|
 |
Negative Residual Variance EFA |
 
|
Message/Author |
|
|
Hello, I am trying to run an EFA on 14 categorical items (some dichotomous, some 3 point scale) with a sample of 342 participants. I have used the WLSMV estimator and Geomin rotation. A one factor solution fits the data poorly (RMSEA = .098, CFI = .928, SRMR = 0.147). The two factor and three factor solutions fit the data well, but each have one item with negative residual variance. A statistics professor recommended using the MLR estimator instead. When I do this, there are no negative residual variances. 1) Is it appropriate to report the model using the WLSMV estimator? or should I report the model using the MLR estimator? 2) Why is my WLSMV estimated model getting negative residual variances? Thank you very much for any help you can provide! |
|
|
When you say the MLR estimator I assume that you mean that you still declare the items as categorical. Does the MLR output actually print residual variances? |
|
|
Hello, Yes the items are still listed as categorical and yes there are residual variances. There is an error message stating "the standard errors of the model parameter estimates may not be trustworthy for some parameters due to a non-positive definite first order derivative product matrix. this may be due to the starting values but also may be an indicator of model non-identification'. I also re-ran the EFA with WLSMV with the item that has negative residual variance removed. When I do this, the 2 factor solution has acceptable fit and no negative residual variance - I am considering just reporting this model. Is this appropriate? Thank you very much for you help! |
|
|
Please send the MLR output to Support. Removing items with negative residual variances is a questionable approach. It isn't necessarily the item with a negative residual variance that is the ultimate cause of the problem and often removing one item makes the negative residual variance show up somewhere else. But if you don't need the item, the decision is yours. |
|
Back to top |
|
|