Rather than needlessly dichotomise or bin your stimuli, you might want to look at the effect of a continuous variable, while controlling for a confounding variable. For example, rather than comparing young, middle-aged, and old faces in a face-judgement task, you might want to vary age continuously. One solution could be minimising mutual information. Just as distribution-wise matching does not assume a distributional shape (e.g., normal distribution), mutual information does not assume anything about the shape of the relationship (e.g., linearity).



Mutual information is a measure of how much information two variables have in common. If we minimise the mutual information between an independent variable (e.g., age) and a control variable (e.g., face width, face height), we can reduce the impact of the control variable on our outcome.

Here’s one short example showing how you might minimise mutual information.

The code uses the infotheo package to calculate mutual information. You can install infotheo with:

install.packages("infotheo")


Title/Link No. Conditions No. Control Variables
01 - Minimising Mutual Information 2 1

Download code for all examples