| Message/Author |
|
|
|
|
I need to run a CFA for items with censored data. I wonder if there is a recommended estimator for this type of CFA? I used MLR but only AIC, BIC and sample-size adjusted BIC are available as fit statistics. I couldn't interpret these outputs the regular way. I tried WLSMV and could get model fit statistics this way. But would WLSMV be a suboptimal estimator?
I would also highly appreciate it if there are any suggestions regarding the interpretation of AIC, BIC and sample-size adjusted BIC from the CFA with MLR. From my limited understanding of these fit statistics, they are for model comparison purposes. Would there be cutoff values or guidelines on how to interpret the fit for the focal model? I was not able to compare my hypothesized model with an alternative model because the alternative model never converge.
Thanks so much! |
|
|
|
|
WLSMV will give results similar to the full-information estimators ML and Bayes.
There are no cutoff values for BIC etc. |
|
|
|
|
Thank you so much for your feedback, Dr. Muthen!
If I may ask a follow-up question, is there a recommended estimator for censored data? Or the choice of estimator does not depend on whether it's censored data or not?
The reason why I ask is that my outcome variable is deviant behavior and the reviewer recommended using censored data CFA. But the model fit was good when I did not use censored data approach and MLR estimator, and the fit was very bad when I used the censored data approach and WLSMV estimator. Would this mean that censored data CFA is not an appropriate approach? Or would this be an issue of choice of estimators?
Thank you! |
|
|
|
|
Q1: See the FAQ on our website: Estimator choices with categorical outcomes
Q2: It does depend on what the outcome is.
You say "the model fit was good when I did not use censored data approach and MLR estimator" - the problem is that a linear model is used for a non-linear situation so the estimates and SEs are a bit distorted if you have strong censoring (say above 25% at the floor or ceiling). I would try to modify the model so that the model fits well as judged by WLSMV.. |
|
|
| Back to top |
