Missing F-statistics of regression

[Junhyun Yun]

Can I ask a question? I am using regression discontinuity designs and ran some regressions. As you can see below, I used the same sample and same weights. In the first equation, the linear and squared terms of the running variable are included in the regression and F-statistic of the regression is reported. However, in the second equation, the cubic term in addition to the linear and squared terms of the running variable is included in the regression and F-statistic of the regression is missing.

1. Could you please let me know the reason?

2. Can I use the estimates of the coefficients and standard errors without any problems?

[Junhyun Yun]

I attach the relevant codes and outputs. I really appreciate your help.

. regress socialsecurity ageeea ageeea2 eea ageeeaeea ageeea2eea if eeadonutregression==1 [pweight=wtcrnh], cluster(hhid)
(sum of wgt is 58,510,996)
 
Linear regression                               Number of obs     =     16,530
                                                F(5, 7181)        =     789.80
                                                Prob > F          =     0.0000
                                                R-squared         =     0.2189
                                                Root MSE          =     .42552
 
                               (Std. Err. adjusted for 7,182 clusters in hhid)
------------------------------------------------------------------------------
             |               Robust
socialsecu~y |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
      ageeea |  -.0000625   .0001282    -0.49   0.626    -.0003137    .0001887
     ageeea2 |  -2.37e-07   1.50e-07    -1.58   0.114    -5.31e-07    5.72e-08
         eea |   .3296937     .04218     7.82   0.000     .2470085    .4123789
   ageeeaeea |   .0003761    .000228     1.65   0.099     -.000071    .0008231
  ageeea2eea |   6.58e-08   2.56e-07     0.26   0.797    -4.36e-07    5.67e-07
       _cons |   .1654308   .0237605     6.96   0.000     .1188532    .2120083
------------------------------------------------------------------------------
 
. regress socialsecurity ageeea ageeea2 ageeea3 eea ageeeaeea ageeea2eea ageeea3eea if eeadonutregression==1 [pweight=wtcrnh], cluster(hhid)
(sum of wgt is 58,510,996)
 
Linear regression                               Number of obs     =     16,530
                                                F(5, 7181)        =          .
                                                Prob > F          =          .
                                                R-squared         =     0.2191
                                                Root MSE          =      .4255
 
                               (Std. Err. adjusted for 7,182 clusters in hhid)
------------------------------------------------------------------------------
             |               Robust
socialsecu~y |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
      ageeea |   .0001067   .0004339     0.25   0.806    -.0007439    .0009573
     ageeea2 |   2.37e-07   1.14e-06     0.21   0.836    -2.00e-06    2.47e-06
     ageeea3 |   3.86e-10   9.02e-10     0.43   0.668    -1.38e-09    2.16e-09
         eea |   .3885139   .0800113     4.86   0.000     .2316683    .5453596
   ageeeaeea |  -.0005564   .0007659    -0.73   0.468    -.0020578     .000945
  ageeea2eea |   1.71e-06   1.98e-06     0.86   0.389    -2.18e-06    5.59e-06
  ageeea3eea |  -2.10e-09   1.59e-09    -1.32   0.186    -5.21e-09    1.02e-09
       _cons |   .1819229   .0476432     3.82   0.000     .0885282    .2753177
------------------------------------------------------------------------------

[Carlo Lazzaro]

Junhyun:
1) see -help j_robustsingular-;
2) you should probably revise your model and your predictors.
As an aside, you would be better off with using -fvvarklist- for categorical variables and interactions than creating them by hand.

[Junhyun Yun]

Thank you for your reply, Mr. Lazzaro. Can I ask more questions?

I read the explanations which appear when I used "help j_robustsingular" command. But I cannot find the reason.

First of all, as you can see in the outputs above, the number of clusters (d) is much larger than the number of coefficients (k). So, my case does not belong to the case where d-k+1>k. Also, the cubic term of the running variable is not zero for almost all of the observations. Therefore, the explanations which appear when I used "help j_robustsingular" command cannot explain my case.

Could you please tell me your guess about the reason? What else should I investigate?



I also really appreciate answers from other people. Thanks.

[Junhyun Yun]

One funny thing is that, when I divided running variables by 10,000, F statistics are reported.