Durbin-Watson test assesses what?

• A. A test for multicollinearity
• B. A test for serial correlation of residuals
• C. A test for heteroscedasticity
• D. A test for normality of residuals

Answer: (B)

Durbin-Watson is about whether the residuals from a regression are serially correlated, meaning an error in one observation is related to the error in the next. This matters because ordinary least squares assumes the errors are independent. The statistic essentially tracks how different consecutive residuals are: if they’re independent, the value hovers around 2. A value much less than 2 indicates positive autocorrelation (errors tend to be similar from one point to the next), while a value much greater than 2 indicates negative autocorrelation (errors tend to flip signs). This test focuses on first-order, adjacent residuals and is not a test for multicollinearity, heteroscedasticity, or normality of the residuals.