Mark Schaffer wrote:
> The R2s depend on the coefficient estimates but not on the estimated
> variance-covariance matrix. It looks like you are using xtreg,fe and areg
> to estimate the same model, and so the coefficients reported by the two
> estimators should be the same. If they are, then you can simply use the
> R2s reported by xtreg,fe.
Whilst I was perfecting my LSDV models, I assumed (both from Mark's and
Kit Baum's comments) that the coefficients from them would indeed be the
same as FE. Not from my data. An example:
. areg edconch lagconch laglabch lagldmch if edmarker==1, absorb(region)
Number of obs =
2187
F( 3, 2173) =
89.91
Prob > F =
0.0000
R-squared =
0.1379
Adj R-squared =
0.1328
Root MSE =
10.752
------------------------------------------------------------------------------
edconch | Coef. Std. Err. t P>|t| [95% Conf.
Interval]
-------------+----------------------------------------------------------------
lagconch | -1.145137 .1207652 -9.48 0.000 -1.381964
-.9083096
laglabch | -.7120315 .1372847 -5.19 0.000 -.9812545
-.4428084
lagldmch | -.8160474 .1272037 -6.42 0.000 -1.065501
-.5665938
_cons | -12.85246 .2432762 -52.83 0.000 -13.32953
-12.37538
-------------+----------------------------------------------------------------
region | F(10, 2173) = 7.705 0.000 (11
categories)
. xtreg edconch lagconch laglabch lagldmch region if edmarker==1, fe
Fixed-effects (within) regression Number of obs =
2187
Group variable (i): pano Number of groups =
304
R-sq: within = 0.1970 Obs per group: min =
1
between = 0.0000 avg =
7.2
overall = 0.1049 max =
11
F(3,1880) =
153.72
corr(u_i, Xb) = -0.0409 Prob > F =
0.0000
------------------------------------------------------------------------------
edconch | Coef. Std. Err. t P>|t| [95% Conf.
Interval]
-------------+----------------------------------------------------------------
lagconch | -1.351399 .1029873 -13.12 0.000 -1.553381
-1.149418
laglabch | -.9368757 .1180488 -7.94 0.000 -1.168396
-.7053552
lagldmch | -1.048965 .1090272 -9.62 0.000 -1.262792
-.8351383
region | (dropped)
_cons | -12.95549 .1934802 -66.96 0.000 -13.33495
-12.57603
-------------+----------------------------------------------------------------
sigma_u | 8.0146543
sigma_e | 8.4675555
rho | .47254255 (fraction of variance due to u_i)
------------------------------------------------------------------------------
F test that all u_i=0: F(303, 1880) = 5.76 Prob > F =
0.0000
Exactly the same results occur if I use individual dummy variables in
-xtreg, fe-.
As they might say in the UK's "Private Eye" magazine, is any difference
between them a coincidence? I think we should be told.
CLIVE NICHOLAS |t: 0(044)191 222 5969
Politics |e: [email protected]
Newcastle University |http://www.ncl.ac.uk/geps
*
* 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/
