Homicide in New York, Los Angeles and Chicago.

• Author: Monkkonen, Eric H.

Homicide rates are understood in large part by comparison. Almost without thinking we compare this year to last, this place to that. Usually we make modest leaps in time and space, taking adjacent sites and time periods in an effort to hold constant otherwise uncontrollable factors. But, in keeping the comparisons modest, we may lose the leverage necessary to make sense of rates. Simply put, the theoretical questions we must address are very different if the United States has always had rates and short term variations similar to those of the present as opposed to completely different ones. If, for example, the highs of 1990 and the lows of 1999 represent a range within which rates have always fluctuated, then the objects to be explained are customary and normal. If, on the other hand, they are extraordinary, or occur only in particular times and places, the explanatory task is very different. Establishing American homicide rates for a wide range of times and places is fundamental to our understanding of homicide. As a beginning of this effort, this paper reports on reconstructed homicide rates from six large and representative cities for 1900, and for what were the nation's two largest cities--Chicago and New York City--over a long span.

In order to compare homicide rates from places separated by long distances in time or space, one must take more care than is customary to make data similar. (1) Comparing this year's count to last year's is perfectly reasonable because the population base is also comparable and extraordinary demographic events would be well known. Over longer time spans, at a minimum, adjusting counts to rates per population base is considered customary and basic. This form is the standard in most homicide studies, with a reporting of deaths per 100,000 population. This form is known as the "crude death rate," and its virtues are clarity, simplicity, and relatively easy construction. (2)

It makes sense to modify this customary approach when there have been large demographic changes, such as an increase in life expectancy or decrease in infant mortality. For example, in the twentieth century, as the United States completed its transition from a youthful, high mortality, high fertility regime to an older, low fertility low mortality one, the proportion of young males, 20-29 years old, in the total population decreased from 9.1% (1900) to 7.7% (1950) to 6.6% (2000). (3) This is approximately a twenty-five percent decrease in one of the most violence prone age and sex groups. Without adjusting for this demographic shift and possibly even greater local variations, one runs the risk of comparing apples and oranges rather than long term rate changes.

The approach taken here is that used in health statistics, standardizing homicide rates per age groups to the distribution of the United States population in the year 2000. The resulting values, age standardized homicides per hundred thousand, can be interpreted as the rates that the target year and place would have had if its base population had the same age distribution as does the whole United States in 2000. (4) This reference population is an arbitrary benchmark set by the Center for Health Statistics: until recently it had been the population in 1940. (5) I use victim rather than offender ages as they are much more completely reported. Obviously, offender ages would be the better data to use, as offenders produce the homicides. Typically, the reporting of victim ages is much more thorough and accurate, in part because a coroner processes victims. Offenders, if caught, are of interest for their actions and culpability, so officials are less assiduous in accurately reporting age. Where evidence on offenders is available, it demonstrates that offenders have a very similar age distribution to victims. For example, in 1995, for the whole United States, the correlation between the 11,760 homicide victim offender pairs over five years old was 0.43, while their mean ages were 32.4 (6) and 29.2 (7) respectively. (8) The exceptions come at the extremes, child victims in particular having no offender analogues. For this reason, and to avoid the poor and probably inconsistent quality of child murder discovery and reporting, I use no victims under five years old.

The technique known as age standardization is designed to compare groups from populations with different age distributions. (9) To calculate these rates, one needs the base population age distribution and the age of homicide victims. The United States Vital Statistics began to collect and publish such information beginning in 1900, but one cannot rely on complete coverage until 1930. At the city level, good coverage is more likely, but, unfortunately, the United States Vital Statistics lumped all violent deaths--e.g., homicide, suicide, and accident--together so that one cannot calculate those due to homicide from these sources. Prior years have some coverage in the United States census, but the coverage is not complete enough for a relatively small numerical category like homicide. The best year, 1850, depended on households recalling previous deaths, thus eliminating recall in the case of victims who had lived alone or died unknown. (10)

Age standardizing practice groups ages on ten year intervals beginning with every five: e.g., twenty-five to thirty-four years old. The year 2000 standard population has been defined by the Center for Health Statistics, and I use their reference population here. Because some health departments grouped their age data on the zero years, e.g., 20 to 29, I have on occasion had to use a slightly different reference population. For the years when ages are grouped on the ten, I have used an appropriate reference population supplied to me by Robert N. Anderson of the Center for Health Statistics. The differences are tiny, but it is important to note that the practice of spanning the zero years smooths the largest region of age heaping: respondents are most likely to round to a zero year, then to fives and threes. (11) Direct age standardization can be done if one has the age rates and population base of the original population: this is the "preferred" method. Otherwise, estimates of the actual age rates have to be created. (12) The computations here are done using Stata 6.0 direct age standardization. (13)

Ideally, the age standardized populations would be male, as men predominate in homicides, both as victims and offenders. Due to data limitations, I use as the population and victim age groups both males and females because sex based age groups cannot always be found or created for the earlier censuses. In order to compensate, I use, as a separate variable, the sex...