Day 7 – It Depends | Data Analysis for Neuroscientists

Topics Covered

Continuous by Categorical Interactions

Learn to model situations where the effect of a continuous variable (like flipper_length_mm) on the outcome depends on a categorical group (like species).

Adding Interaction Terms

Use * or : in R formulas. e.g. body_mass_g ~ flipper_length_mm * species to test if regression slopes differ.

Plotting Real Interactions

Visualize the exact meaning of interaction effects with ggplot2 using geom_smooth(method="lm") colored by group.

Scripts

interaction_penguins.R

A deep dive script on fitting additive vs. interaction models on the penguins dataset, along with plotting the slopes.

Download →

Key Concepts

Main effects vs. Interaction effects: An interaction means the slope of one predictor changes depending on the value of another predictor.
Reference Level Slopes: In a continuous-by-categorical interaction model, the main continuous effect (e.g. flipper_length_mm) is the slope for the reference category.
Interaction Coefficients: Terms like flipper_length_mm:speciesChinstrap represent the difference in slope compared to the reference baseline.
When "it depends" is the right answer: You can statistically prove that an effect is not uniform across all subdivisions in your data.

Homework

Run interaction_penguins.R and carefully read the model summary.
Fit a new interaction model predicting bill_length_mm with the interaction of bill_depth_mm * sex.
Identify if the interaction is statistically significant—does the relationship between bill depth and length depend on sex?
Create an interaction plot showing separate regression lines per sex using ggplot2.
Change the reference level for both sex and species using the relevel() function, then re-run the interaction models. Interpret how the base coefficients adjust!