Person fit? PreviousNext
Mplus Discussion > Growth Modeling of Longitudinal Data >
 Mike Willoughby posted on Thursday, May 30, 2013 - 10:55 am
I am estimating a simple univariate LGC. Model fit was poor, but model modification indices do not indicate plausible changes.

Following Coffman & Millsap 2006, I wanted to consider if a small # of cases were contributing to poor model fit. I used:

file = xxx
save = loglikelihood influence cooks

A small # of cases had very large values for the loglikelhood. However, when I re-estimated the model excluding these cases (using subpopulation command b/c of the complex sampling design), model fit was *unchanged* even though the N reflected the reduction in cases. This was unexpected. My questions:

1. how are the individual LL values related to individual contributions to model chi square, which is what I'm after

2. is there a better way to assess person contribution to model misfit in Mplus?

3. might my use of complex sample (weights/strata) be contributing to lack of changes in model fit despite exclusion of cases w/ large LL values?

4. can you recommend reading for how the individual LL and influence statistics are calculated in Mplus?

Thank you.
 Bengt O. Muthen posted on Thursday, May 30, 2013 - 5:16 pm
Please send to support.
 Mike Willoughby posted on Friday, May 31, 2013 - 8:18 am
In working w/ this longer, I discovered that I had not requested correlations b/w time varying covariates. When I did this, there were no longer cases w/ large contributions to the LL (only highly influential cases). So my problem is solved.

However, I'm still interested in whether you conceive of the individual contribution to LL to be analogous in function to individual contribution to chi square (i.e., large positive values indicate individuals for who the model does not fit well)?

Thank you.
 Bengt O. Muthen posted on Friday, May 31, 2013 - 8:45 am
Yes, they are conceptually the same.
Back to top
Add Your Message Here
Username: Posting Information:
This is a private posting area. Only registered users and moderators may post messages here.
Options: Enable HTML code in message
Automatically activate URLs in message