IV testing after logit svy prefix

[Eugene Chong]

Could someone please help, no one is answering my posts but everyone else's questions get answered.

I just need to know of a way to test for exogeneity and relevance of instruments.

I have survey data, hence the svy prefix. I need to do an IV regression but one of my endogenous variables is a dummy, so normal -regress- shouldn't work. after svy: logit I cannot estimate residuals, nor can I use ivreg2 after svy to conduct the necessary tests.

Does anyone have a solution please??

[Richard Williams]

svy: ivprobit works, Maybe you can see if it will meet your needs.

Unfortunately, a lot of things that work without svy do not work with it:

https://www3.nd.edu/~rwilliam/xsoc73...yCautionsX.pdf

Not getting answers is nothing personal. People may just not know the answer! In this case, I don't know if there is one.

[Eugene Chong]

Hi Richard,

Thanks for responding, it just puzzles me that other people have the same problems as I do but no one seems to respond!

Unfortunately, svy: ivprobit does not work. I'm sorry if I did not specify my problem well enough, essentially I have a continuous dependent variable, so I do not need probit/logit. I simply regress will do. However, I have an ENDOGENOUS regressor which is binary, and while ivregress may suit my needs, I am unable to carry out any sort of tests for suitability (instrument relevance etc.).

Now, I tried the 'manual' way of using svy: probit first, then -predict, residuals- but apparently I am not allowed to do so. Any thought?

Many thanks again for responding, you are quite right in noting that many commands do not work with svy, however, I deem it necessary for my project to account for weights.

[Clyde Schechter]

I deem it necessary for my project to account for weights

If the only reason for using -svy:- is to deal with sampling weights, most estimation commands in Stata will allow you to just specify [pweight = weight_variable_name]. I'm not saying you should ignore strata and PSU's if you have them, but if you don't have them, then there is no need for -svy- just to handle pweights. That may make your life simpler. I don't really know as I don't normally deal with the kind of situation you're describing here.

... it just puzzles me that other people have the same problems as I do but no one seems to respond!

Your frustration at not getting a timely answer is understandable, but objectively this is just not true. If it were true, you would have no need to wait for a response: you could just search the Forum for some other post asking about the same problem and take your answer from there.

When a post goes unanswered for a long time, it usually means either that nobody knows the answer, or the question is unclear and nobody knows what is being asked.

[Eugene Chong]

Dear Clyde,

Thank you for your input, -svy- does indeed make life harder and just weighting using your suggested option gets around the problem; I will have to check if estimates run far from the original stratification.

I did search the forum for posts about the same question, but they have been unanswered! It's the first time I came across a problem StataList does not have the answer to so naturally I panicked a bit.

Thanks again!