The Impact of Changing Demographic Composition on Aggravated Assault Victimization During the Great American Crime Decline

Abstract

The United States experienced a dramatic decline in interpersonal violence rates between the early 1990s and mid-2000s. This decline, however, was much steeper in urban and suburban relative to rural areas. Prior research showed changing demographic composition can account for a substantial amount of change in inequality in victimization rates. We employed National Crime Victimization Survey data and counterfactual modeling to determine if changes in demographic composition—including proportion of population young, unmarried, male, unemployed, and in several income groups—of urban, suburban, and rural areas were partially responsible for changes between 1993 and 2005 in (1) area-specific aggravated assault victimization rates and (2) urban–suburban, urban–rural, and suburban–rural victimization rate ratios. Results showed changes in individual demographic characteristics played a very minor role in changes in area-specific assault rates. The one exception was income, which explained a substantial amount of change in victimization rates across all three areas. Changes in demographic composition explained a greater amount of change in rural relative to urban and suburban victimization rates. Changes in demographic composition across these three area types were also responsible for a small proportion of the large changes in the urban–rural and suburban–rural victimization rate ratios over time.

Keywords

rural crime demographic composition counterfactual analysis

Introduction

Urban violence rates are generally higher than suburban and rural violence, rates but the magnitude of differences between them varies over time. During the great crime decline in the United States, National Crime Victimization Survey (NCVS) data show that between 1993 and 2005 annual aggravated assault victimization rates declined about 75% in urban and suburban areas but only 42% in rural areas. In 1993, urban and suburban aggravated assault victimization rates were, respectively, 163% and 43% higher than rural rates. By 2005, however, the urban rate was only 18% higher than in rural areas, and the rural rate was 69% higher than suburban rates. Violence rates declined markedly in all three areas, but rural residents benefitted substantially less than their urban and suburban counterparts.

While there are several potential methodological and substantive explanations for these trends, we explored whether changes in the demographic compositions of these three area types were partially responsible for changes in rates of aggravated assault victimization rates. Following Thacher’s (2004) substantive and methodological framework, we examined changes in area-specific demographics and aggravated assault incident rates using NCVS data from 1993 to 2005. We then determined the combined effects of demographic composition changes on violence rate changes via counterfactual models. We also addressed whether demographic–victimization risk relationships varied by community type.

Demographic Shifts and Crime Trends

There is an extensive history of linking crime rates to demographic characteristics. The Bureau of Justice Statistics (BJS), for instance, publishes annual victimization rates of various demographic groups by income, race, gender, home ownership, residence area type (i.e., rural, suburban, or urban), and region of the United States (e.g., Truman & Morgan, 2016). Much of this research is descriptive or cross-sectional, with few studies examining the effects of temporal demographic changes on crime rates over time (Meier & Miethe, 1993; Thacher, 2004). Notable exceptions include Cohen and Felson (1979), Steffensmeier and Harer (1999), Levitt (1999a, 1999b), and Thacher (2004), which focused on household activities (e.g., single-adult households and employment outside the home), age composition, or income composition.

There is evidence that demographic changes affect crime rates. Cohen and Felson’s (1979) examination of the impact of changes in age structure and household activities on predatory crime rates post-World War II was a pivotal study of the effects of population composition changes on crime. They found their household activities ratio was consistently positively associated with person and property victimization rates. Steffensmeier and Harer (1999) found age composition had large effects on national crime rates in the 1980s, but effects diminished through the mid-1990s. Levitt (1999a) concluded that while the individual-level association between age and crime is strong, the aggregate-level association between age composition and crime rates is limited. Levitt (1999b) also found property crime victimization became increasingly concentrated among the poor from the 1970s to the 1990s, while homicides went from having an inverse relationship with income to having virtually no relationship. Thacher (2004) advanced Levitt’s (1999a) research by systematically examining whether demographic changes in economic groups accounted for crime rate changes in economic groups over time.

Thacher’s Demographic Change Hypothesis

Using data from the National Crime Survey (NCS) and the NCVS, Levitt (1999a) found from the mid-1970s to the mid-1990s property crime distributions changed across economic groups, with these crimes becoming more concentrated among the poor. Thacher (2004) extended Levitt’s work by (1) accounting for the 1992 NCS/NCVS redesign and (2) examining whether demographic changes in economic groups explained changes in victimization concentration. He found that from the 1970s to 2000s, criminal victimization became more concentrated in poorer economic groups. Victimization rates declined over time for both groups, but the decline was greater among wealthy relative to poor households. Further analysis found differential changes in the demographic composition (e.g., age, marital status, sex, employment, and community size) of income groups accounted for a substantial portion of the change in victimization experienced by these income groups (Thacher, 2004). Compared to the highest income group, the lowest income group had large increases in the proportion of its population that was young, not married, male, and urban. Thacher (2004) provided an important substantive foundation for studying how demographic composition changes affect crime rate changes and how the former can explain differences in crime rate changes between groups. His work also provided a sound methodological framework that can be employed to address changes in the proportion of victimization experienced by groups defined by something other than income, which we do in the present study.

It is important to consider the effect on victimization rates of changing demographics within economic groups (Thacher, 2004). Research on lifestyles and routine activity theory found victimization risks are associated with demographic characteristics such as age, gender, marital status, residential area type, and income. Thus, Thacher (2004, p. 92) proposed a demographic change hypothesis: “[C]hanges in the demographic structure of different income quintiles can explain changes in victimization inequality.” Changes in income group victimization rates can be broken down into two components. First, changes in the demographic structure of each income group partially account for changes in the distribution of victimization rates across income groups. Second, changes in the victimization risk of each demographic group within a given income group account for remaining changes in the distribution of victimization rates across the income groups (i.e., the demographic-victimization risk relationship changes).

Thacher (2004) found support for his demographic change hypothesis—demographic changes in each economic group since 1974 were consistent with increased victimization inequality. For instance, the percentages of high-risk aged individuals (15–24 years old) fell only slightly in the poorest economic group but fell rapidly in the wealthiest economic group. This change coincided with increasing victimization inequality between income groups.

To look at the joint effects of demographics on victimization inequality among income quintiles, Thacher (2004) estimated counterfactual victimization rates using a multistep procedure similar to the Blinder–Oaxaca decomposition technique (Blinder, 1973; Oaxaca, 1973). First, he used negative binomial regression models of demographic characteristics and victimization risks in 1994. He then used the coefficients from 1994 and applied the 1975 demographic data to calculate estimates for the income quintile victimization rates for each type of crime as if demographic compositions had not changed since 1975 (i.e., a counterfactual model). For violent crimes, the 1994 ratio of the victimization rate for the highest quintile to the lowest quintile was 1.65 and the 1975 ratio was 1.18. Applying the model to the 1975 data and calculating a counterfactual victimization rate yielded an equivalent ratio of only 1.36. Thus, 62% of the growth in victimization inequality (i.e., [1.65 –1.36]/[1.65 –1.18]) would have been avoided if the demographic composition of each quintile had remained constant since 1975, given the relationship between demographics and victimization risk changed in the way it actually did. The findings for theft and burglary were similar: 48% and 39%, respectively, of victimization would have been avoided had demographic composition remained the same.

Thacher (2004) did not make direct comparisons of how demographic changes affected victimization between rural, suburban, and urban areas but did use area type as a demographic variable. For violent crimes, the dummy variables for suburban and rural were both significant and their incident rate ratios (i.e., the ratio of victimization risk for those who have the attribute to those who do not) varied, with suburban being .86 and rural being .59 (Thacher, 2004, p. 109). This finding provides two questions. First, are demographic changes in the rural, suburban, and urban populations also associated with changes in violent victimization rates across the three area types? Second, are demographic changes partially responsible for the differential changes in violent victimization rates by area type during the crime decline?

The Link Between Individual-Level Victimization Risk and Aggregate Crime Rates

With little research on the effects of demographic population changes on aggregate crime rates, hypotheses tend to draw on the individual-level victimization risk literature. Thacher (2004) relied on lifestyle exposure and routine activity theories to determine which variables to use and to form his demographic change hypothesis. For example, if the poor are at greater victimization risk than the wealthy, and the concentration of the poor increases in suburban areas, suburban violence rates should increase. Other research has taken this approach, including Levitt’s (1999b) study of whether the changing age structure of the U.S. population partially explains changes in crime rates. He argued that as the fraction of the population with the highest propensity to commit crimes (young adults) rises so to would aggregate crime rates. He found some support for this hypothesis for the years 1960–1980 and 1980–1995 (Levitt, 1999b).

Steffensmeier and Harer (1999) offered a number of possible links between demographic composition changes and aggregate crime rate changes. Of note is their suggestion changes in the age composition of the United States are associated with changes in crime rates due to a shift in the collective conscience. During the 1980s and 1990s, the median age of the population increased due to large cohorts of baby boomers entering middle age. Steffensmeier and Harer (1999) argued this large shift in the age structure changed cultural values in the United States with a new focus on kindness and less emphasis on narcissism, leading to decreased crime rates.

Cohen and Felson (1979) suggested an important causal mechanism linking demographic composition changes and aggregate crime rate changes. Using a routine activity theory approach, they argued demographic changes (e.g., age composition, single-adult households, employment outside the home) in the population can affect predatory crime rates. Changes in the routine activities of individuals that occur at home, in jobs away from home, and in other activities away from home affect the probability that motivated offenders will converge in space and time with suitable targets in the absence of capable guardians. As one example, individuals living in single-adult households and those employed outside the home have little obligation to spend time at home, so they should experience higher rates of predatory crime victimization than their counterparts (Cohen & Felson, 1979). If the proportion of individuals with these lifestyles increases within a given population, predatory crime rates are also expected to increase.

Summaries of research on victimization risk consistently find young adults, males, single persons, and low-income individuals are at higher victimization risk. It is logical to conclude communities that experience increased proportions of young adults, males, single persons, and low-income individuals will have higher crime rates. No research has examined whether the demographic–victimization risk relationship is consistent across urban, suburban, and rural community types, but research shows structural factors do affect crime rates differently in communities of different sizes (Kaylen & Pridemore, 2011, 2013).

Rural, Suburban, and Urban Demographic Changes

The demographic compositions of rural, suburban, and urban areas change over time. While scholars have considered demographic factors as potential explanations for the crime decline (e.g., Barker, 2010; Baumer, 2008; Zimring, 2006), no research made specific rural, suburban, and urban comparisons of the effects of demographic composition on violence rates over time. If victimization risk varies by demographic groups, and demographic compositions changed in rural, suburban, and urban areas, it is likely changes in violence rates are partially explained by changes in demographic compositions (Sampson, 1983, 1986).

The United States experienced demographic changes during the great crime decline. Nationally, the population grew slower in the 2000s than the 1990s, in part because of the aging population and in part because of slower economic expansion in the 2000s (Berube, 2011). In the 2000s, younger baby boomers entered their late 40s, increasing the 45+ population at a rate of more than 18 times that of the under-45 group (Berube, 2011).

Demographic changes within city and suburban sectors of metropolitan areas were closely connected in the 1990s and 2000s (Berube, 2011; Dyson, 2011). As jobs became decentralized—more than 45% of metropolitan jobs were at least 10 miles from the downtown core by 2011—traditional economic distinctions between cities and suburbs blurred (Berube, 2011). In the early stages of suburban expansion, those in most need of the new jobs in suburban areas—poor and unemployed city residents—could not afford to live and work there (Squires, 2002). In the 2000s, housing in suburbs became more affordable due to a combination of policy changes like an aging suburban infrastructure and fair housing laws and subsidies for low-income homeownership (Berube, 2011). In the 1990s, cities were home to over half the metropolitan poor. In the 2000s, more than two-thirds of the increase in the metropolitan poor population occurred in the suburbs (Berube, 2011), due in part to faster suburban population growth (Berube & Frey, 2002).

Rural areas also experienced demographic changes in the 1990s and 2000s. Compared to the 102 largest metropolitan areas and smaller metropolitan areas, rural areas had the highest overall poverty rates between 1990 and 2000 (Berube & Frey, 2002; Weisheit & Donnermeyer, 2000). Despite this, rural areas experienced population growth in the 1990s, increasing by more than 5 million residents compared to an increase of less than 1.3 million in the 1980s. Two-thirds of population growth was due to migration from metropolitan areas (Johnson & Cromartie, 2006). The net migration patterns of rural areas were age selective, with significant out-migration of young adults aged 20–29 and large increases in older adults (Artz, 2003; Johnson & Cromartie, 2006). These age selective migration patterns have long-term consequences for rural areas. Natural decreases in population result from deaths among older cohorts outnumbering births among younger cohorts. In the 1990s, 29% of nonmetropolitan counties experienced natural population decreases compared to 10% in the 1980s (Johnson & Cromartie, 2006).

Research Questions

We tested the demographic change hypothesis across the period of 1993–2005 by asking if changes in the demographic composition of rural, suburban, and urban populations helped account for changes in (1) violent victimization trends among rural, suburban, and urban areas and (2) urban–suburban, urban–rural, and suburban–rural victimization rate ratios. From the lifestyle-exposure perspective (see Meier & Miethe, 1993; Thacher, 2004), we hypothesized risk of violent victimization is highest for those who are unemployed, unmarried, young, male, and low income. Thus, changes in the demographic composition of these characteristics should lead to changes in rates of violence. We also addressed whether demographic–victimization risk relationships varied by community size, hypothesizing changes in income have a larger effect on victimization risk in urban relative to rural areas.

Data and Method

Data

The data for this study came from the NCVS Record-Type household-level, person-level, and incident-level files for the years 1993–2005.¹ We chose 1993–2005 for two reasons. First, the NCVS underwent systematic redesigns in 1992 and 2006 (U.S. Department of Justice, 2011), increasing substantially the challenges of an analysis across a broader range of years. Second, and more importantly, this range of years was a dynamic period encompassing the great American crime decline. During this time, crime went from an almost all-time high in the early 1990s to an almost all-time low in the mid-2000s, all while America experienced major economic and demographic changes.²

Unlike the Uniform Crime Reporting (UCR) Program, the NCVS includes data about incidents not reported to the police. The survey also includes demographic data about individuals who were both victimized and not victimized, making it ideal for our current counterfactual analysis.

NCVS data are aggregated into three categories based on area type—rural, suburban, or urban—using the Metropolitan Statistical Area (MSA) designations provided in the household-level files. These designations are determined by the U.S. Office of Management and Budget and define central city, outside central city, and nonmetropolitan. “Urban,” “suburban,” and “rural” are commonly used to label these areas (e.g., Rennison, Dragiewicz, & DeKeseredy, 2013) and are used here. We merged yearly household-level files with the corresponding incident-level and person-level files to connect the MSA variable to these files.

The dependent variable was aggravated assault incident rate, including incidents classified as aggravated assault with injuries and attempted aggravated assault with a weapon. We chose aggravated assault because of its serious nature (though it is more common than homicide and thus provides more cases) and greater likelihood of being reported to NCVS survey takers and because this study originated from a larger ongoing project on differences in aggravated assault rates in rural areas. We calculated incident rate by dividing counts of victimization incidents by number of survey respondents. If multiple people were victimized in an incident, only one incident was counted unless more than one victim participated in the survey. Incident rates are necessary for this analysis because demographic information about additional victims other than the respondent was not available. The independent variables were employment, marital status, young, female, and household income. See the Appendix for how we operationalized each of these variables.

Methods

Part 1: Aggravated Assault Incident Rates, 1993–2005

We calculated aggravated assault incident rates for each area type and each year by dividing the counts of incidents of aggravated assaults by counts of survey respondents, after the application of survey weights. We also calculated 95% confidence intervals for each rate. Finally, we calculated urban–suburban, urban–rural, and suburban–rural ratios to later compare with similar ratios calculated in Part 3.

Because the NCVS samples the noninstitutionalized U.S. population aged 12 and above, survey weights were used to make estimates for the entire population. Weights adjust for nonresponse and under coverage of certain groups (Rennison & Rand, 2007). Counts of aggravated assault victimization incidents are normally multiplied by incident weights, and counts of respondents are multiplied by person weights. The incident weights are created by dividing the respondent’s person weight by the total number of victims in the incident. For instance, if three people are assaulted in the same incident, the victim weight is divided by three.³

An important aspect of the NCVS data is counting series victimizations. Respondents are first given a screening questionnaire that asks about their victimization experiences over the previous 6 months. The second step in the process is the Crime Incident Report, where specific details about victimization incidents are collected. Individuals repeatedly victimized often cannot distinguish between events. The NCVS addresses this by employing a series victimization protocol. A series victimization is when the respondent reported experiencing the same or similar crime 6 or more times during the previous 6 months and is unable to describe each event in detail. In these cases, the number of reported series victimizations are recorded and details about only the most recent victimization incident is collected (Lauritsen, Owens, Planty, Rand, & Truman, 2012). How series victimizations are included can have a large impact on estimates of national crime rates. The results from Lauritsen, Owens, Planty, Rand, and Truman (2012) suggest using a capping method when including series victimizations in rate calculations. To account for series victimizations in the annual violence rates in the current study, we employed the methods suggested by Lauritsen et al. (2012). Series victimizations were included in rate calculations with a cap of 10 victimizations per 6-month reference period. When respondents reported being unable to recall how many victimizations occurred (but knew it was more than six), we used the modal response of six victimizations.

The NCVS employs a stratified, multistage cluster sample design. Statistics that assume a simple random sample cannot be used for calculating standard errors and confidence intervals with these data because clustering of households results in variances that are smaller than they would be using a simple random sample design. Use of sample weights in calculating rates (i.e., summing the weights) also creates artificially smaller variances. NCVS data include two variables to adjust for the complex sample design: V2117 is a pseudo-strata variable and V2118 is a pseudo-primary sampling unit (PSU) variable.⁴

Part 2: Isolated Effects of Demographic Changes on Aggravated Assault Incident Rate Changes

The second part of the analysis is a bivariate examination of demographic composition changes and aggravated assault incident rate changes from 1993 to 2005 in each area type. We calculated the proportions of the rural, suburban, and urban populations that were young, female, married, employed, and in each income category for the beginning (1993) and the end (2005) of the time period under study. We also calculated the victimization rates for each of the independent variable attributes at each area type for 1993 and 2005. We applied the person weight (V3080) to the respondent data and the incident weight (V4527) to the incident data.

We also examined the effect of area-specific changes in demographic composition on aggravated assault incident rates. First, we calculated percentage change from 1993 to 2005 area-specific rates by subtracting the 1993 rate from the 2005 rate and dividing by the 1993 rate. For each demographic characteristic, we calculated three values for each area type. The first was what the 2005 incident rate would be if demographic composition had not changed since 1993. To do this, we multiplied each group-specific incident rate in 2005 by the group’s 1993 population proportion, then summed all group values. For example, for sex, the 2005 female incident rate was multiplied by the 1993 female population proportion, the 2005 male incident rate by the 1993 male population proportion, and the two values summed. The second value was percentage change from the observed 1993 rate to the hypothetical 2005 rate (i.e., if demographic composition had not changed). The third value was percentage change in actual rate due to changes in demographic composition. We calculated this by subtracting the hypothetical (composition-constant) change from actual change, divided by the actual change. Negative values indicate the incident rate would have grown more than it actually did if composition had remained constant.

Part 3: Combined Effects of Demographic Changes on Aggravated Assault Incident Rate Changes

To determine the contribution of the combined demographic changes on victimization inequality changes among the area types, we conducted two stages of analysis. The first involved modeling the relationship between demographic characteristics and victimization in 2005. We used the negative binomial estimator for several reasons. First, we are modeling a count variable, number of victimization incidents. Second, we selected the negative binomial estimator over the Poisson estimator because of overdispersion of the data. Poisson regression models assume the variance is equal to the mean, while negative binomial regression relaxes this assumption. For our data, the variance was greater than the mean. We employed the coefficients from this model in the second stage of this analysis, the counterfactual models. To calculate estimates of what area-specific victimization rates would be in 2005 if each demographic-victimization relationship changed as they actually did (i.e., 2005 coefficients) but demographic composition of each area type remained the same from 1993 to 2005, we applied 2005 demographic data to the model estimated in the first stage. We multiplied counterfactual incident counts by 1993 person weight (V3080) to get a weighted incident count. Finally, we divided the sum of the weighted incident counts by weighted respondent counts to calculate counterfactual aggravated assault incident rates. We used these rates to calculate urban–suburban, urban–rural, and suburban–rural ratios and compared them with the observed ratios calculated earlier. Substantial changes in ratios indicate demographic composition changes affected victimization unequally across area types. We calculated the percentage of the change in victimization incident rate due to changes in demographic composition by subtracting the hypothetical (composition-constant) change from actual change, divided by actual change. Negative values indicate the incident rate would have grown more than it actually did if composition had remained constant.

Part 4: Challenging the Demographic–Victimization Risk Relationship Assumption

In Part 3, an assumption about the demographic–victimization risk relationship was imposed. Specifically, it was assumed each demographic–victimization risk relationship was the same in rural, suburban, and urban areas. This assumption has not been empirically tested. Research shows some social processes related to violent crime do not operate the same way in rural and urban areas (Kaylen & Pridemore, 2011, 2013). An alternative method for determining the joint effects of demographic changes on victimization rate changes is to calculate a separate model for each area type and then follow the same steps as in Part 3. This method provides more flexibility in the demographic–victimization risk assumption, allowing—but not forcing—these relationships to vary among community sizes.

Results

Part 1: Aggravated Assault Incident Rates, 1993–2005

Figure 1 shows annual aggravated assault incident rates and 95% confidence intervals by area type from 1993 to 2005. Incident rates were highest at the beginning of the period and declined until the early 2000s when they began to level off. Urban rates remained highest for the entire period, with the biggest difference in area type rates in 1993 when the urban confidence interval did not overlap the suburban or rural confidence intervals. Rural areas experienced a large decline in rates from 1996 to 1998, reaching a low of 1.07 (95% CI [0.67, 1.47]), which was much lower than urban and suburban rates. Overall, the relative decline in incident rates from 1993 to 2005 was similar for urban and suburban areas and much lower for rural areas.

Figure 1.

Annual aggravated assault incident rates per 1,000 by area type with 95% confidence intervals, 1993–2005.

The size of the confidence intervals is affected by sample size and amount of variability in the estimate. The 1993 estimates shown here are based on a smaller sample size than the other years because it was part of the redesign overlap period. Only the NCVS sample (as opposed to the NCS sample) is included in these estimates. As a result, the 1993 confidence intervals are expected to be relatively large. Due to budget cuts, the NCVS sample size has been reduced over time. The sample was reduced by 12% in 1996 and by another 4% in 2002 (Rennison & Rand, 2007). The amount of variability in the annual area type estimates also varies, affecting the size of the confidence intervals. The urban confidence interval in 2000, for example, is smaller than the urban confidence interval in 2001. In both of these years, the urban sample was approximately 48,000, but the variability in the estimates for these 2 years was different.

Part 2: Isolated Effects of Demographic Changes on Aggravated Assault Incident Rate Changes

Table 1 shows the demographic composition of each area type in 1993 and 2005. While the percentage of the population who were young increased in urban areas, it decreased slightly in suburban areas (21.8%–21.0%) and more so in rural areas (22.6%–19.9%). The proportion of the population who were married decreased in all area types, while the proportion not reporting marital status increased slightly. The female–male proportions in all area types changed only slightly. The percentage employed increased in urban areas and decreased in suburban and rural areas. Finally, across all three area types, the percentages of the population in the lower income categories decreased while those in the higher income categories increased. The percentage of the population not reporting household income also increased across all three area types, with approximately one quarter of each area type’s population not reporting in 2005.

Table 1.

Percentage of Demographic Composition by Area Type, 1993 and 2005.

	Urban		Suburban		Rural
	1993	2005	1993	2005	1993	2005
Young	23.9	25.3	21.8	21.0	22.6	19.9
Not young	76.1	74.5	78.2	79.0	77.4	80.1
Married	43.4	40.6	56.5	53.6	57.2	54.6
Not married	56.4	58.2	43.3	45.6	42.7	44.8
Marital status unknown	0.3	1.2	0.2	0.8	0.1	0.7
Female	52.8	51.6	51.0	51.2	51.4	51.3
Male	47.2	48.4	49.0	48.8	48.6	48.7
Employed	61.4	62.4	65.2	63.3	59.4	58.9
Not employed	31.1	29.5	27.1	28.0	32.4	34.2
Employment unknown	7.5	8.1	7.7	8.7	8.2	6.9
Income
< $7,500	12.0	4.8	4.9	1.8	11.6	4.9
$7,500–$14,999	14.6	7.2	8.8	4.3	17.2	8.6
$15,000–$24,999	16.2	10.2	13.6	7.2	18.8	12.0
$25,000–$34,999	13.7	9.7	14.4	8.1	16.0	10.2
$35,000–$49,999	14.2	11.7	17.6	12.3	14.0	13.5
$50,000–$74,999	10.0	12.1	16.3	15.6	8.1	14.0
$75,000 or more	6.2	16.5	11.6	26.9	3.2	12.1
Income unknown	13.1	27.9	12.8	23.8	11.2	24.7

Table 2 presents the aggravated assault victimization incident rate per 1,000 for each demographic group in each area type in 1993 and 2005. Aggregating rates by area type and demographic group greatly reduces the sample sizes used to calculate each weighted rate. In urban and suburban areas, the young aggravated assault rate decreased from 1993 to 2005, while the rural rate increased slightly. The not young aggravated assault rates were lower than the young rates at each area type and in each year. From 1993 to 2005, these rates decreased for all area types. The aggravated assault rates for married and not married also decreased across all areas, with rates for unmarried higher than rates for married in each area and in each year. Male and female aggravated assault victimization rates decreased in all area types, with male rates higher than female rates. With the exception of aggravated assault victimization rates for not employed rural, victimization rates for employed and not employed decreased from 1993 to 2005. The aggravated assault rates for urban unemployed were higher than the urban employed rates, the suburban 1993 employed victimization rate was higher than the unemployed rate, the suburban 2005 employed victimization rate was lower than the unemployed rate, and the rural employed victimization rates were higher than unemployed rates. Finally, aggravated assault rates were generally higher for the lower income groups. With one exception (rural $15,000–$24,999), income group victimization rates decreased from 1993 to 2005.

Table 2.

Aggravated Assault Incident Rate per 1,000 by Demographic Composition and Area Type, 1993 and 2005.

	Urban		Suburban		Rural
	1993	2005	1993	2005	1993	2005
Young	21.01	3.39	9.83	2.69	6.24	6.48
Not young	3.96	1.68	2.86	0.62	2.13	0.63
Married	2.23	1.27	1.96	0.40	1.68	1.21
Not married	12.50	2.75	7.41	1.83	4.92	2.51
Marital status unknown	0.00	0.00	33.70	0.70	0.00	0.00
Female	7.77	1.72	2.47	0.85	2.07	1.46
Male	8.35	2.53	6.38	1.28	4.10	2.13
Employed	6.07	2.15	4.29	0.92	3.91	1.74
Not employed	7.65	2.30	3.19	1.15	0.87	1.65
Employment unknown	26.12	1.26	9.31	1.82	5.03	2.92
Income
< $7,500	19.57	1.79	14.82	4.46	4.93	2.60
$7,500–$14,999	6.40	5.93	7.23	5.69	3.29	1.67
$15,000–$24,999	6.86	2.82	4.53	0.55	3.90	4.83
$25,000–$34,999	7.52	2.84	3.70	1.69	1.82	1.06
$35,000–$49,999	6.37	2.82	4.04	1.00	2.31	0.22
$50,000–$74,999	5.12	1.05	3.31	1.01	3.23	2.80
$75,000 or more	2.19	1.12	1.62	0.53	1.65	0.79
Income unknown	8.17	1.43	3.41	0.57	2.31	1.27

Table 3 presents the hypothetical 2005 area-specific aggravated assault incident rates if each of the demographic variable compositions had not changed since 1993, the percentage of the change from 1993 to these hypothetical values and the percentage of the real change in rates between 1993 and 2005 due to the demographic change. Age composition changes had the largest effect on rural aggravated assault rates (accounting for an 8.94% decrease), while urban (1.42% increase) and suburban (0.94% decrease) effects were much smaller. Marital status changes had small effects across all area types, and all percentage changes due to marital status composition changes were positive. Sex composition had the smallest effects on victimization rate changes of all demographics, with a negligible positive percentage change for urban and no percentage change for suburban and rural areas. Employment status changes were fairly small with a negative effect on the change in victimization rate for urban areas and positive effects on the changes in victimization rates for suburban and rural areas. Finally, income composition changes had large negative effects on changes in victimization rates across all area types. Suburban areas experienced the largest percentage change in aggravated assault victimization due to income composition changes (44.34%), followed by urban (27.96%) and rural (17.32%).

Table 3.

Decomposition of Changes in Aggravated Assault Incident Rates by Area Type, 1993–2005.

	Urban	Suburban	Rural
1993 rate	8.04	4.38	3.06
2005 rate	2.11	1.06	1.79
% change from 1993	–73.76	–75.80	–41.50
2005 rate if age composition had not changed	2.08	1.07	1.95
% change from 1993	–74.13	–75.57	–36.27
% of observed change 1993–2005 due to change in age composition	1.42	–0.94	–8.94
2005 rate if marital status composition had not changed	2.10	1.02	1.77
% change from 1993	–73.88	–76.71	–42.16
% of observed change 1993–2005 due to change in marital status composition	0.47	3.77	1.12
2005 rate if sex composition had not changed	2.10	1.06	1.79
% change from 1993	–73.88	–75.80	–41.50
% of observed change 1993–2005 due to change in sex composition	0.47	0.00	0.00
2005 rate if employment status composition had not changed	2.13	1.05	1.77
% change from 1993	–73.51	–76.03	–42.16
% of observed change 1993–2005 due to change in employment status composition	–0.95	0.94	1.12
2005 rate if income composition had not changed	2.70	1.53	2.10
% change from 1993	–66.42	–65.07	–31.37
% of observed change 1993–2005 due to change in income composition	–27.96	–44.34	–17.32

Part 3: Combined Effects of Demographic Changes on Aggravated Assault Incident Rate Changes

Table 4 shows results from the negative binomial regression model using the 2005 data. The young variable is significant in the positive direction as expected. Neither marriage variable (with not being married the reference category) is significant. Female is significant in the negative direction as expected. The employed variable (with not being employed the reference category) is not significant, while the unknown employment status variable is significant in the negative direction. Two of the household income categories were significant compared to the $75,000 or more category, $7,500–$14,999 (positive) and $15,000–$24,999 (positive). Finally, neither area type variable was significant (with rural the reference category).

Table 4.

2005 Aggravated Assault Incidents Regressed on Explanatory Variables.

Independent Variables	Aggravated Assault Incidents
Not young	[Reference]
Young
b	1.232
Standard error (SE)	.319
p	<.001
Not married	[Reference]
Married
b	–.399
SE	.333
p	.233
Marital status unknown
b	–1.181
SE	1.030
p	.253
Male	[Reference]
Female
b	–.437
SE	.211
p	.040
Not employed	[Reference]
Employed
b	.050
SE	.234
p	.831
Employment unknown
b	–.582
SE	.276
p	.037
Income
$75,000 or more	[Reference]
< $7,500
b	.877
SE	.453
p	.055
$7,500–$14,999
b	1.612
SE	.435
p	<.001
$15,000–$24,999
b	1.025
SE	.470
p	.031
$25,000–$34,999
b	.842
SE	.474
p	.078
$35,000–$49,999
b	.561
SE	.378
p	.140
$50,000–$74,999
b	.568
SE	.425
p	.183
Income unknown
b	.222
SE	.373
p	.551
Rural	[Reference]
Urban
b	.153
SE	.402
p	.704
Suburban
b	–.360
SE	.419
p	.392
Constant
b	–7.080
SE	.425
p	<.001
n (unweighted)	158,529
n (weighted)	487,435,471

Table 5 presents the observed aggravated assault incident rates by area type in 1993 and 2005 as well as the counterfactual rates calculated using the 1993 data and the 2005 model. The table also includes the percentage change from the 1993 rate to the 2005 observed and hypothetical rates. Finally, the table includes the percentage change in rates due to demographic composition change. For urban areas, the incident rate decreased 73.76% between 1993 and 2005. The rate would have decreased by a smaller percentage, 66.29%, had demographic composition remained constant. That is, 10.13% of the change in urban victimization rates between 1993 and 2005 is due to changes in the demographic composition. Suburban rates tell a similar story, with 7.53% of the change in suburban victimization rates from 1993 to 2005 due to the suburban demographic composition changing. Finally, the observed rural victimization rate experienced a smaller percentage change from 1993 to 2005, 41.50%, but the victimization rate change due to changing demographic composition is much larger, 37.01%, than in urban and suburban areas. Another difference between rural and nonrural areas is that the rural victimization rate under the hypothetical scenario would have only decreased from 1993 by 26.14%, a much smaller percentage than what was observed relative to nonrural areas.

Table 5.

Observed and Counterfactual Aggravated Assault Incident Rates per 1,000 and Percentage Changes by Area Type in 2005.

	Urban	Suburban	Rural
1993 observed rate	8.04	4.38	3.06
2005 observed rate	2.11	1.06	1.79
2005 counterfactual rate^a	2.71	1.31	2.26
% change from 1993 observed to 2005 observed	–73.76	–75.80	–41.50
% change from 1993 observed to 2005 counterfactual	–66.29	–70.09	–26.14
% change in rates due to demographic change	–10.13	–7.53	–37.01

^aCounterfactual rates are calculated using coefficients from a negative binomial regression model using the full sample.

Table 6 shows the area type ratios for each of the rates. The area type ratios include urban–suburban, urban–rural, and suburban–rural. From 1993 to 2005, the observed urban–suburban ratio increased slightly from 1.84 to 1.99. These values indicate urban areas had higher rates than suburban areas (nearly double in 2005). The increase in the ratio indicates the relative difference in the area type rates increased from 1993 to 2005. The opposite pattern is seen in the urban–rural ratio from 1993 to 2005. This ratio went from 2.63 in 1993, meaning the urban rate was more than two and a half times greater than the rural rate, to 1.18 in 2005, meaning the urban rate was only slightly higher than the rural rate. This indicates the relative difference between the urban and the rural rates decreased dramatically between 1993 and 2005. Finally, the suburban–rural ratio went from 1.43 in 1993, indicating the suburban rate was nearly one and a half times higher than the rural rate, to 0.59 in 2005, meaning the suburban rate was just over half the rural rate, again showing a dramatic change in the suburban–rural ratio.

Table 6.

Observed and Counterfactual Aggravated Assault Incident Rate Area Type Ratios in 2005.

	Urban–Suburban	Urban–Rural	Suburban–Rural
1993 observed	1.84	2.63	1.43
2005 observed	1.99	1.18	0.59
2005 counterfactual^a	2.07	1.20	0.58

^aCounterfactual rates are calculated using coefficients from a negative binomial regression model using the full sample.

The differences between the 2005 observed and the 2005 counterfactual ratios varied in size and direction. The urban–suburban ratio increased from 1.99 to 2.07, indicating the difference in the incident rates for urban and suburban areas would have increased more in the hypothetical scenario than what was actually observed. On the other hand, the urban–rural and suburban–rural ratios were nearly identical in the hypothetical scenario and what was actually observed. In short, aggravated assault incident rate area type ratios changed from 1993 to 2005, but the combined changes in demographic composition played a small role in these changes.

Part 4: Challenging the Demographic–Victimization Risk Relationship Assumption

The results presented in Part 3 were calculated under the assumption that demographic–victimization risk relationships are the same in rural, suburban, and urban areas. This may not be true. We estimated separate regression models for each area and calculated new counterfactual rates and area type ratios based on these models. Table 7 presents results by area type from the negative binomial regression models using the 2005 data. In the suburban and rural models, the age variable was significant in the positive direction as expected. Being married compared to not being married was significant in the negative direction only for suburban areas. The classification of marriage unknown compared to that of unmarried was significant in the negative direction in urban and rural areas. Being female was significant in the negative direction as expected only for rural areas. Employment variables were not significant in any model. Finally, some of the income categories were significant compared to the $75,000 or more category. Many of the income groups are quite small, likely contributing to the lack of significance.

Table 7.

2005 Aggravated Assault Incidents Regressed on Explanatory Variables, by Area Type.

Independent Variables	Urban	Suburban	Rural
Not young	[Reference]	[Reference]	[Reference]
Young
B	.614	1.114	2.560
SE	.359	.317	.822
p	.089	.001	.003
Not married	[Reference]	[Reference]	[Reference]
Married
B	–.493	–.755	.298
SE	.352	.347	.703
p	.163	.031	.674
Marital status unknown
B	–25.713	–.278	–22.350
SE	.270	.992	.646
p	<.001	.780	<.001
Male	[Reference]	[Reference]	[Reference]
Female
B	–.478	–.428	–.536
SE	.267	.285	.259
p	.075	.134	.044
Unemployed	[Reference]	[Reference]	[Reference]
Employed
B	.056	–.032	.337
SE	.301	.395	.568
p	.853	.936	.555
Employment unknown
b	–1.062	–.276	–.419
s.e.	.630	.382	.586
p	.094	.472	.477
Income
$75,000 or more	[Reference]	[Reference]	[Reference]
< $7,500
B	.201	1.772	.627
SE	.787	.634	1.080
p	.799	.006	.564
$7,500–$14,999
B	1.495	2.154	.642
SE	.657	.574	1.118
p	.024	<.001	.569
$15,000–$24,999
B	.812	–.072	1.689
SE	.632	.675	.072
p	.201	.916	.121
$25,000–$34,999
B	.837	1.013	.105
SE	.760	.539	1.117
p	.273	.062	.925
$35,000–$49,999
B	.854	.586	–1.384
SE	.562	.578	1.324
p	.131	.312	.301
$50,000–$74,999
B	–.049	.600	.930
SE	.620	.579	1.129
p	.938	.302	.414
Income unknown
B	.187	.032	.491
SE	.546	.557	.948
p	.732	.954	.607
Constant
B	–6.485	–7.340	–8.177
SE	.527	.516	.989
p	<.001	<.001	<.001
n (unweighted)	40,217	75,756	17,609
n (weighted)	141,476,490	256,924,668	89,034,313

Table 8 presents the observed aggravated assault incident rates by area type in 1993 and 2005 as well as the counterfactual rates calculated using the 1993 data and 2005 models. The table includes the percentage change from the 1993 rate to the 2005 observed and hypothetical rates. The table also includes the percentage change in rates due to demographic composition change. For urban areas, the counterfactual incident rate is 68.2% smaller than the 1993 rate, compared with 73.8% difference in the observed rates; 7.6% of the change in urban observed rates was due to changes in urban demographic composition. Suburban rates are similar. The counterfactual incident rate is 67.6% smaller than the 1993 rate, compared to 75.8% difference in the observed rates; 10.8% of the change in the suburban observed rates was due to changes in suburban demographic composition. Finally, the rural counterfactual incident rate is 27.1% smaller than the 1993 rate, compared to 41.5% difference in the observed rates; 34.7% of the change in the rural observed rates was due to changes in rural demographic composition.

Table 8.

Observed and Counterfactual Aggravated Assault Incident Rates per 1,000 and Percentage Changes by Area Type in 2005.

	Urban	Suburban	Rural
1993 observed rate	8.04	4.38	3.06
2005 observed rate	2.11	1.06	1.79
2005 counterfactual rate^a	2.56	1.42	2.23
% change from 1993 observed to 2005 observed	–73.76	–75.80	–41.50
% change from 1993 observed to 2005 counterfactual	–68.16	–67.58	–27.12
% change in rates due to demographic change	–7.59	–10.84	–34.65

^aCounterfactual rates are calculated using coefficients from negative binomial regression models using separate area type samples.

Table 9 presents the area type ratios. The differences between the 1993 observed and 2005 counterfactual ratios varied in size and direction. The urban–suburban ratio decreased slightly from 1.84 to 1.80 (down 2.2%), indicating relative incident rates for urban and suburban areas would have remained nearly the same in the hypothetical scenario. The urban–rural ratio decreased by a larger amount, from 2.63 in 1993 to 1.15 in the hypothetical scenario (a decrease of 56.3%). The suburban–rural ratio decreased and changed direction, going from 1.43 in 1993 to 0.64 in the hypothetical scenario (a decrease of 55.2%). The percentage change between the 1993 and 2005 hypothetical urban–rural and suburban–rural ratios was about the same, but the suburban–rural ratio changed from >1.0 (suburban rate was higher) to <1.0 (suburban rate was lower). In short, aggravated assault incident rate area type ratios changed from 1993 to 2005, but the combined changes in demographic composition played only a small role in these changes, with the exception of the urban–suburban ratio. Allowing demographic–victimization risk relationships to vary by community type resulted in a substantial change in the urban–suburban counterfactual incident rate ratio compared to the results presented in Part 3.

Table 9.

Observed and Counterfactual Aggravated Assault Incident Rate Area Type Ratios in 2005.

	Urban–Suburban	Urban–Rural	Suburban–Rural
1993 observed	1.84	2.63	1.43
2005 observed	1.99	1.18	0.59
2005 counterfactual^a	1.80	1.15	0.64

^aCounterfactual rates are calculated using coefficients from negative binomial regression models using separate area type samples.

Discussion

We assessed whether demographic changes in rural, suburban, and urban areas accounted for their changes in violent victimization rates from 1993 to 2005. Understanding the role of demographic changes on violence rate changes across different communities is important for criminological research. Following Thacher’s (2004) theoretical and methodological framework for testing demographic change hypotheses, we used NCVS data in a three-step process. First, we calculated rural, suburban, and urban aggravated assault incident rates, showing that although urban rates were highest throughout the study period, the gap between urban and non-urban rates decreased dramatically by 2005. Second, bivariate analyses revealed demographic composition changes affected aggravated assault incident rate changes differently across area types, with age and income having the largest effects. Third, we calculated counterfactual incident rates to determine the percentage of incident rate changes due to combined demographic composition changes. This final step revealed demographic changes affected rural aggravated assault incident rate changes the most. Further analyses supported the bivariate finding that demographic–victimization risk relationships vary by area type, especially for age and income.

Our analyses reveal the drop in violent victimization in the United States from the early 1990s to the mid-2000s was not experienced equally in all area types. Relative to urban and suburban areas, rural areas experienced substantially smaller declines in aggravated assault incident rates. The urban–suburban rate ratio increased 8.2%, the urban–rural rate ratio decreased 55.1%, and the suburban–rural rate ratio decreased 58.7%. This is due to small rate decreases experienced by rural areas and the relatively larger decreases experienced by urban and suburban areas. Our analyses also show changes in the demographic composition of these areas contributed to changes in their victimization rates, although effects were mostly modest. While the effect sizes of demographic composition on changes in urban and suburban victimization rates were similar in magnitude, the effect on victimization rates in rural areas was larger. Our analyses reveal the demographic–victimization risk relationship varies by community size, with changes in demographic composition affecting violent victimization rates in rural, suburban, and urban areas differentially.

Isolated Demographic Effects on Violence Rates

Between 1993 and 2005, rural, suburban, and urban areas in the United States generally experienced small changes in demographic composition but substantial shifts in aggravated assault incidents. In general, we find these incident rate changes were driven in small part by the isolated demographic changes. Thacher (2004) likewise found small demographic changes had small effects on victimization rate changes. When examined individually, age, marital status, sex, and employment composition had relatively small effects on victimization rate changes, while income composition had the largest effects.

The urban population became younger while the suburban and rural populations grew older, with rural areas having the largest percent change (percent young decreased 2.7%). The urban age composition changed in the direction expected by the demographic change hypothesis to lead to a higher victimization rate, but the percentage change in the victimization rate due to the changing age composition was negligible (less than 1%). Age composition changes led to decreases in rural and suburban rates as expected by the demographic change hypothesis. The suburban percentage change in violence rate due to changing age composition was negligible (less than 1%), while percentage change in rural areas was nearly 9%. Aside from income, this percentage change in violence rates due to demographic composition change was the largest.

All three area types experienced decreases in the percentage of the population who were married, ranging from 2.6% (rural) to 2.9% (suburban). Marital status composition changes did affect victimization rates in the hypothesized direction, but changes were small. Changes in the sex and employment compositions of all areas were extremely small. Urban areas experienced a small (1.2%) decrease in the proportion of the population who were female, while suburban and urban areas experienced only negligible changes in sex composition. Likewise, urban areas experienced a slight increase in percentage employed (1%), while suburban and rural areas experienced slight decreases (1.9% and 0.5%, respectively). Both sex and employment composition changes had negligible effects on violence rate changes.

Finally, all areas experienced decreases in the proportion of those reporting household incomes in the lowest categories and increases in the highest categories. These changes are expected to lead to decreased victimization rates according to the demographic change hypothesis. Urban (7.2%) and rural (6.7%) areas had decreases in the lowest income category that more than doubled the suburban decrease (3.1%). On the other hand, suburban areas experienced the greatest increase (15.3%) in the highest income category followed by urban (10.3%) and rural (8.9%) areas. Caution must be taken when interpreting percent changes in income categories because the income unknown category increased by over 10% in all area types. The income composition changes had the largest effects of all demographics on violence rate changes, contributing to decreased rates in all area types. Suburban areas experienced the largest income effects, with 44% of the change in aggravated assault incident rates due to changes in income composition. The change in violence rates in suburban areas due to income composition changes was more than twice that for rural areas and more than 50% that for urban areas. Many of these income groups have very small counts, likely contributing to the lack of significant effects for the income variables.

The isolated demographic effects on violence rates operated as expected by the demographic change hypothesis and lifestyle exposure and routine activity theories, although these effects were generally small. That is, demographic changes tended to be associated with changes in violence rates in the direction expected by theory, but they were relatively small in all areas and contributed minimally to violence rate changes. Thacher (2004) found larger effects of the same isolated demographic changes in the poorest and richest quintiles, ranging from explaining 2% to 72% of the change in violence rates, while our study found percentages between 0% and 44%, with most under 10%. This finding suggests something more than isolated demographic changes contributed to the decreases in rural, suburban, and urban violence rates between 1993 and 2005. It may be these demographic changes do not have large isolated effects but rather operate in concert with other demographic, social, or economic changes to lead to negative consequences. These possibilities should be explored in future research.

Combined Demographic Effects on Violence Rates

Although definitive simple explanations for the great American crime decline remain elusive, our findings illuminate some of the responsible factors. Under the assumption demographic–victimization risk relationships do not vary by community size, the combined contribution of all demographic changes on victimization rates in rural, suburban, and urban areas made the 2005 rates lower than they would have been if the demographic compositions of these areas had remained at 1993 levels. Urban and suburban areas experienced similarly sized benefits from demographic changes, while rural areas experienced much larger benefits. Demographic changes from 1993 to 2005 accounted for 7.5% of the decrease in suburban victimization incident rates and 10.1% of the decrease in urban rates. Alternatively, rural demographic changes accounted for 37.8% of the decrease in rural victimization incident rates.

Our results suggest rural areas are more sensitive to demographic changes than nonrural areas. When looking at the isolated effects of demographic changes on violence rates, one would not immediately expect rural areas to experience the greatest effects from the combination of all demographic changes. However, the isolated effect of age composition on violence rate changes was largest in rural areas, but all other isolated effect sizes were generally similar across area types. The combined effects were calculated using regression models such that, for any one demographic characteristic the effect is calculated while controlling for the other demographic characteristics. Given the substantially larger combined demographic effects in rural areas compared to the isolated effects it is likely co-occurring demographic changes have a greater effect on rural relative to non-rural violence rates.

Demographics and Violent Victimization by Area Type

When we relax the assumption demographic–victimization risk relationships do not vary by community type, the percentage changes in victimization incident rates due to demographic changes are similar to those described above. Regardless of assumptions about demographic–victimization risk relationships, demographic changes had much larger effects on lowering violence rates in rural than non-rural areas.

Results of the regression models presented in Table 7 support the decision to drop the assumption demographic–victimization risk relationships do not vary by community size. The young aggravated assault relationship was positive and significant across all community types, but the magnitude of the relationship varied, with rural areas experiencing the largest effect of being young on violence. Being married versus not married was significant only in suburban areas. Being female versus male was negative and significant in urban and rural areas, at about the same magnitude, but not suburban areas. Differences by area type in income results became clear when modeling the demographic–victimization relationships separately for each area.

Two demographic–victimization risk relationships, age and income, had the largest area type-specific effects. Age composition played an import role in changes in rural violence rates. From 1993 to 2005, rural areas experienced a 12% decrease in the proportion of the population young, with potential ramifications for violence rates. First, young individuals in general have higher risks of victimization than older individuals (Levitt, 1999b; Meier & Miethe, 1993; Steffensmeier & Harer, 1999; Thacher, 2004), so this loss of young individuals is expected to be associated with overall decreased rural victimization rates. Second, the rural change in age composition (fewer young residents) accounted for a larger percentage change in the violence rates (a decrease of nearly 9%) than urban and suburban areas. Third, when looking at the combined effects of demographic changes in each area, the coefficient for young was larger for rural than non-rural areas. This difference in coefficients again indicates age composition has the strongest effect on violence rates in rural areas.

Relative to other demographic changes, income composition changes had larger negative effects on violence rates and affected suburban areas more than urban and rural areas. The percentage of decline in violence rates due to income composition changes was largest in suburban areas. From 1993 to 2005, all areas experienced similar percent decreases in the proportion of the population in the poorest income categories. Suburban areas experienced the largest decreases in the middle-income categories, followed by rural and then urban areas. Finally, suburban areas experienced smaller gains in the percent of the population in the highest income categories than urban and rural areas. This is consistent with prior research and theory suggesting the poor are at greater risk for victimization than the wealthy (Meier & Miethe, 1993; Thacher, 2004).

One reason income composition changes have a relatively small effect on violence rate changes in rural areas might be that poverty affects violence differently in rural relative to nonrural areas. This idea has been proposed in the literature on social structure and crime in rural areas. For instance, Osgood and Chambers (2000) proposed that outside metropolitan areas, poor populations are more stable than average. This conclusion was echoed by Barnett and Mencken (2002), and Roussell, Holmes, and Anderson-Sprecher (2013) suggested the rural poor might not face the same housing pressures as the urban poor, thus decreasing the poverty burden in rural areas that might otherwise be associated with increased crime.

Limitations

The dependent variable for this analysis was aggravated assault. We chose this crime type because of its serious nature and greater likelihood of being reported when being administered the NCVS. However, future research should perform similar analyses with other violent and non-violent offenses to provide a more comprehensive picture. Second, to follow Thacher (2004) closely and to simplify our already complex modeling, we did not include additional controls for variables like ethnicity or region. Additional studies are required to assess any effect of the inclusion of these or other controls. For instance, some of our hypotheses may be more strongly supported in certain regions. Third, due to our interest in the great American crime decline, we limited our analysis to a 13-year period in American history, from 1993 to 2005. This period was also demarcated on both sides by systematic changes to the design of the NCVS (U.S. Department of Justice, 2011). Future researchers in this area should consider employing adjustment weights (Lauritsen, Rezey, & Heimer, 2016) or interpolating data (e.g., Berg & Lauritsen, 2016) to expand the range of years by attempting to overcome these structural breaks.

Conclusion

Our study presents several directions for future research. First, tests of criminological theories in communities of different sizes should relax assumptions about demographic–victimization risk relationships by estimating separate rural, suburban, and urban models. Failing to account for these differential risk relationships could be misleading. Second, violence rate trends are explained by more than just demographic changes. The combined demographic changes in the current study account for only about 35% of the changes in rural violence rates and much smaller amounts in urban and suburban areas. Other structural characteristics, and perhaps interactions between demographic characteristics, should be considered.

Footnotes

Appendix

Table A1.

Independent Variable Descriptions.

Variable Name	Description
Employment	We operationalized employment with three dummy variables: employed, unemployed, and unknown. We used series of two NCVS variables to create these dummy variables. First, respondents were asked in variable V3071 if they were working in the last week. If respondent answered yes, we coded employed as 1. If respondent answered no, they were then asked if they had a job in the last six months (variable V3072). If respondent answered yes we coded employed as 1. If respondent answered no, we coded unemployed as 1. If respondent did not answer the first employment question, we coded employment status unknown
Marital Status	We operationalized marital status with three dummy variables: married, not married, and unknown. We used variable V3015, which asks about current marital status, to create these dummy variables. We coded married 1 if the respondent answered they were currently married. We coded not married 1 if respondent answered they were widowed, divorced, separated, or never married. We coded marital status unknown as 1 if respondent did not answer the question
Young	We dichotomized this variable as young (12–24) and not young (over 24) using V3014, the allocated age variable. The 24-year-old split is the same used by Thacher (2004)
Female	We dichotomized this variable as female or not female using variable V3018, the allocated sex variable
Household income	Household income was measured in the NCVS with variable V2026 and is on an ordinal scale with 14 categories ranging from $0 to $5,000 to greater than $75,000. For our study we combined household income using the categories utilized by the BJS in their publications (e.g., Duhart, 2000): <$7,500, $7,500–$14,999, $15,000–$24,999, $25,000–$34,999, $35,000–$49,999, $50,000–$74,999, and $75,000+. We included an additional income unknown category to account for respondents who did not provide income information

Note. NCVS = National Crime Victimization Survey.

Acknowledgments

We thank Lynn Addington and Chunfeng Huang for their comments on and critiques of earlier drafts of this manuscript. We thank David Thacher for his guidance with the counterfactual analysis.

Declaration of Conflicting Interests

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The authors received no financial support for the research, authorship, and/or publication of this article.

Notes

References

Artz

(2003). Rural area brain drain: Is it a reality? Choices, 4, 11–16.

Barker

(2010). Explaining the great American crime decline: A review of Blumstein and Wallman, Goldberger and Rosenfeld, and Zimring. Law & Social Inquiry, 35, 489–516.

Barnett

Mencken

F. C.

(2002). Social disorganization theory and the contextual nature of crime in nonmetropolitan counties. Rural Sociology, 67, 372–393.

Baumer

E. P.

(2008). An empirical assessment of the contemporary crime trends puzzle: A modest step toward a more comprehensive research agenda. In Goldberger

A. S.

Rosenfeld

(Eds.), Understanding crime trends: Workshop report (pp. 127–176). Washington, DC: National Academy Press.

Berg

M. T.

Lauritsen

J. L.

(2016). Telling a similar story twice? NCVS/UCR convergence in serious violent crime rates in rural, suburban, and urban places (1973–2010). Journal of Quantitative Criminology, 32, 61–87.

Berube

(2011). The state of metropolitan America: Suburbs and the 2010 Census. Speech presented at the Suburbs and the 2010 Census: National Conference. Retrieved from http://www.brookings.edu/research/speeches/2011/07/14-census-suburbs-berube

Berube

Frey

W. H.

(2002). A decade of mixed blessings: Urban and suburban poverty in Census 2000. The Brookings Institution Census 2000 Survey Series: 1–18. Retrieved from http://www.frey-demographer.org/reports/R-2002-7_DecadeofMixedBlessings.pdf

Blinder

A. S.

(1973). Wage discrimination: Reduced form and structural estimates. Journal of Human Resources, 8, 436–455.

Catalano

S. M.

(2006). Criminal victimization, 2005 (Bureau of Justice Statistics Bulletin, NCJ 214644). Washington, DC: U.S. Department of Justice, Office of Justice Programs.

10.

Cohen

L. E.

Felson

(1979). Social change and crime rate trends: A routine activity approach. American Sociological Review, 44, 588–608.

11.

Duhart

D. T.

(2000). Urban, suburban, and rural victimization, 1993-1998 (Bureau of Justice Statistics Special Report (NCJ 182031)). Washington, DC: U.S. Department of Justice, Office of Justice Programs.

12.

Dyson

(2011). The role of the demographic transition in the process of urbanization. Population and Development Review, 37, 34–54.

13.

Johnson

K. M.

Cromartie

J. B.

(2006). The rural rebound and its aftermath: Changing demographic dynamics and regional contrasts. In Kandel

W. A.

Brown

D. L.

(Eds.), Population change and rural society (pp. 25–49). Dordrecht: Springer.

14.

Kaylen

M. T.

Pridemore

W. A.

(2011). A reassessment of the association between social disorganization and youth violence in rural areas. Social Science Quarterly, 92, 978–1001.

15.

Kaylen

M. T.

Pridemore

W. A.

(2013). Social disorganization and crime in rural communities: The first direct test of the systemic model. British Journal of Criminology, 53, 905–923.

16.

Lauritsen

J. L.

Owens

J. G.

Planty

Rand

M. R.

Truman

J. L.

(2012). Methods for counting high-frequency repeat victimizations in the National Crime Victimization Survey (Bureau of Justice Statistics Technical Series Report (NCJ 237308)). Washington, DC: U.S. Department of Justice, Office of Justice Programs.

17.

Lauritsen

J. L.

Rezey

M. L.

Heimer

(2016). When choice of data matters: Analyses of US crime trends, 1973–2012. Journal of Quantitative Criminology, 32, 335–355.

18.

Levitt

(1999a). The limited role of changing age structure in explaining aggregate crime rates. Criminology, 37, 581–598.

19.

Levitt

(1999b). The changing relationship between income and crime victimization. Federal Reserve Bank of New York Economic Policy Review, 5, 87–98.

20.

Meier

R. F.

Miethe

T. D.

(1993). Understanding theories of criminal victimization. Crime and Justice, 17, 459–499.

21.

Oaxaca

R. L.

(1973). Male-female wage differentials in urban labor markets. International Economic Review, 14, 693–709.

22.

Osgood

D. W.

Chambers

J. M.

(2000). Social disorganization outside the metropolis: An analysis of rural youth violence. Criminology, 38, 81–116.

23.

Perkins

C. A.

Klaus

P. A.

Bastian

L. D.

Cohen

R. L.

(1996). Criminal victimization in the United States, 1993 (Bureau of Justice Statistics Report (NCJ 151657)). Washington, DC: U.S. Department of Justice, Office of Justice Programs.

24.

Rennison

C. M.

Dragiewicz

DeKeseredy

W. S.

(2013). Context matters: Violence against women and reporting to police in rural, suburban and urban areas. American Journal of Criminal Justice, 38, 141–159.

25.

Rennison

C. M.

Rand

(2007). Introduction to the National Crime Victimization Survey. In Lynch

J. P.

Addington

L. A.

(Eds.), Understanding crime statistics: Revisiting the divergence of the UCR and NCVS (pp. 17–54). New York, NY: Cambridge University Press.

26.

Roussell

Holmes

Anderson-Sprecher

(2013). Community characteristics and methamphetamine use in a rural state: An analysis of preincarceration usage by prison inmates. Crime & Delinquency, 59, 1036–1063.

27.

Sampson

R. J.

(1983). Structural density and criminal victimization. Criminology, 21, 276–293.

28.

Sampson

R. J.

(1986). Effects of inequality, heterogeneity, and urbanization on intergroup victimization. Social Science Quarterly, 67, 751–766.

29.

Squires

G. D.

(2002). Urban sprawl: Causes, consequences, & policy responses. Washington, DC: The Urban Institute.

30.

Steffensmeier

D. J.

Harer

M. D.

(1999). Making sense of recent U.S. crime trends, 1980 to 1996/1998: Age composition effects and other explanations. Journal of Research in Crime and Delinquency, 36, 235–274.

31.

Thacher

(2004). The rich get richer and the poor get robbed: Inequality in U.S. criminal victimization, 1974-2000. Journal of Quantitative Criminology, 20, 89–116.

32.

Truman

J. L.

Morgan

R. E.

(2016). Criminal victimization, 2015 (Bureau of Justice Statistics Bulletin, NCJ 250180). Washington, DC: U.S. Department of Justice, Office of Justice Programs.

33.

U.S. Department of Justice. (2011). National Crime Victimization Survey 2010 codebook (Office of Justice Programs, Bureau of Justice Statistics. ICPSR21303-v1). Ann Arbor, MI: Inter-University Consortium for Political and Social Research.

34.

Weisheit

R. A.

Donnermeyer

J. F.

(2000). Change and continuity in crime in rural America. In LaFree

(Ed.), The nature of crime: Continuity and change, criminal justice (pp. 309–357). Washington, DC: Office of Justice Programs, National Institute of Justice.

35.

Zimring

F. E.

(2006). The great American crime decline. New York, NY: Oxford University Press.