Message/Author |
|
carla hard posted on Tuesday, November 20, 2012 - 5:09 am
|
|
|
dear dr. muthén, when doing likelihood-based (!) difference testing/strictly positive difference testing using MLR as described on your webpage/webnotes - how do I name it correctly? as far as I know, satorra and bentler developed the method for difference testing/strictly positive difference testing using correction factors, but with MLR the correction factors of yuan & bentler are provided- is this right? what is the difference between the satorra-bentler and the yuan-bentler correction factors? and what is the correct name especially when doing it likelihood-based? thanks in advance carla |
|
|
See the following paper: Satorra, A., & Bentler, P.M. (2010). Ensuring positiveness of the scaled difference chi-square test statistic. Psychometrika, 75, 243-248. |
|
|
See also the more applied Bryant-Satorra article that appeared in the SEM journal recently: http://www.statmodel.com/download/BryantSatorraInPressSEM2011.pdf |
|
carla hard posted on Wednesday, November 21, 2012 - 4:22 am
|
|
|
Thank you for the quick answer. I´m just wondering if the correction factors provided by MPlus when using the MLR-Method are the same like the correction factors used by Satorra and Bentler in their articles. |
|
|
Yes. |
|
Shiny7 posted on Friday, November 28, 2014 - 2:15 am
|
|
|
Dear Dr. Muthen, first of all: thank you so much for your professional suppor being so valuably. I´ve got a further question concerning Difference-Tests for nested models using MLR-estimator (simple mulitple regression analysis). Mplus usually provides two ways for difference testing: Chi-square and loglikelidhood. a) Which one do I have to prefer? Or are they equivalent? b) When I do NOT get Chi-square statistics because of df being 0, can I readily take the LL-Difference Test? Thank you in advance for your reply. Shiny |
|
|
a. They will give the same results. b. Yes. |
|
Shiny7 posted on Friday, November 28, 2014 - 12:43 pm
|
|
|
Perfect, thank you very much! |
|
Back to top |