Premium
A model selection tool in multi‐pollutant time series: the Granger‐causality diagnosis
Author(s) -
Pitard A.,
Viel J. F.
Publication year - 1999
Publication title -
environmetrics
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.68
H-Index - 58
eISSN - 1099-095X
pISSN - 1180-4009
DOI - 10.1002/(sici)1099-095x(199901/02)10:1<53::aid-env335>3.0.co;2-g
Subject(s) - multicollinearity , econometrics , autoregressive model , granger causality , model selection , feature selection , pollutant , causality (physics) , statistics , mathematics , computer science , regression analysis , machine learning , ecology , physics , quantum mechanics , biology
When studying the effects of atmospheric pollutants on health, multicollinearity often impedes the simultaneous study of various pollutants levels or entails a large imprecision in the estimation of their parameters. Hence there is a need for methods to reduce the dimensionability of the set explanatory variables. Causality analysis provides a way to explore relationships between explanatory variables in a preliminary analysis and then to select an adequate explanatory variables subset. An approach initially developed in econometrics and based on causality Granger's definition is used. Granger's causality tools which deal with stationary series links, include the vector autoregressive (VAR) process. Their applicability to the epidemiologic field is explored and an extension of the method providing a selection model rule is proposed. As an example, we used this method in a first step to describe the relationships between daily pollutants (NO, NO 2 , O 3 ), and in a second step to assess the proper effects of these pollutants on children's lengths of hospital stay. The results of this study add some evidence to the significant effect of nitrogen oxide on length of hospital stay. Copyright © 1999 John Wiley & Sons, Ltd.