st: Change in coefficient sign with Pooled Fractional Probit

[Rijo John <rmjohn@gmail.com>]

P.S. I am reposting it hoping to get an answer.


Hi Statalist,

I am trying to estimate a Pooled Fractional Probit (PFP) model for my
panel data which has 51 panels and 10 years. My dependent variable is
a fraction strictly between zero and 1 (no zeros or ones). I am using
the Fractional Probit model developed by Papke and Wooldridge.

here is what I am doing.

Specification 1)
glm Y X1 X2 X3 X4 X5 aX1 aX2 aX3 aX4 aX5, link(probit) fam(bin)
cluster(Clustvar)

where aX's are the time averages of all the X's. This is how PFP is
implimented I guess.

Specification 2)
glm Y X1 X2 X3 X4 X5 Yeardummies, link(probit) fam(bin) robust

This is the implementation of fractional logit/probit model in cross
sectional data. I just used this here by adding year dummies.

The variable of interest to me is X1 and the rest are all control
variables. I am getting expected and significant results for most of
the coefficients.

However between the above 2 specificaitons, the cofficient sign for
the variable X1 changes from specification 1 to 2 both being
significant. A simple linear regression also returns the same sign as
the one in specification 2. How can I decide which specification to go
with?

Any help will be appreciated.

Thanks,
Rijo.
*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/