I have two categorical values in an normal linear model and the mean value specification is that they both have an effect but there is no interaction. In R, this was modelled as lm(x~ A + B).

I would now like to test whether one of the factors (A) has an effect at all. I am unsure how to do this. Do I compare my model A + B to the model which only assumes B has an effect (so, a F-test between lm(x ~ A + B) and lm(x ~ B))?

Or, do I compare the model where I assume A does have an effect and compare it to the model where we assume there isn't any effect at all (so, F-test between lm(x ~ A) and lm(x ~ 1))?

What's "best", or what differences are there?

  • $\begingroup$ what are your scale values for A and B and which regression you have tried. Give output. $\endgroup$ – Subhash C. Davar May 12 '16 at 10:31
  • 2
    $\begingroup$ Possible duplicate of How to test hypothesis of no group differences? $\endgroup$ – Erik May 12 '16 at 10:38
  • $\begingroup$ The specific question stated at the end of this post seems different to me. I think this could stay open. $\endgroup$ – gung May 12 '16 at 11:20
  • $\begingroup$ Yes, I think my question has more to do with the specific combinations of models and F-tests and their differences, as opposed to a larger debate about whether F-tests are even the way to go. $\endgroup$ – Michaelel May 12 '16 at 11:43

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.