URLs in message
Action: