2
$\begingroup$

I found out that my panel data 'suffers' from heteroscedasticity by doing the test described on Stata FAQ, Testing for panel-level heteroskedasticity and autocorrelation. I understand that the $H_0$ for lrtest is homoskedasticity and I'm rejecting this. Please correct me if I'm wrong here.

I'm carrying out a fixed effects logistic regression with xtlogit and I found that the vce option should be used to correct for heteroskedasticity. Still, I don't understand when to use vce(bootstrap) or vce(jackknife) ?

What's the difference between the two? How can I decide which one to use?

EDIT: Some info on the data size:

   Number of observations: 116304
   Number of groups 2549
   Obs per group: min = 1
                  avg = 45.6273
                  max = 20
Improve this question
$\endgroup$
4
  • 2
    $\begingroup$ Aren't logit models heteroscedastic by definition? You explicitly model the mean-variance relationship through the logit link and binomial variance function. $\endgroup$
    – AdamO
    Nov 14, 2013 at 23:30
  • 2
    $\begingroup$ The FAQ you link to seems to refer to linear models, and you have a non-linear (logit) model. So, I don't think the test is relevant. $\endgroup$
    – Maarten Buis
    Nov 15, 2013 at 10:58
  • 1
    $\begingroup$ Thanks! I just found your explanation for heteroscedasticity in probit models here @MaartenBuis : stata.com/statalist/archive/2010-11/msg00996.html $\endgroup$
    – yumba
    Nov 15, 2013 at 11:12
  • 1
    $\begingroup$ For clarification, does this mean that I don't worry about autocorrelation as well? $\endgroup$
    – yumba
    Nov 15, 2013 at 11:29

0

Reset to default

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.