R squared with Fixed Effects. xtreg or Areg. Within (xtreg) or Adj (areg). Which to use?

[Victor Stuart-Smith]

I know there have been similar questions with answers yielding different results.

I am running a fixed effects model with LCR as the dependent variable.
The first adjusted R squared states 0.687 for areg. The second is 0.239 for xtreg.
Which should I use?

Thanks
Any contribution is greaty appreciated.

areg lcr lnta nim roaa assetg llrtgl depositg lev gdpg LCRpolicy if lcrlarge==1, absorb(id) rob
> ust

Linear regression, absorbing indicators         Number of obs     =        464
Absorbed variable: id                           No. of categories =        116
                                                F(   9,    339)   =       5.43
                                                Prob > F          =     0.0000
                                                R-squared         =     0.7710
                                                Adj R-squared     =     0.6872
                                                Root MSE          =     0.3797

------------------------------------------------------------------------------
             |               Robust
         lcr |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
        lnta |  -.7698172    .282831    -2.72   0.007    -1.326142   -.2134925
         nim |  -24.20668   12.03465    -2.01   0.045    -47.87867   -.5346989
        roaa |   5.610156   8.106008     0.69   0.489    -10.33425    21.55456
      assetg |  -.0094219   .2088364    -0.05   0.964    -.4202003    .4013565
      llrtgl |  -.0354481   1.150923    -0.03   0.975    -2.299297    2.228401
    depositg |  -.3610365   .1272425    -2.84   0.005    -.6113207   -.1107523
         lev |  -.0903418   1.275511    -0.07   0.944    -2.599255    2.418571
        gdpg |   1.219889   1.909962     0.64   0.523    -2.536982    4.976759
   LCRpolicy |   .7254302   .1662699     4.36   0.000     .3983796    1.052481
       _cons |   14.79609   4.933593     3.00   0.003     5.091784     24.5004
------------------------------------------------------------------------------

xtreg lcr lnta nim roaa assetg llrtgl depositg lev gdpg LCRpolicy if lcrlarge==1, fe robust

Fixed-effects (within) regression               Number of obs     =        464
Group variable: id                              Number of groups  =        116

R-sq:                                           Obs per group:
     within  = 0.2399                                         min =          4
     between = 0.0580                                         avg =        4.0
     overall = 0.0642                                         max =          4

                                                F(9,115)          =       6.41
corr(u_i, Xb)  = -0.7731                        Prob > F          =     0.0000

                                   (Std. Err. adjusted for 116 clusters in id)
------------------------------------------------------------------------------
             |               Robust
         lcr |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
        lnta |  -.7698172   .3059003    -2.52   0.013    -1.375747   -.1638876
         nim |  -24.20668   11.80274    -2.05   0.043    -47.58563   -.8277353
        roaa |   5.610156    7.56064     0.74   0.460    -9.366018    20.58633
      assetg |  -.0094219   .1908474    -0.05   0.961     -.387454    .3686101
      llrtgl |  -.0354481   1.087143    -0.03   0.974    -2.188868    2.117972
    depositg |  -.3610365   .1177785    -3.07   0.003    -.5943331   -.1277399
         lev |  -.0903418   1.000216    -0.09   0.928    -2.071578    1.890894
        gdpg |   1.219889   1.263876     0.97   0.336    -1.283607    3.723385
   LCRpolicy |   .7254302   .1858142     3.90   0.000     .3573681    1.093492
       _cons |   14.79609   5.398516     2.74   0.007     4.102672    25.48952
-------------+----------------------------------------------------------------
     sigma_u |  .90287735
     sigma_e |   .3797107
         rho |   .8497133   (fraction of variance due to u_i)
------------------------------------------------------------------------------

[FernandoRios]

I think you need to answer first, what do you want to use the R2 for.
The easiest approach would be to report the Standard R2 and the within R2. with a foot note on the adj r2.

[Victor Stuart-Smith]

When you say standard R2, do you mean the areg R2 or the overall xtreg R2?

[FernandoRios]

Standard R2 (not adjusted) you get from a reg. It corresponds to R2=1-SSR/SST
Then within R2 follows the same principle, but SSR and SST are estimated after partialling out the fixed effects.

[Victor Stuart-Smith]

so sorry, you said: 'you get from a reg' - is that 'a reg i.e. -reg-' or '-areg-' Thanks

[Victor Stuart-Smith]

I shall use areg?

[Carlo Lazzaro]

Victor:
whereas -areg- and -regress- both give back the same -Rsq- and -adj-Rsq-, for -xtreg- the interesting -Rsq- is the -within-one.
You can access -adj_Rsq- after -xtreg,fe- typing:

di e(r2_a)

.

However, I don't think it has the same meaning as in -regress- or -areg-.