Dear all,
I am currently working in Stata 13.0 with an xtologit model. I would like to verify that a random-effects model is appropriate via the Hausman test.
I have read in the manual that a hausman test is a postestimation command for xtologit. I have two questions and would appreciate your help.
1) I am unclear on how to do the hausman test after xtologit, since to the best of my knowledge I cannot specify fe and re. The closest instructions I found were for xtlogit (http://www.stata.com/statalist/archi.../msg00669.html).
Is it legitimate to do the hausman with an xtlogit in my case or is there a more adequat option?
2) Building on the link above, Prob>chi2 = 0.0040 and the author states at the end that "In this example, we reject the null hypothesis that the unobserved
individual level effects are uncorrelated with the other covariates. This implies that we should use the fixed-effects estimator instead of the
random-effects estimator". When I do the test for my model, I get Prob>chi2 = 0.37. Hence, I would choose a random effects model. Is this correct? (The reason for the question is, that I
have the impression that it is the other way round for xtreg, i.e., Prob>chi2 = 0.37 would support the fixed effects model.)
Many thanks in advance.
Benjamin Horvath
I am currently working in Stata 13.0 with an xtologit model. I would like to verify that a random-effects model is appropriate via the Hausman test.
I have read in the manual that a hausman test is a postestimation command for xtologit. I have two questions and would appreciate your help.
1) I am unclear on how to do the hausman test after xtologit, since to the best of my knowledge I cannot specify fe and re. The closest instructions I found were for xtlogit (http://www.stata.com/statalist/archi.../msg00669.html).
Is it legitimate to do the hausman with an xtlogit in my case or is there a more adequat option?
2) Building on the link above, Prob>chi2 = 0.0040 and the author states at the end that "In this example, we reject the null hypothesis that the unobserved
individual level effects are uncorrelated with the other covariates. This implies that we should use the fixed-effects estimator instead of the
random-effects estimator". When I do the test for my model, I get Prob>chi2 = 0.37. Hence, I would choose a random effects model. Is this correct? (The reason for the question is, that I
have the impression that it is the other way round for xtreg, i.e., Prob>chi2 = 0.37 would support the fixed effects model.)
Many thanks in advance.
Benjamin Horvath
