Bookmark and Share

Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

st: Seemingly unrelated logistic regression (SUR logit) vs. -suest-

From   "Hogerbrugge, M.J.A. (Martijn)" <>
To   "" <>
Subject   st: Seemingly unrelated logistic regression (SUR logit) vs. -suest-
Date   Mon, 21 Oct 2013 16:38:45 +0000

Dear Stata users,

I am currently analysing 4 binary DVs for which I have reason to believe that they are correlated. To complicate things, I have a large number of missings for one of the DVs.

If I want to run a seemingly unrelated logistic regression, would it be somehow possible to include all cases, even those which have a missing on one of the DVs? 

If not, I believe the -suest- might be a good alternative, as it allows the user to compare models/equations that use different/unequal sample sizes. But does -suest- really take into account that DVs are correlated? In other words, to what extent is the -suest- command really a Seemingly Unrelated ESTtimation?

The latter question is triggered by my observation that the SEs of the coefficients of the separately estimated logistic regressions are identical to the ones that are printed after combining the estimations of the separate models using -suest-. Although identical SEs seem to occur only when I use the -vce cluster- option. Is the simultaneous use of logit models with clustered data using the -suest- postestimation not possible? Or is it correct that the SEs no longer differ from single estimations if the -vce cluser- option is used?

I look forward to any advice regarding these issues.

With kind regards,
Martijn Hogerbrugge
*   For searches and help try:

© Copyright 1996–2018 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   Site index