Analysis of Survival Data Using a Partially Linear Single Index Survival Model via Accelerated Failure Time Model

• Activity Number: 257
• Type: Contributed
• Date/Time: Monday, August 5, 2013 : 2:00 PM to 3:50 PM
• Sponsor: Section for Statistical Programmers and Analysts
• Author(s): Asheber Sewalem*+ and Antony F. Desmond and Radhey S Singh and Xuewen Lu
• Companies: AAFC-CDN and University of Guelph, Department of Mathematics and Statisitcs and University of Guelph, Department of Mathematics and Statisitcs and University of Calgary
• Keywords: Survival, PLSISM,Weibull, Bootstrap and Dairy Cattle

Abstract

In many practical situations the linear model is not complex enough to capture the underlying relationship between the response variable and its regressors. This paper explores this association in dairy cattle breeding data using the partially linear single-index model via the accelerated failure time model in addition to the ordinary Weibull model. Calves survival data were used this study. Each calf record contains the following information, survival time till weaning, type of birth (TB), calving ease score (CE), season of birth (SEAS), origin of farm or herd (HRD), number of treatments received (NTR), weight at the time of event taken (BWT), total volume of colostrums (TVOL) and serum total protein (TP) g/dl. Calves that survived up to 120 days were considered as censored. BWT, TVOL and TP were included in the nonparametric vector in the PSLISM model. The results show that the estimates of the parametric component are similar in the two models. However, the estimates of the nonparametric component differ from paramteric analysis. This difference may be attributed largely to the nonlinearity of the estimated function indicating the standard linear model does not adequately capture the underlying association between the response and regressors in this study.