New Methods for Studying Event Occurrence

Abstract

Early intervention researchers often want to know whether and when critical events occur. Questions about event occurrence can be difficult to answer because of censoring; not everyone in the sample experiences the target event during the period of data collection. Yet for findings to be unbiased and representative, statistical analyses must include both censored and uncensored cases. In this paper, we show how methods of survival analysis are ideally suited for studying event occurrence. Using data describing entry into preschool among children with special needs, we introduce the fundamental statistical quantities—the hazard and survivor function—-and we show how to construct formal statistical models of event occurrence over time. We then identify the different types of predictors that can be included in hazard models and comment on the importance of these models for early intervention research. Finally, we describe how researchers who fail to adopt a survival analysis perspective can be led astray, producing erroneous conclusions about the predictors of event occurrence.