The dose-response function approach to modelling the health effects of air pollution

Following from research conducted in the USA there is renewed interest in measuring the health impacts of air pollution. Direct methods of valuing the impact of poor air quality on health are likely to suffer from a variety of problems, chiefly a poor appreciation of the biological impacts of air pollution. However, indirect methods fail to account for averting behaviour and consequently provide only a lower bound on true willingness to pay. Dose-response functions have been estimated using different sorts of statistical models with different statistical properties; and there are arguments for resorting to Poisson regression techniques rather than conventional ordinary least squares routines. The results which emerge may depend upon the nature of the data with time series data failing to capture chronic health effects. The main issue, however, is the overall competence of the statistical studies themselves. In particular there are concerns that the means used to deal with the problem of multicollinearity have biased the results; similarly for the technique of filtering the data to deal with seasonal fluctuations. There are also concerns about the dynamic specification of the models. Most empirical work is likely to involve the use of meta-analyses to combine studies drawn from other countries. But because dose-response functions rest upon chosen behaviour they depend upon baseline factors which underlie individual analyses. Given the magnitude of the claims currently being made concerning the health effects of poor air quality, further research is clearly warranted and should focus on panel data analyses to deal with the problem of multicollinearity, the dynamic specification of the models and on the extent of avertive behaviour. Copyright (C) 1996 Elsevier Science Ltd.