Questions: Least Squares Estimation

5 questions to test your understanding

Score: 0 / 5
Question 1 Multiple Choice

A data scientist says: 'We can't use least squares regression here — our residuals clearly aren't normally distributed.' Is this objection valid?

AYes — least squares is only mathematically valid when errors follow a normal distribution
BYes — without normality, the slope and intercept formulas give different results
CNo — least squares gives the minimum sum of squared residuals regardless of the error distribution; normality is only needed for certain inferential guarantees like confidence intervals
DNo — least squares is always optimal regardless of the error distribution, so normality never matters
Question 2 Multiple Choice

Why does least squares minimize the sum of *squared* residuals rather than, say, the sum of absolute residuals?

ASquaring residuals is required by the central limit theorem
BSquaring ensures all residuals are positive so they don't cancel out
CSquaring yields a smooth, differentiable objective function with a unique closed-form solution, and it penalizes large deviations more heavily than small ones
DSquared residuals correspond exactly to the variance of the errors, which makes the estimator unbiased
Question 3 True / False

A regression model with R² = 0.95 is expected to make accurate predictions for new observations drawn from the same population.

TTrue
FFalse
Question 4 True / False

Least squares estimates are particularly sensitive to outliers because the squaring of residuals causes large deviations to contribute disproportionately to the objective function.

TTrue
FFalse
Question 5 Short Answer

Explain why minimizing squared residuals rather than absolute residuals is a deliberate design choice with real consequences, not just an arbitrary convention.

Think about your answer, then reveal below.