Data mishap forces retraction of paper on gun laws, domestic killings

The authors of a 2018 paper on the effects of gun laws on domestic violence have retracted the article after discovering errors in their analysis and replaced it with a clean version. The new study shows that some gun laws — particularly ones that keep firearms out of the hands of violent offenders, even if their offenses don’t involve domestic assaults — do seem to reduce the incidence of domestic killings.

The paper, which appeared last November in the American Journal of Epidemiology and received some press coverage, including this piece in the New York Times, looked specifically at whether laws that keep guns away from people convicted of violent crimes beyond domestic abuse reduce the number of intimate partner homicides. It also considered the effect of laws that covered dating partners and not simply spouses or former spouses. The first author is April Zeoli, of Michigan State University. Zeoli has published other papers on the topic and delivered a TEDMED talk on it as well.

According to the abstract of the article:

We conducted a quantitative policy evaluation using annual state-level data from 1980 through 2013 for 45 US states. Based on the results of a series of robust, negative binomial regression models with state fixed effects, domestic violence restraining order firearm-prohibition laws are associated with 10% reductions in IPH. Statistically significant protective associations were evident only when restraining order prohibitions covered dating partners (-11%) and ex parte orders (-12%). Laws prohibiting access to those convicted of nonspecific violent misdemeanors were associated with a 24% reduction in IPH rates; there was no association when prohibitions were limited to domestic violence. Permit-to-purchase laws were associated with 10% reductions in IPHs. These findings should inform policymakers considering laws to maximize protections against IPH.

But after the findings were published, the researchers learned that they had used incorrect dates of implementation for some of the laws they’d analyzed. After correcting those dates, the picture looked different.

As the retraction notice explains:

We re-ran our statistical models with the corrected law data, and there were differences from our original publication in the estimates for a number of the laws that did not change either the direction of the incident rate ratio or the statistical inference. We also found 2 changes in estimates that strengthened the support for our hypothesis that firearm restrictions that include a broader set of domestic violence offenders are associated with larger reductions in intimate partner homicide. Specifically, the confidence interval for the association of domestic violence restraining order restrictions that do not cover ex parte orders with firearm intimate partner homicide rates changed to indicate no clear association (incident rate ratio (IRR) = 0.95, 95% confidence interval (CI): 0.84, 1.07), and the confidence interval for the association of the inclusion of a relinquishment law in the domestic violence restraining order firearm restriction with intimate partner homicide changed to indicate an association (IRR = 0.88, 95% CI: 0.81, 0.97). We also found that the federal misdemeanor domestic violence firearm prohibition was associated with reduced rates of total intimate partner homicide (IRR = 0.92, 95% CI: 0.86, 0.98) and firearm intimate partner homicide (IRR = 0.88, 95% CI: 0.81, 0.96). Additionally, the confidence intervals for the revised incident rate ratio estimates eliminated the support for our hypothesis that laws requiring permits to purchase handguns reduce intimate partner homicides (95% CI: 0.85, 1.28).

Given these changes, we retract our original article and apologize for our error.

Ichiro Kawachi, chair of the Department of Social and Behavioral Sciences at Harvard, and the editor at AJE who worked on the paper, told us that the journal heard from the authors about the “data error” — which proved to be crippling:  

They told us that that they had mis-coded the gun control laws for one of the states in their sample. When they re-ran their analyses with the corrected data, the substance of their conclusions were unchanged, but it affected all of the numbers in their tables & texts. The total number of corrections ran to over 100 corrections, which we deemed to be too unwieldy to publish as an erratum. Some of the point estimates including those reported in the Abstract) were also changed.

Therefore, we advised the authors to retract the original paper and immediately re-publish the cleaned & corrected version (which we have done).

The new version is available here.  The new abstract states:

Statistically significant protective associations were evident only when restraining order prohibitions covered dating partners (−13%) and ex parte orders (−13%) and included relinquishment provisions (−12%). Laws prohibiting access to those convicted of nonspecific violent misdemeanors were associated with a 23% reduction in IPH rates; there was no association when prohibitions were limited to domestic violence. These findings should inform policymakers considering laws to maximize protections against IPH.

Gun safety group noted errors

Zeoli told us that her group learned about the errors from

a staff member at the organization Everytown for Gun Safety who also pays quite close attention to these laws. We conducted additional independent legal research to verify that errors were made and to ensure accuracy of the new data. We then re-ran the models with the corrected data and contacted the American Journal of Epidemiology with what we had found, and retracted and replaced the journal article.

In the New York Times piece in March, Zeoli called Everytown’s estimates on the links between background checks and decreased killings a “’back of the napkin” calculation.'”

Zeoli added that the corrected findings still support the hypothesis that curtailing guns in a targeted way can reduce the slayings of domestic partners.

They also support the hypothesis that laws that explicitly require relinquishment of firearms by those prohibited from gun possession by domestic violence restraining orders are associated with decreases with intimate partner homicide. These relinquishment laws are associated with a 12% reduction in IPH and 16% reduction in firearm IPH.

Our third hypothesis that laws that established a system of accountability to prevent the transfer of guns to prohibited persons, such as permit to purchase laws, was not supported in the corrected analysis, however. This is a change in the findings from the original analysis, in which permit to purchase laws were significant and protective.

Meanwhile, Michigan State put out a press release to promote the paper in which it … failed to mention the retracted predecessor.

Like Retraction Watch? You can make a tax-deductible contribution to support our growth, follow us on Twitter, like us on Facebook, add us to your RSS reader, sign up for an email every time there’s a new post (look for the “follow” button at the lower right part of your screen), or subscribe to our daily digest. If you find a retraction that’s not in our database, you can let us know here. For comments or feedback, email us at team@retractionwatch.com.