← Glossary

Multicollinearity

When independent variables move together so tightly a regression can't tell their effects apart

In one line

Multicollinearity happens when independent variables you put into a regression (say, spend on multiple channels) always move up and down together, so the model can't tell which one is actually driving the outcome.

Why it matters

This is a common trap in marketing data. If you always scale two channels together, revenue going up can't be cleanly attributed to one or the other — coefficients become unstable and can even flip sign. It's especially common in Marketing Mix Modeling, where channel-level contribution is being regressed.

How to spot it

Go deeper

See how multicollinearity distorts MMM results in What is MMM.

Related:What Is MMM: Measuring Channel Contribution Beyond Attribution Models