C19. Fisher scoring and IRLS

Topic: Introduction to generalised linear models · more questions on this topic
‹ C18 C20 ›

What is Fisher scoring, and how is it related to iteratively reweighted least squares in GLMs?

Worked answer (click to show)

Definitions/notation. With score S and nonsingular expected information I, Fisher scoring iterates \beta^{(t+1)}=\beta^{(t)}+ I(\beta^{(t)})^{-1}S(\beta^{(t)}). For GLMs this can be written as iteratively reweighted least squares (IRLS): repeatedly update working responses and weights, then solve the resulting weighted least-squares problem.