Manhattan Murder Mystery.

• Date: 22 June 2020
• Author: Greenberg, David F.

Franklin E. Zimring, The Great American Crime Decline (New York: Oxford University Press, 2007)

Franklin E. Zimring, The City That Became Safe: New York's Lessons for Urban Crime and Its Control (New York: Oxford University Press, 2012)

Patrick Sharkey, Uneasy Peace: The Great Crime Decline, the Renewal of City Life, and the Next War on Violence (New York: W.W. Norton & Company, 2018)

IN TYPICAL MURDER MYSTERIES, ONE OR MORE CORPSES ARE FOUND, AND the challenge is to determine who killed them. The Manhattan murder mystery of the last quarter century is of a different sort: It concerns the people who never became corpses even though they were expected to die. We do not know who those people are, because our expectations of deaths are not for particular individuals; they are for the collectivity of people living in Manhattan (or passing through). The unexpected survivors themselves do not know who they are, and yet we have good reason to believe that they exist. Based on past experience, it could reasonably be expected that a certain number of people in the city would be killed annually in recent years. Yet they were not. Worse than that, top-flight criminologists and expert criminal justice policy analysts predicted that violent crime rates would soon soar as a new generation of young Black male superpredators came of age. However, fewer--far fewer--actually died. Thus, we have among us the walking living, unknown to themselves or anyone else. The mystery is why they are alive.

The mystery is not restricted to Manhattan; it extends to all five of New York City's boroughs and to many other cities in the United States and other countries (Baumer & Wolff 2014, Griffiths & Norris 2019). It also includes the people who were not raped or assaulted and the property that was not stolen. It is these puzzles of the missing crimes--the crimes not committed--that have engaged criminologists for 20 years (Barker 2010, Blumstein & Wallman 2000, Levitt 2004). Franldin Zimring takes up the challenge in his book The City That Became Safe (2012) (henceforth TCTBS). Solving the mystery, in this case, does not mean identifying the individuals who were expected to be killed or stolen from but were not, or who did not kill when they should have. It means figuring out why the expected murders, assaults, and thefts did not occur.

Zimring's undertaking builds on The Great American Crime Decline (henceforth TGACD), his earlier effort to understand the national drop in crime that took place in the 1990s (Zimring 2007). In that book, he addresses the New York pattern, but only briefly. In the new study, more attention is given to a city whose crime trend has been heralded as exceptional--an amazing deviation from the national pattern.

In the usual pattern of book reviewing, reviews appear soon after the book's publication. In this instance, the delay is longer. The lapse allows us to assess Zimring's arguments more fully in light of more recent work. In so doing, I will pay particular attention to a newer book by urban sociologist Patrick Sharkey (2018), Uneasy Peace: The Great Crime Decline, the Renewal of City Life, and the Next War on Violence.

I encourage readers to read both publications, as they complement one another. The two books systematically collect and analyze data to support extended arguments in clear and lively prose. Graphs and tables are deployed liberally, without the technical apparatus of statistics that many non-specialist readers would find off-putting.

Zimring's Strategy

To provide background for the more recent book, I begin with some remarks on Zimring's earlier study. My remarks are written from the point of view of someone who uses statistical modeling techniques to study spatial and temporal patterns in crime and criminal justice. Zimring begins by arguing that, starting in the early 1990s, index crime rates began to fall in the United States in a manner that was quite different from earlier trends, in which crime rates rose or fell for just a few years at a time. Annual drops in the 1990s were individually unremarkable, but because they persisted for a longer period of time, they had a larger cumulative effect.

When viewed in the context of a longer time frame, the 1990s drop is, in fact, not entirely unique. In 1933, U.S. homicide rates began a sustained descent that, apart from a modest upsurge during the Second World War, continued until 1955, bringing about a reduction by a factor greater than two, comparable to the more recent drop (Goertzel et al. 2013, Sasinoski 2011). The trend can be seen in Figure 1.

An understanding of the recent drop could conceivably have been achieved by considering evidence of earlier long-term trends, but this was a path not taken, possibly for lack of data for the earlier years. Its potential lies in the likelihood that each decade is not totally unique. Factors that promote or prevent crime in one era may well have done so at other times as well.

In distinguishing the sustained drop from the more common short-term fluctuations (which he misleadingly calls "cycles" (1)), Zimring observes that the latter can easily occur by chance. Though not elaborated in the text, the observation is a valid one. Potentially, crime rates can be affected by lots of shocks, none of them very strong and none of them enduring. Trying to chase them down can be a fruitless enterprise. However, a series of randomly occurring, short, independent shocks can, once in a while, produce what appears to be a sustained directional effect--one that persists over a number of time points. It is improbable that a long drop arose in this way, but it is not impossible.

This possibility can be checked statistically. I carried out such a test (the generalized least squares [GLS] version of the Augmented Dickey-Fuller unit root test) using time series data for each of the seven index crime rates for the United States, for the years 1946-2013. (2) For each offense, the patterns were consistent with a unit root process--one in which annual changes were completely random.

However, the statistical power of the test is low, which implies that this finding is inconclusive. (3) This does not mean that Zimring's project of identifying causes of the change is misguided. It does suggest that his project of distinguishing short-term from long-term changes may be difficult to carry out. Below, I will present evidence that the annual changes in crime rates were not totally random.

To advance our understanding of what did--or did not--explain the drop of the 1990s, Zimring eschews the statistical modeling adopted by economists and quantitative criminologists, pointing out (to a degree correctly in my view) that these models sometimes require strong assumptions whose validity is uncertain and difficult to assess. Whether avoiding this method altogether is a satisfactory approach is not so clear. I personally think not. In any event, Zimring examines the possible impact of several candidate causes, one at a time. In one chapter, Zimring considers the expansion of prison populations, demographic change, and improvement in the economy as possible explanations and finds that, although they may have contributed to the decline, they still leave a substantial fraction of the decline unexplained. (4)

Though Zimring's discussion of these topics is smart and insightful, some of the details allow room for quibbles or comments. Zimring dismisses the fizzling out of the crack epidemic in a single sentence, by pointing out that other illegal recreational drugs were still being marketed. However, the relevant process for crack may have been the stabilizing of the market rather than its disappearance. The market for heroin was older, and by the 1990s it may have already stabilized.

Zimring also takes up the possibility of regression to the mean--and rejects it. He was right to do so, but I wish he had said more about this, because Harcourt and Ludwig (2006) have argued, in an essay that no doubt appeared too late for Zimring to take into account, that mean reversion, not arrests on marijuana charges, could have explained the New York decline.

However, Zimring does not demonstrate that regression to the mean cannot explain the crime drop. Actually, Harcourt and Ludwig (ibid.) use the phrase "mean reversion" to refer to something different from what most statisticians mean by this phrase. In standard statistical usage, regression to the mean is a selection effect. Suppose a crime rate was level over the long run but fluctuated randomly from year to year about that level. Someone who began tracking the crime rate in a year when it was unusually high would probably see lower rates in subsequent years but would actually be seeing only random fluctuations about a level. Someone who thought that the introduction of COMPSTAT in the early 1990s led to the crime drop would need to consider that COMPSTAT was introduced at a point in time when crime rates were exceptionally high, with subsequent drops being due to mean reversion.

Harcourt and Ludwigs (ibid.) analysis, however, proceeds by introducing a lagged dependent variable into their model. It has a negative effect on change in the crime rate. This is a way to model a homeostatic process--the sort of self-regulating process that governs the operation of a thermostat (Kessler & Greenberg 1981). In the criminological context, this might mean that an unusually large increase is crime in the 1970s and 1980s set in motion social...