Message/Author |
|
Tongyun Li posted on Wednesday, September 25, 2013 - 3:20 pm
|
|
|
Hi, I'm trying to run a multiple group logistic regression. I hope to compare the model without constraints and the model for which the regression coefficients are constrained to be equal across groups. The problem is that I have to use TYPE=COMPLEX with population weights and replicate weights to draw inference because the data is from a large-scale survey. It seems that DIFFTEST doesn't work with TYPE=COMPLEX. Is there another way to do model comparison test if I have to use TYPE=COMPLEX and plug in weights in my analysis? Thank you very much! |
|
|
It sounds like you are using WLSMV if you are using DIFFTEST. WLSMV give probit regression not logistic regression. You can use MLR and do difference testing and obtain logistic regression. |
|
Tongyun Li posted on Wednesday, September 25, 2013 - 3:28 pm
|
|
|
The estimator I used is WLSMV. |
|
|
So use MLR instead. |
|
Tongyun Li posted on Wednesday, September 25, 2013 - 3:39 pm
|
|
|
Thank you, Dr.Muthen! As I also have missing data (which I coded as 99), which estimator is better in my situation? |
|
|
MLR has better missing data handling. |
|
Tongyun Li posted on Wednesday, October 02, 2013 - 12:24 pm
|
|
|
Dr.Muthen, I've tried MLR, but it doesn't work because it's not allowed with TYPE=COMPLEX and replicate weights, and my outcome variable is categorical. Does that mean WLSMV is the only choice for my analysis? Thank you! |
|
|
WLSMV is your only option for categorical and replicate weights. |
|
Back to top |