Pooled probit,random effect probit and fixed effects

[Elias Aloyce]

I am trying to run a pooled probit, random effects probit and fixed effects and producing them on the same table. Anyone who knows how can I go with Code on this please help.

[Jared Greathouse]

// Note that I use esttab

// Download estout from ssc


qui eststo m1: probit y x

qui xtset idvar timevar

qui eststo m2: xtprobit y x i.idvar // note that there's no Stata implementation for this

// see h xtprobit

qui eststo m3: xtprobit y x, re

cls


esttab m1 m2 m3, compress

[Elias Aloyce]

Thank you

[Elias Aloyce]

Thank you , but on qui estate m2 , the stata doesn’t respond, what could be a problem

[Elias Aloyce]

Correction Is qui essto m2

[Jared Greathouse]

What does Stata say when you do

qui eststo m2: xtprobit y x i.idvar

[Elias Aloyce]

After putting industry dummy and time dummies, after storing
qui eststo m2: xtprobit y x i.idvar, the stata does not respond to any command that I want to enter.

[Jared Greathouse]

It isn't that it isn't responding, it's that me running it quietly obscured some problems. When I took the quietly prefix away, I started seeing that numerous groups in the below example were perfectly predicting (failure, I think). So here's the fix for you. Without seeing your data, you likely want to use a logit model.

webuse union, clear
xtlogit union age grade i.not_smsa i.year, fe

It estimates the fixed effects quickly (quicker than probit which lacks this option) as well as, for all practical purposes, gives you the same results as probit will. If using an xtlogit model doesn't do what you want, then post your example data as well as the output you received.

[Elias Aloyce]

Can I used this
xtlogit union age grade i.not_smsa i.year, fe

for estimation of also Pooled Probit and random effect probit????

[Jared Greathouse]

No. Pooled analysis is panel data without unit or time fixed effects. m1 i estimated was pooled, m2 was fixed effects, and m3 was random effects.

You can't used FE with probit, so assuming you'd want to be consistent, I'd just advise you to use only logit analysis since it does all of what you want without needing to switch between logit and probit.

[Clyde Schechter]

You can't used FE with probit,

You can't do that with official Stata commands. There are, however, some user-written programs that estimate fixed-effects probit models. As I almost always use logit rather than probit modeling, I have no experience with any of those user-written commands and I have no idea whether they are any good or not. But if anybody is interested in trying them out, -search fixed effects probit- will provide links.

[Jared Greathouse]

As someone who also doesn't use probit or logit very much, I should've added the disclaimer that it's perfectly possible that someone's made user written commands to use FE with probit.

In fact, a possible resource for this might be Hilbe and Hardin's book on GLMs and their extensions, but I'm unsure of this.

[Elias Aloyce]

Thank you, I still have challenge when I include industry dummy and time dummie in (fe )and (re) by this code, it store but i can not move on to reproduce table of pooled probit,random effect probit and fixed effects, Is there other alternative or there is a problem somewhere
Code:
// Note that I use esttab // Download estout from ssc qui eststo m1: probit y x qui xtset idvar timevar qui eststo m2: xtprobit y x i.idvar // note that there's no Stata implementation for this // see h xtprobit qui eststo m3: xtprobit y x, re cls esttab m1 m2 m3, compress

[Jared Greathouse]

Why can't you reproduce them? What error code was issued or what did Stata do that you didn't want it to do?

[Elias Aloyce]

In the random effect for probit code
qui eststo m3: xtprobit y x, re Am I not supposed to put time dummy like
qui eststo m3: xtprobit y x i.year re??