explanatory variables statistically not significance

[EKUBAN Jeffrey]

the panel models fails. what do i do?. I did lag and it has failed too. for instance; regression showed this-reg GDPpercapita Trade Inflation Grosscapitalformation Accountage Madereceivedigitalpayment

Source | SS df MS Number of obs = 19
-------------+---------------------------------- F(5, 13) = 1.09
Model | 68.9702477 5 13.7940495 Prob > F = 0.4096
Residual | 164.080581 13 12.6215832 R-squared = 0.2959
-------------+---------------------------------- Adj R-squared = 0.0252
Total | 233.050829 18 12.9472683 Root MSE = 3.5527

-------------------------------------------------------------------------------------------
GDPpercapita | Coef. Std. Err. t P>|t| [95% Conf. Interval]
--------------------------+----------------------------------------------------------------
Trade | .0501887 .0399098 1.26 0.231 -.0360311 .1364085
Inflation | -.2635678 .1630886 -1.62 0.130 -.6158992 .0887636
Grosscapitalformation | .1894215 .1297907 1.46 0.168 -.0909743 .4698173
Accountage | -.1040997 .2257085 -0.46 0.652 -.5917132 .3835138
Madereceivedigitalpayment | .1425778 .2513621 0.57 0.580 -.4004569 .6856126
_cons | -3.702757 4.429473 -0.84 0.418 -13.27205 5.866537
-------------------------------------------------------------------------------------------

then the lag showed this; reg GDPpercapita L1_Accountage L2_Trade L3_Inflation
note: L3_Inflation omitted because of collinearity

Source | SS df MS Number of obs = 3
-------------+---------------------------------- F(2, 0) = .
Model | 2.02966639 2 1.01483319 Prob > F = .
Residual | 0 0 . R-squared = 1.0000
-------------+---------------------------------- Adj R-squared = .
Total | 2.02966639 2 1.01483319 Root MSE = 0

-------------------------------------------------------------------------------
GDPpercapita | Coef. Std. Err. t P>|t| [95% Conf. Interval]
--------------+----------------------------------------------------------------
L1_Accountage | .0312481 . . . . .
L2_Trade | -.0221909 . . . . .
L3_Inflation | 0 (omitted)
_cons | 3.028148 . . . . .
-------------------------------------------------------------------------------
then random effects;
xtreg GDPpercapita Trade Inflation Accountage Ruralpopulation Grosscapitalformation, re

Random-effects GLS regression Number of obs = 23
Group variable: country_id Number of groups = 5

R-sq: Obs per group:
within = 0.1580 min = 3
between = 0.9232 avg = 4.6
overall = 0.2573 max = 5

Wald chi2(5) = 5.89
corr(u_i, X) = 0 (assumed) Prob > chi2 = 0.3170

---------------------------------------------------------------------------------------
GDPpercapita | Coef. Std. Err. z P>|z| [95% Conf. Interval]
----------------------+----------------------------------------------------------------
Trade | .0541794 .0352419 1.54 0.124 -.0148934 .1232523
Inflation | -.1752834 .1445983 -1.21 0.225 -.4586909 .1081241
Accountage | .0503954 .0645909 0.78 0.435 -.0762005 .1769913
Ruralpopulation | .3618812 1.141951 0.32 0.751 -1.876301 2.600064
Grosscapitalformation | .1863113 .0980751 1.90 0.057 -.0059124 .3785351
_cons | -6.224283 5.189902 -1.20 0.230 -16.3963 3.947737
----------------------+----------------------------------------------------------------
sigma_u | 0
sigma_e | 3.4812246
rho | 0 (fraction of variance due to u_i)
---------------------------------------------------------------------------------------

please, how do i solve this?

[Richard Williams]

Welcome to Statalist.

Your output is very hard to read. The Statalist FAQ explains how to use code tags which will make the output much clearer.

What do you mean when you say the model fails? You don’t get significant results? If so, that may just be the way life is. You theory may be wrong or your data may not be good enough to test it.

if you can be clearer on what you are doing and what your concern is we may be able to help you more.

[Clyde Schechter]

Although I agree your output is hard to read, it is easy enough for me to pick out the sample sizes. In the first model you have N = 19, with 5 predictor variables. In the second one you have N = 3 with 3 predictor variables (and hence, no statistical inference statistics at all)--this one is a true "failure." And in the third you have N = 23 with 5 predictors. These data sets are severely underpowered for the questions you are trying to answer with them. With samples this small, you would struggle to get "statistically significant" results even for something as large as the correlation between height and weight in human adults.

You need better data. Also, you need to choose a model that is most appropriate to your specific research question and stick with that. Swapping models to get a "significant" result isn't science--it's called p-hacking, and it is increasingly viewed as scientific misconduct. So first get an adequate data set. Then review the models that might be suitable, learn about and understand the differences between them, and then pick one that actually answers your exact research question and whose assumptions are at least reasonably close to being met in your data. Then live with its results, whether they are what you hoped for or not.

[EKUBAN Jeffrey]

Although I agree your output is hard to read, it is easy enough for me to pick out the sample sizes. In the first model you have N = 19, with 5 predictor variables. In the second one you have N = 3 with 3 predictor variables (and hence, no statistical inference statistics at all)--this one is a true "failure." And in the third you have N = 23 with 5 predictors. These data sets are severely underpowered for the questions you are trying to answer with them. With samples this small, you would struggle to get "statistically significant" results even for something as large as the correlation between height and weight in human adults.

You need better data. Also, you need to choose a model that is most appropriate to your specific research question and stick with that. Swapping models to get a "significant" result isn't science--it's called p-hacking, and it is increasingly viewed as scientific misconduct. So first get an adequate data set. Then review the models that might be suitable, learn about and understand the differences between them, and then pick one that actually answers your exact research question and whose assumptions are at least reasonably close to being met in your data. Then live with its results, whether they are what you hoped for or not.

Thank you