I am looking for the best way to depict a concave, quadratic association. I'm using logistic regression to measure the association between affect and military advancement (yes/no).

The primary predictor centered on the mean squared was significant, and it is graphically clear that Y increases as X increases only to a point, before leveling off and decreasing. I was able to determine at which level of X, Y reaches its maximum by adding the mean to the linear term (centered on the mean), divided by two times the quadratic term (mean + (b1/2b2).

Are results from logistic regression uninterpretable if there is a curvilinear/quadratic association between X and Y?

What would be the best way to determine the slope at different levels of X for a binary outcome? I would like to compare how the slope differs at varying levels of X (low vs. medium vs. high), and specifically how it decreases at higher levels of X.



  • $\begingroup$ Hi Emily, Welcome to CV. Note that your username, identicon, & a link to your user page are automatically added to every post you make, so there is no need to sign your posts. In fact, we prefer you don't. Also, there's no need to say "thank you" at the end of your post - it might seem rude at first, but it's part of the philosophy of this site (tour) to "Ask questions, get answers, no distractions" and it means future readers of your question don't need to read through the pleasantries. $\endgroup$ Mar 13, 2018 at 15:23

1 Answer 1


The resulting coefficients of logistic regression are difficult to interpret in general. The coefficient when not squared is the change in log odds for every 1 unit increase in your IV. So if that IV is squared it is still interpreted the much the same way. The coefficient is equal to the change in log odds for every 1 unit increase in the squared IV (controlling for the other predictors in your model including the linear version of the IV).

Because logistic regression follows a logistic curve interpreting the slope as is would be strange. I think your best bet would be to transform to percent likelihood, so that its more interpretable and then use your low, medium, high method. Either that or just graph it, that will provide the most detail.


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.