Ying Qing Chen

On Measuring Functional Attribution

Statistical methods and inferences often concern measuring association between an outcome variable and covariates. In medical studies, for example, logistic regression models are frequently used to study an association between a disease outcome and its covariates of risk factors. Without taking into account the distribution of covariates, however, the association itself is yet to sufficiently measure a collective population impact due to the covariates. In public health literature, researchers have used the notion of attributable risk to measure the population impact of covariates, but mostly for binary outcome variables. In this article, we extend this notion and propose more general time-varying attributable risk functions (ARF) to measure the so-called functional attribution in time-to-event analysis. We further develop nonparametric and semiparametric model-based procedures to estimate, compare and project the ARFs. In addition to numerical simulation studies, our developed methods are applied to an HIV/AIDS behavior intervention trial conducted by the HIV/AIDS Prevention Trial Network (HPTN).