RE : st: Fixed-effects models and coefficient for a constant variable

["PETITT Barbara" <barbara.petitt@cote-azur.cci.fr>]

Thank you very much, Vera, I downloaded your article and the stata ado, as your suggestion is a direct solution to my problem. However, I ran xtfevd on my data, but the results I am getting are very strange.

I have panel data for 185 firms over three years. My DV is the firm's return on assets. I have three IV: group (dummy, time invariant), diversification (time-varying) and the interaction group*diversification (time-varying. I also have some control variables: age (time-varying), size (time-varying), two year dummies and six industry dummies (time invariant). So, I defined:

xtfevd roa1 groupc tdec interactionc age size y88 y89 bas cap cdr cns ftb tex, invariant(groupc bas cap cdr cns ftb tex) robust

The results are:

panel fixed effects regression with vector decomposition, robust standard errors

degrees of freedom fevd    =      357           number of obs       =      555
mean squared error         = .0015608           F( 15, 357)         = 1.04e+09
root mean squared error    =  .039507           Prob > F            =        0
Residual Sum of Squares    = .8662438           R-squared           = .7425136
Total Sum of Squares       = 3.364232           adj. R-squared      = .6004273
Estimation Sum of Squares  = 2.497988

------------------------------------------------------------------------------
             |             fevd_robust
        roa1 |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
      groupc |   .0516898   .0010607    48.73   0.000     .0496038    .0537759
        tdec |  -.0675595   .0045422   -14.87   0.000    -.0764923   -.0586266
interactionc |   .1571639   .0009579   164.07   0.000     .1552801    .1590477
         age |  -.0127348   .0049431    -2.58   0.010    -.0224561   -.0030135
        size |   .0941073          .        .       .            .           .
         y88 |   6.30e-11   .0010791     0.00   1.000    -.0021222    .0021222
         y89 |  -.0194528   .0000104 -1872.31   0.000    -.0194733   -.0194324
         bas |   .0112112   .0000198   566.14   0.000     .0111722    .0112501
         cap |   -.005021    .000019  -264.95   0.000    -.0050583   -.0049837
         cdr |  -.0952531   .0000187 -5107.38   0.000    -.0952898   -.0952164
         cns |  -.0215131   .0000257  -836.08   0.000    -.0215637   -.0214625
         ftb |   .2322076    .000028  8284.75   0.000     .2321524    .2322627
         tex |   .1149538   .0000216  5329.38   0.000     .1149114    .1149962
         eta |          1   .0000292        .   0.000     .9999426    1.000057
       _cons |   .1745182   .0000163        .   0.000     .1744861    .1745502
------------------------------------------------------------------------------

The problem seems to be in the robust standard errors. Do you have any idea why? Thanks.

-----Message d'origine-----
De : owner-statalist@hsphsun2.harvard.edu [mailto:owner-statalist@hsphsun2.harvard.edu] De la part de Vera Troeger
Envoy� : mercredi 11 octobre 2006 09:39
� : statalist@hsphsun2.harvard.edu
Objet : Re: st: Fixed-effects models and coefficient for a constant variable


Hello,

I don't want to advertise myself here but there is an article supposed to be in a special inssue in Political Analysis (on panel data) dealing with exactly this problem. We propose a procedure called fevd (fixed effects vector decomposition) that produces coefficients for time invariant variables in a fixed effects framework. There is also a stata ado called xtfevd.ado (a beta version) and help-file (xtfevd.hlp) to estimate the proposed procedure in stata. the paper and the ado can be downloaded from:

http://www.polsci.org/pluemper/xtfevd.htm

there is of course the Hausman-Taylor method and its variants but this is much less efficient as shown in the paper.

Vera


PETITT Barbara schrieb:
> Hello,
>
> I have data for 185 firms over three years (therefore 555 
> observations). One of my dependent variables is a dummy variable that 
> is constant over the three years. I cannot use a random-effects model, 
> as the Hausman test is rejected. But when I use a fixed-effects model, 
> I cannot estimate the coefficient for my dummy variable, which is 
> problematic, as it is one of the most important variables of my model. 
> Is there any fix to this problem? How is it possible to estimate the 
> coefficient for a variable that does not vary over time? Thank you.
>
> *
> *   For searches and help try:
> *   http://www.stata.com/support/faqs/res/findit.html
> *   http://www.stata.com/support/statalist/faq
> *   http://www.ats.ucla.edu/stat/stata/
>
>
>

-- 
Vera E. Troeger
Lecturer in Political Science
Government Department
University of Essex
Wivenhoe Park
Colchester CO4 3SQ
UK

phone: +44 (0)1206 872509
email: vtroe@essex.ac.uk


*
*   For searches and help try:
*   http://www.stata.com/support/faqs/res/findit.html
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/

*
*   For searches and help try:
*   http://www.stata.com/support/faqs/res/findit.html
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/