Correlations among a larger number of...
Message/Author
 EFried posted on Tuesday, April 02, 2013 - 9:45 am
Dear Prof Muthens,

I am aiming to investigate whether the effects of a number of risk factors (x) are differentially associated with various medical symptoms (y).

Correlations among x, and correlations among y, are moderate to high (0.2-0.6). Correlations between x and y are low (maximum 0.2).

I want to compare two models: one in which the effects of all x on all y are freely estimated, and one in which the effects are constrained (see example below).

Model I:
y1 ON x1 x2;
y2 ON x1 x2;

Model II:
y1 ON x1 (1);
y1 ON x2 (2);
y2 ON x1 (1);
y2 ON x2 (2);

(1) I want to allow for correlations between all x. Is
x1 x2;
the right way to do this? Does it have consequences regarding if model I or model II is preferred?

(2) The unconstrained model has perfect fit (e.g., with 4y and 7x and 42 DF). Is that a sign that something might be wrong with the model?

(3) What test do I perform to compare Model I and Model II -- is it correct to use

Thank you
 EFried posted on Tuesday, April 02, 2013 - 9:48 am

"What test do I perform to compare Model I and Model II -- is it correct to use a LLH test?
 Bengt O. Muthen posted on Tuesday, April 02, 2013 - 10:19 am
1. The correlations among the x's are free as the default without mentioning their means, variances, or covariances.

2. If you have 42 df and perfect fit, something strange is going on. Also, Model I looks like it has 0 df. Perhaps you are inadvertently uncorrelating the x's; check your output.

3. A likelihood-ratio chi-2 test comparing Model I and II is fine. I don't know what a LLH test is.
 EFried posted on Tuesday, April 02, 2013 - 10:50 am
Bengt, thank you very much for the quick answer. Running a simplified model with much fewer variables:

MODEL:
d_s1_m ON Sex;
d_s1_m ON ZNeuro;
d_s3_m ON Sex;
d_s3_m ON ZNeuro;

This results in the perfect fit summarized below. What am I missing?

Thank you!

9 DF
LLH H0 -2940.091
LLH H1 -2940.091

AIC 5898.182
BIC 5945.495

Chi-Square Test of Model Fit
Value 0.000
Degrees of Freedom 0
P-Value 0.0000

RMSEA (Root Mean Square Error Of Approximation)
Estimate 0.000
90 Percent C.I. 0.000 0.000
Probability RMSEA <= .05 0.000

CFI/TLI 1.000

Chi-Square Test of Model Fit for the Baseline Model
Value 164.887
Degrees of Freedom 5
P-Value 0.0000

SRMR (Standardized Root Mean Square Residual)
Value 0.000
 Linda K. Muthen posted on Tuesday, April 02, 2013 - 12:10 pm
You model has zero degrees of freedom.

Chi-Square Test of Model Fit
Value 0.000
Degrees of Freedom 0
P-Value 0.0000

Model fit cannot be assessed for a model with zero degrees of freedom.
 EFried posted on Wednesday, April 03, 2013 - 4:49 am
Thank you. When running 10 regressions in MPLUS as listed above, allowing all variables to correlate with each other, do I need to adjust for type-I error manually, or does MPLUS do that automatically?
 Linda K. Muthen posted on Wednesday, April 03, 2013 - 2:19 pm
No.
 EFried posted on Sunday, April 07, 2013 - 4:09 am
Sorry --- "No." regarding "do I need to adjust for type-I error manually", or "No." regarding "or does MPLUS do that automatically"?

Thank you
 Linda K. Muthen posted on Sunday, April 07, 2013 - 6:12 am
No, nothing needs to be done.
 EFried posted on Sunday, April 07, 2013 - 6:14 am
Great, thank you!