|
Activity Number:
|
369
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Tuesday, August 4, 2009 : 2:00 PM to 3:50 PM
|
|
Sponsor:
|
Section on Nonparametric Statistics
|
| Abstract - #305502 |
|
Title:
|
Cox-McFadden Marginal Likelihood for Clustered Proportional Hazards
|
|
Author(s):
|
Jan Ondrich*+
|
|
Companies:
|
Syracuse University
|
|
Address:
|
Center for Policy Research, Syracuse, NY, 13066,
|
|
Keywords:
|
Cox marginal likelihood ; proportional hazards ; clustered observations ; duration baseline ; discrete choice ; GEV models
|
|
Abstract:
|
Cox (1972) describes a marginal likelihood for proportional hazards with independent durations that eliminates the duration baseline from the estimation. My study examines marginal likelihood for proportional hazards with clustered observations. I show that the problem of eliminating the duration baseline is isomorphic to the problem of relaxing the assumption of independent errors in an extreme-value stochastic utility model. McFadden (1978) describes discrete-choice GEV models with extreme-value disturbances for which the copula is consistent with stochastic utility maximization. My paper presents necessary and sufficient conditions for the duration baseline to be eliminated from the marginal likelihood for clustered proportional hazards. A sufficient condition is that the probability that the first ordered failure time is the first observed ordered failure time be a GEV probability.
|
