Alice Frye posted on Thursday, October 29, 2015 - 11:42 am
Hi, I'm estimating latent and observed path models that including mediating pathways, using bootstrapping.
I had the impression that it was not possible to use an MLR estimator as part of this process, since when I try to I get an error message saying that MLR estimator is not available with bootstrapping. A reviewer has told me that it is possible, and that I should be using an MLR estimator rather than an ML estimator.
When you use ML + bootstrapping you get bootstrap standard errors for ML parameter estimates. MLR gives MLR SEs for ML parameter estimates. This means that trying to do MLR with bootstrap is a contradiction and we therefore don't allow it. Either you want bootstrap SEs or MLR SEs.
I'm testing competing SEM models (stability model, normal causation, reversed causation and reciprocal causation) using Likert-type scored items only as indicators for my latent variables. Which estimator would you recommend using, MLM or MLR? Depending on the estimator I use, I get different result in terms of which model is best. Many thanks!
Hello! I am running CFAs in Mplus using ordinal data (frequency of PTSD symptoms on 5-point Likert scale); we have decided to treat these data continuously due to skewness of our data and a precedent set in the literature with this measure. There are no missing data. We ran CFAs examining model fits using the MLM estimator and MLR estimator, which yielded slightly different chi-square values between the two. I was under the impression that these two estimators should yield the same values in the presence of NO missing data. Are the chi-square values equivalent across the MLM and MLR estimators?