[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

Re: st: selection and clustered s.e.

From   "Stas Kolenikov" <>
Subject   Re: st: selection and clustered s.e.
Date   Tue, 14 Aug 2007 07:58:47 -0500

You would need a Heckman-style model if your zeros are created in a
different way from non-zeros. May be what you need is a Poisson model
of the kind.

Note that in general the likelihood modeling of the flows data is
tricky, as those are dyadic data, not panel data (i.e., they will have
the sender effect and the receiver effect, and they won't be the same
for a given unit), and their likelihoods are pretty weird. Look up the
works of Peter Hoff from University of Washington, he had a number of
papers, and I think he was writing up a book on this topic, too.

On 8/13/07, Rosa Sanchis-Guarner <> wrote:
> Hi!
> I am running a migration equations with data for 15 years. I have 2162
> flows (which I consider the individuals), for every year. I did a FE
> estimation, but as I am highly interested on the time-invariant
> variables, I also estimated using OLS and clustering the s.e. by
> individual. My problem is that I have some zero flows, and the
> equation is estimated in logs, so when I do the OLS I just drop them.
> I don't have many, but one option is to use a Heckman model to control
> for the zeros in the first step and to explain the flows in the
> second.
> How can I combine the panel structure with the heckman model?
> I have two options:
> - do a FE estimation (with the desadvantage that I lose the info on
> the time-invariant variables, very important in my model)
> - do OLS with clustered s.e.
> I have been thinking that actually Heckman two-step just includes the
> inverse mill-ratio in the second step, so I could actually calculate
> it and include it in a OLS with cluster s.e. Does this make sense? How
> would I do it in the FE estimation?
> Any suggestions accepted!
> Thank you very much!
> --
> Rosa Sanchis
> University of Barcelona
>  "If we knew what it was we were doing, it would not be called
> research, would it?" - Albert Einstein
> *
> *   For searches and help try:
> *
> *
> *

Stas Kolenikov, also found at

Small print: Please do not reply to my Gmail address as I don't check
it regularly.
*   For searches and help try:

© Copyright 1996–2017 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index