I want to know how to interpret the quadratic-by-quadratic interaction (e.g., X^2*W^2). I have looked for textbooks or scholarly articles on this issue, but all I found is “Statistics methods and applications” a book written by Pawel Lewicki and Thomas Hill (2012), which just briefly noted that:

“… the interactions by the quadratic components can be interpreted as half the difference between the quadratic main effect of one factor at the respective settings of another; that is, either the high or low setting (quadratic by linear interaction), or the medium or high and low settings combined (quadratic by quadratic interaction).” (p.223)

I wonder if there would be an expert who can help me to understand how to interpret the quadratic-by-quadratic interaction, hopefully with a bit more detailed explanation than Lewicki and Hill (2012) and with an example.

Let’s say we would like to know the effects of providing financial incentives (=X) and interesting tasks (=W) on workers’ happiness (=Y).

Equation: Y = a + b1X + b2X^2 + b3W + b4W^2 +b5XW + b6X^2W + b7XW^2 + b8X^2W^2

In this case, for the quadratic-by-linear interaction terms, I think, depending on their signs, we could say, for example, for b6X^2W, that providing interesting tasks (W) to workers can flatten or steepen the financial incentives―happiness curve.

However, I am unsure how to interpret b8X^2W^2 in this case.

Thank you so much for your help in advance!

  • 2
    $\begingroup$ I would choose an interpretation suitable for the context: that is, it would depend on the other terms involving $X$ and $W$ that appear in the model, as well as on the objective, the data, and the audience. If you have a particular model in mind, could you please share that with us? $\endgroup$
    – whuber
    Dec 28, 2022 at 15:25
  • $\begingroup$ Thank you whuber for your suggestion! I've included an example and an equation :-) $\endgroup$
    – user376288
    Dec 28, 2022 at 23:58
  • $\begingroup$ Part of the understanding you seek might reside in a theoretical analysis of multivariate polynomial models, such as the one I gave at stats.stackexchange.com/a/408855/919. $\endgroup$
    – whuber
    Dec 29, 2022 at 2:11


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.