Association for Public Policy Analysis & Management

Currently, most work on matching strategies for causal inference focus on binary treatments. Building on this work, we develop a sequential matching strategy to create hypothetical two-factor experiments from observational data using a generalized version of the propensity score.  The resultant data set allows the estimation of conditional causal effects of one factor at different levels of the second factor, as well as the usual estimation of main effects and interactions, under explicit assumptions. We apply this matched-sampling method to environmental health time-series to estimate the joint effects of air pollution and temperature on cardiovascular mortality in Munich, Germany assuming that Nature has conducted this reconstructed randomized experiment for us. Prior work has shown that health effects resulting from climate change, via events such as extreme temperature conditions, could be catastrophic, especially when coupled with bad air pollution, since epidemiological studies have associated both air pollution and extreme temperature with mortality.  We find that, under our assumptions, air pollution does appear to cause increased cardiovascular-related mortality in the days following the event, and furthermore that the consequence of air pollution on a day with extreme temperature appears even more severe than an unconditional analysis of air pollution alone would indicate.