0
$\begingroup$

My hypothesis is: Company size (CSIZE) is positively correlated with Return on Assets (ROA).

I create a Pearson correlation matrix in Stata using pwcorr csize roa and it returns a correlation of 0.2 with a 5% significance level.

Now I also want create a multiple regression to test Company Age (CAGE) and Company Size (CSIZE) on Return on Assets (ROA).

I create a linear regression model in Stata using regress roa cage csize

I returns low R^2=0,007 and low F-statistic 1,08. Neither independent variable had significant T-statistic either.

My question is how to interpret such a result: The low R^2 means my model does not explain much of the variation. Plus it is not significant. But is that only for the combined effect of both independent variables (CAGE+CSIZE) on the dependent variable (ROA)? The individual t-test was also insigificant, so what about the result from my Pearson correlation?

I am having a hard time seeing how linear regression and Pearson correlation relates to each other, and what implications the results of both have on my hypothesis.

$\endgroup$
  • $\begingroup$ It isn't possible for a Pearson correlation to be 1.2, because Pearson correlations are limited to [-1, 1]. Did you make a typo? Likewise, should "Plus it is significant" be "Plus it is not significant"? $\endgroup$ – Kodiologist May 20 '16 at 23:55
  • $\begingroup$ Ah yes you are right. I just took a random number to illustrate. It's just an example. I guess that backfired :-) it has been corrected now. $\endgroup$ – Esben May 21 '16 at 7:51
  • 1
    $\begingroup$ You will probably get better answers if you use the real numbers you obtained. $\endgroup$ – Kodiologist May 21 '16 at 11:26
2
$\begingroup$

First, I think regress is the multiple linear regression in stata, not correlate. Second, "theoretically," the relationship (or model) between your response and covariates cannot weaken (lower $R^2$) when you add variables (for more reading on this, I recommend google searching "degrees of freedom"). Third, if in the first case you mentioned, if correlation is .2, then $$R^2 = (.2)^2 = .04 < .07$$ meaning the response was explained somewhat better when you added the aditional covariate.

Nevertheless, if the above doesn't apply (because you made up the numbers for the sake of example), the behavior you observed can occur if your two covariates (age and size) are highly correlated, resulting in some numerical instability. Certainly company size and age may be highly correlated, particularly in more traditional sectors of the economy. For more on dealing with this issue, Google "colinearity" and perhaps even "pre-conditioning."

$\endgroup$

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.