Are the increases in murders in major cities like Chicago, Milwaukee, and New York City indicative of a broader trend in American cities? That’s the conclusion encouraged by a front-page New York Times article, Murder Rates Rising Sharply in Many U.S. Cities. It’s a scary story, conjuring images of the high-crime 1990s and fueling speculation about an ostensible “Ferguson Effect” — the unsubstantiated notion that, as The Times put it, “less aggressive policing has emboldened criminals.” This is badly misleading and, at a time when criminal justice reform is making notable bipartisan advances, it’s also dangerous.
Of course, The Times isn’t an academic journal, and its story wasn't meant to be a rigorous analysis of a big database; it was a glimpse into a current conversation with some new numbers. Still, it's worth taking a closer look at those numbers.
My own analysis of publically available homicide statistics for a broader selection of cities yields conclusions that are rather different from those stated or implied in the Times article. The differences are related to how cities were selected and the way the data were interpreted.City Selection
It is not clear how the cities examined by the Times were chosen. The article included ten cities with populations ranging from over 8 million (New York) to just over 317,000 (St. Louis). But there are 60 U.S. cities with estimated 2014 populations in that range. The Times included only four of the 20 most populous U.S. cities. The authors do not explain how those cities were chosen, leaving readers to assume that the findings presented are representative of a broader increase in homicides across U.S. cities. That does not appear to be the case.
In just a few hours, I was able to locate publically available data to support similar analyses for 16 of the 20 most populous cities, and the results, summarized below, suggest a much less pervasive increase than one might infer from the Times analysis.Interpretation of Statistics
First, not all of the increases cited by the Times are statistically reliable; that is, some of them are small increases, or are based on small numbers of cases, such that the observed increases could have occurred by chance alone. Among the 16 top-20 cities for which I found publically available data, only three experienced statistically reliable increases. Only one of the top-20 cities included in the Times’ sample, Chicago, experienced an increase that was statistically significant. Five of the smaller cities included by the Times did experience statistically reliable increases, but what of the other 35 cities with populations in that range?
Even where a statistically reliable increase has been experienced, a single year-to-year increase does not necessarily imply a meaningful trend. Often, such changes fall within the range of normal year-to-year fluctuations. For example, I was able to obtain historical data on year-to-year changes in homicide counts for Chicago, the only top-20 city in the Times analysis that had a statistically significant increase from 2014 to 2015. From 2009 to 2010, homicides increased 5.1 percent. The next year, however, there was a 13.1 percent decrease. The year after that, a 28.5 percent increase, and then decreases of 16.4 and 3.4 percent in 2013 and 2014, before homicides climbed back up 11.3 percent in 2015. Looked at over a longer time period, the numbers do not demonstrate a stable trend.
Thus, neither the Times analysis nor my own yields compelling evidence that there has been a pervasive increase in homicides that is substantively meaningful. It seems premature to be discussing broad explanations and long-term solutions for what may not be a broad or long-term phenomenon. And yet the spike in a few cities has already prompted speculation that the numbers reflect the increased availability of guns, or the demoralization of police.
Of course, the lack of compelling evidence of a broad-based increase does not prove that no such increase is occurring. Trends in homicide rates (and crime rates generally) are extremely important topics that warrant further investigation. But before we begin to speculate about causes and potential remedies, we need a more comprehensive understanding of the prevalence and location of increases in homicide rates that actually depart from normal fluctuations. This suggests a need for analyses that span several years, for as many medium- and large-sized jurisdictions as possible. It would also be useful to undertake such analyses for related crimes, such as non-fatal shootings, non-fatal stabbings, and aggravated assaults, since the difference between those and homicides may often be a matter of random luck, the type of weapon readily available, or the distance to the nearest emergency room.
Only then will we be in a position to undertake rigorous efforts to explain the problem and explore potential remedies.
Bruce Frederick is a senior research fellow at the Vera Institute of Justice.