Structural Modeling of Epidemiological Time Series

摘要

The paper considers the problem of developing a predictive model for mortality in Los Angeles County as a function of temperature and carbon monoxide (CO), a primary pollutant generated by auto emissions. The authors use a time series structural component model that includes the fixed regressors temperature, squared temperature and pollution along with a random component emulating some possible common disturbed (i.e. noisy) periodicity. They compare the results of fitting the above parametric models to the results of simple nonparametric smoothing techniques designed to display mortality in profile as a function of pollution and temperature. It is shown that there are substantial mortality fluctuations introduced by both temperature and CO gradients, particularly in the over 65 year old group. The consistency exhibited by the parametric and nonparametric profiles means that they can use either to predict the average mortality by age-group and location as a function of temperature and CO level.