I am using indicator regression with an interaction term to compare the slopes to two groups.

Here is the form of the equation:

$Y = \beta_0 + \beta_1 X_1 + \beta_2 X_2 + + \beta_3 X_1 X_2$

$X_2 $ is my dummy variable (1 or 0).

Based on how I set up the regression:

When $X_2 = 0$, then the $slope_0 = \beta_1 $.

When $X_2 = 1$, then the $slope_1 = \beta_1 + \beta_3 $.

What I would like to determine is whether there is a test to determine whether $slope_1$ is significantly different than zero? I know I can split the dataset into two (based on the dummy variable) and regress separately to determine whether $slope_1$ is significantly different than zero, but could I could determine this without the separate regression? Is there a test for this in R?

  • $\begingroup$ The two most obvious alternatives are to reparameterize so that your sum of coefficients in the original formulation becomes a single parameter under the reparameterized formulation (and from there you can just look at the p-value of the resulting coefficient estimate in the subsequent regression output), or you can perform a test of the general linear hypothesis. Both approaches (for a slightly different hypothesis) are discussed here. $\endgroup$
    – Glen_b
    Nov 28, 2016 at 11:02
  • $\begingroup$ Thanks, that seems like what I am looking for. Just to confirm, in the re-parameterization, the coefficients will not have much (physical) meaning, but the p-values will tell you whether the re-parameterized coefficients are significant or not? $\endgroup$
    – ken
    Nov 30, 2016 at 22:07
  • $\begingroup$ Sometimes it's the case that the reparameteried coefficients don't have a direct physical meaning (though they're usually interpretable as sums or differences of coefficients), So sure, you might in some situation say "well, $\beta_1+\beta_3$ doesn't have direct physical meaning" but ... "$\text{slope}_1$ when $X_2=1$" seems like a readily understood concept. $\endgroup$
    – Glen_b
    Nov 30, 2016 at 22:15
  • $\begingroup$ Possible duplicate of Is there a hypothesis test for B1 > B2 in multiple regression? $\endgroup$ Aug 9, 2018 at 17:25
  • $\begingroup$ Possible duplicate of Testing whether two regression coefficients are significantly different (in R ideally) $\endgroup$
    – mdewey
    Aug 10, 2018 at 12:31


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Browse other questions tagged or ask your own question.