Say we have an OLS regression of $y$ on $x_1$ and $x_2$, where both $x_1$ and $x_2$ are independent from each other, and create the following regression: $$ y= \beta_0 + \beta_1 x_1 + \beta_2 x_2 + \beta_3 x_1 x_2 $$

How would one go about showing that the coefficient on interaction term, $\beta_3$, should be zero?

Regressing $x_1$ on $x_2$ will give a coefficient of 0, since $\hat{\beta}=cor(Yi,Xi)⋅\frac{SD(Yi)}{SD(Xi)}$, but I am not sure how to fit this in to what I am trying to show above, or if this is the right way of showing that.


2 Answers 2


Next to Dave's +1 numerical example, consider an applied one, explaining wages as a function of covariates like ability, gender etc.

It is well-documented that ability and gender both have an influence on wages (i.e., $\beta_1,\beta_2$ both are nonzero), because being more able is good for your earnings and there is such a thing as wage discrinimation. Also, we may assume that gender and ability are independent - women are no more/less able than men in general/on average.

That, however, does not rule out that there is an interaction with ability ($\beta_3\neq0$). For example, there might be more wage discrimination for high-skilled women than for low-skilled women, where wages for the latter group may involve less individual negotiation due to wage-setting negotiated via unions and smaller glass-ceilings effects etc.


This is not correct, and it is possible to simulate a counterexample, such as with R code.

x1 <- c(0,1,0,1,0,1,0,1)
x2 <- c(0,0,0,0,1,1,1,1)
y <- x1 + x2 + x1*x2 + rnorm(length(x1))
summary(lm(y ~ x1*x2))

This has independent $x_1$ and $x_2$ but also results in a nonzero coefficient on their interaction.


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.