Abstract:
|
Double truncation often arises in survival data, when the data is only observed if it falls within a particular time interval. If we do not account for double truncation, the estimator of the survival function will be biased. In this presentation, we first show that the standard Kaplan-Meier analyses are in general biased in analyzing doubly truncated data through simulations. Second, we compare a semi-parametric and non-parametric estimator of the survival function. The semi-parametric estimator assumes that the truncation times follow a parametric distribution, and makes no distributional assumptions about the survival times. The non-parametric estimator makes no distributional assumptions about the truncation or survival times. Through extensive simulations, we show that the semi-parametric estimator is unbiased and is more efficient than the non-parametric estimator when the distributions of the truncation times are correctly specified. However unlike the non-parametric estimator, the semi-parametric estimator is not robust to misspecification of the truncation times and yields a biased estimate of the true survival times. We apply both of these methods to mental health data.
|
ASA Meetings Department
732 North Washington Street, Alexandria, VA 22314
(703) 684-1221 • meetings@amstat.org
Copyright © American Statistical Association.