For Online Publication: Appendix of “Breaking Bad: How Health Shocks Prompt Crime”

To further validate our approach, we conduct an exact matching between pre-diagnosis individuals who, in the same (calendar) year, are of the same age and gender but are diagnosed as far apart as possible.1 Specifically, Online Appendix Table I.III Panel a compares the averages of seven key covariates (pre-diagnosis crime, survival probability decline, total income, earnings, homeownership, financial wealth, marital status, and education) between people who will be diagnosed in 1 year and people who, in the same year, are of the same age and gender but will be diagnosed only 10 years later, which is the longest gap available by construction (as we retain only 10 years before cancer in our sample). For comparison, Online Appendix Table I.III Panel b shows unmatched samples. Furthermore, Online Appendix Figure I.2 plots the average crime rate, total income, marital status, and education over time for the two groups (cancer in 1 and 10 years, respectively). As can be seen from Online Appendix Table I.III and Online Appendix Figure I.2, the two samples show very similar averages and trajectories with age. While differences in several covariates are statistically significant, in economic terms the differences between the two groups are small. Importantly, we find no significant difference in the average crime rate before diagnosis. This evidence supports the argument that individuals diagnosed many years apart are ex ante observationally equivalent in terms of the relevant covariates. And, therefore, the timing of the cancer diagnosis is as good as random for the purpose of our analysis.