Hi, Is there a way to get standardized coefficients and fit indices (RMSEA, CLI, etc.) with MLR? I am comparing models and selecting the best with BIC but because I have an interaction model using the xwith command I don't seem to have these available. Thanks.
With TYPE=RANDOM; the variance of y varies with the values of x. This precludes the calculation of standardized coefficients and chi-square and related fit statistics. In this case, nested models are tested using -2 times the loglikelihood difference which is distributed as chi-square.
Hi, I am testing the direct and interaction effects of two second-order continuous latent variables (f1 and f2) on another continous latent variable (f3), and the mediation effect of this variable on a binary (f4) dependent variable with a single indicator. My model is roughly as follows (I have left out BY commands for f5 through f10 for simplicity):
f4 BY y1; f1 BY f5 f6 f7; f2 BY f8 f9 f10; f3 BY x1-x6; f3 ON f1 f2; f1xf2 | f1 xwith f2; f3 ON f1xf2; f4 ON f3 f1 f2 f1xf2
I have two questions:
1. I understand that I need to use either MLR or WLSMV estimators in a random effects model with numerical integration. Is this correct?
2. I also understand that no fit indices are provided when testing continuous latent variable interactions. If my understanding is correct, then how can I understand whether my model is good or ill fitting for the full model?
I will greatly appreciate any suggestions and guidance.
2. No absolute fit statistic is available for this model. I would make sure the fit is good for the model without the interaction and then check that the interaction is significant. See a recent paper in Psychometrika on fit for interaction models by Moojart and Satorra.
While testing a model with a single-indicator, binary dependent variable, I have run into an issue that I can't figure out.
My model is as follows: VARIABLE: CATEGORICAL ARE y1; ANALYSIS: ESTIMATOR=MLR; MODEL: f1 BY x1-x3; f2 BY x4-x6; f3 BY x7-x9; f4 BY x10-x12; f5 BY x13-x15; f6 BY x16-x18; f7 BY x19-x25; f8 BY f1-f3; f9 BY f4-f6; f10 BY y1; f7 ON f5 f6; f10 ON f7;
The model runs fine with a single indicator continuous dependent variable, I get the following message when I use the binary DV:
THE MODEL ESTIMATION DID NOT TERMINATE NORMALLY DUE TO A NON-ZERO DERIVATIVE OF THE OBSERVED-DATA LOGLIKELIHOOD.
THE MCONVERGENCE CRITERION OF THE EM ALGORITHM IS NOT FULFILLED. CHECK YOUR STARTING VALUES OR INCREASE THE NUMBER OF MITERATIONS. ESTIMATES CANNOT BE TRUSTED. THE LOGLIKELIHOOD DERIVATIVE FOR PARAMETER 72 IS 0.16653809D+00.
I have tried increasing the MITERATIONS up to 1000, but the result remained the same. I would greatly appreciate some insight regarding what I might be doing wrong.
Second, for such a model, is there a way to get the indirect effect of f8 (via f7) on f10?
No, we don't give them automatically. You can use MODEL CONSTRAINT to do this.
yao lu posted on Tuesday, March 13, 2012 - 7:37 pm
Following up your post above (July 26, 2010-7:40am),how can Model Constraint calculate standardized coefficients when using Type = Random? I read the Model Constraint section in the manual, but still not quite get it.
Could you please provide me the coding for calculating standardized coefficients?
I am currently testing a cross-level interaction. According to the discussions absolute statistic fits for those sorts of models (Type = Random;) are not available. You proposed to make sure the fit is good for the model without the interaction.
Which part of the model do I need to change to test the same model without the interaction.
This is my model:
Variable: usevar = code S1 cs bo MG; cluster = code; within = cs; ! between = S1 MG;
analysis: type = twolevel random; model: %within% beta1j | bo on cs; %between% bo Beta1j ON S1 MG; bo with beta1j; output: sampstat;