Re: st: xtreg vs reg w/ dummies

[David Harless <dwharles@vcu.edu>]

Scott Hankins wrote:
********************************************************************************
I have estimated what I belive to be the same model from an econometric point of
view but Stata doesn't give me the same results.

The 1st uses "xtreg" to estimate a fixed effects model (along with "xi" for the
year effects). The 2nd uses "regress" and "xi" to generate dummy variables for
the individual fixed effects as well as year effects.

The only difference between the two routines is the coefficient on pitt is
"dropped" when I use xtreg. I thought this was a multi-colinearity problem until
I ran the regression with "regress" and dummies. It may still be a
multi-colinearity problem as I do not believe the estimated coefficient is correct.

Why does one command estimate this coefficient when the other won't? I assume
there is something subtle (or not so subtle) going on behind the scenes with Stata.

thanks

scott

**************  1st model  *****************
xi: xtreg csec i.year pitt breech previous medicaid if att_issue>=1994 ,fe
i(new_attid)
i.year            _Iyear_1994-2004    (naturally coded; _Iyear_1994 omitted)

Fixed-effects (within) regression               Number of obs      =      1002
Group variable (i): new_attid                   Number of groups   =       353

R-sq:  within  = 0.3679                         Obs per group: min =         1
       between = 0.0303                                        avg =       2.8
       overall = 0.1626                                        max =        10

                                                F(14,635)          =     26.40
corr(u_i, Xb)  = -0.0949                        Prob > F           =    0.0000

------------------------------------------------------------------------------
        csec |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
 _Iyear_1995 |  -.0457742   .0561129    -0.82   0.415    -.1559636    .0644151
 _Iyear_1996 |  -.0237076   .0557849    -0.42   0.671    -.1332528    .0858377
 _Iyear_1997 |  -.0281403   .0561732    -0.50   0.617     -.138448    .0821675
 _Iyear_1998 |  -.0322946   .0566514    -0.57   0.569    -.1435413    .0789521
 _Iyear_1999 |  -.0382227   .0567723    -0.67   0.501    -.1497069    .0732615
 _Iyear_2000 |  -.0388285   .0567053    -0.68   0.494    -.1501811    .0725242
 _Iyear_2001 |  -.0312434   .0567966    -0.55   0.582    -.1427752    .0802885
 _Iyear_2002 |    .027376   .0568078     0.48   0.630    -.0841779    .1389299
 _Iyear_2003 |  -.0026329   .0570527    -0.05   0.963    -.1146676    .1094018
 _Iyear_2004 |   .0051664   .0572499     0.09   0.928    -.1072556    .1175883
        pitt |  (dropped)
      breech |   .7840754   .0496142    15.80   0.000     .6866476    .8815032
previous_c~c |   .3019584   .0425659     7.09   0.000     .2183714    .3855454
    medicaid |    -.03466   .0199792    -1.73   0.083    -.0738933    .0045733
       _cons |   .1505913   .0561206     2.68   0.007      .040387    .2607956
-------------+----------------------------------------------------------------
     sigma_u |  .23549195
     sigma_e |  .09962159
         rho |  .84820555   (fraction of variance due to u_i)
------------------------------------------------------------------------------
F test that all u_i=0:     F(352, 635) =     7.22            Prob > F = 0.0000
...

******************************************************************************

This case is curious because pitt is dropped in the -xtreg- regression, but the 
F-statistic numerator degrees of freedom read as if it is not dropped:
F(14,635)          =     26.40

With the dropped variable the numerator degrees of freedom should be 13, not 14. Why would 
14 be listed in this circumstance?

Dave Harless
*
*   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/