For the better part of two decades, Columbia University law professor Bernard Harcourt has been on a personal crusade against Broken Windows policing, criticizing both its theoretical underpinnings and its policy applications. A close look at Harcourt’s work, however, reveals not only the weaknesses of his arguments but also his lack of attention to other research findings that conflict with his own. His portrayal of Broken Windows policing, it turns out, is fundamentally inaccurate and incomplete. In effect, Harcourt creates and then fights a paper tiger.
By way of background, Broken Windows is a policing tactic that emphasizes the police management of minor offenses. The authors of the Broken Windows hypothesis—George L. Kelling and the late James Q. Wilson—always maintained that Broken Windows policing should encourage proper discretion on the part of officers. Kelling in particular has discussed the importance of discretion when it comes to maintaining order, as in a recent article in Politico, where he indicates that arrest should be the last option when managing minor offenses.
Recent events in American cities have led Broken Windows critics to suggest an unfortunate link between the tactic and high arrest rates. In the Huffington Post, Harcourt rejects Kelling’s claim that Broken Windows was never meant to be a high-arrest strategy. As evidence, Harcourt cites a research report on NYPD policing written in 2001, in which Kelling and I use misdemeanor arrests as a proxy variable for disorder management. For Harcourt, our use of this variable is undeniable proof that Kelling calls for a high number of arrests as part of a Broken Windows policing strategy.
What Harcourt fails to mention, however, is that we use misdemeanor arrests for the specific purpose of that report’s quantitative analyses. Quantitative analyses of this sort require stable measures over a long period of time, and when using publicly available data, only so many variable options exist—none of which are ideal for capturing complex concepts like disorder management. Because of the limitations of this variable, the report’s qualitative analyses were designed to determine the extent to which officers relied on arrests when managing minor offenses. The results of these analyses showed that Broken Windows policing (as performed by NYPD during the 1990s) involved a great deal of discretion on the part of officers. When it came to disorder, arrest was actually the exception—police were far more likely to handle minor offenses informally.
Harcourt must know that these were the results of the report. Even a casual look at the document shows that the NYPD’s Broken Windows policing involved a high amount of officer discretion. To conclude, as Harcourt does, that Kelling encouraged a high-arrest strategy clearly misrepresents Kelling’s arguments.
Harcourt has misled on this issue for years. In 2005, in an online Legal Affairs debate on Broken Windows with Harcourt, University of Michigan professor David Thacher pushed back. “I am sorry to put the point bluntly,” he wrote, addressing Harcourt, “but I think your repeated references to Kelling and Sousa’s use of misdemeanor arrests as a proxy measure—I’d argue a flawed one—in a piece of quantitative research is disingenuous.” Thacher went on: “Your continued claim that ‘Broken Windows boils down to the aggressive use of misdemeanor arrests’ seriously muddies public understanding of what order maintenance is, and it encourages the dangerous confusion of ‘order maintenance’ with ‘zero tolerance.’”
And Harcourt’s own research on the matter is deeply flawed. In his Huffington Post article, Harcourt continues his critique of our 2001 report by stating that its conclusions—that Broken Windows policing had an impact on crime in New York—are contradicted by his own analyses, conducted with Jens Ludwig in 2006. But Harcourt fails to mention the findings of other academics. Writing in Criminology in 2007, Steven Messner and his colleagues reach conclusions consistent with our 2001 results. Their findings “support the Kelling and Sousa general claim that the implementation of new policing policies played a nontrivial role in the homicide decline in New York City.” Importantly, Messner and his colleagues specifically mention discrepancies between their analyses and those of Harcourt and Ludwig. They find these discrepancies “perplexing,” especially since they used similar statistical strategies. The failure to replicate research in this manner is a serious blow to the integrity of any study. It’s no wonder that Harcourt doesn’t mention it when he touts his New York analyses.
Nor does Harcourt acknowledge that the overall body of research on Broken Windows policing, while admittedly inconsistent, tends to favor the benefits of order-maintenance policing. In a recent review in the prestigious Journal of Research in Crime and Delinquency, Anthony Braga and colleagues statistically combined the results of the most rigorous academic studies and found that, overall, police efforts designed to manage disorder reduce serious problems in communities. They conclude that “the results of this systematic review and meta-analysis lend some credibility to the NYPD’s claim that disorder policing was influential in reducing crime in New York City over the course of the 1990s.” Harcourt is either unaware of this research or chooses not to acknowledge it.
Harcourt will likely continue his assault on Broken Windows. But an inspection of much of his writing reveals that his arguments are based heavily on intentional misrepresentation, questionable research, and omission of relevant information. The media outlets that continue to accommodate Harcourt’s promotion of this misleading agenda should take note.