Stalking the Elusive Homicide: A Capture-Recapture Approach to the Estimation of Post-Reconstruction South Carolina Killings

Anyone who hopes to trace quantitative historical changes in populations faces the problem of incomplete and biased records. This is at least as much a problem for statistics of crime as for other kinds of statistics, and it plagues historical homicide research. As with other historical trends, those of homicide are derived from actual counts from some set of sources, including coroners’ records, indictments, arrests, and newspaper accounts. But what proportion of the original incidents were recorded and–if so–still exist? Adding sources usually increases the count, but how closely does it approach the true count? This cannot be known directly. One is caught between the facts that counting is the major means open to us to understand crime "as part of the sweep of history" (Monkkonen 1980: 53) and that there are gaping "holes in the historical record" that leave a large "dark figure" of crime (Lane 1992: 30). How much does the latter cancel the former? The available count of homicides for any period before the development of regular death or crime reporting likely will undershoot the true number substantially. This will make earlier periods appear less violent than they were, create false upward trends, overstate real upward trends, and mask downward trends.¹

Assuming very few false reports of killings, counts provide the lowest estimate of the number in a particular place and time.² That in itself can be useful. Rates derived from simple counts make it pretty clear that killings declined precipitously across much of Europe during the past half-millennium, and that the U.S. South and West long have been more violent than the North.³ But when, on the basis of colonial court records, Bradley Chapin (1983) tells us that homicides (and other crimes) were generally rare in the 1600s, the argument is merely plausible. The same is true of the thesis that killings of blacks were often unrecorded by southern county officials. Lacking complete records, one must look for help to patch some of those holes in the record.

The problem of missing or hard-to-count people or things has long interested demographers, who have devised methods that may be helpful for the type of problem faced here. In this paper I will discuss the uses and limitations of one such method that seems particularly suited to historical data. Originally named after its inventors, the method of C. Chandra Sekar (later Chandrasekar) and W. Edwards Deming (1949) matches records from different lists of items to estimate the total number of such items in a population. I will show the applicability of the method for estimating the number of unrecorded or lost homicides by applying it to data from post-Reconstruction South Carolina.

Like most southern states, South Carolina was relatively late to collect and publish statewide crime or death statistics. In the late 1880s the South Carolina Attorney General first began publishing the annual number of murder prosecutions (there were so many that Senator Benjamin Tillman was outraged that state newspapers actually printed the figures; see Lancaster (S.C.) Ledger 6 January; 13 February 1904), but actual counts of homicide victims were not available until 1916, when the state entered the U.S. Death Registration Area. Five years later the registration compilers at the Bureau of the Census (1924: 97—108) extensively, and unfavorably, contrasted South Carolina’s death-registration procedures with those of Vermont. There is, however, a "window" on South Carolina homicides, opened during the years 1877—78 by Horace V. Redfield (1880), a journalist who meticulously compared the amount of homicide in northern and southern states and who can be credited as the first person to explore the thesis of southern violence systematically. Because none of the southern states he studied published statistics of crime or death, Redfield drew on articles in those...