RE: st: RE: xtabond2 estimation and observations used

[DE SOUZA Eric <eric.de_souza@coleurope.eu>]

The purpose of asking you to include all four time dummies was to see which ones would be dropped on account of collinearity.  Also it was not clear to me what exactly you had done in your original regression.

I don't know how Stata chooses which variables to drop when it encounters collinearity. I have not been able to deduce the rule if there is one. 
But I have noticed that when running -xtabond2- with time dummies the two extremes are dropped, such as here:
. xtabond2 n L.n L(0/1).(w k) yr*, robust small noleveleq gmmstyle(L.n w k) ivstyle(yr*)
yr1976 dropped due to collinearity
yr1984 dropped due to collinearity

Eric de Souza 
College of Europe 
Brugge (Bruges), Belgium 
http://www.coleurope.eu

-----Original Message-----
From: owner-statalist@hsphsun2.harvard.edu [mailto:owner-statalist@hsphsun2.harvard.edu] On Behalf Of webgeeky
Sent: 18 August 2013 16:31
To: statalist@hsphsun2.harvard.edu
Subject: Re: st: RE: xtabond2 estimation and observations used

I created  four time dummies (e.g. t1 = 1 if the observation is for period 1), and entered t* in the xtabond2 model. I've also included a lagged dependent variable (LDV) in the model. The message is:

t1 dropped due to collinearity
t4 dropped due to collinearity

I noticed that t1 and t4 are also dropped when I use xtreg with a LDV.
However, when the LDV is excluded, only t4 is dropped.

So I guess the LDV is the cause of the "problem" here. A question is, why are the extremes time periods dropped with LDV included? It is more intuitive if t1 and t2 are dropped.

Webber Geelang

On Sun, Aug 18, 2013 at 8:56 PM, DE SOUZA Eric <eric.de_souza@coleurope.eu> wrote:
> Since we do not know how you have created your time dummies, nor what instruction you entered, it is difficult to answer.
>
> Create four time dummies, t1 to t4, and enter them into your command as t*. Then see what happens and report, if necessary.
>
>
> Eric de Souza
> College of Europe
> Brugge (Bruges), Belgium
> http://www.coleurope.eu
>
>
>
> -----Original Message-----
> From: owner-statalist@hsphsun2.harvard.edu 
> [mailto:owner-statalist@hsphsun2.harvard.edu] On Behalf Of webgeeky
> Sent: 17 August 2013 16:34
> To: statalist@hsphsun2.harvard.edu
> Subject: st: xtabond2 estimation and observations used
>
> I've a question regarding the mechanism of xtabond2. I've read Roodman
> (2009) in Stata Journal, but there are some aspects that I couldn't figure out.
>
> I have a balanced dataset that involves 4 time periods. I estimated a one-step difference GMM using xtabond2 with a lagged dependent variable (one-period lag) and the "nolevel" option. I included 3 time period dummies (t1, t2, t3) in the model.
>
> The output from the estimation shows that the number of observations per group (min/avg/max) is 2. The total number of observations is half that of my sample size. These figures are consistent with what the one-step difference GMM, where observations in the first two periods are dropped.
>
> What puzzle me is that there is an estimated coefficient for the 2nd time period (t2), while t1 is dropped from the model. I would like to know why t2 is being estimated even though the observations used in the estimation should be those in t3 and t4?
>
> Thanks!
> *
> *   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/

*
*   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/