0

I have used a modified poisson regression to model a binary outcome because I want to obtain relative risks, rather than odds ratios. The predicted probabilities are of interest, but many are beyond the range [0, 1].

Is there an appropriate way to constrain these to occur in the range [0, 1]? If a solution exists, I am keen to apply it to standard glm models and models generated using glmnet in R.

C_H
  • 95
  • 8
  • 3
    Your mathematically impossible results show forcibly why this approach is invalid. Why don't you just do logistic regression with a log link? – whuber Jun 29 '18 at 00:17
  • 3
    You could consider a binomial model with a different link function than the logistic. – Glen_b Jun 29 '18 at 02:24

0 Answers0