Message: "note: multiple positive outcomes within groups encountered" and different coefficients

[Damian Rozenberg]

Dear all,

I ran into the message "note: multiple positive outcomes within groups encountered" when running a xtprobit model. I estemate multiple models with each time more independent variables included. So the first model is defined as follows:

xtlogit Initialreturndummy lnofferprice lnDealsize Greenshoepercentage OffertoOpendummy DistanceIPOandfilingdatein Bookrunnerdummy Rankingdummy Segmentdummy, fe

.

If I run the second model I get the aforementioned message and the coefficients change to very large values, e.g. from 0.0744 to -4.33490. The second model is defined as:

xtlogit Initialreturndummy lnofferprice lnDealsize Greenshoepercentage OffertoOpendummy DistanceIPOandfilingdatein Bookrunnerdummy Rankingdummy Segmentdummy lnTotalAssets PEVCdummy Phoneticdummy lnSharesoffered Techdummy CompanyAge, fe

A few observations (4/5) have been omitted as well. Does anyone have an idea how I can fix this?

[Carlo Lazzaro]

Damian:
my guess is that there's no enough outcome variation within panels to make the MLE estimator converge.

[Damian Rozenberg]

Hi Carlo Lazzaro ,

Thanks for your reply. Could there be something wrong with my data?
What is now the best thing to do? I have to estimate a binary model with fixed effects per country and if I'm not mistaken a probit model does not allow fixed effects. Therefore, if have to use a logit model since a LPM has some downsides.

[Carlo Lazzaro]

Damian:
I thought that -xtprobit- read -xtlogit-.
That said:
1) -xtlogit- allows conditional fixed effect, because of incidental parameters bias (https://www.sciencedirect.com/scienc...04407699000445);
2) I'd take a look at your model and double-check whether its specification is correct.

[Damian Rozenberg]

Hi Carlo,

Thanks for your help! I think I figured it out now.