[daniel klein]

I think you are better of with the RE approach, even if you are not interested in changes over time, as this model makes best use of the information in your data. Note that the RE estimator takes the correlated errors, i.e. the clustering of occasions in individuals, into account by design, which is another advantage - actually it is the reason it makes the best of your data.

If you still want to go with a cross section, just keep the values for one interview only. In that case, I guess a test for auto-correlation makes little sense, as auto-correlation in cross sections can only be due to clustering/sampling design, for which you should correct anyway.

Best
Daniel

[Katharina Meyer]

Hey Carlo,

I did run both regressions:

1)
xtlogit UG_W career_W social_W payment_W sex, re

Number of obs = 332
Number of groups = 140
Wald chi2(4) = 95.19
Prob > chi2 = 0.0000

This is the result of the likelihood ratio test:
Likelihood-ratio test of rho=0: chibar2(01) = 91.79 Prob >= chibar2 = 0.000


2)
logit UG_W career_W social_W payment_W, vce(cluster id)

Number of obs = 332
Wald chi2(4) = 36.10
Prob > chi2 = 0.0000
Pseudo R2 = 0.2088


If I understand it right I could use the pooled model?

Best Katharina

[Katharina Meyer]

the 2nd regression is of course
logit UG_W career_W social_W payment_W sex, vce(cluster id)

[daniel klein]

Based on the LR test you can reject the null-hypotheses, that the unit-specific variance is 0 at p = 0.000. So, no, based on this test you cannot use a pooled model. Whether it makes much a difference in your situation, I do not know, but since I still see no reason at all to not use the RE model, I will no longer contribute to this discussion.

Best wishes
Daniel

[Carlo Lazzaro]

Katharina:
I agree with Daniel's diagnosis of your issue.
As from

Likelihood-ratio test of rho=0: chibar2(01) = 91.79 Prob >= chibar2 = 0.000

you should go -xtlogit, re-; running a pooled logit model would be wrong, instead.
Hence, you should focus on your concerns about using -xtlogit, re - (that, admittedly, I do not understand, as Stata does all the calculations for you) when it is the 1st choice for analysing your data.