Methods

Sample

The sample consists of the nonmetropolitan counties in Florida, Georgia, Nebraska, and South Carolina.¹ The standard unit of analysis for research in the urban setting has been neighborhoods no more than a few miles across. This conception of community does not generalize very well for rural settings, where population density is much lower. The county is a convenient unit of analysis for the study of community influences on rural crime rates because both arrest data, taken from the Federal Bureau of Investigation's Uniform Crime Reports (UCR), and population characteristics, from U.S. Bureau of the Census population reports, are available at the county level. The county is also a common unit of analysis in rural research of all types because counties typically have strong internal economic and governmental structures. It should not be forgotten, however, that most counties include several distinct communities. The county level of analysis was necessitated by the availability of data, but it is not ideal.

The analysis was limited to counties not included in metropolitan statistical areas by the U.S. Bureau of the Census. These counties lack cities with populations of 50,000 or more, and less than 50 percent of their population resides in metropolitan areas with a population of 100,000 or more. Thus, residents of these counties live in smaller cities, towns, and open country rather than in moderate to large cities and their suburbs.

The study sample included 264 counties with populations ranging from 560 to 98,000. Although these nonmetropolitan counties are much larger geographic units than areas analyzed in community-level research on crime in metropolitan settings, they are of equal or smaller size in terms of population. The average total population of these nonmetropolitan counties was roughly 10,000, which is comparable to the smallest units used in research on urban neighborhoods (Sampson, Raudenbush, and Earls, 1997; Warner and Pierce, 1993). This sample compares favorably with those in studies of urban areas in terms of the number of communities, the size of the populations, and the variety of communities included.

Measures

Delinquency. UCR data (Federal Bureau of Investigation, 1998) were used to measure each county's delinquency rate. These data are the logical starting point for analyses of crime and delinquency in rural areas, and previous community-level studies of rural crime have relied on the same source. No measure of crime or delinquency is perfect, and criminologists have long been concerned about potential biases in crime rates based on official records, especially that arrests might reflect the behavior of law enforcement officers more than the behavior of offenders. Fortunately, findings relating social disorganization to arrests have been replicated by more recent studies that measured offending through citizen calls for police assistance (Warner and Pierce, 1993), self-reports of victims (Sampson, 1985; Sampson and Groves, 1989), and self-reports of offenders (Elliott et al., 1996).

This study's measure of delinquency was the per capita arrest rate of juveniles ages 11–17 in each county, pooled over the 5-year period from 1989 through 1993. The outcome measures were as follows: rates of arrest for homicide, forcible rape, aggravated assault, robbery, weapons offense, simple assault, and the UCR Violent Crime Index, which comprises the first four offenses. The study considered the full spectrum of violent offenses (capturing a large range of offense seriousness) for which recording is comparable across the four states. This approach provided a rich pool of information for establishing the consistency of the findings.

Table 1 presents descriptive statistics for all of the measures, calculated separately for each state. Rates of arrest for serious violent offenses in nonmetropolitan counties in Florida and South Carolina are considerably higher than in Georgia and Nebraska. Differences are less consistent for simple assaults. Some of these inconsistencies, such as the extremely low rate of simple assault, compared to the Violent Crime Index, in Florida, suggest that police and citizens give less attention to minor offenses in areas with high rates of serious offenses (as noted by Stark, 1987).

Explanatory variables. Data from the 1990 census provide measures for most of the explanatory variables (U.S. Department of Commerce, 1992). As is standard in research on communities and crime, the measure of residential instability was the proportion of households occupied by persons who had moved from another dwelling in the previous 5 years (Sampson, 1985; Warner and Pierce, 1993). Ethnic diversity was measured in terms of the proportion of households occupied by white versus nonwhite persons. Ethnic diversity was computed as the index of diversity, which reflects the probability that two randomly drawn individuals would differ in ethnicity (Blau, 1977). Family disruption was indexed by female-headed households, expressed as a proportion of all households with children. Low economic status was defined as the proportion of persons living below the poverty level. Proximity to urban areas was coded "1" for counties adjacent to a metropolitan statistical area and "0" for counties nonadjacent, based on census classifications (U.S. Government Accounting Office, 1989).

Also included in the analysis was the number of youth ages 10–17, which is the population at risk for juvenile arrests. Population size serves as a proxy measure for population density because the two variables are so strongly correlated that they are effectively indistinguishable. Because states may differ in their statutes and in the organization, funding, and policies of their justice systems, it was important to make sure that differences among states were not confused with the contributions of the explanatory variables. Therefore, the analysis controls for differences among states in arrest rates for each offense.

Data Analysis

The outcome of interest in this study is the arrest rate, defined as the number of arrests in a county divided by the size of the juvenile population. Standard statistical methods of analyzing crime rates are inappropriate for these data because the population sizes are small relative to the arrest rates, so only very crude estimates of arrest rates are available for the counties with the smallest populations. This problem is resolved with a specialized statistical technique (negative binomial regression) that takes into account the contribution of population size to the accuracy of arrest rates.²

Tables 2 and 3 present two versions of the relationships of the explanatory variables to delinquency rates. Table 2 considers each explanatory variable separately, controlling only for overall differences among the states. Table 3 presents the second estimate of each relationship, which controls for all other explanatory variables. The first estimate reflects the overall association of the variable with the rate of juvenile violence (the bivariate relationship), and the second estimate reflects only the association that cannot be accounted for by the other variables (the multivariate relationship). Comparing tables 2 and 3, one can see that the patterns of results are essentially the same, with the magnitude of the relationships typically somewhat higher for the bivariate relationships, and somewhat fewer of the multivariate relationships reaching statistical significance.

Tables 2 and 3 express the relationships in terms of the proportional change in the rate of arrests associated with an increase in the variable.³ Most of the explanatory variables reflect proportions of the population, such as the proportion living in poverty. The tables indicate the change in arrest rate associated with a 10-percent increase in each explanatory variable. For instance, the first entry in table 2 indicates that the arrest rate of juveniles for violent offenses will average 45 percent higher (e.g., 145 versus 100 per 100,000) for counties with 25-percent residential instability than for counties with 15-percent residential instability.

Overall, the analysis found that one or more of the social disorganization variables were significantly associated with arrest rates for all of the violent offenses except homicide. Low numbers of homicides limited the researchers' ability to detect differences in the homicide rates; 69 percent of the counties in the sample recorded no homicides during the 5-year study period.

Community Correlates of Rural Youth Violence	OJJDP Bulletin • May 2003