Problem with command goprobit

[Jean Jacques]

Hi, I'm trying to use the command goprobit and I'm facing two problems for which I would appreciate some help:

The first is that when I do

goprobit score i.quarter_year i.country_code

I receive factor-variable and time-series operators not allowed. I think this is because the command was written before Stata started to support factor variable notation. As I won't rewrite the command I guess the only (or at least the best) option I have is to create dummy variables for "quarter_year" and "country_code". Is that correct? The problem ls that "country_code" has around 200 different values. Moreover, I may need to include variables with millions of different values. What would be the best way of approaching this?

Second, if I run

oprobit score nights if country_code == 4

it takes just a couple of seconds to show the output (I have 400 obs) while if I run

goprobit score nights if country_code == 4

it keeps running for ages. I started to write this message an hour ago, while I ran the command and it still didn't output any result. Any thoughts about this?

Thanks a lot!

[George Ford]

oprobit and goprobit are different models (the latter more complex), so that likely explains the time. As a user-written command, it may not be as efficient as Stata commands in general.

have you tried gologit2?

[Richard Williams]

Yes, use gologit2 with link(p) instead. For more info, see

https://www3.nd.edu/~rwilliam/gologit2/index.html

i would also add the autofit option or constrain coefficients some other way. Unconstrained gologit models aren’t worth much in my opinion. If an unconstrained. model is the best you can do, you might as well use the better known mlogit.