13_AS_2_lec_a

Comparing survival functions introductory example a

Info iconThis preview shows page 1. Sign up to view the full content.

View Full Document Right Arrow Icon
This is the end of the preview. Sign up to access the rest of the document.

Unformatted text preview: Maximum Likelihood with censoring The likelihood as a function of the discrete hazard function Kaplan-Meier (product limit) estimator Nelson-Aalen estimator 4 Comparing survival functions Introductory example A hypothesis test Special cases 15/42 Actuarial Statistics – Module 2: Non-parametric methods Estimating the lifetime distribution: non-parametric approach The likelihood as a function of the discrete hazard function Discrete hazard function Suppose F (t ) corresponds has positive probability masses at and only at the points t1 < t2 < . . . < tk (we’ve just shown this had to be the case). Define the discrete hazard function as λj = Pr [T = tj |T ≥ tj ] , (1 ≤ j ≤ k ) . Then S (t ) = 1 − F (t ) = (1 − λj ) j :tj ≤t 15/42 Actuarial Statistics – Module 2: Non-parametric methods Estimating the lifetime distribution: non-parametric approach The likelihood as a function of the discrete hazard function Proof Note that since t0 = 0 we have P (T > t0 ) = P (T > 0) = 1 Then P (T > t1 ) = P (T = t1 ) P (T > t1 ) =1− = 1 − λ1 P (T > t0 ) P (T > t0 ) More generally 1 − λj = P (T > tj ) P (T > tj ) = P (T ≥ tj ) P (T > tj −1 ) and therefore (by induction) S (t ) = P (T > t ) = j :tj ≤t 16/42 P (T > tj ) = P (T > tj −1 ) (1 − λj ) j :tj ≤t Actuarial Statistics – Module 2: Non-parametric methods Estimating the lifetime distribution: non-parametric approach The likelihood as a function of the discrete hazard function Now let us rewrite the total likelihood as dj k k F (tj ) − F tj− 1 − F tj− L= − 1 − F tj j =0 j =1 dj cj [1 − F (tjl )] l =1 where F (0) = 0 and d0 = 0. This can be simplified since j −1 1 − F (tj− ) = 1 − F (tj −1 ) = S (tj −1 ) = (1 − λi ) = i :ti ≤tj −1 (1 − λi ) i =1 and j 1 − F (tjl ) = 1 − F (tj ) = S (tj ) = (1 − λi ) = i :ti ≤tj 17/42 (1 − λi ) i =1 Actuarial Statistics – Module 2: Non-parametric methods Estimating the lifetime distribution: non-parametric approach The likelihood as a function of the discrete hazard function Now note that F (tj ) − F tj− 1 − F tj− = λj and that nj = dj + cj + nj +1 forj = 1, · · · , k Hence, k [λj ]dj [1 − λj ]nj −dj L= j =1 which is in ‘binomial’ form and has maximum likelihood estimator ∧ λj = 18/42 dj nj (1 ≤ j ≤ k ) Actuarial Statistics – Module 2: Non-parametric methods Estimating the lifetime distribution: non-parametric...
View Full Document

Ask a homework question - tutors are online