3

So I have a test subject A which consist of a binary variable 0 and 1. I want to see if this binary variable is decided by some independent variables (plus interactions) which are continuous, which in my case I apply normal logistic regression to see if I can model it.

However after I apply a treatment on test subject A, I want to see if: - Does the binary variable odds ration effect change after treatment? - Does the independent variables effects change on the binary variable after treatment

Should I do two different logistic regression models (before and after) and compare its parameters with each other?

kjetil b halvorsen
  • 63,378
  • 26
  • 142
  • 467
arezaie
  • 157
  • 9
  • So, $y$ is a function of treatment $D$ and covariates $X$. You want to know whether $\partial y/\partial X | D = 1$ is the same as $\partial y/\partial X | D = 0$? – generic_user Feb 08 '17 at 14:57
  • Yes, most likely covariates X have likely changed as well after treatment. – arezaie Feb 08 '17 at 15:00
  • 2
    You probably want a more authoritative answer from someone who uses these methods more, but you might consider a mixed logit. Basically include both before and after data in the model, include a subject-wise random effect, and include interactions between the before/after variable and each of the X's. – generic_user Feb 08 '17 at 15:09
  • Any thoughts if I add the before and after effect into one logistic model. I then subject my dependent and independent variables to a dichotmous variable which descirbes 0 for before and 1 for after treatment? – arezaie Feb 08 '17 at 15:43

0 Answers0