How to compare nested clogit with robust or clustered standard errors

[Charlie Hammond]

Stata/MP 14.2. No access to internet.

My data is longitudinal, with 128 individuals and around 700 consecutive daily observations per individual. There can be many zeros per individual or there might be many non-zero outcomes, just depends on the case.

How can I compare the goodness of fit of two nested models fitted with clogit with robust or clustered standard errors?

lrtest is not an option according to Stata.

I have found nothing on the internet or in the documentation.

[Andrew Musau]

A Wald test is still valid with clustering. See especially the second link.

https://www.statalist.org/forums/for...-robust-errors
https://www.stata.com/support/faqs/s...od-ratio-test/

[Charlie Hammond]

I guess where I get confused with the Wald test is how to compare two models with only one variable missing from one of them. The Wald test just gives the p-value found in the standard output.

So when modifying my model I should remove two or more variables at a time, then test them for joint significance using Wald?

Thanks Andrew!

This was helpful also.

https://stats.idre.ucla.edu/stata/fa...test-in-stata/

[Andrew Musau]

Yes, as illustrated in the link. You test for the joint significance of the additional variables.

webuse lbw
logit low age ptl ht
est sto m1
logit low age ptl ht smoke i.race
est sto m2
lrtest m1 m2
est restore m2
testparm smoke i.race

Res.:

. lrtest m1 m2

Likelihood-ratio test                                 LR chi2(3)  =     10.18
(Assumption: m1 nested in m2)                         Prob > chi2 =    0.0171

. est restore m2
(results m2 are active now)

. testparm smoke i.race

 ( 1)  [low]smoke = 0
 ( 2)  [low]2.race = 0
 ( 3)  [low]3.race = 0

           chi2(  3) =    9.31
         Prob > chi2 =    0.0255

[Charlie Hammond]

Thanks Andrew!