Bayesian Random Exposure Poisson Regression Models for Evaluating the Safety Impact of Cellphone Visual-Manual Tasks

• Activity Number: 496
• Type: Contributed
• Date/Time: Wednesday, August 12, 2015 : 8:30 AM to 10:20 AM
• Sponsor: Transportation Statistics Interest Group
• Author(s): Youjia Fang* and Feng Guo
• Companies: Virginia Tech Transportation Institute and Virginia Tech
• Keywords: Bayesian ; random exposure ; Poisson regression ; simulation ; model fitting ; visual-manual task

Abstract

Poisson regression is a state-of-practice method for modeling event data. Exposure is a critical component, as it is the base for normalizing event frequencies to event rate.Traditionally, exposure is considered as fixed and obtained by direct observation or estimation. However, when exposure is difficult to observe and the estimation is required, an uncertainty exist with the estimation process. Failure to incorporate the uncertainty could lead to biased estimation and jeopardize the validity of statistical inference. This paper developed a Bayesian random exposure method to accommodate the uncertainty associated with the estimation of exposure. The posterior of the exposure reflects the randomness associated with exposure, and the posteriors of regression parameters inherently incorporate the uncertainty of the exposure. Simulation studies showed that random exposure method successfully incorporated uncertainty of exposure and achieved better model fitting performance than traditional fixed exposure model. We implemented proposed method to Cellphone Pilot Analysis study data. Results showed that text-related visual-manual tasks are associated with increasing driving risk.