Message/Author 


I am working on the development of a measurement instrument measuring relationship closeness. i have found that there is good fit for the hypothesized model at the baseline measurement point. I have available data for the same participants 4 months later, after they have participated in a facilited group intervention. i would hypothesize that they would have higher scores on this measure at that point (having been in the intervention together) but would hope that there would not be differential item functioning (if this is truly a reliable set of items). is it appropriate to use a MIMIC model in this waywith the covariate representing time of measurement (basleine vs. postintervention)? 

bmuthen posted on Monday, February 09, 2004  8:12 am



I would not use MIMIC here because it would not take into account that the measures at the two time points are correlated  you have nonindependent observations. Instead, I would suggest a 2time point CFA, so if your instrument has 10 items, you would analyze 20 items in one run. Then test item invariance across the two time points. 

AdamCarrico posted on Monday, February 14, 2005  7:08 am



I am reviewing a manuscript that utilizes MIMIC to examine treatment effects over 2 timepoints. From my reading of the MPlus handouts, this would seem to be an inappropriate stategy. Specifically, MIMIC does not account for nonindependence of the measures over time. Is this correct? 


With two timepoints, an intercept only model could be estimated. This would be fine. Adding a covariate of treatment to this model would also be okay. It's is difficult to answer your question without seeing exactly what they are doing. 

AdamCarrico posted on Monday, February 14, 2005  12:24 pm



The authors used a MIMIC model to create two latent factors of distress at: Baseline (T1) and Change (T2). Then they used treatment as a covariate to predict the change latent factor of distress (T2) to examine intervention effects. 


This sounds fine. 

Anonymous posted on Tuesday, June 14, 2005  11:10 am



I got a negative estimate for a covariate in MIMIC model using the two factors extracted from EFA. But the simple correlation analysis gave positive correlation between the covariate and the first and second factor (these two variables were created based on the items that load on the factor). I'm confused with the contradictary results, but there seems no mistake in both calculations. Thank you. 

bmuthen posted on Wednesday, June 15, 2005  7:39 am



Perhaps the CFA measurement model (without covariates) does not fit sufficently well as compared to the EFA. 

J.W. posted on Sunday, March 01, 2009  8:09 am



A couple of questions on factor covariance estimates in MIMIC model. 1) Are the estimated factor covariances in a MIMIC model reported in the Model Results section of Mplus output residual covariances? 2)Tech4 ESTIMATED COVARIANCES FOR THE LATENT VARIABLES in a MIMIC model are different from the residual covariances in the Model Results section; they are close, but not identical, to those estimated from the corresponding CFA model without covariates. How to interpret those covariances? Thank you. 


Covariances are estimated for exogenous factors. Residual covariances are estimated for endogenous factors. TECH4 provides model estimated covariances not residual covariances. If you estimate a model where all factors are exogenous, you will obtain the same factor covariances as are given in TECH4. 


Hi, I did a CFA with 3 factors and all the indicators of each factor are significant. And, I did also a MIMIC from this analysis with four covariates. And there, I saw that the indicators relative to the factor 2 become non significant. The covariates also are not significant in this factor. What can explain that? Should I need to respecified my model, If yes, how? Thank you 


Did you regress the factors on the covariates or the factor indicators? 


I regress the factor on the covariates. 


When covariates are included in the model, the sample statistics to which the model is fit are no longer just the means, variances, and covariances among the factor indicators. The covariances between the covariates and the factor indicators are also sample statistics to which the model is fit. This can change the estimates of the factor loadings. 

Back to top 