0

Let's say I have two models:

lifeexpF~group + log(ppgdp) + group:log(ppgdp)

lifeexpF~log(ppgdp) + group:log(ppgdp)

group is a factor

I understand that the first model is the general model, and it gives a different slope and intercept for each group. For the second model, I think the slopes are all different, but the intercepts are the same; however, I'm not quite sure as I am very new to regression and R.

Haitao Du
  • 32,885
  • 17
  • 118
  • 213
mistersunnyd
  • 595
  • 3
  • 13
  • What is your question? You might be interested in the answers at http://stats.stackexchange.com/questions/11009, especially that by [Frank Harrell](http://stats.stackexchange.com/a/11080/919). – whuber Feb 24 '17 at 22:16
  • I'm not exactly sure what adding "group" to the model does. – mistersunnyd Feb 24 '17 at 22:54
  • 1
    Although your question is not very clear, the difference between both models is the inclusion/exclusion of a main effect term from a model with interaction consisting of the term. In other words, you have included the main effect term for "group" in the first model which is the "advisable"/"right" thing to do if the term is part of a significant interaction term. Refer to the link @whuber provided to see a detailed explanation – godspeed Feb 24 '17 at 22:55
  • 1
    I read through the responses in the link, and most answers say that the main effects term should be included if there's an interaction that uses the main effects term. I guess my question is: what happens to the model when the main effects term is omitted even though it is part of an interaction? – mistersunnyd Feb 24 '17 at 23:20
  • Yes, the intercepts are the same in the second model. Specifically, the expected value is the same for all groups when ppgdp=1. That's probably a pretty odd constraint on the model, and so I think you would have to have a very good reason to use the second form of the model rather than the first. – The Laconic Feb 25 '17 at 03:30

0 Answers0