Statalist


[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: st: Inverse Mills in clustered (multilevel) cross-sectional panel data


From   Austin Nichols <austinnichols@gmail.com>
To   statalist@hsphsun2.harvard.edu
Subject   Re: st: Inverse Mills in clustered (multilevel) cross-sectional panel data
Date   Mon, 7 Sep 2009 16:51:59 -0400

Erkko Autio<erkko.autio@imperial.ac.uk> :
A lot of techniques assuming normal distribution of error terms will
give biased estimates, but that doesn't stop folks from using
them--however, if you want an IMR you are asking for normal densities
and cumulative probabilities. Plus maybe your theoretical model
demands interactions between various groups of variables.  Probably
you were looking for a dynamic selection model, none of which is
easily estimated in Stata afaik; see e.g. 8.6.2 in
http://www.stata.com/bookstore/aopd.html
or 17.7.2 in
http://www.stata.com/bookstore/cspd.html
or references in this thread:
http://www.stata.com/statalist/archive/2005-06/msg00456.html
and see what kind of model you need to program for your case.  Or
switch to panel IV using -xtivreg- if you can--sounds like you've got
the sample size for it!

On Mon, Sep 7, 2009 at 11:41 AM, Erkko Autio<erkko.autio@imperial.ac.uk> wrote:
> I am trying to use inverse Mill's ratio to control for self-selection in
> clustered data.
>
> My problem is that I have multilevel data, clustered by year and country.
> The basic dataset comprises interviews with some 900 000 individuals in
> nearly 60 countries over 10 years.
>
> Specifically, I am assuming that self-selection of individuals into a given
> economic activity is influenced by both individual-level variables (such as
> age, gender, attitudes), as well as country-level variables (e.g.,
> taxation). As behaviours may be conditioned by context (the same individual
> would behave differently under different taxation regimes, for example), the
> error terms will no longer be normally distributed, and techniques assuming
> normal distribution of error terms will thus give biased estimates.
>
> My selection equation would thus consist of both individual and
> country-level variables, as would my regression equation.
>
> Hence my problem. Normally, inverse Mill's ratio is computed using probit
> models. However, Stata has no multi-level probit. It does have xtmelogit,
> which can be used for multi-level data. If I use xtmelogit instead of, say,
> xtprobit to compute inverse Mill's ratio? Will xtmelogit result in biased
> estimates?
>
> Alternatively, is there a way to do a probit in cross-sectional panel data
> (i.e., one-off interviews of randomly sampled individuals in lots of
> countries in many consequtive years, individuals not followed over time)?
>
> Thank you for any suggestions.
>
> Erkko Autio
> -
*
*   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/



© Copyright 1996–2014 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index