Abstract:
|
The present paper uses a new data set from Toronto to investigate the health effects of air pollution. Statistical methods for investigating the relationship between air quality and health are complicated by two related problems. Firstly, there are numerous explanatory variables which may be relevant (e.g., various pollutants, weather conditions, interactions between the two, and lags). Secondly, various transformations (especially as relating to threshold effects) of the explanatory variables are possible. Allowing for all such factors implies a model with hundreds or thousands of potential explanatory variables. These issues are partially addressed in the recent literature by using Bayesian model averaging. One component of the statistical model used in the present paper draws on ideas from this literature. A second component of the statistical model draws on ideas from the nonlinear time series literature. A nonlinear parametric model, which allow thresholds effects of various sorts (and at various time lags) to be estimated from the data, is used. Empirical results indicate the importance of Bayesian model averaging and of allowing for nonlinearities and threshold effects.
|