SOME METHODS TO ADDRESS COLLINEARITY AMONG POLLUTANTS IN EPIDEMIOLOGIC TIME-SERIES

The aim of this paper is to provide accurate estimation methods for re gression models used in epidemiological time series to deduce quantita tive morbidity relationships. Such models often include highly correla ted variables (pollutant levels and climatic conditions) as well as la gged and unlagged values of the same variables (which also show a high collinearity due to the stochastic dependency of consecutive measurem ents). We first describe some methods to detect and assess multicollin earity. We recall the drawbacks of usual methods of estimation, and th en after briefly mentioning traditional solutions, we explore three al ternative methods accounting for multicollinearity: Sclove's estimatio n; Almon's method; and a combination of Almon's method and principal c omponents procedure. We compare these methods in obtaining efficient e stimators on environmental epidemiological data (children's hospital a dmissions as dependent variable and unlagged and lagged values of outd oor temperature, SO,, NO and CO as explanatory variables). (C) 1997 by John Wiley & Sons, Ltd.