Co-Pollutant Bias in Air Pollution IV Designs

This paper studies air pollution IV estimates when instruments move several pollutants simultaneously and provides diagnostic tools for interpreting them. We distinguish two interpretations: a stimulus effect of the named pollutant holding co-pollutants fixed, and a proxy effect of the pollution bundle indexed by that pollutant. An empirical evaluation shows that adding controls, swapping the focal pollutant, or jointly instrumenting multiple pollutants can still leave nontrivial bias. We derive conditions under which single- and multi-pollutant IV estimates support either interpretation and provide a framework for quantifying bias when they do not. A crime application illustrates the framework.