Potential confounding factors: The panel study has the advantage of comparing measurements within individual over different time periods. The study design minimizes potential effects of confounding from factors that do not vary over time. The potential for time-dependent confounding from time-varying factors such as smoking is a distinct possibility. However, such factors are unlikely to change substantially in such a short time period. It is known that smoking status, age and gender are determinants of lung function. These factors may modify individual's PEF response to changes in air pollution levels. Thus, we conducted stratified analysis by gender, smoking status and age and also adjusted those factors and their interactions with air pollution in the models. Further adjustment for lymph nodes swelling and presence of cold symptoms did not alter the association (data not shown). The beta coefficients were very similar between models with and without these two variables.