xtreg, r(2001) insufficient observations

[Elias Zurfluh]

Hello,

I am currently writing my thesis on STEM gaps, my dataset contains 106'000 independent observations. I described my data as a panel data set using "xtset OBSNUM". However, when I try to run a regression for example as the following one:

xtreg lnearn c.age i.kid##i.sex

the r(2001) insufficient observations error message appears. I do have some missing values for lnearn, but dropping them doesn't change anything...

Does anyone know what I am doing wrong ?

Thanks for your help,
Elias

c.f. attachment for dataex

[Nick Cox]

Thanks for attempting dataex but please use it as specified, to copy and paste the results into the forum software. Images are not nearly as helpful as you hope: they don't allow copy and paste and (in your case for me, and perhaps for others) the image may be unreadable any way.

[Elias Zurfluh]

* Example generated by -dataex-. For more info, type help dataex
clear
input float lnearn long(OBSNUM sex kid) byte age long(married degree) byte sector float stemcat
11.608235  99671 0 0 40 1 6 2 0
10.668956  92121 1 1 40 1 6 1 0
11.0021  27204 1 1 34 1 6 2 0
11.617286   5203 0 0 46 0 5 2 0
9.903487  19266 1 1 32 1 6 3 0
.  67306 1 1 71 0 5 0 0
11.407565  46965 0 1 56 1 5 3 1
10.434115  59541 1 1 35 1 5 3 0
11.289782  40288 1 1 58 0 5 1 0
11.289782  47151 1 0 38 0 5 2 2
10.99541  36556 1 0 43 1 6 1 1
.  75303 0 0 74 0 5 0 0
10.08581  88199 1 0 74 1 6 1 0
10.518673  16216 1 0 42 0 6 3 1
11.512925  46218 0 0 40 0 7 1 1
10.094108  54854 1 1 36 1 6 0 0
11.695247  31147 0 1 37 1 6 3 1
12.323855  77221 1 0 32 0 6 3 1
11.91839 102601 1 1 52 1 5 3 1
9.903487   1579 1 1 34 1 6 1 0
11.326596  67003 0 0 31 0 5 2 1
11.350407  56278 1 1 38 1 5 3 0
.  56630 0 1 37 0 5 0 0
11.561716  71363 1 0 32 0 5 1 0
12.733755  97740 1 0 30 1 6 3 1
12.100712  32735 0 1 38 1 7 3 1
11.751943 105248 1 0 61 0 7 1 0
12.380026  32479 0 1 37 1 7 3 0
11.225244  54875 1 1 31 0 5 3 0
11.472103  30889 1 1 41 1 5 3 0
11.695247  38476 1 1 45 0 6 1 0
11.225244  51800 0 0 37 0 5 3 1
10.819778  70229 1 0 32 0 8 3 0
. 101395 0 1 69 1 6 0 0
11.695247  59353 0 1 57 1 5 3 0
11.289782  23177 1 1 37 0 6 2 0
. 102154 1 1 32 1 5 2 1
12.206073  22403 0 1 39 1 6 3 0
.  18936 1 0 68 1 6 0 0
12.542545  22318 0 0 41 1 6 3 0
10.714417  48712 1 1 33 1 5 3 0
12.556163  35418 1 1 31 1 5 3 1
12.429216 100700 0 1 40 1 6 3 0
12.388394  31194 0 1 37 1 7 3 0
11.589887  83422 1 1 44 1 5 1 0
11.81303  74846 0 0 35 0 7 3 1
11.964  54828 1 1 43 1 7 3 2
12.206073  83125 0 1 58 1 6 3 1
11.81303  93977 0 1 33 1 5 3 0
7.313221  15072 0 0 59 0 5 0 0
10.596635  61363 0 0 50 0 5 3 0
10.325482  63799 0 1 35 1 8 0 0
11.225244 103236 0 0 34 0 5 3 0
11.456355  76049 1 1 32 1 7 3 0
11.695247 104104 0 1 38 1 6 2 0
11.512925  73044 0 0 57 1 5 3 0
11.184422  73519 1 0 75 0 6 0 0
11.512925  54690 0 1 32 1 8 2 0
12.7367  64745 0 0 32 0 8 3 0
8.699514  10390 0 0 72 0 6 3 0
10.12663  28437 1 0 35 1 6 1 2
10.714417  23614 1 1 35 0 6 3 0
11.695247  30328 0 0 60 1 5 3 1
10.91509 101608 1 1 36 0 6 1 0
. 102393 1 0 75 1 5 0 0
11.15625   2938 1 0 33 0 6 3 0
9.903487  25240 1 0 30 0 6 3 1
12.468437   8639 0 0 33 0 7 3 1
11.580585  65520 1 0 46 0 6 1 0
10.714417   9329 0 1 38 1 6 2 2
11.407565 103863 1 1 35 1 5 3 1
11.711777  70853 0 0 52 0 5 3 0
6.633318  53471 0 0 73 0 6 0 0
11.429543 105735 1 0 33 0 8 2 0
12.061047  63114 0 1 52 1 5 3 0
10.819778  95921 0 1 60 1 7 1 0
10.878047  42682 1 1 33 1 6 1 0
12.15478 104852 0 1 36 1 5 3 1
.  57430 0 0 65 1 6 0 0
10.12663  21534 0 1 33 1 6 3 0
11.15625  70291 1 1 40 1 6 1 0
10.714417   1457 1 1 48 0 6 2 0
11.184422  43601 1 1 37 1 5 1 0
11.429543  17004 1 1 35 1 6 3 0
.  62450 0 0 66 1 6 0 0
11.804326  47418 0 1 38 1 6 3 1
11.407565   5710 1 0 71 0 6 3 0
11.635143  93449 0 1 38 1 6 3 0
11.77529  21005 1 0 33 1 5 3 0
11.925035  90559 0 1 39 1 6 3 1
12.0137  64212 1 0 33 0 6 3 0
11.350407  74471 1 1 39 1 5 3 1
11.512925  25293 0 1 36 1 5 3 1
12.388915  41036 0 0 61 1 7 3 0
11.385092  79374 0 1 39 0 5 3 1
12.08954  18851 0 0 59 1 5 3 0
11.571195  96938 0 1 57 1 5 3 0
.  47890 0 0 59 1 6 0 0
11.751943  44539 0 1 45 1 5 3 1
9.903487   9255 0 1 46 1 8 3 0
end
label values sex genderlabel
label def genderlabel 0 "M", modify
label def genderlabel 1 "F", modify
label values kid yesnolabel
label values married yesnolabel
label def yesnolabel 0 "N", modify
label def yesnolabel 1 "Y", modify
label values degree degreelabel
label def degreelabel 5 "1", modify
label def degreelabel 6 "2", modify
label def degreelabel 7 "3", modify
label def degreelabel 8 "4", modify
label values sector sectorlabel
label def sectorlabel 0 "L", modify
label def sectorlabel 1 "educational institution", modify
label def sectorlabel 2 "government", modify
label def sectorlabel 3 "business/industry", modify
label values stemcat stemlabel
label def stemlabel 0 "others", modify
label def stemlabel 1 "GEMP", modify
label def stemlabel 2 "LPS", modify

[George Ford]

Your fixed effect is a dummy for every observation?

[Rich Goldstein]

at least in your example data, each OBSNUM occurs once and once only - in what sense is this panel data? unless you have more observations in which OBSNUM is repeated, this is not panel data and that is the reason for the error message

[Elias Zurfluh]

Oh ok yes thanks. The data was however collected in 4 different years although individuals aren't observed multiple times is it possible to include time fixed effect in any way ?

What would the best method be ?

[Carlo Lazzaro]

Elias:
as per your example, you do not have a panel dataset. Therefore, I'd go:

. regress lnearn i.sex i.kid c.age##c.age i.married i.degree i.sector i.stemcat

      Source |       SS           df       MS      Number of obs   =        90
-------------+----------------------------------   F(13, 76)       =      4.55
       Model |  38.1598991        13  2.93537686   Prob > F        =    0.0000
    Residual |  49.0383721        76  .645241739   R-squared       =    0.4376
-------------+----------------------------------   Adj R-squared   =    0.3414
       Total |  87.1982713        89  .979755857   Root MSE        =    .80327

------------------------------------------------------------------------------------------
                  lnearn | Coefficient  Std. err.      t    P>|t|     [95% conf. interval]
-------------------------+----------------------------------------------------------------
                     sex |
                      F  |   .1154282   .2121206     0.54   0.588    -.3070466    .5379031
                         |
                     kid |
                      Y  |  -.2431639   .2246946    -1.08   0.283    -.6906819    .2043542
                     age |   .1103801   .0736976     1.50   0.138    -.0364014    .2571616
                         |
             c.age#c.age |  -.0012263    .000741    -1.66   0.102    -.0027021    .0002494
                         |
                 married |
                      Y  |   .3504836   .2210022     1.59   0.117    -.0896805    .7906477
                         |
                  degree |
                      2  |  -.0921419   .2025288    -0.45   0.650     -.495513    .3112291
                      3  |   .6133596   .2903066     2.11   0.038      .035164    1.191555
                      4  |   .0651655   .3919277     0.17   0.868    -.7154263    .8457572
                         |
                  sector |
educational institution  |   1.426503   .4612709     3.09   0.003     .5078021    2.345203
             government  |   1.864617    .477633     3.90   0.000     .9133282    2.815905
      business/industry  |    1.85534   .4197086     4.42   0.000     1.019418    2.691262
                         |
                 stemcat |
                   GEMP  |   .2034165   .2102973     0.97   0.336    -.2154269    .6222598
                    LPS  |  -.5708022   .4456755    -1.28   0.204    -1.458442    .3168376
                         |
                   _cons |   7.084807     1.7402     4.07   0.000     3.618898    10.55072
------------------------------------------------------------------------------------------

[Jeff Wooldridge]

You have a repeated cross section. If you don’t know the year an observation was collected, you can’t include time fixed effects.

[Elias Zurfluh]

Thanks a lot Carlo !

Jeff, thanks for your answer. I can identify which year each individual was observed in yes. However, I was unable to use the year in xtset because I only have 4 different dates (2013, 15, 17, 19), therefore many individuals share the same year. Is there any solution for this ? Or is the best to simply include a year dummy in the regression ?

[Carlo Lazzaro]

Elias:
if you have a repeated cross-sectional study there's no need to -xtset-, unless you want to use the -xtsum- command.
If different individuals share the same year, this is not an issue.
I would add an -i.time- predictor in the right-hand side of your regression equation.