Re: st: Estimating Wooldridge's CML for Dynamic Probit model: On right path?

[Jeffrey Wooldridge <jmwooldridge60@gmail.com>]

in my paper. Instead, I proposed the full blown Chamberlain device
along with the intitial condition. I think it can matter, and
simulation evidence seems to suggest that for certain covariate
processes using the time averages can lead to serious bias. Whether it
is a problem in practice, I don't know. I've found that for static
models using the simplification rarely changes anything, especially
when the focus is on partial effects. But once the initial condition
is controlled for, this very likely changes the partial correlations
between the heterogeneity and the time path of the covariates. Even
though it burns up degree of freedom it is worth putting the entire
history of all covariates on each data line. These are then averaged
out when computing the average marginal effects. My talk at the Stata
meetings in Chicago in 2011 has an example -- cleaned up thanks to
Austin Nichols.



On Mon, Feb 11, 2013 at 2:02 PM, Nicholas Dadzie
<dadzie.4@buckeyemail.osu.edu> wrote:
> Hello Statalisters,
>
> I am estimating a dynamic probit model to assess household crop choices. Am interested in the dynamics since previous cultivation of a crop might affect the choice of a crop in current period. From the reading i have done, Wooldridge's estimator is suitable
>  to estimate this model (Though there is Orme and Heckman's (1981) procedures). This procedure requires generating new variables (from the 1st period- initial values of the dependent variable and time-averaged versions of the household variables ) and including
>  these in Stata's xtprobit. the inclusion of the time-averaged household variables is consistent with Mundlak (1978) and Chamberlain's (1984) suggestion in order to allow correlation between the unobserved effects and the household variables.
>
> I have gone ahead to do this but want feedback on whether the code is correctly specified.
> Any comments, thoughts or suggestions are welcome.
>
> bysort u_hhid: egen hhvar1_ = mean(hhvar1)    //   hhvar are vars in first stage to be included in the main regression
> bysort u_hhid: egen hhvar2_ = mean(hhvar2)
> bysort u_hhid: egen hhvar3_ = mean(hhvar3)
>
>
> gen yvar_r1 = yvar                         // yvar - dependent var (binary var)  and round is the time var
> replace yvar_r1 = . if round != 1
> bysort u_hhid: egen yvar_1 = min(yvar_r1)
>
>
> xtprobit yvar yvar_lag hhvar1 hhvar2 hhvar3 yvar_1 hhvar1_ hhvar2_ hhvar3_
>
>
> [this code yields decent but not the best results (weakly signifcant results in parameters of interest) ]
>
> Thanks in advance,
>
> Nick Dadzie
> Ohio State Uni
>
>
>
>
>
>
>
>
>
>
> *
> *   For searches and help try:
> *   http://www.stata.com/help.cgi?search
> *   http://www.stata.com/support/faqs/resources/statalist-faq/
> *   http://www.ats.ucla.edu/stat/stata/

*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/