Received: 2 October 2016 Revised: 16 October 2017 Accepted: 17 October 2017 DOI: 10.1002/env.2484 RESEARCH ARTICLE Accounting for uncertainty in source-specific exposures in the evaluation of health effects of pollution sources on daily cause-specific mortality Eun Sug Park 1 Man-Suk Oh 2 1 Texas A&M Transportation Institute, College Station, TX 77843-3135, U.S.A. 2 Department of Statistics, Ewha Womans University, Seoul 120-750 South Korea Correspondence Eun Sug Park, Texas A&M Transportation Institute, 3135 TAMU, College Station, TX 77843-3135, U.S.A. Email: e-park@tamu.edu Man-Suk Oh, Department of Statistics, Ewha Womans University, Seoul 120-750, Korea. Email: msoh@ewha.ac.kr Funding information Health Effects Institute, Grant/Award Number: R-82811201; National Research Foundation of Korea, Grant/Award Number: 2016R1A2B4008914 Assessment of source-specific health effects has received growing attention in air pollution epidemiology over the past decade. Regardless of inherent uncertainty in the assessment of source-specific exposures, only a handful of previous stud- ies coped with model uncertainty in source apportionment and/or accounted for exposure measurement error in the estimation of health effects, all under normal health outcome models. We present a source-specific health effects evaluation approach within a Bayesian framework that can handle both parameter uncer- tainty and model uncertainty in source apportionment under Poisson health outcome models for low daily mortality count data. While the use of a Poisson health outcome model is apparently more appropriate for low daily mortality count data for which normal approximation is not justified, it introduces addi- tional complexity in estimating model uncertainty. We handle this complexity by introducing appropriate latent variables. The proposed method is illustrated with simulated data and daily ambient concentrations of the chemical composi- tion of fine particulate matter (PM 2.5 ), weather data, and counts of deaths from pneumonia in older adults (≥65 years of age) in Houston, Texas, from January 2002 to August 2005. KEYWORDS exposure measurement error, model uncertainty, mortality from pneumonia, multipollutant approach, PM health effects 1 INTRODUCTION In air pollution epidemiology studies, the time series studies are often employed for estimating the acute effects of air pollution on health. In time series studies, each observation of the outcome could be a count indicating the number of deaths or hospital admissions that occurred on day t. With time series of counts, the most commonly used model is the log-linear Poisson model, taking the health outcome variable to be Poisson with mean including the terms for the exposure of interest (e.g., air pollution levels) and other potential confounders (Peng & Dominici, 2008). Estimating health effects of source-specific exposures rather than pollutant-specific exposures has received increas- ing attention in air pollution epidemiology over the past decade (see, e.g., Bell et al., 2014; Dominici, Peng, Barr, & Belle, 2010). From a regulation standpoint, assessing the health effects of specific sources or group of sources (i.e., source-specific health effects) is more advantageous than assessing the health effects of individual pollutants Environmetrics. 2018;29:e2484. wileyonlinelibrary.com/journal/env Copyright © 2017 John Wiley & Sons, Ltd. 1 of 15 https://doi.org/10.1002/env.2484