![Sample our Law journals, sign in here to start your FREE access for 14 days]({"type":"image" , "src" : "/sda/5942/TOC-Subject-Sample-2021-Law.jpg"})
ABSTRACT
Mass shootings have a strong impact on public discourse and perception, affecting more than their direct victims. We use data on charitable contributions and criminal activity in the U.S. over the last decade to identify and quantify the effect of mass shootings on prosocial and antisocial behavior. We find that the effect of mass shootings on prosocial behavior, measured primarily by monetary contributions, is positive and statistically significant. However, the directly affected localities react to mass shootings differently than their neighboring communities, decreasing their charitable contributions. Additionally, we are unable to find a statistically significant effect of mass shootings on antisocial behavior, as measured by various crime rates. Furthermore, we show that mass shootings are different than any other type of criminal behavior, including all other violent offenses, in terms of its effect on prosocial behavior.
Acknowledgments
We are grateful to René Bekkers and Pamala Wiepking for very helpful comments and suggestions, and to Ariel Karlinsky for his excellent research assistance. We have also benefited from comments from participants of numerous seminars and conferences, including those of the 2018 ARNOVA conference in Austin, the 2019 ERNOP conference in Basel, and the 2019 Midwest Political Science Association conference in Chicago. Claude Berrebi is grateful for the warm hospitality of the Woodrow Wilson School at Princeton University and Princeton’s Empirical Studies of Conflict Project (ESOC) while he was working on this research.
Disclosure statement
No potential conflict of interest was reported by the author(s).
Notes
1. Mass shootings are rare and distinctive events compared to the background frequency and death toll associated with single-victim or “targeted” firearm homicides. In the U.S. for example, “random/rampage” shootings are responsible for a small fraction of 1% of firearm homicides (Shultz et al., Citation2014).
2. https://obamawhitehouse.archives.gov/blog/2015/10/01/watch-president-obamas-statement-shooting-oregon.
3. https://edition.cnn.com/2018/05/18/politics/trump-texas-school-shooting/index.html
6. Author’s calculations, cross state correlation in the 2012 presidential elections. p <.05 in both estimates.
9. As some of our key covariates are only available at the state level, this is our main level of analysis. However, since both mass shootings and charitable contributions are available at a more granular level, we also employ several drill downs to the county level.
10. https://www.irs.gov/statistics/soi-tax-stats-individual-income-tax-statistics-zip-code-data-soi.
11. We followed SOI’s recommendations and instructions to aggregate the data to the county level.
12. As mentioned earlier, our data is available at a more granular level; therefore, we employ further analyses at a higher resolution at the county level as well.
13. Form 1040, Schedule A
14. 48 mainland states and the District of Columbia.
15. Though the mechanism for reverse causality between contributions and mass shooting fatalities seems unlikely, we lag the explanatory variables both to ascertain the chronological order of the events, in which mass shooting events occurred prior to contributions, and in order to better address endogeneity concerns.
16. In order to alleviate any concern, we have also re-estimated all our models with other ranges (in 500 KM intervals) and the results are robust to the choice of threshold radius.
17. Due to insufficient variation in mass shootings at the county level in our data, we are unable to include county fixed effects. In order to control for geographic variation, we include state fixed effects. In addition, we cluster our standard errors at the county level as recommended by Abadie et al. (Citation2017).
18. We have estimated models similar to those in which also include data on religiosity Based on data from the American Religious Census for the year 2010. The religiosity variables are positive and statistically significant, and do not alter the magnitude and significance of our mass shootings variables. Additionally, they do decrease the magnitude of our political variable, yet those remain positive and statistically significant as well. The results are available upon request.
19. As Washington DC is not a state, it has no governor. Instead, we use voting data for the office of mayor of the District of Columbia, as it’s mayor is considered equivalent in many aspects to a state governor.
20. For example, in our sample period, Texas had four gubernatorial elections: 2002, 2006, 2010, 2014. New Hampshire had seven gubernatorial elections: 2002, 2004, 2006, 2008, 2010, 2012, 2014.
21. For example, while California voted for Democratic candidates in all presidential elections in our sample period, both Republican and Democratic governors held office during this time.
22. Violent crimes include murder, rape, robbery and aggravated assault. Property crime includes burglary, larceny and motor vehicle thefts.
23. The difference between the coefficient of mass shootings without the inclusion of crime and the inclusion of all and property crimes is also statistically insignificant with p = 0.47 for both.
24. Contemporaneous models were also estimated. The results are virtually identical and are available upon request from the authors.
25. One should be careful interpreting this finding as a relationship between mass shootings and charitable giving on individual level, since the data is aggregated on the state level and is subject to “ecological Fallacy” risk.
26. Unlike the monetary data which was retrieved from administrative datasets, volunteering data is based on surveys and questionnaires and therefore may be subject to survey biases.
27. The inclusion of other major events into our model, such as natural disasters, doesn’t alter our results, and the coefficients for mass shootings remain qualitatively similar.