Announcement

Collapse
No announcement yet.
X
  • Filter
  • Time
  • Show
Clear All
new posts

  • Interaction terms in logit regression Computing real odd after estimation

    Hi there, I am writing this logit model in stata: logit prop_FS_cat i.Path##need..., where Path is housing pathways (with 5 pathways) and need (0 vs. 1) is level of need for mental services... After the estimation, I've estimated the probabilities (margins) at need=0,1. Now, I would like to compute the (real) odd ratio for the interaction terms with confident interval (CI). I can compute the odd ratios manually using the estimated probabilities, but without CI. So, I've tried the lincom command, applying the formula for odd ratio (lincom (_b[1.need#5.Path]/(1-_b[1.need#5.Path]) - ( _b[1.need#1.Path]/(1- _b[1.need#1.Path]))). It doesn't work. I think the lincom only works for sum or difference operators. Please do you know whether a way to process with such estimation with Stata? Thanks a lot

  • #2
    Thanks. I think I fixed it. The odd ratios estimated in the model for the interaction terms are kind off ratio of "odd ratios". Since I have here simple interaction terms (two categorial variables), I just changed the reference category of the variable level of need to deal with it. Though, for more complex interactions, this trick won't work.

    Comment

    Working...
    X