Weighted Least Squares

Also known as · WLS

OLS after dividing every term by hi\sqrt{h_i}, where Var(ui)=σ2hiVar(u_i)=\sigma^2 h_i, which is inverse-variance weighting: noisy observations get less weight. OLS is the equal-weight special case, and WLS is itself a special case of Generalized Least Squares.

When to use

Use it when the form of the Heteroskedasticity is known: the transformed model satisfies A4, so the estimates are efficient (BLUE), though the coefficients and R2R^2 change. Check that the leftover error variance collapses to a constant.

See Lecture 12.

Appears in