WLSMV with missing data PreviousNext
Mplus Discussion > Missing Data Modeling >
 Eric Teman posted on Friday, June 22, 2012 - 5:51 pm
On p. 7 of the user's manual, it says essentially that if WLS is used, missingness is not allowed to be a function of observed outcomes. Does this mean WLSMV requires MCAR data?
 Bengt O. Muthen posted on Friday, June 22, 2012 - 8:39 pm
Not if you have covariates - they are allowed to influence missingness.
 Bengt O. Muthen posted on Friday, June 22, 2012 - 8:39 pm
And, you can do multiple imputation followed by WLSMV.
 Eric Teman posted on Wednesday, July 04, 2012 - 4:56 pm
If there are a mix of categorical and continuous variables and WLSMV is used during the imputation phase of multiple imputation,

1. Does the multivariate probit model only apply to the categorical missing values or to continuous as well?

2. Does the multivariate probit model assume normality?
 Linda K. Muthen posted on Thursday, July 05, 2012 - 10:38 am
Multiple imputation is done using the Bayes estimator.

1. Only categorical.
2. Yes.

See the following paper on the website under Papers/Bayesian Analysis:

Asparouhov, T. & Muthén, B. (2010). Multiple imputation with Mplus. Technical Report. Version 2.
 Sarah  posted on Wednesday, August 27, 2014 - 6:48 am
Dear Drs. Muthen

I have a query regarding missing data when using WLSMV estimator. I conducted a CFA to validate a single latent variable construct of father involvement using data from sweep 4 of a longitudinal survey. My sample size was 363.

I then used this latent construct of father involvement at sweep 4 as a predictor of child well-being at sweep 5 (child well-being was a four factor model which I had previously validated by conducting a CFA). I also include a range of control variables for child well-being. The sample size for the full model was 525.
Is the sample size larger because only cases with data missing on X variables are removed? How are estimates calculated based on 525 cases when only 363 cases provided information for the measurement model?

I hope my query is clear.

Many thanks for your help.

 Bengt O. Muthen posted on Wednesday, August 27, 2014 - 4:46 pm
I assume that you don't use factor score estimates from sweep 4 to compute the father involvement variable but rather use a CFA with all those indicators. The analysis excludes those with missing on X's and missing on all the Y's, so sweep 5 must have added subjects with at least one sweep 5 Y.
Back to top
Add Your Message Here
Username: Posting Information:
This is a private posting area. Only registered users and moderators may post messages here.
Options: Enable HTML code in message
Automatically activate URLs in message