0
$\begingroup$

I am trying to estimate vote shares of different parties. So, suppose I have 4 parties, each having its own column in the data set. Hence, the sum would be tending to 1. Now, if a party does not contest in a particular area, then the vote share for that area is '.'. So, I have 4 equations -

sureg (voteshare_party1 voteshare_party2 voteshare_party3 vote share_party4 = X_variables i.fixedeffectvariable1 i.fixedeffectvariable2)

What I want to do is remove NAs separately for each equation, like it would have happened with reghdfe. But I need to use sureg because the error terms are correlated. Do you have any suggestions for this? Or any research paper which did something similar?

I am using Stata. In R, I used systemfit() but the estimates I am getting from Stata and R vary widely. I am happy to use any method given my estimates are consistent.

I really appreciate any suggestions / literature on this. And am happy to share more context.

$\endgroup$
1
$\begingroup$
  1. You don't need to use sureg. The benefit of SURE is smaller standard errors, but you could just estimate separate regressions. This is what I would advise. Are the independent variables the same in each regression? If so, then there is no efficiency gain in SURE at all, and separate regressions is what you should use.

  2. Your syntax for sureg seems incorrect, you want: sureg (depvar1 varlist1) (depvar2 varlist2) . . . (depvarN varlistN ).

  3. You could replace missing with 0, that's imperfect, but something to think about.

$\endgroup$
10
  • $\begingroup$ Thank you Michael! 1. But I have been advised to use SURE as our error terms are correlated. The independent variables are the same. But if we run separate regressions, then number of observations change, depending on the number of NAs in each independent variable column. 2. When the list of X's is the same, we can use the short notation (stata.com/manuals/rsureg.pdf). Nonetheless, I also used the one that you mentioned to double check. 3. Wouldn't replacing with 0s be misleading as the party did not contest election from that particular area, and give us biased results? $\endgroup$ Jul 31 at 2:11
  • $\begingroup$ Do you know of any literature wherein people did something like this? $\endgroup$ Jul 31 at 2:15
  • $\begingroup$ "The independent variables are the same. " Then SURE and OLS give the same standard errors. This has been discussed on this board. economics.stackexchange.com/questions/45753/… $\endgroup$ Jul 31 at 12:24
  • $\begingroup$ Thank you for sharing this! :) $\endgroup$ Jul 31 at 15:02
  • $\begingroup$ However, my OLS and SURE estimates are coming out to be different $\endgroup$ Aug 1 at 13:57

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.