Need help with code

[Ismoil Ismoilov]

Dear all,
I am running panel data regression analysis with -xtscc command where N=9 and T=30.

xtset id year
xtscc GDPpercapitagrowth Grosscapitalformation lagGovconsump GDPpercapita_initial Populationgrowth Secondaryschoolenrollment Tradegdp Inflation Naturalresourcesrent Domesticcredit RuleofLaw, fe

The problem is most of the independent variables are found insignificant, which is unreal. Because in other scientific papers these variables were found to be significant.
Can you suggest some measures/commands to run a proper regression analysis?

Thank you.

[Carlo Lazzaro]

Ismoil:
welcome to this forum.
How could interested listers ever reply positively to your query if you fo not post what Stata gave you back, too (as per FAQ)? 😀

[George Ford]

It is not uncommon for FE to eat variables.

[Ismoil Ismoilov]

Ismoil:
welcome to this forum.
How could interested listers ever reply positively to your query if you fo not post what Stata gave you back, too (as per FAQ)? 😀

Code:

. xtscc gdp_pcg grosscapform lagGovconsump gdpppc_ini pop_g sec tradegdp cpi resourcerentgdp domcredit    R
> uleofLaw, fe

Regression with Driscoll-Kraay standard errors   Number of obs     =       115
Method: Fixed-effects regression                 Number of groups  =         9
Group variable (i): id                           F( 10,     8)     =     70.16
maximum lag: 2                                   Prob > F          =    0.0000
                                                 within R-squared  =    0.2559

---------------------------------------------------------------------------------
                |             Drisc/Kraay
        gdp_pcg | Coefficient  std. err.      t    P>|t|     [95% conf. interval]
----------------+----------------------------------------------------------------
   grosscapform |    .154158   .0743587     2.07   0.072    -.0173135    .3256294
  lagGovconsump |   .2753659   .3609152     0.76   0.467     -.556906    1.107638
     gdpppc_ini |   1.165284    1.49813     0.78   0.459     -2.28941    4.619979
          pop_g |  -.7596739   1.102854    -0.69   0.510     -3.30286    1.783512
            sec |  -.1463251   .0620392    -2.36   0.046    -.2893878   -.0032624
       tradegdp |   .0581031   .0302076     1.92   0.091    -.0115558     .127762
            cpi |  -.0570464   .0303555    -1.88   0.097    -.1270462    .0129534
resourcerentgdp |   .3030473   .1289455     2.35   0.047     .0056983    .6003962
      domcredit |  -.1260692    .061233    -2.06   0.073    -.2672727    .0151342
      RuleofLaw |   2.230128   2.992813     0.75   0.477    -4.671312    9.131568
          _cons |          0  (omitted)
---------------------------------------------------------------------------------

Dear Professor, thank you for your warm greetings. I'm new here and need some time to get to know all of the rules.

Here are the result that I got from regression, where 9 countries during 30 years

[Carlo Lazzaro]

Ismoil:
1) did other paper report on the very same regression specification on a very same sample size with the same number of max. lags?
2) what happen if you reduce the number of predictors?
3) did you already check if your model is correctly specified?

Last but not least, please call me Carlo, like many on (and many more off) this forum do. Thanks.

[Ismoil Ismoilov]

Ismoil:
1) did other paper report on the very same regression specification on a very same sample size with the same number of max. lags?
2) what happen if you reduce the number of predictors?
3) did you already check if your model is correctly specified?

Lastly, please call me Carlo, like many on (and many more off) this forum do. Thanks.

Carlo, thank you for your response and attention.

1) I am trying to read as much as possible. But most of them did research with more countries (N).
2) I haven`t tried yet. I`ll try it.
3) I took this model (almost all DV and IV) from the specific research paper and ran regression according to it. The difference is that paper did research with more N(9) and T(30), while mine is remarkably small. Therefore, I am struggling with the proper code

[Carlo Lazzaro]

Ismoil:
from your reply I surmise what follows:
1) your sample is what it is (and lagging does not help in this respect, as it eats observations);
2) the smaller the sample, the lower the number of predictors to be plugged in the right-hand side of your regression equation. The Portnoy's suggested ratio (https://projecteuclid.org/euclid.aos/1176346793) states that [(constant+predictors)^2]/sample size should ideally go to 0 as the sample size goes to infinity;
3) the difference with the sample size reported on the paper and yours is relevant and, as we know, sample size has a bearing on statistical significance. However, non-significanct coefficients are as informative ad their significant counterparts. They simply require more time and effort to understand the reason why they do not reach statistical significance, considering the p-value as a yardstick (but not a totem) and keeping in mind that, as frequently reminded on this forum, the difference between statistical significant and non-significant results is (oftentimes) non-statistical significant.

[Ismoil Ismoilov]

Ismoil:
from your reply I surmise what follows:
1) your sample is what it is (and lagging does not help in this respect, as it eats observations);
2) the smaller the sample, the lower the number of predictors to be plugged in the right-hand side of your regression equation. The Portnoy's suggested ratio (https://projecteuclid.org/euclid.aos/1176346793) states that [(constant+predictors)^2]/sample size should ideally go to 0 as the sample size goes to infinity;
3) the difference with the sample size reported on the paper and yours is relevant and, as we know, sample size has a bearing on statistical significance. However, non-significanct coefficients are as informative ad their significant counterparts. They simply require more time and effort to understand the reason why they do not reach statistical significance, considering the p-value as a yardstick (but not a totem) and keeping in mind that, as frequently reminded on this forum, the difference between statistical significant and non-significant results is (oftentimes) non-statistical significant.

I got it. Thank you