FE results are inconsistent with literature

[Eni Jana]

I hypothesize that if IQ in Norway improves It may -vely affect the EP (not revenue) of competing countries. To test I used: xtreg loilexport L.linst lexrate lrol lgdp i.yr, fe robust. Where loilexport is the logged value of oil export revenue, L.linst is the Lagged institutional quality of Norway. (logged), Lexrate xchange rate per year (logged), Lgdp gdp growth rate (logged), lrol is each countries IQ per year.


my results shows that “IQ of the competing countries has a +ve but nt statistically significant effect on their oil EP." which is inconsistent with literature.

• Could it be because I used export revenue instead of export growth rate (on 2nd thought) ?
• I intentionally omitted certain time fixed effects like global oil prices and entity ones like proximity of trading countries, production levels which are very material to their export performance or could I be using the code wrong or wrong FE method altogether

My apologies for the multiple questions.
Thanks,
Eni
(For some reason, I couldnt past the pic of my results so I had to upload as link. Apologies for the inconvenience).

[Carlo Lazzaro]

Eni:
welcome to this forum.
Some comments on your post:
1) you could not exclude that IQ of the competing countries coefficients is zero (use -test- to check for that). Therefore, your results are inconclusive;
2) what above might be due to your limited sample size;
3) please use CODE delimiters (not pics/screenshots) to share what you typed and what Stata gave you back (as per FAQ). Thanks.

[Eni Jana]

Hi Carlo,
Thank you for your warm & prompt response and my apologies I'm still learning the ropes here

I am using xtreg in stata 18.5

 xtreg loilexport L.linst lexrate lrol lgdp i.yr, fe robust
note: 10.yr omitted because of collinearity.

Fixed-effects (within) regression               Number of obs     =        423
Group variable: count                           Number of groups  =         47

R-squared:                                      Obs per group:
     Within  = 0.1917                                         min =          9
     Between = 0.0001                                         avg =        9.0
     Overall = 0.0000                                         max =          9

                                                F(11, 46)         =      76.20
corr(u_i, Xb) = -0.6500                         Prob > F          =     0.0000

                                 (Std. err. adjusted for 47 clusters in count)
------------------------------------------------------------------------------
             |               Robust
  loilexport | Coefficient  std. err.      t    P>|t|     [95% conf. interval]
-------------+----------------------------------------------------------------
       linst |
         L1. |   2.977051   1.067377     2.79   0.008     .8285317     5.12557
             |
     lexrate |  -.2857933   .0988856    -2.89   0.006    -.4848398   -.0867469
        lrol |   .1913978   .2223174     0.86   0.394    -.2561038    .6388994
        lgdp |   .2358087   .0238246     9.90   0.000     .1878523     .283765
             |
          yr |
       2011  |  -.0391254   .0198687    -1.97   0.055    -.0791191    .0008683
       2012  |  -.0218212   .0456835    -0.48   0.635    -.1137772    .0701349
       2013  |  -.1404138    .040063    -3.50   0.001    -.2210564   -.0597713
       2014  |  -.1670421   .0510283    -3.27   0.002    -.2697567   -.0643275
       2015  |  -.1726409   .0625733    -2.76   0.008    -.2985944   -.0466875
       2016  |  -.0400879   .0352523    -1.14   0.261    -.1110472    .0308713
       2017  |  -.0446002   .0286752    -1.56   0.127    -.1023203    .0131199
       2018  |          0  (omitted)
             |
       _cons |   4.361708   .6173764     7.06   0.000     3.118994    5.604422
-------------+----------------------------------------------------------------
     sigma_u |  1.4566089
     sigma_e |  .18397277
         rho |  .98429826   (fraction of variance due to u_i)
------------------------------------------------------------------------------

.
end of do-file

Please find a snippet of my data set and the alterations I have made.

* Example generated by -dataex-. For more info, type help dataex
clear
input float Inst long(count yr) float(linst loilexport lexrate rol_2 lrol lgdp)
1.87 1  1 .6259384  7.06732 4.2856536 1.1700001   .1570038 3.9550824
1.89 1  2 .6365768 6.931472 4.3093214 1.1800001   .1655145  4.021774
1.86 1  3 .6205765 7.196687  4.289637      1.15  .13976192  3.988984
1.95 1  4 .6678294 7.137279  4.350794      1.19  .17395335  3.977811
1.97 1  5 .6780335 7.087574 4.3741207      1.31   .2700271  3.981549
2.02 1  6 .7030975 7.094235 4.3892503       1.2   .1823216 4.0091496
1.98 1  7 .6830968 7.129298 4.6120467      1.06  .05826885  3.992681
   2 1  8 .6931472 7.098376 4.6953764 1.0799999  .07696097  4.005513
1.98 1  9 .6830968 7.091742   4.70926 1.0699999  .06765859  3.960813
1.92 1 10 .6523252 7.005789 4.7586637      1.19  .17395335  3.958907
1.87 2  1 .6259384  7.49053  4.373616       .76 -.27443686  3.948548
1.89 2  2 .6365768 7.462789 4.5208097       .73  -.3147107   4.01458
1.86 2  3 .6205765  7.36328 4.5425496       .72   -.328504   3.99765
1.95 2  4 .6678294 7.434258  4.558812        .7  -.3566749 4.0866485
1.97 2  5 .6780335 7.439559 4.5697503       .71  -.3424903 4.0244584
2.02 2  6 .7030975 7.401842  4.588024  .9299999 -.07257075 4.0221324
1.98 2  7 .6830968 7.466228  4.787992       .97 -.03045918  3.950089
   2 2  8 .6931472 7.452403  5.097791         1          0  3.879913
1.98 2  9 .6830968 7.374002  5.111506       .96 -.04082195   3.92888
1.92 2 10 .6523252 7.271008  5.532836         1          0 3.9056025
end
label values count count
label def count 1 "Algeria", modify
label def count 2 "Angola", modify
label values yr yr
label def yr 1 "2009", modify
label def yr 2 "2010", modify
label def yr 3 "2011", modify
label def yr 4 "2012", modify
label def yr 5 "2013", modify
label def yr 6 "2014", modify
label def yr 7 "2015", modify
label def yr 8 "2016", modify
label def yr 9 "2017", modify
label def yr 10 "2018", modify

Where loilexport is the logged value of oil export revenue for 10 years

L.linst is the Lagged institutional quality of Norway. (logged)

Lexrate exchange rate per year (logged)

Lgdp gdp growth rate also logged (I couldn’t lay hands on the actual gdp per year)

lrol is each countries IQ per year.
My sample size is 470 by the way (47 countries over 10years)

[Carlo Lazzaro]

Eni:
thanks for using CODE delimiters.
What you are experiencing might be caused by the limited variation in IQ across the observations included in your sample.
As we know, to work out properly, the -fe- estimator needs a remarkabe within panel variation (once time-invariant veriables are wiped out).

[Eni Jana]

Hi Carlo,
I appreciate your feedback.
What do you mean by limited variation in my IQ? I know usually the IQ variables are introduced into the model separately to avoid multicollinearity.
And please what do you suggest I do to fix this problem??

[Carlo Lazzaro]

Eni:
I meant that a limited within panel (within nation) variation of IQ across years might be responsible for your "weird" result.
As yours is not my research field I do not have a qualified opinion about the way IQ is introduced in the model.
That said, as far as multicollinearity in concerned, it is a real issue in a limited number of instances (see, for a humorous coverage of this topic Chapter 23 of A Course in Econometrics — Harvard University Press).
Eventually, you may want to test if -re- is better suited in your case (see the community-contributed module -xtoverid-, as -hausman- does not support non-default standard errors).
As a sidelight, results are what they are and, once you have ruled out major flaws in data entry and/or in your methodological approach, you should live with them.