Estimating the Accuracy of Historic Homicide Rates: New York City and Los Angeles

Even though it declined in the last decade of the twentieth century, homicide remains an American problem of extraordinary importance. Yet our empirical knowledge is remarkably shortsighted. Simply put, most researchers focus on the past decade or two–usually for reasons having to do with convenience, not theory–and ignore the longer term. However, recent work has shown that past lethality is inherently recoverable, and that there is every reason to expect that comparable homicide rates across time and place should be used to set current research in context (Gurr 1981; Johnson and Monkkonen 1996; Ylkiangas forthcoming; Eisner 2000). This paper builds on some of my recent research and responds to the challenge of a recent paper by Douglas Eckberg (1998), who has shown that not only can we recover the past, but that we can even estimate missing data counts.

In the late nineteenth century, criminal statistics were often taken as transparent, as accurate measures of reality. Whether using imprisonment figures or criminal-court convictions, almost no one questioned whether or not the measures precisely captured underlying behavior. Alolphe Quetelet, an exception, was a century ahead of himself when in 1835 he analyzed measured crime as an index (Quetelet 1842 [1835]: 82). The increasing influence of criminologists brought Quetelet’s view into the mainstream in the early twentieth century. In the 1930s, criminologists began to discuss the underlying behavior, referring to it somewhat mysteriously–even evocatively–as the "dark figure." "Crime," Edwin H. Sutherland stated authoritatively–and in an authoritative place, the report of the President’s Research Committee on Social Trends (1933: 1123)–"cannot be measured directly." He urged a sophisticated understanding of crime measures as indexes. Throughout the post—World War II era, criminologists have tried to get more accurate, if still indexed, measures of reality, and the current usage of highly sophisticated survey instruments in the National Crime Survey reflects the best social science technology (Chilton and Jarvis 1999; Sparks 1981). These surveys ask carefully selected samples of people about their victimization experiences: the results show levels of underreporting of crime and are considered an excellent gauge of criminal victimization.

Murder, however, has always been a bit different, simply because of its universal seriousness, the existence of institutions to deal with death, and the difficulty with which it is hidden. It has been a cliché from at least as early as Chaucer’s time: "Murder will out [Mordre wol out], that see we day by day" ("The Nun’s Priest’s Tale," line 15058). Shakespeare repeated this conventional wisdom in The Merchant of Venice: "Truth will come to sight; murder cannot be hid long" (2.2). More than pretentious statements, these old ideas contain a solid sense. If criminologists did not accept the Chaucerian point of view, they did insist that the more serious the crime, the more likely it was to be reported and measured (see, for example, Little 1961). In either case, murder comes out as measurable. There are, of course, several problems in the measurement of murder: infanticides, for example, have almost certainly been underreported. No doubt some murders have been perfectly hidden (Emmerichs 1999; Nadel 1999). The line between intentional and accidental homicides is sometimes blurred. Some kinds of murder are tolerated, or at least go unpunished. With the exception of infanticide, however, we know that silence was not always the rule, so at the moment, most researchers reconstructing long-range homicide rates exclude murders of children under five for the sake of consistency and comparability.

Although the measure of murder remains the very best criminal violence index, there is a problem: one needs an even better measure if one is to compare different places and times, for the idea that the indexes are comparable in levels across places has never been proposed. In other words, the index for New York City will be incomparable...