Margins after ivprobit

[Tricia Chia]

Hi,

When I use the margins, dydx(explanatory variable) command after an ivprobit estimation, the marginal effect reported is the same as the coefficient in my ivprobit regression. Does anybody know why this is the case?

Thanks!

[Richard Williams]

Check out the post-estimation help for the command. When margins is used after ivprobit, xb is the default option. You probably want the pr option, but read the help.

[Tricia Chia]

Hi Richard, thank you so much for your help! it was very helpful.

I am now having the problem where when i run an ivprobit regression and then execute the command "margins, dydx(x) predict(pr) post", it takes hours to give me the marginal value. Do you by any chance know how I can make the command run quicker?

Thanks!

[Jeff Wooldridge]

I've noticed that margins takes somewhat long after ivprobit, but that means more like 5 seconds with 6,000 observations and about 8 covariates. It shouldn't take hours unless the data set is very large.

It is easy to use the two-step control function method and then use the margins command after probit -- which runs very quickly -- but then you need to use the bootstrap for the standard errors. If you show code then I can likely help with that.

[Richard Williams]

If you are willing to go without the standard errors, the -nose- option on margins can speed things up.

[Tricia Chia]

Hi Jeff,

my code is

ivprobit fulltime ageh i.health i.religion ethnicity marriage i.la_group year (imm = IV) if highskill == 1 & sex == 2, vce(cluster la_group)

i.la_group refers to local authority dummies and I have 271 local authorities in my sample. I am thinking that this could be the reason why the regression takes such a long time. Do you have any thoughts?

Thanks!

[Jeff Wooldridge]

Tricia: Yes, the 270 group dummies are likely causing computational problems, but it's more serious than that. There's likely a serious incidental parameters problem. How many observations per local authority do you have?

We can't really give you good advice unless you tell us much more about your problem, including showing us Stata commands and output.

[Abdullah Selim]

hi
I have a similar issue. I use ivprobit and estimation takes 30 seconds or so. However, when I try margins I cannot get the results for hours. I have 1 million observations.
Do you have any advice for this
thanks

[DANIELLE MCPHERSON]

I am having this issue too and was wondering if I needed to use an alternative to margins. I have approx. 6.9M observations and is running margins after the IV Probit regression. I am typing margins, dydx(*)...which has been running indefinitely and hasn't give me an answer. I event went as far as trying to run margins, dydx(EXT_CRIMEV) which is the endogenous variable only and I still can't get results...its still processing hours after the fact. Is there another way to interpret the results of an IVPROBIT regression besides margins?

[Joro Kolev]

Everybody is describing the problem in abstract terms, nobody shows an example data where the problem can be replicated, only one person showed the command she actually typed (and what she showed is the maximum likelihood version of -ivprobit-)... This all seems like a test of the psychic abilities of the people on this board...

Have you tried to see whether the problem still arises if you do -ivprobit, twostep- and then margins?