Content area
Full text
psychometrikavol. 80, no. 3, 811833 September 2015doi: 10.1007/s11336-014-9413-1
http://crossmark.crossref.org/dialog/?doi=10.1007/s11336-014-9413-1&domain=pdf
Web End = http://crossmark.crossref.org/dialog/?doi=10.1007/s11336-014-9413-1&domain=pdf
Web End = http://crossmark.crossref.org/dialog/?doi=10.1007/s11336-014-9413-1&domain=pdf
Web End = http://crossmark.crossref.org/dialog/?doi=10.1007/s11336-014-9413-1&domain=pdf
Web End = RATIONALE AND APPLICATIONS OF SURVIVAL TREE AND SURVIVAL ENSEMBLE METHODS
Yan Zhou
UNIVERSITY OF CALIFORNIA, LOS ANGELES
John J. McArdle
UNIVERSITY OF SOUTHERN CALIFORNIA
Classication and Regression Trees (CART), and their successorsbagging and random forests, are statistical learning tools that are receiving increasing attention. However, due to characteristics of censored data collection, standard CART algorithms are not immediately transferable to the context of survival analysis. Questions about the occurrence and timing of events arise throughout psychological and behavioral sciences, especially in longitudinal studies. The prediction power and other key features of tree-based methods are promising in studies where an event occurrence is the outcome of interest. This article reviews existing tree algorithms designed specically for censored responses as well as recently developed survival ensemble methods, and introduces available computer software. Through simulations and a practical example, merits and limitations of these methods are discussed. Suggestions are provided for practical use.
Key words: survival trees, random forests, survival analysis, statistical learning, recursive partitioning..
1. Introduction
Survival analysis is a branch of statistical methods for investigating event occurrence whether events occur and when events occur. Survival tree and survival ensemble methods are statistical learning techniques adapted to right-censored survival data. The counterparts of these techniques for more general categorical and continuous outcomesClassication and Regression Trees (CART; Breiman, Friedman, Olshen, & Stone, 1984), bagging (Breiman, 1996) and random forests (Breiman, 2001), are better known and have promising merits (Strobl, Malley, & Tutz, 2009). There is a strong motivation for the adaptation of these methods to the survival contexts, because questions about the occurrence and timing of events arise throughout psychological and behavioral sciences (see Singer & Willett, 1991, 2003), especially in longitudinal studies. For example, researchers investigating the course of alcohol abuse are interested in the onset of the disorder (DeWit, Adlaf, Offord, & Ogborne, 2000) as well as post-treatment relapse (Mertens, Kline-Simon, Delucchi, Moore, & Weisner, 2012). Industrial and organizational psychologists study the rate and timing of employee turnover (e.g., Morita, Lee, & Mowday, 1993). Developmental psychologists ask the attainment of developmental milestones, for instance, the age of the acquisition of gender labeling...