Consider a simple linear regression with an interaction term:

$Y=b_0 + b_1X +b_2Z+b_3XZ$

where $X$ is continuous and $Z$ is a dummy.

I want to find out whether $X$ has a significant impact on $Y$ when $Z=1$

$b_3$ is negative and significant but $b_1$ is positive and not significant.

The total effect of $X$ on $Y$ when $Z=1$ (i.e. $b_1$ + $b_3$) is negative. How can I determine whether this total effect is significant or not?

  • $\begingroup$ One thing you could do is to "center" the $Z$ variable such that it has a value of 0 where there was 1 before, that way you will get effect of $X$ and $Y$ for when $Z$=0 (and 1 before transformation). $\endgroup$ – user2974951 Jan 7 '19 at 11:39

You should use an F-test for the linear hypothesis $H_0: b_1 + b_3 = 0$. You can implement that test on R by using the command linearHypothesis from the car package.

  • $\begingroup$ Thanks! I believe this is what is also called a Wald test? $\endgroup$ – user6441253 Jan 7 '19 at 11:52
  • $\begingroup$ Almost. The Wald is fairly similar to the F-test but uses different distributional assumptions that make it not as high-powered as the F-test. In other words, the Wald test is less likely to reject the null than the F-test. $\endgroup$ – Romain Ferrali Jan 10 '19 at 14:43

Your Answer

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

Not the answer you're looking for? Browse other questions tagged or ask your own question.