Online Program

Survival Analysis of Longitudinal Event History Data from Complex Samples (302532)

 View Presentation

Keywords: Discrete time models for event data, Kaplan-Maier Curves, Cox Proportional Hazards models, Complex Samples, Weighting; Pseudo-Maximum Likelihood; Survival analysis

Survival analysis, or event history analysis as it is often labeled in the social sciences, includes a range of statistical methods for analyzing the time at which “failures” or events of interest occur. In the context of population-based survey research, survival analysis problems arise in three primary ways: (1) through longitudinal observations on individuals in a panel survey; (2) by administrative record follow-up of survey respondents and (3) from retrospective survey measurement of events and the times that events occurred. This paper will present the general methodology for the application of survival analysis methods to these three forms of event history measurement involving complex sample surveys and population-based administrative records systems. Models and methods to be covered in this presentation include Kaplan-Maier estimation of survivor functions, the Cox Proportional Hazard Model, and Discrete Time Models for the Logistic and Complementary Log-Log link function. The presentation will focus on model specification, estimation, evaluation and inference to the population survival model. The methods will be illustrated through applications to two large data sets that include longitudinal measurement of the onset of adverse mental health conditions and related health outcomes. Examples will be drawn from the National Comorbidity Survey Replication (NCS-R), a 2002 national survey of mental health in the U.S. adult population and Army STARRS, a large ongoing longitudinal survey and administrative records based study of suicide and adverse mental health outcomes among U.S. Army soldiers. Software illustrations will employ SAS, Stata, SUDAAN and the R system programs for the analysis of complex sample survey data.