Isn't it quite unrealistic to assume that they both affect each other in the same way? A one-unit increase in price both increases sales by B1 and the amount that a one-unit increase in ads increase sales by B3. But what if a one-unit increase in ads increases sales by B2 as well as the amount that a one-unit increase in prices increase sales, but by different amount than B3?
Hi, great video. I have one question, though, and am hoping you might give a clear answer. Imagine the coefficient on B3 were equal to, let's say, 50. This would tell us that the effect of one X on Y becomes stronger as values of the other X increase. However, could you give a single sentence that tells us exactly what the value of 50 means? Is 50 a difference in coefficients on one of the X variables as the other X changes values?
Hi there, I'd say you're right :) it may be helpful to distinguish X1 X2, etc. so: The value of B3 in the example of @BenLambert indicates that an increase in X1 leads to a [insert value of B3; e.g. 50] points stronger (if negative ⇒ weaker) change in Y for every additional unit increase in X2.
Hello :) Does this method for interaction effect work even if you get 3 positive valued regression coefficients, or does two of the values have to be negative and the interaction coefficient positive for this modell to give consistent results? Can it be done with all 3 coefficients as positive or do they have to "cancel" each other out in the way you explain?
All coefficients can be positive! (reverse-recoding [i.e. x'=max-xi] one of the independent variables would already turn around the sign of the coefficient)