Modeling exposures to air pollutant mixtures: Application of copula methods to the RIOPA data


Distributions of personal exposures to volatile organic compound (VOC) mixtures are modeled using multivariate methods including copulas. The data used are the personal VOC exposures collected for 306 participants in the RIOPA study. Factor analyses identified five groups of compounds: gasoline-related compounds (Mix 1); terpenes (Mix 2); chlorinated cleaning solvents (Mix 3); chlorine disinfection by-products (Mix 4); and foam emissions (Mix 5). Multivariate distributions of each group were fitted to four candidate copulas (Gaussian, Gumbel, Clayton, Frank). Gaussian copulas best fit Mixes 1, 3 and 4, and Clayton copulas best fit Mixes 2 and 5. Binary mixtures mostly followed Clayton copulas. This work indicates that co-exposures to extreme concentrations may be overestimated using standard multinormal distributions. More generally, copulas can model dependency structures with great flexibility given the many families of copulas available, and they allow any forms of marginal distributions, which is advantageous over conventional multinormal distributions.

Publication Title

12th International Conference on Indoor Air Quality and Climate 2011

This document is currently not available here.