Standard error of difference in a proportion (Two samples)

[Marc Asensio]

Dear all,

I've checked similar posts in this forum but none seems to address my particular issue.

Say that I want to compute the standard error of the difference in a proportion for a certain estimate (Marital status, measured at Wave 1) using two samples of panel nature (so, for respondents from Wave 1 v. respondents from Wave 2 - the later being the same as in wave 1 but less due to attrition).

Is there a particular Stata function that computes this? Or is there a way like in R to create a function (I think the answer here is no, so there should be a way in Stata)?

Thanks in advance for your help.

[Daniel Shin]

One option would be to bootstrap. How are you dealing with missing data (attrition)?

[Marc Asensio]

One option would be to bootstrap. How are you dealing with missing data (attrition)?

I am assessing the non-response bias by computing the difference in the estimates of the respondents I had at time Wave 1 compared to Wave 2 - so I guess that is my way. But it looks extremely complicate to do in STATA for how "simple" it should be.

[Daniel Shin]

Any reason why you can't use "simpler" approaches like using estimates from xtlogit or clogit?

[Tiago Pereira]

Margins (with the appropriate options and/or additional commands, e.g., nlcom) allows us to calculate the difference between risks, accounting for the correlation between waves. It depends on the approach you have used. Can you provide us with more detail? Perhaps an example?