The F model statistic is missing

[Miguel A. Duran]

Stata does not report the F model statistic of a regression where I am clustering on firms' loans (i.e., I use the command regress with the option cluster(loan)). According to the potential causes that Stata suggests, this problem seems to be due to the fact that there are regressors that, for some loans (i.e., the clustering unit), are nonzero for just one observation.

If this is the reason, I am not quite sure about how serious the problem is, that is, I am not quite sure about how safe is to stick to this type of clustering.

Will anyone please help me?

Although clustering at the loan level is my preferred option and makes more sense from a theoretical point of view, I could also cluster at the firm level. If I do this, the F statistic is reported and qualitative results do not seem to change substantially.

Thanks in advance.

Miguel A.

[Carlo Lazzaro]

Miguel:
unfortunately, you provide a handful of details which are far from being substantive for a helpful reply.
Just to start with: how many observations your dataset is composed of? How many clusters did you specify?
At the top of that, posting what you typed and what Stata gave you back (as per FAQ) is, as always, the best approach to increase the chances of getting useful advice.

[Miguel A. Duran]

I am sorry, Carlo, for not having been more specific. I copy below what I typed and Stata's output. If you needed further information, please let me know

 regress profit $var $v $va $co dquarter* dprimarysiccode*, cluster(loan)
note: dquarter14 omitted because of collinearity
note: dquarter22 omitted because of collinearity
note: dprimarysiccode6 omitted because of collinearity

Linear regression                                      Number of obs =    1320
                                                       F( 50,   328) =       .
                                                       Prob > F      =       .
                                                       R-squared     =  0.3189
                                                       Root MSE      =  2.5281

                                     (Std. Err. adjusted for 329 clusters in loan)
----------------------------------------------------------------------------------
                 |               Robust
          profit |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
-----------------+----------------------------------------------------------------
            var1 |   2.930124   .9270396     3.16   0.002     1.106431    4.753817
            var2 |   2.601272   .9786344     2.66   0.008     .6760806    4.526464
            var3 |  -.7260777   .5822838    -1.25   0.213     -1.87156    .4194043
            var4 |   .2361716   .3038575     0.78   0.438    -.3615837    .8339269
              v1 |  -.3084606   .2894533    -1.07   0.287    -.8778797    .2609585
              v2 |  -.3997945   .5286378    -0.76   0.450    -1.439743    .6401539
              v3 |    -.17686   .5711966    -0.31   0.757    -1.300531     .946811
              v4 |  -.2230113   .2356749    -0.95   0.345    -.6866363    .2406138
              v5 |   .0037224     .00605     0.62   0.539    -.0081793    .0156241
              v6 |  -.0323875   .8291812    -0.04   0.969    -1.663572    1.598797
              v7 |   .0016892   .0006226     2.71   0.007     .0004645    .0029139
             va1 |  -3.894708    1.05408    -3.69   0.000    -5.968318   -1.821098
             va2 |   .0415696   .0097497     4.26   0.000     .0223898    .0607494
             va3 |  -.0000769   .0001707    -0.45   0.653    -.0004127     .000259
             va4 |  -.2637492   .1790804    -1.47   0.142    -.6160403    .0885419
             va5 |   4.729815    2.07896     2.28   0.024     .6400368    8.819592
             va6 |   -.000897    .000319    -2.81   0.005    -.0015245   -.0002695
             co1 |  -.1050434   .2750097    -0.38   0.703    -.6460488     .435962
             co2 |  -.2048562   .1194779    -1.71   0.087    -.4398959    .0301836
             co3 |  -.7027358   .1715464    -4.10   0.000    -1.040206   -.3652658
             co4 |   .4666311   .3722271     1.25   0.211    -.2656226    1.198885
             co5 |   .0617743   .5128242     0.12   0.904    -.9470652    1.070614
       dquarter2 |   2.035726   .5081596     4.01   0.000     1.036063    3.035389
       dquarter3 |   2.299586   .4906962     4.69   0.000     1.334277    3.264895
       dquarter4 |   2.127063   .4638295     4.59   0.000     1.214607    3.039519
       dquarter5 |   2.269416   .4376293     5.19   0.000     1.408501     3.13033
       dquarter6 |    2.71931   .4312285     6.31   0.000     1.870987    3.567632
       dquarter7 |   1.749113   .4407634     3.97   0.000     .8820337    2.616193
       dquarter8 |   2.297448   .4301087     5.34   0.000     1.451328    3.143567
       dquarter9 |   1.641382   .3369831     4.87   0.000     .9784608    2.304303
      dquarter10 |   1.254673   .3136797     4.00   0.000     .6375946    1.871751
      dquarter11 |   1.415793   .3008758     4.71   0.000     .8239031    2.007683
      dquarter12 |   1.422007   .2692813     5.28   0.000     .8922704    1.951743
      dquarter13 |  -.2926386   .7360086    -0.40   0.691    -1.740531    1.155254
      dquarter14 |          0  (omitted)
      dquarter15 |  -.2757125   .2807557    -0.98   0.327    -.8280215    .2765966
      dquarter16 |  -.8899223   .7168913    -1.24   0.215    -2.300207    .5203625
      dquarter17 |     -.1407     .27029    -0.52   0.603    -.6724206    .3910207
      dquarter18 |  -.2458511   .2548749    -0.96   0.335    -.7472468    .2555446
      dquarter19 |  -.0109272   .2495392    -0.04   0.965    -.5018265     .479972
      dquarter20 |  -.0609755   .2270182    -0.27   0.788     -.507571    .3856199
      dquarter21 |  -.0935288   .2160449    -0.43   0.665    -.5185373    .3314796
      dquarter22 |          0  (omitted)
      dquarter23 |   .1068122   .2298405     0.46   0.642    -.3453353    .5589597
      dquarter24 |   .0068374   .3223044     0.02   0.983    -.6272071     .640882
      dquarter25 |   -.038453   .2614575    -0.15   0.883    -.5527982    .4758923
      dquarter26 |    .103531   .2759368     0.38   0.708    -.4392982    .6463603
dprimarysiccode2 |   .0405262   .6851097     0.06   0.953    -1.307237    1.388289
dprimarysiccode3 |  -.2166358   .6868839    -0.32   0.753    -1.567889    1.134618
dprimarysiccode4 |   .0046873   .7897625     0.01   0.995    -1.548951    1.558326
dprimarysiccode5 |  -.3472507   .6968732    -0.50   0.619    -1.718156    1.023654
dprimarysiccode6 |          0  (omitted)
dprimarysiccode7 |   .3899533   .6950502     0.56   0.575    -.9773654    1.757272
dprimarysiccode8 |   .0384064   .7028253     0.05   0.956    -1.344208     1.42102
           _cons |  -3620.177    850.051    -4.26   0.000    -5292.417   -1947.938
----------------------------------------------------------------------------------

[Carlo Lazzaro]

Miguel:
thanks for providing more details.
No wonder that the F test value does not creep up: perhaps you have really too many clusters vs predictors and/or some variables (possibly loans, as you previously stated) does not vary across observations..
I would also consider a more parsimoniuos model: I would find very hard to convey effectively to any audience the results of a 50-predictor OLS!

[Marcos Almeida]

Hello Miguel,

Carlo pointed out the main aspects.

Indeed, quite recently, he and a couple of Forum members presented very helpful comments so as to shed light on this matter (http://www.statalist.org/forums/foru...o-f-test-value).

In short, there are several features that may entail a lack of the "overall F-test", such as: too many regressors, too many clusters, ranking mismatch, unbalanced clusters, cluster with only one observation which is zero, etc.

It seems the background theory is related to a matrix that is not positive definite. Here is an interesting approach on the issue of being or not being a positive definite matrix: http://www2.gsu.edu/~mkteer/npdmatri.html

Hopefully that helps.

Best,

Marcos