This just means that once you control for other variables that are also likely correlates of GDP per capita, the incremental effect of GINI above and beyond the other variables is not strong enough to yield a significant p-value.

It is important, however, to take into consideration that your sample size (given that it's at the country-level) is only 57 and thus you'll generally need a pretty big effect for the p-value to dip below the conventional 0.05.

Whether it's actually 'bad' or not depends on how you approach research. I personally don't think non-significance is all that bad as it's showing the lack of an effect, which can still be important. However, there is a publication bias where it's difficult to publish non-effects.

Certainly there are ways to p-hack to get a significant effect, although I wouldn't recommend any of them. If you're mainly interested in GDP per capita predicted by GINI, there are enough datasets out there online where you can cover at least 99% of the world population (and sample increased to around 180-200). Is there a reason why you're only looking at those 57 countries?

Hello Kostas

More important than p-values are effect sizes and their uncertainty. I avoid p-values whenever I can.

I suggest caution with your predictors. Average income and GDP per capita have population in the denominator, which you also included as a predictor, i.e., total population.

As a general reminder, graph your data -- but I gather you have done this.

All-in-all, Hohjin Im gave valuable advice, right above.

Best of luck

Aderivaldo

As Prof Im has suggested you have probably gotten the wrong model for your data. I would suggest that you try some other predictors. Perhaps adaptive lasso regression might be of help. Google this term and you can find description and programs. Also you could look at Jared Lander, R for everyone available from the z-library I believe. Best wishes, David Booth

It means the coefficients are not significant.

It could be due wrong model or wrong explanatory variables.

It looks like in the first model where you have all your variables, p value for the GINI index is actually less that .05 (it looks like p = .007), indicating that your IV is significantly related to your DV or that it explains unique variance in the DV (if you use the traditional interpretation of p values of course in the null hypothesis testing context).

p values for each predictor variable/ IV in the regression model are expected to change based on other variables entered in the model, so it would be expected that p value for your IV in the model where it is the only predictor would be different from p value in the model with multiple predictors.

Also, if you want to look at whether GINI index predicts your DV above and beyond your covariates, then I would recommend running a hierarchical regression, where you would enter all covariates in your first step/ block 1 and then add your variable of interest (GINI Index) at the second step/ block 2.

Finally, I would also recommend checking your VIF and Tolerance numbers in the regression output to assess for multicollinearity problems. You have Average education and Higher Education - I would check correlation between these two variables, and if it is high, then you want to drop one of these and only use one of these variables in your regression model, as they may essentially be providing you with the same information. You want to avoid multicollinearity in your regression models.

Have you looked at plots of your data? I'd guess that many of your variables have skewed distributions and that the relationships among them aren't linear. You probably need to consider re-expressing some or all of your variables. then look for outliers. A collection of countries almost always contains outliers. You'll want to set those aside and deal with them specially. Your regression has a strong F statistic, but weak t's--that's a warning sign for collinearity. Check for that and consider omitting some of the predictors. Finally, once you fix all of those issues, a partial regression plot of the predictor of particular interest to you (I assume the Gini coefficients) can help you visualize what is going on.

If all the p-values are high, this could be a problem of ill-conditioned X matrix. Hence, you might want to start by checking for the presence of multicollinearity among the explanatory variables, before proceeding into other econometric tests.