Nonlinear Interaction Effects: Summary of Recommendations (from Mize 2019)
From Mize 2019:
- Include a product (interaction) term in the model if you want to investigate whether the effect of one independent variable is contingent on the level of a second independent variable
- Ignore the coefficient of the product term: It does not necessarily provide accurate information about the significance, magnitude, or even the direction of the underlying interaction effect on the predictions
- Plot the predictions to determine the nature of the underlying interaction effect on the metric of interest
- Determine the size and significance of the effects of interest using marginal effects, not regression coefficients
- Use tests of second differences (whether two marginal effects are equal) to determine whether an interaction effect is significant for specific values of interest of your independent variables
- Absent substantive or theoretically interesting values of the focal independent or control variables to test the interaction effect, use average marginal effects to summarize whether there is an interaction effect present on average in the data
How I used to feel when teaching interactions and cross-model comparisons in nonlinear models:

ASR Editors’ 2018 “Guidelines for Quantitative Submissions”
“The case is closed: don’t use the coefficient of the interaction term to draw conclusions about statistical interaction in categorical models such as logit, probit, Poisson, and so on.” (Mustillo, Lizardo, and McVeigh 2018 ASR)
Additional resources on interactions
- I cover nonlinear interaction effects in greater detail in Mize 2019
- There is also accompanying example Stata and R code for the article
- This is written as easy-to-follow template code
- In Stata, mecompare is now the easiest and most flexible way to test for interaction; this page recreates the Mize (2019) examples using mecompare
- There is also accompanying example Stata and R code for the article
- See my 2023 workshop slides for coverage of miscellaneous advanced topics in interaction effects. E.g.,
- 3-way interaction effects
- Interaction effects with many category nominal IVs
- Nominal X nominal interaction effects
- Statistical power in the context of interaction
- Joint/overall tests of significance
- Etc.