I have two models: one predicting the sales for a cheap product and another predicting the sales of a more expensive product (both of the same company). Let us say that both models look like this:

Sales = a + b*advertising 

I want to test whether the effect of advertising on sales for one brand is higher/lower than the effect of advertising on sales for the other brand.

Does anyone know how I can formally test whether the difference is statistically signifcantly different?

A hint I got from someone is that I might consider using only one model. I have no clue however.

Many thanks in advance.

  • $\begingroup$ Have a look at this question. Does this help? $\endgroup$
    – lanenok
    Dec 7, 2014 at 17:48
  • $\begingroup$ Is this time series data? Are these 2 products substitutes? $\endgroup$
    – dimitriy
    Dec 8, 2014 at 8:59

1 Answer 1


This can be addressed by introducing a dummy variable and an interaction term.

Let $Y = \text{Sales}$

$A = \text{Advertising}$

$X = \left\{ \begin{array}{lr} 1 & \text{if Brand A}\\ 0 & \text{if Brand B} \end{array} \right.$

So we have continuous outcome $Y$, continuous covariate $A$, and dummy variable $X$.

We wish to determine whether the effect of advertising ($A$) on sales ($Y$) is the same between the two values of $X$. Graphically, this corresponds to the null hypothesis that regression lines for $X=0$ and $X=1$ are parallel.

We can test this hypothesis using the interaction model:

$$Y = \beta_0 + \beta_1 A + \beta_2 X + \beta_3 AX$$ which can also be written as $$Y = \beta_0 + \beta_1 A + X(\beta_2 + \beta_3 A)$$.

If $X=0$ (Brand B):

$$Y = \beta_0 + \beta_1 A$$

If $X=1$ (Brand A):

$$ \begin{align} Y &= \beta_0 + \beta_1 A + \beta_2 + \beta_3 A\\ &= (\beta_0 + \beta_2) + (\beta_1 + \beta_3) A \end{align} $$

Our null hypothesis is that the slopes of the two lines are the same, so $$\begin{align} H_0 : \beta_1 &= \beta_1 + \beta_3\\ \Leftrightarrow \beta_3 &= 0 \end{align}$$

This means that to address whether the effect of advertising on sales is the same between brands, we can perform a t-test for $\beta_3$ in the interaction model.


Your Answer

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

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