

Log likelihood vs chi square differen... 

Message/Author 

Ger_Wel posted on Friday, October 20, 2017  10:21 am



I have some related questions, hopefully you can help me out: a) I am trying to find out when to use chi square difference testing and when log likelihood difference testing (using the SatorraBentler Scaled ChiSquare). Are these interchangeable options? If not, when should I use one or the other? b) In my case, I am testing a multilevel mediation model and I am adding predictors stepbystep. In the first steps the model is saturated, so the chi square is zero. Can I use the log likelihood difference test here? c) In an old post http://www.statmodel.com/discussion/messages/12/27.html#POST18651 Linda (May 27th 2008),states that the log likelihood has to be multiplied by minus 2, does that still apply or is that outdated? d) I want to use the log likelihood difference test to check if a (level 1) predictor adds a significant amount of explained variance. Is this possible this way, or is there a better option? e) Can the number of parameters (for the log likelihood diff test) only be found by identifying the highest parameternumber in the Tech1 output, or is it also somewhere in the regular MPLUS output? I am looking forward to your reply. Best, Gerald. 


a) they are interchangeable  see http://www.statmodel.com/chidiff.shtml b) No. c) 2 logL is used in connection with likelihoodratio chisquare. d) Good question for SEMNET. e) That's shown at the top of the Model Results section of the output. 


Dr. Muthen, Do you by chance have a citation for the page on chi square difference testing? 


The SatorraBentler Psychometrika article referred to there would be a good reference. 


Thank you! 

Back to top 

