Course contentsShow
← Data Science
Lesson 580 of 2,14515. Regression DiagnosticsFree lesson

What is Multicollinearity?

Define multicollinearity as high correlation between predictors and understand why it causes problems in regression.

What is Multicollinearity?

You'll learn: What multicollinearity is and why it creates problems when building regression models with multiple predictors.

The Core Idea

Multicollinearity occurs when two or more predictor variables in a regression model are highly correlated with each other. In other words, some of your independent variables are not truly independent—they move together in a predictable way.

Think of it like this: Imagine you're trying to predict house prices using both "square footage" and "number of rooms" as predictors. These two variables are likely highly correlated—bigger houses tend to have more rooms. When this happens, the regression model struggles to separate out the individual effect of each predictor because they're essentially telling overlapping stories about the same underlying information.

Why Multicollinearity Causes Problems

When predictors are highly correlated:

  1. Unstable coefficients: Small changes in your data can lead to large swings in the estimated regression coefficients
  2. Inflated standard errors: The uncertainty around coefficient estimates increases dramatically
  3. Difficulty interpreting effects: You can't confidently say "holding all else constant, X increases Y by..." because the correlated predictors don't truly vary independently
  4. Reduced statistical significance: Even though your overall model might fit well (good R-squared), individual predictors may appear non-significant

The regression model hasn't "broken"—predictions may still work reasonably well—but you lose the ability to understand and trust the individual contribution of each predictor.

Key Takeaway: Multicollinearity is high correlation between predictor variables that makes it difficult to isolate individual predictor effects, leading to unstable and unreliable coefficient estimates even though the overall model fit may seem fine.