Degrees of freedom in DIFFTEST
Message/Author
 Peter Ji posted on Friday, March 27, 2009 - 9:50 pm
Hello,

I am using DiffTest to compare the fit of a one factor, 6 correlated factor model, and a second order factor model. My colleague and I were wondering about the fact that the changes in df are not consistent across the waves when we conduct the DiffTest.

For Wave 1
One factor compared to 6 correlate factor model: chi-square difftest = 108.22, df = 13

6 correlated factor model compared to 2nd order factor model: chi-square difftest = 28.88, df = 8

2nd order factor compared to one factor model: chi-square difftest = 90.35, df = 6

For Wave 2
One factor compared to 6 correlate factor model: chi-square difftest = 229.57, df = 12

6 correlated factor model compared to 2nd order factor model: chi-square difftest = 9.70, df = 8

2nd order factor compared to one factor model: chi-square difftest = 197.14, df = 5

For Wave 3
One factor compared to 6 correlate factor model: chi-square difftest = 247.49, df = 12

6 correlated factor model compared to 2nd order factor model: chi-square difftest = 16.78, df = 8

2nd order factor compared to one factor model: chi-square difftest = 219.08, df = 5

We expected that the df's would be the same because the model structures are the same across the waves. Is there a cite for why the different df's occured?

Thank you.
Peter
 Linda K. Muthen posted on Saturday, March 28, 2009 - 10:15 am
With WLSMV, you should be looking only at the p-value for both model results and DIFFTEST. If you want to see degrees of freedom as expected, use WLSM.
 Peter Ji posted on Saturday, March 28, 2009 - 1:38 pm
thank you.
one follow-up. does it matter that the diff test is run with WLSMV vs. WLSM? Will reviewers take note that the df is not as expected because of the DiffTest used WLSMV vs. WLSM?
 Linda K. Muthen posted on Saturday, March 28, 2009 - 2:49 pm
DIFFTEST is only for estimators ending in MV. If you use WLSM, you need to use the scaling correction factor that is given and complete the difference test by hand. The difference in the number of free parameters between the two models is the same as the difference in the degrees of freedom.
Back to top
Add Your Message Here
Post:
Username: Posting Information:
This is a private posting area. Only registered users and moderators may post messages here.
Password:
Options: Enable HTML code in message
Automatically activate URLs in message
Action: