The F or chi2 model statistic has been reported as missing

[Clara Joan Joachim]

Hello, I have problem with my xtreg fe with robust. I tried hausman and fe shows the best model comparing to re. I put robust option because there seem to be heteroskedastic problem. However, there is my F result shows only dot.

xtreg cdi temp rain fcy nfcy cl lnfer lngni lnexp irrig, fe robust

Fixed-effects (within) regression               Number of obs     =        264
Group variable: ID                              Number of groups  =          8

R-squared:                                      Obs per group:
     Within  = 0.4251                                         min =         33
     Between = 0.0033                                         avg =       33.0
     Overall = 0.0464                                         max =         33

                                                F(7,7)            =          .
corr(u_i, Xb) = -0.1725                         Prob > F          =          .

                                     (Std. err. adjusted for 8 clusters in ID)
------------------------------------------------------------------------------
             |               Robust
         cdi | Coefficient  std. err.      t    P>|t|     [95% conf. interval]
-------------+----------------------------------------------------------------
        temp |   .0016434   .0160311     0.10   0.921    -.0362642     .039551
        rain |  -.0000255   .0000214    -1.19   0.272     -.000076     .000025
         fcy |   .0009383   .0014402     0.65   0.535    -.0024671    .0043437
        nfcy |   .0008511   .0005483     1.55   0.165    -.0004454    .0021476
          cl |   .0005852     .00182     0.32   0.757    -.0037184    .0048888
       lnfer |  -.0295356   .0308141    -0.96   0.370    -.1023993    .0433282
       lngni |   .0086028    .012025     0.72   0.498    -.0198319    .0370375
       lnexp |  -.3012621    .077288    -3.90   0.006    -.4840192   -.1185049
       irrig |   .0024558   .0024765     0.99   0.354    -.0034002    .0083117
       _cons |   .6024456   .6345898     0.95   0.374    -.8981208    2.103012
-------------+----------------------------------------------------------------
     sigma_u |  .15618413
     sigma_e |  .04816762
         rho |  .91314837   (fraction of variance due to u_i)
-----------------------------------------------------------------------

Also, I have read that if T > N, I should use xtgls command. However, I am kind of confuse with the usage of xtgls (as some used it for remedy version of Random Effect with heteroskedasticity and autocorrelation). Anyone can help to explain this? Thank you v.much

[Clyde Schechter]

First, understand that with -xtreg, fe-, vce(robust) is understood as vce(cluster ID). You have only 8 clusters, so you have only 7 residual degrees of freedom. But you have 10 predictors. You cannot test 10 predictors simultaneously when there are only 7 residual degrees of freedom. The model F statistic is a test of the hypothesis that all of the coefficients in the model are zero. Usually this hypothesis is of no interest at all, so the inability to obtain this statistic is not a problem.

In any case, with only 8 clusters, you should not be using robust standard errors. They require larger numbers of clusters to be valid. A side "benefit" is that by reverting to non-clustered standard errors, your residual degrees of freedom will depend on the number of observations, not the number of clusters. So you will get that F-statistic back, although, as I said, there is probably no use for it anyway.