Course contentsShow
← Data Science
Lesson 581 of 2,14515. Regression DiagnosticsFree lesson

Symptoms of Multicollinearity

Identify warning signs: high R² but insignificant coefficients, unstable estimates, and wrong coefficient signs.

Symptoms of Multicollinearity

You'll learn how to spot multicollinearity by recognizing its telltale warning signs in your regression model.

The Problem It Creates

When predictors in your regression model are highly correlated with each other (multicollinearity), the model struggles to isolate their individual effects. This confusion manifests in specific, recognizable symptoms that should raise red flags during your analysis.

Four Key Warning Signs

1. High R² but Insignificant Coefficients

Your model explains a lot of variance (high R-Squared), yet individual t-tests show most predictors aren't statistically significant. It's like a team winning games, but you can't tell which players are actually contributing—they're too intertwined to separate.

2. Unstable Coefficient Estimates

Adding or removing one observation, or swapping in a similar predictor, causes coefficient values to swing wildly. The model's estimates become unreliable and hypersensitive to small changes in your data.

3. Wrong Coefficient Signs

You know from domain knowledge that age should increase insurance costs, but your model shows a negative coefficient. Multicollinearity can flip signs because the correlated predictors are "stealing" each other's explanatory power.

4. Large Standard Errors

The standard errors of your coefficient estimates balloon, making confidence intervals extremely wide. The model is basically saying, "I'm very uncertain about these values"—a direct result of not being able to separate correlated predictors.

Why This Matters

These symptoms don't mean your data is bad—they mean your model can't reliably attribute effects to individual predictors. Predictions may still work, but interpretation becomes dangerous.

Key Takeaway: Watch for high R² with insignificant predictors, unstable estimates, wrong signs, and inflated standard errors—these are your early warning system for multicollinearity problems.