INDOOR AIR QUALITY: MULTIVARIATE ANALYSES OF THE RELATIONSHIP BETWEEN INDOOR AND OUTDOOR AEROSOLS (POLLUTION, RESIDENTIAL, WISCONSIN, OHIO)

A unique multivariate data set incorporating simultaneous indoor and outdoor measurements of sixteen air contaminants at ten homes has been used to investigate the contribution of outdoor concentrations to indoor aerosol variability, and to characterize indoor source contribution to the indoor concentrations. The data were available from an earlier field study of particle and gas concentrations outside and inside five homes in each of two cities: Portage, Wisconsin and Steubenville, Ohio. Three distinct multivariate statistical techniques were used sequentially in the research, successively building on the results and interpretations as they developed. Cluster analysis was selected as the initial method for partitioning the variables into subgroups comprised of highly intercorrelated variables. It was used to explore interrelationships and discover patterns in the data set. For most of the ten home sites studied, four to five clusters were formed, containing eight to fourteen of the measured variables. Significant site-to-site variability was evident in both cities, however within sites, indoor clusters had similarities to the outdoor clusters. Principal component analysis was next performed on the Portage data, reduced in dimension to avoid problems of singularity in the data matrix. Dominant influences on the composition of both the indoor and outdoor aerosol were revealed by both the cluster and principal component analyses. This similarity is important because hierarchical cluster analysis cannot distinguish secondary associations among variables. The principal component analyses results were used to attribute predominant indoor and outdoor sources, including cigarette smoke, wood stove, road dust, and urban combustion sources. Finally, multiple regression analysis was preformed to relate outdoor pollutant concentrations to a composite index of the indoor aerosol as represented by the orthogonal rotations of the indoor principal components. Special treatment of extreme value observations was required to reduce their influence on the factor structure and the regression results. The method for identifying extreme values may have application in other principal component analyses of air quality data. The final regression analyses showed that the influence of outdoor sources can be identified in the indoor data, in particular for the older homes, presumably having higher rates of air infiltration. The research reported here indicates that this multivariate analysis framework is preferable to single univariate analysis (e.g., indoor/outdoor ratio analysis) in evaluating the influence of outdoor aerosols and indoor sources on indoor air quality data.