Week 1 Project: Risk of Violent Crime Victimization Assignment Resources: Access the Project?Resources graphic to below?for the article used in this assignment. From the South Universi

Violence and Victims, Volume 27, Number 5, 2012

© 2012 Springer Publishing Company 635 http://dx.doi.org/10.1891/0886-6708.27.5.635

Risk of Violent Crime Victimization During Major Daily Activities

Andrew M. Lemieux, PhD Netherlands Institute for the Study of Crime and Law Enforcement (NSCR)

Marcus Felson, PhD Texas State University

Exposure to risk of violent crime is best understood after considering where people are, what they do, and for how long they do it. This article calculates Americans’ exposure to violent attack per 10 million person-hours spent in differ- ent activities. Numerator data are from the National Crime Victimization Survey (2003-2008) estimates of violent incidents occurring during nine major everyday activities. Comparable denominator data are derived from the American Time Use Survey. The resulting time-based rates give a very different picture of violent crime victimization risk. Hour-for-hour, the greatest risk occurs during travel between activities. This general result holds for demographic subgroups and each type of violent crime victimization.

Keywords: routine activities; lifestyle theory; risk of violence; epidemiology of violence; opportunity for violence

Crime opportunity theories are extremely important for studying how violent crime victimization distributes across time and space. These theories give special atten- tion to how victims and offenders converge. Both lifestyle theory (Hindelang,

Gottfredson, & Garofolo, 1978) and the routine activity approach (Cohen & Felson, 1979) explain this convergence as a function of noncriminal activity patterns. Specifically, the daily movements of individuals and populations through time and space create or diminish opportunities for violent crime to occur. Lifestyle theory focuses mainly on risky personal choices, such as engaging in activities away from home after dark or spending time near youth settings. The routine activity approach gives greater weight to conventional daytime activities, such as work and school, which expose participants to crime opportunities and risks (Roman, 2004). Similar versions of crime opportunity theory were postulated by Dutch and British criminologists around this time indicating the international importance of the link between routine activities and crime (see Mayhew, Clarke, Sturman, & Hough, 1976; van Dijk & Steinmetz, 1980, respectively).

Over time, lifestyle theory and the routine activity approach have been treated as com- plementary (or even synonymous) because they emphasize the impact of everyday activity patterns. Both theories relate victimization risk to the quantity of time people spend in risky settings. Among others, Eck, Chainey, and Cameron (2005) employed these theories

636 Lemieux and Felson

to comprehend how illegal behaviors cluster. Research on “dangerous places” and “hot spots” has repeatedly shown that violent crime concentrates in and around particular places (Block & Block, 1995; Kautt & Roncek, 2007; Roncek & Bell, 1981; Roncek & Faggiani, 1985; Roncek & Lobosco, 1983; Roncek & Maier, 1991; Sherman, 1995; Sherman, Gartin, & Buerger, 1989; Weisburd, 2005). Theoretically, people and populations spending more time in such places should have a higher risk of victimization. Unfortunately, victimization research has been plagued by a limited ability to quantify respondent exposure to risk on a large-scale national basis and instead has been forced to rely on summary measures of risk (Mustaine & Tewksbury, 1998). For example, early research estimated lifestyle exposures from female labor force participation, marital status, age, and sales at eating and drinking establishments (Cohen & Cantor, 1981; Cohen & Felson, 1979; Messner & Blau, 1987).

In this article, we draw from the epidemiology literature to reintroduce an alterna- tive option for measuring and comparing population exposures to risk of violent crime victimization in the United States. This alternative approach adjusts for the time exposed to risk in different major activities. Such adjustment can do more than improve measure- ment precision; it can reverse findings that neglect how much time is spent in settings where risk of violent crime is relatively high. Yet our purpose for writing this article is not methodological, but rather to improve our understanding of violent victimization by taking into account where people are and what they are doing.

EXPOSURE AND VICTIMIZATION

Several victimization studies quantify lifestyles with frequency counts of how respondents use their time. A few questions embedded in a victimization survey can serve this purpose by asking how many nights a week or month respondents spend on certain activities away from home. For example, the British Crime Survey and Canadian General Social Survey victimization supplement have used this approach in the past. The valid ranges of answers for such questions are 0–7 nights (per week) and 0–31 nights (per month). Frequency measures such as these have been used to measure exposure to several types of crime risk, including violent crime victimization (Clarke, Ekblow, Hough, & Mayhew, 1985; Felson, 1997; Gottfredson, 1984; Kennedy & Forde, 1990; Miethe, Stafford, & Long, 1987; Mustaine, 1997; Sampson & Wooldredge, 1987). Counts of nights out are very use- ful for building predictive models, often with logistic regressions, but have unfortunately produced some mixed and confusing results about how victimization relates to lifestyles.

In 1998, Mustaine and Tewksbury expressed doubt about counting nights spent away from home while ignoring what activities occurred while away. They developed a 95-item instrument to collect specific information on the daily activities of college students in eight American states. Although their interest was property crime rather than violence, they demonstrated with a logistic regression model that actual hours out did not predict college student victimization very well. On the other hand, they found that victimization is more a function of which locations and activities students selected. For example, victimization risk increased for those who went out to eat more often but decreased for those who went out to play basketball. Beyond the victimization literature, other studies have also shown specific exposure to risk measures are important and useful predictors of delinquency (Osgood & Anderson, 2004; Osgood, Wilson, O’Malley, Bachman, & Johnston, 1996).

Although measuring what people do when away from home seems obvious after the fact, it is not so easy to accomplish without a substantial questionnaire, and such elabora-

Risk of Violent Crime Victimization 637

tion is not currently available from a large-scale national survey. The idea of measuring detailed time use and detailed victimization in the same survey was discussed and dis- carded three decades ago as too long, cumbersome, and expensive (Gottfredson, 1981, pp. 721–722; Skogan, 1981, 1986). Even with the advanced tracking technology of today’s world, this is an enormous task that would produce a vast amount of data. Herein lies the complexity of quantifying “exposure to risk” and the practical rationale for using general time use measures such as demographic proxy variables and frequency counts. To date, no national study has yet collected sufficient lifestyle detail to meet the challenge offered by lifestyle and routine activity theories. Given this roadblock, we seek an alternative approach to disaggregate and comprehend lifestyle exposure to violent crime risk.

THE DENOMINATOR DILEMMA: TIME-ADJUSTED VICTIMIZATION RATES

Ratcliffe (2010) explains the denominator dilemma as “the problem associated with iden- tifying an appropriate target availability control” (p. 12). In demography and epidemi- ology, this is the classic problem of figuring out what population is exposed to risk to make appropriate comparisons. The denominator dilemma has been recognized for more than 40 years in criminal justice research. Indeed, many scholars have argued crimi- nologists’ reliance on population-based rates neglects the actual opportunity structures of many crimes and can produce misleading and even incorrect findings (Harries, 1981; Sparks, 1980; Stipak, 1988). Early attempts to overcome the problem include Leroy Gould’s auto theft work (1969), which calculated rates using the number of automobiles in the denominator, whereas Sarah Boggs (1965) investigated several alternative denomi- nators for exposure to risk.

The general denominator issue was taken into account by Cohen and Felson (1979) and articulated by Ronald V. Clarke (1984). Although there may be different ways to approach the appropriate denominator issue, the larger problem is the uncritical acceptance of sim- ple residential population as the default denominator for crime rate comparisons. As Stipak (1988) wrote, “Exclusive reliance on population-based crime rates stems more from blind tradition than from logic or merit” (p. 258). To illustrate this, we might note that tourist cities have a substantial influx of persons that can be offenders or victims of crime, who are not contained in the traditional denominator such as a census population (Lemieux & Felson, 2011). Using a nontourist example, the movements of a resident population dur- ing the week and on weekends will alter the number of occupied households at any given moment (Harries, 1981)—a topic taken up by Andresen and Jenion (2010) in studying ambient populations. Thus, when describing victimization risk using rates, researchers must select denominators carefully.

In 1984, Stafford and Galle suggested studying unequal exposure to victimization risk by looking beyond population-based rates. They noted that the conventional victimization rate V/Pt (victimizations per 100,000 population during year t) is an inadequate measure because the denominator only controls for population size. Those spending a great deal of time in a dangerous setting are treated no differently from those spending very little time there. That contradicts a central tenet of lifestyle theory and the routine activity approach. Stafford and Galle (p. 174) suggested a more defensible, adjusted rate:

V / (P 3 E)t (where E accounts for the population’s exposure to risk during year t)

638 Lemieux and Felson

This calculation of victimization risk takes into account both population size and a more direct measure of population exposure. Their suggestion reflects epidemiological and demographic thinking that proves useful in this article. The important point is that people spend very unequal amounts of time in different activities, thus distorting estimates of how much risk one activity generates compared to another. Time-adjusted rates take this into account and thus produce a better measure of risk exposure.

The question now is “how do you quantify exposure to enable time-adjusted rate calculations?” The answer is the person-hour. The person-hour is a useful measure for determining how much time individuals or a population spends in a specific place or activity. For example, a person who sleeps at home for 8 hours a night 7 days a week spends 56 person-hours per week in that activity. Aggregating this measure to a population, if 100 persons had the same sleeping pattern, this group would spend 5,600 person-hours per week sleeping. Unlike frequency counts or demographic proxies, the person-hour is a direct measure of time use that enables researchers to calculate time-adjusted rates.

A few examples of time-adjusted rate calculations are already found in the crime literature. Cohen and Felson (1979) combined time use and victimization data from the United States to describe the relative risk of three broad place categories accounting for the unequal durations of time spent in each. The place categories were at home, on the street, and elsewhere. They calculated the number of victimizations per one billion person-hours spent in each location for the American population as a whole. They estimated that the population’s risk of being assaulted by a stranger was 15,684 victimizations per billion person-hours spent on the street, but only 345 for equivalent time spent at home; a ratio of 45:1 (see Cohen & Felson, 1979; Table 1, panel D). A second exception found in the literature is auto crime research by Clarke and Mayhew (1998), which calculated the amount of time cars were parked in different set- tings to compare the relative risk of each. They found that risk increases sharply when cars are in public places; parking in a public lot was more than 200 times more risky than using a pri- vate garage. The rate was reported as the number of car crimes per 100,000 cars per 24 hours parked in a location. A third research exception is found in a series of papers by Andresen and colleagues, who calculated crime rates in British Columbia, Canada, for the ambient population as an alternative to the residential population (Andresen, 2010, 2011; Andresen & Brantingham, 2008; Andresen & Jenion, 2008, 2010). This takes into account the major shift of population as people leave their residential area to go to work, school, or leisure settings. Despite these three exceptions, most studies of the relative risk of violent crime have neglected time adjustment, despite major differences in time spent in various places and activities.

In the field of epidemiology, researchers have long been accustomed to adjusting for time exposed to adverse conditions, including pollution, secondhand smoke, danger in sports, as well as risky consumer products and workplaces (see Barnoya & Glantz, 2005; Cai et al., 2005; Dasgupta, Huq, Khaliquzzaman, Pandey, & Wheeler, 2006; de Löes, 1995; Hayward, 1996; Messina, Farney, & DeLee, 1999; Starr, 1969). In his analysis of consumer product injuries, Hayward (1996) clearly showed that time adjustment makes a difference when describing the relative risk of activities such as riding a bike or using an electric hedge trimmer. Without time adjustment, bicycling appeared to be the most dangerous activity. However, accounting for both the participant population and time spent, bicycling dropped to the seventh most injurious. The most dangerous product per person-hour of use proved to be the electric hedge trimmer, with a time-adjusted injury rate five times higher than bicycles. Put simply, short periods spent using this tool are extremely dangerous compared to other household products. Thus, time-adjusted rates can produce a vastly different picture of risk than incident counts or population-based rates.

Risk of Violent Crime Victimization 639

THE CURRENT STUDY

This study reconsiders how we measure routine exposures to the risk of violent crime in the United States as a whole. Using two national-level data series, we calculate risk for nine broad activity categories, including six destination activities and three transit activities (movement between destination activities). These rates are adjusted for the amount of time people spend participating in each of the nine activities, helping us to compare the exposure to risk. Although this approach is common in epidemiological studies, it was not possible in the past to apply it to violent crime given the limited daily activity data accompanying vic- timization and crime data. A newer data source—the American Time Use Survey—allows us to overcome earlier limitations of denominator data. The purpose of this research is not to compare individuals or families but rather to comprehend the relative exposure to violence in different daily activities, taking into account hours exposed to risk.

This approach is not comparable to the Federal Bureau of Investigation (FBI)’s “crime clock,” which divides the number of crimes by the number of seconds in a year. A crime clock uses the same denominator for every calculation. We use a different denominator for each activity category because unequal amounts of time are spent in each. The ideal approach would use a unified national survey of victimization and time use for both victims and nonvictims. Such a study would enable easy risk calculations for individuals and facilitate logistic regression models of the victimization process (see Mustaine and Tewksbury, 1998). Given that no such survey is found in the United States or elsewhere, we instead follow the lead of epidemiologists, drawing numerator and denominator data from separate sources (see Hayward, 1996).

This multi-dataset approach is not new in criminology where conventional crime rates are usually calculated using two different sources of information. For example, it is common to use Uniform Crime Report data in the numerator and census population data in the denominator even when calculating age-specific arrest rates or comparing one city to another. The main contribution of this study is to draw denominator data from a time use source not usually employed by crime and victimization researchers. Because the American Time Use Survey (ATUS) and National Crime Victimization Survey (NCVS) both use a stratified, multistage sampling strategy and weight estimates to the national level, it was possible to harmonize these data and calculate meaningful rates. Table 1 compares the NCVS and ATUS respondents by dichotomized age, sex, and race, indicating substantial demographic consistency between the two surveys as well as among the six annual samples.

We report rates as the number of violent victimizations per 10 million person-hours. These rates can be used to (a) determine which activity is the most dangerous hour for hour, (b) compare the relative danger of one activity to another, (c) make comparisons among demographic groups, and (d) make future international and longitudinal compari- sons as time use and victim surveys continue to develop. Although we cannot provide a predictive analysis for individuals, we will be able to assess whether the overall findings hold within major demographic subgroups.

In shifting away from an individual analysis, we face at least three limitations: (a) our numerator and denominator data come from different individuals, who are not interviewed simultaneously; (b) we cannot use log-linear analysis or other multivariate methods to predict victimization risk at the individual level; and (c) activity categories are not perfectly matched between our two data sources. Despite these imperfections, we believe this analysis produces results that are important, useful, and robust. We consider a

640 Lemieux and Felson

population’s exposure to risk in different activities even though we lack full details about the individual’s exposure compared to other individuals. The sections that follow describe our data sources and how they were matched to produce time-adjusted victimization rates.

Numerator Data

The NCVS estimates on an annual basis the number of violent victimizations occurring in different everyday activity categories. During an NCVS interview, victims are asked, “What were you doing when the incident (happened/started)?”; NCVS variable V4478. The choices included the following nine broad activity categories including travel to dif- ferent destinations:

1. Sleeping 2. Other activities at home 3. Working 4. Attending school 5. Shopping or errands 6. Leisure activity away from home 7. Going to or from school 8. Going to or from work 9. Going to and from some other place.

During the study period (2003–2008), 93.6% of violent crime victims indicated the inci- dent in question happened during one of these nine activity categories (U.S. Department of Justice’s Bureau of Justice Statistics, 2005, 2006a, 2006b, 2008a, 2010, 2011). The other options available to respondents were “don’t know” or “other”; however, these victimiza- tions are excluded from the present analysis.

Between 2003 and 2008, the NCVS performed 1,273,942 interviews, which captured 9,220 separate violent incidents. Of these, 7,264 incidents are included in this analysis; some data were removed to match the numerator and denominator data, as explained later in this article. Twenty types of violence are included in this analysis, ranging from verbal threats of

TABLE 1. Demographic Composition of National Crime Victimization Survey and American Time Use Survey Samples, 2003–2008

% Male % White % Aged 15–29

NCVS ATUS NCVS ATUS NCVS ATUS

2003 47.6 43.7 82.3 83.5 17.2 18.6

2004 47.6 43.8 82.1 84.1 17.5 18.4

2005 47.8 42.9 82.4 82.9 17.5 19.1

2006 48.0 42.6 83.0 82.0 17.6 19.2

2007 48.1 43.3 82.8 81.6 17.8 18.7

2008 48.1 44.4 82.7 80.8 17.7 18.4

Note. From National Crime Victimization Survey (NCVS) Person Record-Type Files and American Time Use Survey (ATUS) Activity Summary Files.

Risk of Violent Crime Victimization 641

assault to completed rapes. We begin by analyzing all types of violent crime combined and later separate violent crimes into five broad categories (see Appendix) to assess the robust- ness of the findings.

Weights provided in the NCVS incident-level extract file allow us to estimate the inci- dence of violence in the United States for each activity category. Similar estimates were produced for each demographic subgroup. To produce time-adjusted rates, we employ additional data from the ATUS.

Denominator Data

The ATUS officially began collecting data about the routine activities of Americans in 2003. The survey and sample were specifically designed to provide information about time use at the national level; additional information concerning the rationale for and history of the ATUS can be found on the survey’s Website (http://www.bls.gov/tus/overview.htm). The ATUS is a unique survey that uses computer-assisted telephone interviewing (CATI) to create time use diaries for the day before each interview. The ATUS asks respondents to detail where they were, what they were doing, and with whom, over a 24-hour period beginning at 4:00 a.m. the preceding day (Fisher, Gershuny, & Gauthier, 2011). Because the study is spread over the year and has a large sample, these snapshots combine to pro- vide a substantial general picture of time use for the population of the United States.

During the study period (2003–2008), 85,645 individuals were interviewed by the ATUS. Respondents reported 1,971,368 separate activity records that were classified into nearly 400 categories—far more than the nine types of activity delineated in the NCVS. An activity record refers to one activity performed by a single person. For example, sleeping from 8:00 a.m. to 10:00 a.m. constitutes a single activity record. When the respondent gets out of bed and showers from 10 a.m. to 10:15 a.m., this is classified as a separate activity record. The number of activity records reported by each person was not evenly distributed. Some persons reported 10 or fewer records, whereas others reported more than 50. When summed, these activity records produce the total number of hours respondents spent in each activity category. Although a single respondent’s reports are not representative for that one person’s annual experience, the total sample’s reports cover and represent what the American population does in the course of the year.

Like the NCVS, ATUS data files contain weights that enabled us to make national time use estimates. Two component variables were quantified: (a) the daily participant popula- tion for different activities and (b) the mean participation time. Together these produced an estimate of how many person-hours the American population spent in the nine NCVS activity categories each year. To ensure the validity of our time-adjusted rates, it was nec- essary to reconcile the two surveys, taking into account their different levels of detail; this procedure is described in the following section.

Reconciling Discrepancies Between the Two Data Sources

To match these data sources, ATUS activities were recoded to match the nine broad NCVS categories because it was not possible to make the NCVS time use variable more specific. This means the detailed picture of American life the ATUS provides was not captured in this analysis because of NCVS limitations. For example, the numerous home activities detailed by the ATUS were subsumed under two categories: “sleeping” and “other activi- ties at home.” Fortunately, 99.8% of the original ATUS data were amenable to recoding. The final denominator data include 1,967,356 activity records for the 6 years. The average

642 Lemieux and Felson

person-hours per day spent in each of the nine activity categories was sleeping (8.60), other activities at home (8.10), working (8.07), at school (4.90), leisure (2.94), shopping (1.54), to or from other (1.21), to or from work (0.73), and to or from school (0.58). It is important to note here that the participant population of each activity varied; that is, although most Americans slept, only a small proportion attended school. Thus, the total time spent in each activity is dependent on (a) the participant population and (b) the aver- age person-hours spent in the activity per day. This is accounted for in the time-adjusted rates reported in the section that follows (see Table 2).

Demographic features of the samples also needed to be reconciled. The NCVS sample included Americans residing outside the United States, active-duty military personnel, and persons younger than 15 years of age—all of whom were removed to achieve compat- ibility with the ATUS. We also omitted incidents classified as series crimes, which is a standard procedure for making NCVS estimates (see U.S. Department of Justice, Bureau of Justice Statistics, 2008b, p. 459). Future analyses could include these crimes; however, in this analysis, the aggregated, national level approach does not enable us to tease out the individual factors associated with repeat victimization. After these exclusions, the numera- tor data include 7,264 violent incidents for the years 2003–2008.

Table 2 outlines how NCVS and ATUS estimates are used to calculate the time-adjusted rates presented in the sections that follow. These calculations are not as difficult as they may look but do require attention to detail. For example, multiplications by constants are needed to generalize from 1 day to 365 days as well as to arrive at a rate per 10 million person-hours. Activities must be harmonized to make sure numerator and denominator apply as closely as possible to the same activity. Thus, to get the denominator in terms of person-hours shopping (D), we multiply the population of shopping participants (B) by the average time spent shopping per participant per day (C). That product is then multiplied by 365 to cover the time shopping in a year. The numerator data consists of the number of victimizations while shopping (A). However, that fraction is too small to work with, so we

TABLE 2. Example of How Activity-Specific Time-Adjusted Violence Rates Were Calculated: The Risk of Violence While Shopping, United States, 2003

Component Estimated from the Surveys Source National Estimate

(A) Violent victimizations while shopping (incidence count)

NCVS, 2003a 238,530

(B) Average daily population of shoppers (participants)

ATUS, 2003b 133,893,190

(C) Average time spent shopping (person-hours)

ATUS, 2003b 1.42

(D) Total time spent shopping in 2003 (B) 3 (C) 3 365 69,551,975,288

(E) Time-based rate of violence (Victimizations per 10 million person-hours)

(A) 3 10 million (D)

34.3

aNational Crime Victimization Survey (NCVS) Incident-Level Extract File, 2003. bAmerican Time Use Survey (ATUS) Activity File, 2003.

Risk of Violent Crime Victimization 643

multiply it by 10 million to produce a smaller index number. For comparison purposes, we use the same standard rate for all activities: the risk of violent victimization per 10 million person-hours engaged in a given activity.

RESULTS

Basic Pattern

We begin with basic violence risk calculations for the American population in general. Table 3 shows the annual time-adjusted violence rate for all nine activities from 2003 to 2008. The mean, standard deviation, and coefficient of variation (CV) are reported for each activity cat- egory. We do not report the standard error of our time-adjusted rates as this calculation would be very complex because the numerator and denominator come from different sources. Yet the coefficient of variation tells us that most statistics in this study display considerable sta- bility from year to year. For this reason, we average the 6 years for subsequent tables.

Compared to every other activity, sleeping (row 1) is the safest activity overall; other activities at home are the second safest activity (row 2). Thus the results strongly uphold a major premise of the routine activity approach and lifestyle theory: being at home is safer than being away from home. Interesting, however, is that by disaggregating at-home activi- ties into two categories, the results indicate that on an hour-for-hour basis, being awake at home is nearly 11 times more risky than being asleep. Although the risk of a violent victimization while sleeping is very low, it is not zero.

On the other hand, activities away from home do not fit a clear and single pattern. The apparent risk of violence during activities away from home differs from one activity to the next (rows 3–6, Table 3). This supports our earlier suggestion and that of Mustaine and Tewksbury (1998) that broad lifestyle measures (such as activities away from home) do not adequately measure risk. Consider that working and shopping are relatively safe among activities away from home, in