I have a problem with how to test the difference of two models, one is a unconstrained model, the other one is constrained, with a variable (or more) coefficient forced to be equal. The code is as following:
* assign the survey design
svyset [w=weights12], psu(sdmvpsu) strata (sdmvstra)
*Set constraints
constraint 1 [1=2]:hyperchol
constraint 2 [1=2]:hypertension
*unconstrained model
svy: mlogit db_cat hypertension hyperchol
*preserve the unconstrained model
estimates store unconstrained
*constrained model
svy: mlogit db_cat hypertension hyperchol, constraints(1)
*test the significance between constrained and unconstrained model.
lrtest unconstrained
----------------------
So when I use lrtest, of course it is error, "lrtest is not appropriate with survey estimation results". But how to use test to do this comparison? I searched for this but most of posts are about set coefficient to 0, not equal between two models.
I need the code to do the similar thing( comparison between models) like the lrtest when there is no survey design.
Thank you very much!
Hi Xin,
There is a Stata FAQ post on this (link). "lrtest" is simply not an option, so you'll have to think of an alternative approach. The Stata FAQ post suggests "test". So run an unconstrained model and "test hypertension == hyperchol" might suit your purpose.
- Rich
Ref:
http://www.stata.com/support/faqs/statistics/likelihood-ratio-test/
Example:
sysuse auto, clear
foreach x in weight length {
qui su `x'
gen z`x'=(`x'-`r(mean)')/`r(sd)'
}
qui su mpg
gen lowmpg=mpg
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Statistics
- Statistical Software