Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.

# Re: st: question about the interaction term

 From David Hoaglin To statalist@hsphsun2.harvard.edu Subject Re: st: question about the interaction term Date Wed, 24 Apr 2013 23:52:05 -0400

```For the moment, let's set aside the question of whether the
interaction term is significant.

We can write out the model in the two cases (B=0 and B=1).
If B=0, you have y = constant + A
If B=1, you have y = constant + A + B + A*B
That is, when B=0, the model provides an intercept and a slope against
A; and when B=1, it provides a separate intercept (constant +
coefficient of B) and slope against A (coefficient of A + coefficient
of A*B).  The test that you quoted says that this slope does not
differ significantly from 0.  That result is not inconsistent with
having a negative coefficient for A that differs significantly from 0.

David Hoaglin

On Wed, Apr 24, 2013 at 10:46 PM, ZHVictor <victerzj2@hotmail.com> wrote:
> I am running probit model with an interaction term. The regression is like this
> y=constant+A+B+A*B
> A is a continuous variable and B is a dummy variable with either zero or one.
> A has a significant and negative coefficient but A*B has an insignificant and positive coefficient.
> However, when I write "test A+A*B=0" on stata, stata reports "Prob > chi2 =    0.1718". That means when B=1, the coefficient of A becomes insignificant. However, When B=0, the coefficient of A is significant.
> However, I think if interaction term is insignificant, A should have the same negative and significant coefficient no matter whether B is zero or one.
> I feel the question is a little bit stupid but I just can not figure out why.
> Any comment is welcome.