Track 9 · Correlation and Regression · Lesson 11 of 15
Multicollinearity
Recognise multicollinearity, measure it with variance inflation factors, and explain why it makes individual coefficients unstable while leaving predictions intact.
An estate agent fits house prices on floor area and number of bedrooms and is puzzled: bedrooms, obviously valuable, gets a small and non-significant coefficient — and adding bathrooms even turns it negative. Nothing is wrong with the arithmetic. The predictors carry almost the same information, and the model cannot tell them apart.
Multicollinearity means that some predictors are strongly correlated with each other (or one is nearly a combination of several others). Each slope in a multiple regression is estimated from the variation in its predictor that the other predictors do not already explain. When little such variation is left, the slope rests on thin evidence: its standard error balloons and its value can swing wildly between samples or model specifications.
Included with a Formula Quest membership
It has 7 graded exercises with feedback, hints and worked solutions. After it you will be able to:
- Recognise multicollinearity from correlated predictors and unstable coefficients
- Calculate and interpret a variance inflation factor
- Explain why multicollinearity inflates standard errors but leaves R² and predictions intact
- Choose a sensible response: accept, simplify, combine predictors, or collect better data