Week 4 Lecture Slides (1)

# If j 1 actuarial statistics week 4 the cox regression

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

This is the end of the preview. Sign up to access the rest of the document.

Unformatted text preview: ikelihood estimate of β . (for a justiﬁcation, see K&M, section 8.8, in conjunction with Technical Note 2 on page 258) Finally, since ˆ ˆ S0 (t ) = exp{−Λ0 (t )} we have 35/45 T ˆ ˆ S (t ) = S0 (t )exp(bZ ) . Actuarial Statistics – Week 4: The Cox Regression Model Diagnostics for the Cox regression model 1 Introduction 2 Main assumptions 3 On the proportionality of hazard rates 4 Estimation of the regression parameters β 5 Hypothesis tests on the β ’s 6 Estimation of the full survival function 7 Diagnostics for the Cox regression model 35/45 Actuarial Statistics – Week 4: The Cox Regression Model Diagnostics for the Cox regression model Diagnostics for the Cox model We will discuss two types of diagnostics: 1 goodness-of-ﬁt via Cox-Snell residuals 2 appropriateness of the proportionality assumption via graphica techniques Notation: Cj : right censoring time δj = 0 means censoring, that is, Xj = 36/45 Cj Tj if δj = 0 . if δj = 1 Actuarial Statistics – Week 4: The Cox Regression Model Diagnostics for the Cox regression model Cox-Snell residuals The Cox-Snell residuals are deﬁned as p ˆ ˆ ej = − log(S (Xj ; zj )) = Λ0 (Xj ) exp( T bk zjk ) k =1 where b is the MLE of β , and ˆ Λ0 (t ) is Breslow’s estimator of the cumulative baseline hazard rate. If the model is correct and the b ’s (MLEs) are close to the true values of β , then the Cox-Snell residuals ej ’s behave as a censored sample from a unit exponential distribution. 37/45 Actuarial Statistics – Week 4: The Cox Regression Model Diagnostics for the Cox regression model Proof 1 − S (Tj ; Zj ) ∼ Uniform (0,1) So S (Tj ; Zj ) ∼ Uniform (0,1) Deﬁne Ej = − log(S (Tj ; Zj )). Then Ej ∼ Exponential(1) Deﬁne C j = − log(S (Cj ; Zj )). Notice that − log(S (Xj ; Zj )) = − log(S (Tj ; Zj )) if Tj ≤ Cj i .e . Ej ≤ C j − log(S (Cj ; Zj )) if Tj > Cj i .e . Ej > C j Consider Ej as a new lifetime r.v. and C j the corresponding censoring time. Then − log(S (Xj ; Zj )) is just a right censored, exponentially distributed version of Ej with censoring time C j . 38/45 Actuarial Statistics – Week 4: The Cox Regression Model Diagnostics for the Cox regression model To check the goodness of ﬁt of a Cox regression model, we need to check wether the Cox-Snell residuals: p ˆ ˆ ej = − log(S (Xj ; zj )) = Λ0 (Xj ) exp( T bk zjk ) k =1 behave as samples from a unit exponential random variable. To check whether ej behaves as a sample from a unit exponential, compute the Nelson-Aalen estimator of the cumulative hazards ˆ rate of Ej s : ΛE (ej ) ˆ A plot of ΛE (ej ) versus ej should be roughly straight line through the origin with a slope of 1 39/45 Actuarial Statistics – Week 4: The Cox Regression Model Diagnostics for the Cox regression model Example 2.5 2.0 1.5 1.0 0.5 0.0 estimated cumulative hazard function of the residuals 3.0 For the Cox regression model, a Cox-Snell residual plot is given as below. Comment on the overall ﬁt of the Cox model. 0.0 40/45 0.5 1.0 1.5 residuals 2.0 2.5 3.0 Actuarial Statistics – Week 4: The Cox Regression Model Diagnostics for the Cox regression model Assessment of the proportionality assumption To check for proportional haza...
View Full Document

## This document was uploaded on 04/03/2014.

Ask a homework question - tutors are online