0
$\begingroup$

I have a linear regression model, where the outcome variable diagnosis is a categorical variable representing a tumor being either "malignant" or "benign":

lm(diagnosis ~ radius + perimeter + compactness + smoothness, ...)

I'm not sure how to interpret the regression coefficients for the binary categorical variable here. For instance, the variable smoothness has a regression coefficient of $16.9805981$, how would we interpret this? Something like: a 1 unit increase in smoothness.stderr increases the probability of diagnosis=1 (i.e. malignant) by 16.9805981?

$\endgroup$
1

1 Answer 1

0
$\begingroup$

If "diagnosis" is coded as either 0 or 1, then your interpretation is in the ballpark. There are issues with causality, ceteris paribus, and estimates vs actual, but you have the right idea.

But the result is indeed strange, so I would guess that you have something like 0,100 instead of 0,1 for "diagnosis"; or that the "smoothness" variable has a very small sdev so that a one unit increase is out of bounds; or maybe that there is extreme multicollinearity, which might imply a one unit increase is out of bounds, ceteris paribus.

Or maybe there is some other mistake, like reading the output wrong.

Finally, you should use logistic regression here.

$\endgroup$

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.