1
$\begingroup$

I have a question about correlation.

My research has 8 independent variables out of which 3 are continuous variables and 5 are dichotomous/binary variables. I have to check multicollinearity before running the regression model through correlation. can I use Pearson's correlation and point biserial correlation simultaneously, and show the results in the same table?

Please help!

$\endgroup$
1
  • $\begingroup$ Welcome to Cross Validated! I’ve voted to reopen this question, since point biserial correlation is a correlation between a continuous variable and a binary variable, and this is reasonable evidence to answer the question in the affirmative. // A bigger question is why you feel you must check for multicollinearity. What if it is not there? What if it is there? $\endgroup$
    – Dave
    Oct 31, 2022 at 20:46

1 Answer 1

1
$\begingroup$

Point biserial correlation (magnitude) is Pearson correlation (magnitude) between a continuous variable and a binary variable that is encoded with numbers (e.g., as $0$ and $1$). Consequently, feel free to combine “regular” Pearson correlation and point biserial correlation in one table as if they were synonymous, since point biserial correlation really is a Pearson correlation.

I say that point biserial correlation is equal in magnitude to Pearson correlation because the numerical encoding of a categorical variable is ambiguous. Should dogs be coded as $0$ and cats as $1$? Should dogs be coded as $1$ and cats as $0$? You get the same Pearson correlation magnitude either way, but the sign will flip. Fortunately, this does not matter to an assessment of feature multicollinearity.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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