• DOI:10.2105/AJPH.2009.187476
• Corpus ID: 20167981

Flaws in study of firearm possession and risk for assault.

@article{Wintemute2010FlawsIS,
  title={Flaws in study of firearm possession and risk for assault.},
  author={Garen J. Wintemute},
  journal={American journal of public health},
  year={2010},
  volume={100 6},
  pages={
          967-8; author reply 968-9
        }
}

• G. Wintemute
• Published 1 June 2010
• Psychology
• American journal of public health

The study by Branas et al.1 contains errors in design and execution that make it difficult to determine the meaning of their findings.

Their study assessed risk for being assaulted and then shot, a compound outcome event whose second element (being shot) is not inevitable given the first (being assaulted). Persons who were assaulted but not shot are not studied. We do not know whether any association between firearm possession and their outcome measure applies to assault, to being shot given an assault, or both.

The study does not control for time and place. The authors invoke stray bullets to argue that residents of Philadelphia, Pennsylvania, are at equal risk for being shot, no matter where they are and what they are doing. This ignores the fact that violence is not randomly distributed and is unfair to Philadelphia.

The control group is inappropriate, as was probably guaranteed by its selection from all adult Philadelphians. There were large differences between case participants and control participants in prior criminal history and alcohol or drug involvement, all of which influence gun-carrying behavior and risk for violent victimization. Personal and geographic differences compounded one another: 83% of shootings occurred outdoors, yet while those shootings were occuring, 91% of control participants, arguably at lower risk already for personal reasons, were indoors. A list could easily be made of likely differences between case participants and control participants that were not addressed.

The problems with geography and control selection are not insurmountable. A classic study of alcohol use among adult pedestrian fatalities in Manhattan enrolled the first 4 pedestrians reaching the site where the fatality occurred “on a subsequent date, but on the same day of the week and at a time as close as possible to the exact time of day of the accident [italics retained]”2(p657) as control participants for each case participant.

Branas et al. have omitted critical detail from their results. Assaults can be independent of any prior relationship between perpetrator and victim—a would-be robber spies a prospect emerging from a bar—or can occur in the context of, and perhaps because of, some prior relationship. The association between gun possession and risk of being assaulted or shot may differ greatly between these 2 types of encounters. Attacks by strangers are common, accounting for 50.5% of robberies and aggravated assaults reported by males and 34.7% of those reported by females.3 The authors should present separate results for assaults independent of and related to prior personal involvement between victims and shooters.