Message/Author |
|
Phil Wood posted on Wednesday, September 21, 2016 - 8:17 am
|
|
|
Hi folks, I have two models based on the same set of simulated data. One of which has a very significant chi square (416) and nine degrees of freedom, and the other has a very nonsignificant chi square (7.9) and 8 degrees of freedom. Yet, the BIC for the first model is smaller (10217.309) than for the second (12302.872). 1000 replications of this data yield this pattern across all replications. My question: The second model contains some nonlinear constraints which are functions of ther parameters in the model. Does Mplus count these as additional parameters when computing the penalty for the BIC? Thanks! |
|
|
No. BIC uses what is reported in the output as the number of free parameters. Send your example to support@statmodel.com It looks to me that what you are describing is mathematically impossible. |
|
Phil Wood posted on Wednesday, September 21, 2016 - 1:08 pm
|
|
|
I'll do that. Perhaps the likelihood function is the problem, given that I'm programming nonlinear constraints, but you'll see that shortly. Thanks so much- Phil |
|
Phil Wood posted on Wednesday, September 21, 2016 - 5:43 pm
|
|
|
Just so that people know- the problem was a syntax mistake. (It's sometimes frustrating to see the "send it to us" message with no follow-up. Apologies! |
|
Back to top |