Interaction plot ordered probit

[Elko Klijn]

Dear all,
I have a few questions regarding my research. I ran a first stage ordered probit regression in which I saved the residuals. These residuals are brought into my second stage ordered probit regression.

a. Which STATA code exists to make sure that the error terms in the first stage and second stage are not correlated? Or is there a better model (I use survey data, so ivprobit or xtabond2 seems not really appropriate.)

b. How do I obtain the marginal effects as well as its significant for an ordered probit model? i.e. what is the STATA code?

c. How do I make interaction plots for ordered probit regression analyses in stata? I can do them for probit, but am not sure how to do it for ordered probit.

Regards and have a nice weekend!
Elko

[Stephen Jenkins]

(a) You have ordered probit models at both steps. Be sure that your 2-step approach provides estimates with appropriate properties. (Have you checked the literature about this? I am unaware of relevant papers, but I am a non-specialist.) In the meantime, you could look at your data using a model for either (1) ordered probit selection and continuous outcome -- see http://www.stata-journal.com/sjpdf.h...iclenum=st0123, or (2) a binary selection and ordered probit outcome -- see help heckoprobit. Otherwise, you might have to write your own code to maximize the relevant likelihood for your general model. If I were you, I'd probably use (2) to take a first look.
(b) margins is your friend, at least for (2). Be clear precisely what marginal effect you are seeking in this context.
(c) what is an "interaction plot"? margins combined with marginsplot is probably what you're looking for. Be sure to specify your interactions appropriate using factor variable notation ("i." and ".c" prefixes, etc.)