0
$\begingroup$

I'm having a bit of a hard time understanding why the 'multiple correlation' coefficient within a simple linear regression (i.e., 1 predictor) isn't identical to the coefficient of determination. I understand that they're respectively denoted by $R$ and $R^2$, and thus the transformation from one to another is should be $\sqrt{R^2}$...but my mental barrier is that there is no combination of predictors to produce a the $R$ and therefore the two would be equivalent. Many thanks in advance.

$\endgroup$
2
$\begingroup$

If you have a linear regression model which regresses an outcome variable Y on a single predictor variable X, then you can only talk about a simple coefficient of determination $R^2$ which describes the percentage of variability in the Y values which can be accounted for by X. The coefficient $R^2$ can be shown to be equal to the squared value of the (Pearson) correlation coefficient $r$ between Y and X: $R^2$ = $r^2$. Recall that $r$ can range between -1 (perfect negative association between Y and X) and 1 (perfect positive association between Y and X).

If the linear regression model has multiple predictors, then it makes sense to talk about a multiple coefficient of determination $R^2$ which describes the percentage of variability in the Y which can be accounted for by the multiple predictors. The multiple correlation coefficient is simply defined as the square root of the multiple coefficient of determination and is a measure of how well Y can be predicted using a linear combination of the multiple predictors. By virtue of how it is defined, the multiple correlation coefficient can range between 0 and 1, where 1 indicates that Y can be perfectly predicted by a linear combination of the multiple predictors (which includes an intercept) and 0 means that the multiple predictors cannot be used to predict Y. See https://en.m.wikipedia.org/wiki/Multiple_correlation for more details.

$\endgroup$
  • $\begingroup$ Hi Isabella, thanks for your detailed answer. I'm with you that it doesn't make any sense in the context of simple regression. Nonetheless I had a university-level psychometrics test ask this very question. Just trying to figure out how to best navigate this scenario with the prof. I propose that given how it's defined, a multiple correlation in the context of simple regression, if it should be anything...should be identical to the coefficient of determination. Your thoughts? $\endgroup$ – Connor G Mar 8 at 19:40
  • 1
    $\begingroup$ There is no multiple correlation in the context of simple regression, because there are no multiple predictors! $\endgroup$ – Isabella Ghement Mar 8 at 20:46

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.