Natural disasters, salience and public support for climate change policy

For example, in 1992 Congress passed the Residential Lead-Based Paint Hazard Reduction Act requiring the disclosure of the potential presence of lead-based paint and hazards before the sale of a home. Effective June 1,1998 California’s Natural Hazards Disclosure Act requires sellers of real property in state-mapped hazard areas to provide potential buyers with a “Natural Hazard Disclosure Statement.” In 2019 67,572,768 participants registered in the Great ShakeOUT Earthquake Drills worldwide (https://​www.​shakeout.​org/​glb_​participants.​php) with the goal of providing participants information regarding the relative risks of living in an earthquake-prone region.

https://​www.​pewresearch.​org/​science/​2019/​11/​25/​u-s-public-views-on-climate-and-energy/​.

https://​www.​nytimes.​com/​2018/​07/​28/​opinion/​sunday/​republicans-climate-change.​html.

Please see https://​climatecommunica​tion.​yale.​edu/​visualizations-data/​ycom-us/​ for more discussion on this issue.

Link to data: https://​www.​fema.​gov/​about/​openfema/​data-sets. The version of the dataset that we utilize in this paper was retrieved from fema.gov on April of 2019.

An overview and a link to the 2018 data are available at: http://​climatecommunica​tion.​yale.​edu/​visualizations-data/​ycom-us-2018.

We also rely on two additional assumptions even though not explicitly stated. First, the sample is i.i.d. and second, there is common support between control and treatment groups. Specifically, \(\Pr \left( {{\text{ND}} = 1|x} \right) \in \left( {0,1} \right)\). We do check for common support in the data and find no violations of this second assumption. Additionally, see Supplemental Appendix Figure A1 for plots of estimated propensity scores and Supplemental Appendix Table A1 for estimated marginal effects (evaluated at the mean) associated with the estimated probit model used for matching.

Additionally, covariate balance was assessed before and after matching by looking at the standardized differences in the raw and matched samples, respectively. Overall, matching achieved balance according to the standardized differences in covariates; in absolute value the differences are not “large.” We follow Rosenbaum and Rubin (1985) and consider any absolute standardized difference greater than 0.20, in absolute value, to be large. A covariate balanced table highlighting the covariate balance for each of our matching specifications can be found in Supplemental Appendix Table A2. Note, in two instances for the measure for coastal counties we find a standardize difference greater than 0.20 in absolute value (see the baseline and specification dropping adjacent controls with the inclusion of power plant counts in Supplemental Appendix Table A2). While this is not ideal, the stability of results across the various specifications provides reassurance that this is not a main driver of the results. Further, and as can be seen in Supplemental Appendix Table A1, the marginal effects, evaluated at the mean, associated with the coastal county indicator are arguably small and statistically insignificant in all first stage probit results.