0
$\begingroup$

enter image description here

Our variables are insignificant. What are the reasons? But Adjusted R square is good Percentage. [![enter image description here][2]][2]

[![enter image description here][3]][3]mgur.com/CQtdk.png

$\endgroup$
  • $\begingroup$ Not all your variables fail to reach some arbitrary level of statistical significance so your question rests on a false assumption. $\endgroup$ – mdewey Aug 22 '17 at 10:45
  • 1
    $\begingroup$ GDP is the hands-down winner here. Perhaps that's a foregone conclusion and you'd get a more interesting analysis if you scaled by GDP and looked to see what the other predictors explain. Note that the output omits one of the most important pieces of information, the sample size. I wouldn't want to use a model with 8 predictors without a sample much, much larger than 8. $\endgroup$ – Nick Cox Aug 22 '17 at 11:10
  • 1
    $\begingroup$ I'd expect that some of these predictors are better handled on log scale. $\endgroup$ – Nick Cox Aug 22 '17 at 11:11
  • $\begingroup$ Isn't the variable "GDP Million in Rupee" significant (p-Value<0.05)? $\endgroup$ – Julian Aug 22 '17 at 11:24
1
$\begingroup$

If the model fits well (good R^2), but the variables cannot be estimated reliably (not significant), here's a couple of things that may be the case

  • (See comment below) One of your variables is actually 'very significant'. If what you are predicting is already very well explained by this one variable, then you can get a good (adjusted) R^2, even with the other, 'insignificant' variables also in the model.
  • You have few data points compared to the number of variables. Then the fit is very good, but estimates are noisy.
  • Your explanatory variables have "a lot of" correlation. If this is the case, then many coefficients fit well.

In both cases, it can help to remove explanatory variables, in particular the ones that covary with other variables.

$\endgroup$
  • $\begingroup$ Allright Nick, good point, didn't intend to be mean but I guess I was. Thanks, I'll remove it. $\endgroup$ – Gijs Aug 22 '17 at 11:02
  • 1
    $\begingroup$ The VIFs are fairly low, making the second bullet possible but probably not among the likeliest explanations. The relatively small sizes of the t statistics suggest the first bullet does not apply, either. Note that the question refers to adjusted $R^2$, which does compensate for the number of variables. First on your list ought to be to note that one variable is clearly significant, according to just about any standard, and that alone could account for a high adjusted $R^2$. $\endgroup$ – whuber Aug 22 '17 at 17:35
0
$\begingroup$

First, as others point out, one variable is highly significant. Also, one is close.

Second, you may have collinearity. VIF is not the best measure of this. It's better to use condition indexes, as I showed in my dissertation and as David Belsley showed in much more detail in Conditioning Diagnostics: Collinearity and Weak Data in Regression. If you do have collinearity, there are a number of solutions, perhaps the most relevant here is ridge regression.

Third, although it's a relatively minor point, I would transform "remittances" and "gold" by dividing them by 1,000,000. This will make the output easier to understand.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.