xtlogit how do I test for autocorrelation

[Hien Nguyen]

Hi all,

How do I test for auto correlation after using the xtlogit command in stata and if autocorrelation exit how do I treat it?

Hien

[Joseph Coveney]

Try

search qic

Assumes that you're doing

 xtlogit , pa

[Jeff Wooldridge]

It's very hard to both test and correct if you are using FE or RE. There are other options. You need to be very specific on this listserve if you want to receive useful advice.

[Hien Nguyen]

Dear Professor Wooldridge,

This is my problem, I am using panel data set with 3 years and each year has 3000 individual. When I run

xtserial empstat_d age age_sq sex_d marst_d ethnicvn_d edattain_d indgen_d

It shows

Wooldridge test for autocorrelation in panel data
H0: no first-order autocorrelation
F( 1, 2999) = 75.622
Prob > F = 0.0000

So it mean there is serial correlations in my data set...When I run

xtlogit empstat_d age age_sq sex_d marst_d ethnicvn_d edattain_d indgen_d, fe vce(robust)
trying to correct for autocorrelation

vcetype 'robust' not allowed
r(198);

same with random effect

[Hien Nguyen]

I am not sure understand the pa option after xtlogit either population average how does that work, I didn't learn about that when I was studying my econometric class....

[Shakira Foster]

Dear Mr Wooldridge, dear Statalist community,

We are currently analyzing a dataset about the German Automotive Industry from 2015-2019. The dataset contains data from 10 OEMs and supplier throughout the 5 years. With the help of a logistic regression we analyze our model which contains a binary dependent variable and eight independent variables (of which 7 are binary scaled and one is metrically/continuously scaled). From the literature we have seen that one assumption that has to be tested is autocorrelation. However, we are not sure how to conduct the test with mainly binary variables. Do you know if autocorrelation is even possible when analyzing binary (in)dependent variables and one continuously scaled variable?

I am really desperate and would really appreciate your help.

Thank you already in advance!

Best Regards,

Shakira

[Jeff Wooldridge]

How large is N? Hopefully not N = 10, as that would be very small. Are you use FE or RE, or pooled logit?

[Shakira Foster]

Thank you for your reply! Our sample size is n =3308. And we are using FE. However we're not sure if its a pooled logit- in how far does it make a difference?