What does multicollinearity mean in regression analysis?

• A. Occurs when predictor variables have no correlation
• B. Occurs when two or more predictor variables are highly correlated independent of the outcome
• C. Occurs when there is heteroscedasticity
• D. Occurs when the sample size is too small

Answer: (B)

Multicollinearity is when two or more predictor variables are highly correlated with each other, independent of the outcome. In regression, we want predictors to provide unique, additive information about the response. When predictors carry overlapping information, it becomes hard to tell which variable is truly driving the observed effect, so the estimated coefficients become unstable and their standard errors inflate. This can lead to wide confidence intervals and changes in the apparent sign or magnitude of a predictor across samples, even if the model’s overall fit isn't drastically wrong. It’s about redundancy among the predictors, not about how they relate to the outcome itself. This is different from heteroscedasticity, which concerns non-constant residual spread, or from issues caused merely by a small sample size (though small samples can worsen problems once multicollinearity is present).