0

Do I have to check linearity assumption of logistic regression (Log odds of outcome with continuous variable)

using graph of predicted logodds (by logistic regression) and continuous variable

Or

using graph with real logodds and continuous variable ?

I think the latter is right In univariate setting, the first one should be perfect linear. Is it right?

Koh Heebyung
  • 1
  • 1
  • I think it might be easier to check that the deviance residuals have no patterns. – Demetri Pananos Sep 04 '19 at 16:36
  • Thank you for reply. But I wonder whether logodds should be made by predicted probability of regression , or each level's real log odds on continuous variable. – Koh Heebyung Sep 04 '19 at 16:42
  • See https://stats.stackexchange.com/search?q=Hosmer+Lemeshow. – whuber Sep 04 '19 at 16:46
  • 1
    @whuber, the Hosmer & Lemeshow test is not considered valid. What do you have in mind there? – gung - Reinstate Monica Sep 04 '19 at 18:14
  • @gung We have a couple hundred posts referring to this test: among them they provide an indication of what one might consider doing to test goodness of fit. I did not intend my comment to constitute or even suggest a recommendation of that test. – whuber Sep 04 '19 at 18:26
  • See https://stats.stackexchange.com/questions/212867/problems-with-zero-values-when-testing-for-linearity-of-the-logit, https://stats.stackexchange.com/questions/169348/how-should-i-check-the-assumption-of-linearity-to-the-logit-for-the-continuous-i?rq=1 – kjetil b halvorsen Feb 21 '20 at 03:00

0 Answers0