0

I have three independent variables x1, x2, x3 which are proportions summing up to 1, so I am only using x2 and x3 as the independent variables. The dependent variable is 3 level categorial variable.

I also have two treatments:

  • In treatment 1, the distinct values of the proportions are 0, 1/3, 2/3, 1.
  • In treatment 2, the distinct values are 0, 1/8, 2/8, 3/8, 4/8, ... , 1.

The different distinct values are by design. The two treatments have the same number of observations. Thus in treatment 1, the bulk of obs is at 1/3 for example, which concerns me.

I want to run mixed (or random coefficient) logit and multinomial logit to check the different effects of x1, x2, x3 across the treatments. I am currently using interaction terms of the treatment variable and the proportion variable. Also clustered robust standard errors are used.

My questions is whether it is the right to use joint regression. Should I just run two separate regressions for two treatments?

kjetil b halvorsen
  • 63,378
  • 26
  • 142
  • 467
Jasmin
  • 43
  • 4
  • Unless you have very good reasons for separate models, usually it is best to use one common model. See https://stats.stackexchange.com/questions/373890/separate-models-vs-flags-in-the-same-model – kjetil b halvorsen Oct 05 '21 at 14:37

0 Answers0