Hi Statalist,
I am running a probit regression, using survey weights via the code below, and want to test (and control) my model for heteroskedasticity. From reading though a selection of sources online, my understanding is that using survey weights creates standard errors that are robust to heteroskedasticity.
From reading though a selection of sources online, my understanding is that using survey weights creates standard errors that are robust to heteroskedasticity. However I have also read from previous statalist threads that robust standard errors are not sufficient when using probit/logit models, and have seen hetprob recommended instead.
So I have decided to run a regression using hetprob (Code below), but am not sure if using it with the svy prefix will produce inaccurate results due to it already adjusting for heteroskedasticity. Is it a case of one or the other? Or is it fine to go ahead and use the following code:
Thanks,
Mike
I am running a probit regression, using survey weights via the code below, and want to test (and control) my model for heteroskedasticity. From reading though a selection of sources online, my understanding is that using survey weights creates standard errors that are robust to heteroskedasticity.
Code:
svyset psu, strata(Cluster) weight(wt_int) vce(linearized) singleunit(certainty)
So I have decided to run a regression using hetprob (Code below), but am not sure if using it with the svy prefix will produce inaccurate results due to it already adjusting for heteroskedasticity. Is it a case of one or the other? Or is it fine to go ahead and use the following code:
Code:
svy: hetprob y x1 x2 x3 x4 x5, het(x1 x2 x3 x4 x5)
Mike
