|
|
 |
| Chi-sq vs LL conclusions in MLR diff ... |
 
|
| Message/Author |
|
|
|
|
I am currently comparing some nested models using the MLR difference testing methods outlined on the MPlus website, but have been getting substantively different conclusions on whether there is a difference in model fit when using the two approaches. Here's a summary to demonstrate what I mean:
Comparison model:
Chi-square: 4316.384
MLR scaling correction: 1.2008
df: 985
Log-Likelihood L0: -312276.28
MLR scaling correction: 1.5509
free parameters:573
Nested Model:
Chi-square: 5008.031
MLR scaling correction:
df: 1192
Log-likelihood: -312687.487
MLR scaling correction: 1.7539
free parameters: 366
---
TrD (chi-square) = 690 on 207 df (significant)
TrD LL = 2.655 on 207 df (not-significant)
I have a large sample size, so could this, or something else, be responsible for the differing conclusions?
My previous understanding was that the two methods were inter-changeable so are there any references that examine situations in which their performance differs? |
|
|
|
|
| Check your computations again and if you don't get it right, post your steps. |
|
|
|
|
Thanks Bengt,
I checked my workings again, and found I had made a mistake in the formula on my excel file in the Log-likelihood calculation when calculating cd. Now they produce almost identical values, so problem solved.
Not sure how I didn't spot that in my checks!
Thanks again |
|
|
| Back to top |
|
|
