|
|
TWOLEVEL RANDOM slope model differenc... |
|
Message/Author |
|
Andy Luse posted on Wednesday, October 25, 2017 - 7:58 pm
|
|
|
I am comparing a twolevel model without random slopes to one with random slopes. I'm using the Satorra & Bentler method linked off the homepage to do LL difference testing since I am using MLR. Based on the formulas, I use the LL and correction values for H0 in the first model and the LL and correction values for H1 in the second model. The problem is that I only get LL and correction values for H0 in the random slope model. How would I compare these two? |
|
|
You have to be careful when using LL to test zero variance of a random effect, see http://amstat.tandfonline.com/doi/abs/10.1080/01621459.1987.10478472#.WfJ5YmhSzIU There is extensive amount of literature on that topic and some is Bayesian, see https://www.researchgate.net/publication/232064621_Bayesian_tests_of_measurement_invariance and tech16 in the Mplus user's guide Our simple rule of thumb is that if the T-test value > 5 - it is safe to conclude significance. You can also use BIC as an alternative. We always give the correction number with the MLR estimator - add estimator=mlr; in the analysis command of the input. If you still don't get a correction number send your data and input to support@statmodel.com |
|
Back to top |
|
|