With homicide as the outcome, model fit improves when the search variable is interacted with percent African American, percent Latino, and percent of city residents age 15 to 24. The “Ferguson effect” appears to be spreading from the University of Missouri to other campuses. First, if a rise in crime were to have generated a rise in searches around police violence, on the idea that criminal justice issues in general became more salient, it stands to reason that it should also have generated a rise in searches for the term crime itself and that crime searches should then be associated with police violence-related searches. So it is with her most recent reporting on the demonization of our nation’s cops and the predictable outbreak of lawlessness in our big cities. (2016) ask whether the trajectory of crime rates in 81 cities was altered by the events of August 2014 in Ferguson. After the death of Michael Brown, the hashtag became a rallying point for protesters who streamed into Ferguson and for others who were already demanding criminal justice reform in their communities (Cobb 2016). Arrests for minor offenses dropped off the most, which they see as consistent with the idea of a Ferguson effect.

The strong version of what I have called the Ferguson Effect—a drop in proactive policing leading to rising crime—is the only explanation for the crime increase that matches the data. The year 2015 also saw a non-negligible increase in violent crime in some U.S. cities. For Now, Florida Judge Stays in Proper Lane on Elections.

/ Meaningful discussion of this question begins with recognition of how far crime rates fell in the decades following the early 1990s—what Zimring (2007) calls “the great American crime decline.” At that time, the country was in the midst of a crack cocaine epidemic that fueled a wave of gun violence. Because concern about police violence, as indexed by Google search patterns, only goes very high in response to events in Baltimore, we think it could be appropriate to rename the Ferguson effect hypothesis the Baltimore effect.18 Measured by Google searches, concern about police violence rose dramatically between 2014 and 2015. Relatedly, so far as we are able to tell (Google AdWords was not designed for social science research and aspects of the algorithm are not transparent), search numbers are not adjusted for repeated searches from the same IP address. Analyzing data on 43 large U.S. cities, we find that violent crime was higher and rose more in cities where concern about police violence was greatest. Table 3. In any case, given the tens of millions of law-enforcement interactions that occur each year, these numbers are anything but an epidemic. A related possibility, suggested to us by Patrick Sharkey, is that in the aftermath of police violence and then protests in cities like Baltimore and St. Louis—protests that were met with shows of force by local authorities—young people’s perceptions of the legitimacy of core social institutions changed. They find that in the period of time between the Ferguson events and the events surrounding the arrest and death of Freddie Gray, there was no meaningful change in crime rates in Baltimore—only “trendless fluctuation” (p. 2). On Monday, Tim Wolfe of the University of Missouri resigned because he had not been lenient enough. Harcourt 2001; Sampson 2012). The MCCA data in fact cover more large police jurisdictions than this, but we have excluded from our data set most sheriff’s departments with urban policing responsibilities (e.g., the Los Angeles County Sheriff’s Department) because the geographic boundaries of their jurisdictions do not match up with the municipal boundaries in our Google search data. For more information view the SAGE Journals Article Sharing page. Not surprisingly, fear of crime in Chicago is high.

A fourth surge—one that displays more heterogeneity across cities—comes around August 2015, as a second wave of Ferguson protests broke out and Black Lives Matters protestors successfully shut down a campaign event for Senator Bernie Sanders. While this would seem to be prima facie evidence that public concern about police violence is indeed tied to violent crime, we turn next to random effects models that allow us to analyze variation across cities as well and to break out the contribution of our adjustment variables.19. Finally, Rosenfeld (2016) turns to the Ferguson effect idea, fleshing out in particular the legitimate authority interpretation. On this interpretation, the increase in crime that occurred in the spring and summer of 2015 would reflect not a demonization of the police but a boiling over of long-simmering racial tensions as some city residents—those not given to protesting—took out their frustrations on one another. A fellow activist soon created a Twitter hashtag around that phrase.
Third and finally, we address implications for future research. Neil Gross is the Charles A. Dana Professor of Sociology at Colby College. Table 1 displays the results from a fixed effects negative binomial regression in which Google search activity from 2014 and 2015 is used to predict variation in homicide rates and all other violent crimes within the 43 cities in our sample. More police per capita was also positively associated with homicide, while more young people was associated with less homicide. On the other hand, Americans who are already well informed about police brutality, either from personal experience or exposure to other sources of news and information (e.g., social media), might not need to do any Googling. On Tuesday, USC’s student body will vote on a controversial campus resolution urging $100 million be spent on mandatory diversity classes to create an “inclusion climate” on campus. The second problem raises questions about replicability. Our measure of concern about police violence is the frequency of searches in 43 large U.S. cities for phrases that signal an interest in the topic and/or a personal commitment around it. The police have again pulled back in our big cities — with predictable results. We double-checked the procedures used to generate these files by having research assistants replicate searches for select cities within the current window. For Hausman test statistics, we used ordinary least squares models with logged dependent variables. A Lot. This could be because of a saturation effect, as discussed previously. Often bundled together with “community policing,” problem-oriented policing requires cooperation and trust between citizens and the police, who must be viewed as a legitimate source of authority. It also seems plausible that people with more strident anti-police views might Google phrases like I hate cops in order to connect with online communities of the likeminded and that the views of such persons would not show up in traditional polling or survey data. 13The latest numbers here are from 2012 and reported at http://www.governing.com/gov-data/safety-justice/police-officers-per-capita-rates-employment-for-city-departments.html. Moving on to robberies, one observes small upward movement over time.

We think there are two takeaways from our regression models. Our outcome variables are violent crimes reported to the police in 2014, 2015 (full year and first half year), and the first half of 2016 in the same 43 cities.11 Like Rosenfeld, we draw on crime information compiled by the Major Cities Chiefs Police Association (MCCA).12 Although not every large city is included in the MCCA data—New York City is not, for example (at least in this time period)—the cities represent a large swath of urban America, with a combined population of almost 33 million. While the findings on public concern in no way represent proof of the Ferguson effect hypothesis—not least because we have no data on the activities of the police—alongside the findings of Morgan and Pally (2016) and Rosenfeld (2016), they provide empirical warrant for further study of it.
The list is thus centered on relatively abstract topics rather than specific people or events, whereas the latter (e.g., Samuel DuBose shooting—referring to a man killed in 2015 by a University of Cincinnati police officer) might be more common search targets.

The average gain across all the other cities in the sample was 2 percent. As we discuss in more detail in the following, a comparison of first half 2015 to first half 2016 homicide numbers for the 43 big cities in our sample shows that while some cities experienced larger spikes, the average increase across all cities was 16 percent.