Skip to main content
SearchLogin or Signup

The moving home effect: A quasi experiment assessing effect of home location on the offence location

Journal of Quantitative Criminology; https://doi.org/10.1007/s10940-011-9161-7

Published onJul 09, 2020
The moving home effect: A quasi experiment assessing effect of home location on the offence location

Abstract

Past research has found the journey to crime (JTC) distance from an offender’s home tend to be short (Rossmo, 2000). Theoretical reasons why most JTCs are short may be due to that offenders’ routine activities take place closer to their home (Brantingham and Brantingham, 1993), or that offenders rationally choose targets nearer to their home (Bernasco and Block, 2009). Most current studies examining the JTC ignore the fact that offenders tend to move frequently. Awareness space and targeting patterns of offenders may change dramatically over time, and moving one’s home residence may signify an expansion of this awareness space. By treating when an offender moves as an experiment, this study finds that when offenders move they tend to commit crimes in locations farther away from past offences than would be expected without moving. This finding suggests that the home has an impact on where an offender will choose to commit a crime, independent of offence, neighborhood, or offender characteristics. The effect is small though, suggesting other factors may play a larger role in influencing where offenders choose to commit crime.

1. INTRODUCTION

Crime pattern theory dictates that where an offender commits crime is influenced by their awareness space (Brantingham and Brantingham, 1993). One of the obvious nodes from which an offender’s awareness space diffuses through repeated exposure is around their home residence (Brantingham and Brantingham, 1993; Cohen and Felson, 1979). One of the most consistent pieces of evidence for this theory is that the journey to crime (JTC) distance from an offender’s home tend to be of a short distance (see Rossmo, 2000 for a review).

Offenders tend to be itinerant (Davies and Dale, 1996; Wiles and Costello, 2000), and so awareness of surroundings may increase when relocating their residence to different neighborhoods within that same urban environment. Because the majority of the JTC literature focuses on crime trips instead of within offender patterns (Townsley and Sidebottom, 2010), little attention has been paid to the development of awareness space over time (Bernasco, 2010a). This study aims to address an under examined phenomenon that may indicate expansion of awareness space, when an offender moves. Studies that examine intraoffender behavior often ignore that the home base of offenders tends to change frequently. Studies that acknowledge this shortcoming often use the only available address or the address the offender was living at during the majority of their offences (Levine, 2007; Townsley and Sidebottom, 2010; Warren et al., 1998).

Two exceptions to this are recent studies in which Bernasco and colleagues show an offender is more likely to commit crimes in discrete areas nearby to both the current home residence and nearby past residences for the crimes of burglary, robbery, motor vehicle theft, and assault (Bernasco, 2010a; Bernasco and Kooistra, 2010). This research builds on those past findings of Bernasco and colleagues by demonstrating that an offender commits crime in different locations when he/she moves, which is a different question than what characteristics attract offenders to those locations. Specifically the question this research aims to answer is if the location where an offender commits crime conditional on where they live, independent of other personal or environmental factors.

2 EXPLAINING OFFENCE PATTERNS IN SPACE

This review categorizes literature explaining offending patterns in space into three areas; gravity models, offender as a forager, and the JTC. The review then ends with illustrations for why the JTC distance from the home being typically short may not necessarily be indicative of a causal effect of the home location on where an offenders’ commit crime, particularly if offenders tend to live in areas that have abundant opportunities to commit crime nearby.

2.1 GRAVITY MODELS OF CRIME

Gravity models of crime include studies examining the environmental characteristics that attract or repel offenders (Bernasco, 2010b; Bernasco and Block, 2009; Clare et al., 2009; Hessling, 1992; Pettiway, 1985; Smith, 1976). Areas may present crime generators that attract offenders from outside areas, or have attributes that act as barriers to committing criminal activity.

Empirical findings are consistent that the environments offenders travel to have higher frequencies of vulnerable targets, such as other delinquents, prostitutes, or drug dealers (Bernasco and Block, 2009; Rengert, 2004). Barriers to travel can take the form of physical, social, or psychological barriers. Physical barriers, such as highways, fences, or rivers, present an impediment to travel, although they are often not entirely intraversable (Clare et al., 2009; Rengert, 2004). Social barriers may exist in the form of racial segregation that prevent interaction between offenders and victims of different races, and hinder travel to ethnically different environments (Bernasco and Block, 2009; Jacobs, 2010; Pettiway, 1985). Informal social controls or the collective efficacy of the community may prevent criminal activity in specific neighborhoods (Bernasco and Block, 2009). While these environmental characteristics are permeable barriers to travel, they increase perceived costs of travelling to and through these areas.

2.2 OFFENDER AS A FORAGER

A popular analogy to describe offending patterns in space is to equate predatory criminal behavior with that of a foraging animal (Bernasco, 2010a; Bernasco 2010b; Bernasco and Block, 2009; Davies and Dale, 1996; Jacobs, 2010; Johnson et al., 2009). While spatial and temporal constraints place restrictions on the overall dispersion of criminal acts, within those constraints offenders use a rational calculus to determine optimal targets. The dispersion of an offender’s criminal acts likely expand in time though. An offender may perceive increased risks in victimizing multiple targets in close proximity (Johnson et al., 2009) and make a rational decision to offend elsewhere. An offender may also return to previous areas that have proven to be fruitful in the past (Bernasco, 2010a; Bernasco and Kooistra, 2010; Davies and Dale, 1996; Johnson et al., 2009).

The offender as a forager model has been used to explain the boost effect of victimization, in which the risk of victimization to an individual or a location is elevated after a prior victimization. Particularly salient is the fact that spatial and temporal proximity to a prior criminal offence increases risk of victimization (Johnson et al, 2007; Ratcliffe and Rengert, 2008; Short et al., 2009). This phenomenon is more easily explained by the same offender(s) victimizing multiple targets in a foraging manner than it is by other theorized factors, such as risk heterogeneity (Short et al., 2009).

2.3 THE JOURNEY TO CRIME (JTC)

The JTC literature consists of studies that examine distance or direction to crime from an offender’s home (Capone and Nichols, 1976; Goodwill and Alison, 2005; Rengert et al., 1999; Rossmo, 2000; Townsley and Sidebottom, 2010). A consistent finding within the JTC literature is that distance from an offender's home address is a strong predictor of where an offence occurs (Bernasco and Block, 2009; Capone and Nichols, 1976; Rossmo, 2000). The theory why offenders are so bounded by their residence is steeped in rational choice and routine activities theory. As distance increases, so do the costs to committing the criminal act. Several examples of costs of travel are increased time devoted to travelling and psychological discomfort for offenders’ in unknown areas (Jacobs, 2010; Rengert and Wasilchick, 2000).

Although environmental and personal characteristics appear to influence the distance an offender travels to commit their crime (Pettiway, 1985; Rengert, 2004; Rossmo, 2000; Tita and Griffiths, 2005), offenders consistently show a strong spatial bias towards committing offences close to their home (Bernasco and Block, 2009; Brantingham and Brantingham, 1993; Capone and Nichols, 1976; Rengert et al., 1999; Rossmo, 2000; van Koppen and de Keijser, 1997). This distance decay relationship also appears to be consistent for all types of crime as well (Hesseling, 1992; Rossmo, 2000; Smith, 1976). For this study’s sample this same distance decay effect is demonstrated by plotting the histogram of the journey to crime distances for 40,691 criminal events, displayed in Figure 1.1

Figure 1: Distance Decay in the Journey to Crime Histogram

<p>Figure_1.JPG</p>

Figure_1.JPG

Caption: The most consistent finding in the JTC literature is that offenders tend to commit crimes near their homes. The distribution of the JTC distance is typically described as a distance decay pattern, and my sample of the JTC distance follows this same pattern.

While all of the models of offender behavior as presented above (gravity models, offender as a forager, or the JTC from ones home) are not necessarily mutually exclusive, each process has different implications for how we understand offender decision making. One of the most consistent findings in the above literature is that the short distance between the home location and the offence location is taken to be evidence that the home location has a strong effect on where an offender will choose to commit their crime. The next section presents reasons why the distance decay effect in the JTC may not be indicative of the home location having an effect on where offender’s commit crime, and present a more appropriate research strategy to identify whether the home location has a causal effect on the crime location.

2.4 WHY A SHORT JTC MAY NOT BE INDICATIVE OF A HOME EFFECT

Figure 2 displays several hypothetical examples of the spatial patterns of offenders over time. These figures are used to display potential examples of offender behavior over time, in particular what happens to their offence location when they move, and how the behavior is or is not consistent with the home location having an effect on the crime location. In Figure 2 offence locations are represented by a triangle, home locations represented by a square, events at the first home colored light grey, events at the second home colored dark grey, and the JTC represented as a solid line between the home and offence location.

Figure 2: Hypothetical JTC Scenarios When an Offender Moves

<p>Figure_2.png</p>

Figure_2.png

Caption: Figure 2A represents a scenario in which the home location has no effect on the crime location. Figure 2B represents a scenario in which the home location has a large effect on the crime location. Figure 2C represents a scenario in which crime locations are entirely determined by the distribution of opportunities within the environment, represented by a circle.

Figure 2A represents a scenario in which the offence location has no relationship to the home location. In this figure, even though one moves their home address, the crime occurs in the exact same location. So in Figure 2A, the JTC distance simply varies due to the change in the home location relative to that past offence location. If offenders are simply attracted to targets in specific areas of the city, rational decisions to offend in these areas may not change. Also moving may only affect one of an offender’s many nodes around which crime takes place and other areas of routine activities may be unchanged. This scenario is not consistent with the home location having an effect on the crime location.

Figure 2B represents a scenario in which the offence location is heavily influenced by the home location, as the offence location is always in the immediate vicinity of the home. While the offence location changes, essentially proportional to the distance of the move, the JTC distance is stable. If the home location has an effect on the crime location, this scenario should be observed in empirical data. That is to say, the offence location should perceivably change when an offender moves if the home location has an effect on the crime location.

A third possibility is that when an offender moves, their awareness of potential targets increases, and their JTC may increase or decrease depending on suitable and attractive targets in the immediate area. If an offender moves to an area of the city in which attractive targets are plentiful in the immediate area, the JTC may decrease. Whereas if an offender moves to an area in which attractive targets are scarce in the immediate area, their JTC may increase. This scenario is depicted in Figure 2C, with the only potential location of targets within the circle. In this example the JTC simply varies relative to opportunities to commit crime in the environment.

These hypothetical scenarios depict why the JTC distance being typically short, which is interpreted as the home having an effect on the crime location, cannot identify if offender behavior is influenced by home location, or if it is simply reflective of factors in the environment. In particular, if offence locations are largely determined by opportunities within the environment and offenders tend to live in areas in which opportunities are abundant, the short JTC could simply be a spurious artifact of the opportunity structure.

3. RESEARCH QUESTION

This work will address one research question. When an offender moves, does the location where they commit crime change? This question is examined to demonstrate the causal impact the home location has on the offence location. Regardless of other reasons why offenders may choose to commit crimes in certain locations, if the crime location does not change when an offender moves it can hardly be said the home location has a causal impact on the offence location.

3.1 THE INTER-OFFENCE DISTANCE (IOD)

The dependent variable of interest in this research design is the inter-offence distance (IOD). The inter-offence distance is the distance between sequential offence locations committed by the same offender. The IOD was chosen over the JTC because the JTC does not necessarily change when the offence location changes when an offender moves. Figure 3 depicts a hypothetical offender who has committed four crimes from two different home locations. Crime locations are shown as triangles, and home locations are shown as squares. In Figure 3 the JTCs are displayed as solid lines, and the IOD is displayed as a dashed line (with the darker dashed line the IOD when the offender moved his/her home location). Offences are sequentially ordered according to their orientation on the y axis.

Figure 3: The JTC and the inter-offence distance (IOD) for a hypothetical offender

<p>Figure_3.JPG</p>

Figure_3.JPG

Caption: This figure represents why when an offender moves, the JTC is not representative of the new offence location compared to past offence location, and hence why I chose the IOD as my dependent variable to assess the moving home effect on the crime location.

As Figure 3 shows, the offences take place in different locations when the offender moves, but the JTC distance is the same for all offences. Figure 3 displays a hypothetical relationship between moving and the IOD. Although variation in where a crime is committed even when the offender lives at the same residence would be expected, if the distance between offences is greater when an offender moves to a different residence, compared to when they did not move, it can be said that the moving of the offender’s home had an effect on where the crime took place relative to where the previous crime took place. This approach treats when an offender moves as an experiment, and the control condition is the expected IOD when an offender maintains a permanent residence.

Figure 4: The Difference Between the IOD and the JTC

<p>Figure_4.JPG</p>

Figure_4.JPG

Caption: These scatterplots demonstrate the discrepancy between the difference in the JTC for two sequential offences and the IOD. In particular, when an offender moves, the JTC difference has a much more ambiguous relationship to the IOD (shown in Figure 4B). This means that when an offender moves, the JTC is not necessarily indicative of the offender committing crime in the same or different locations than prior offences.

Figure 4 contains a scatter plot of the change in JTC distance between two sequential offences on the x axis and the IOD on the y axis, and demonstrates how these two constructs are not unique in this sample, particularly when an offender moves. In the left panel, Figure 4A, the scatter plot is for events in which the offender did not move (i.e. two offences committed at the same home address). In the right panel, Figure 4B, the scatter plot is for events in which the offender did move (two offences committed from different home addresses). Examining the left panel, one can see the change in the JTC distance and the IOD have a strong correlation. The relationship between the IOD and the change in the JTC distance can be described in geometric terms (for non-moving events). If you consider both the JTCs and the IOD as sides of a triangle, the IOD can never be smaller than the difference between the JTCs, and the IOD is only larger than the JTC difference because of a change in direction between the two offences in comparison to the home address. This relationship does not hold when an offender moves though, and one can see the reduction in the correlation when examining the right scatter plot, Figure 4B. The concentration of points in the lower left corner of the scatter plot suggests both changes in JTC distance and IOD tend to be small, but one can see many anecdotal observations in which this is not true. Particularly the larger number of points towards the upper left of the scatter plot suggests that the JTC distance may stay constant, but where offenders commit crime changes appreciably. One can see many instances of the obverse situation as well, with points low on the y axis and towards the right on the x axis. These suggest the JTC distance changed, but the two offences occurred in close proximity in space. If one wants to examine the historical trends in where an offender commits crime, the JTC is not an appropriate construct, especially if the offender moves during their history.

4. OTHER MEASURES, METHODS, AND DATA

4.1 INDEPENDENT VARIABLES

The independent variable of interest is whether the offender moved or the distance of that move. Whether an offender moved is determined by sequentially ordering an offender’s offences according to time, and if the address listed for the current offence is different from the address listed for the prior offence the offender is recorded as having moved. When the distance between homes is used, the distance will be zero if an offender did not move, and if the offender listed a different home address the distance between the previous address and the current address is calculated. 2 All distances are calculated using Euclidean geometry (straight line or crow flight distance) and are displayed in meters.

Other control measures included the type of crime, the time in between sequential offences, the census tract(s) the current and the previous offences were committed in, and the census tract(s) the current and previous home locations resided in. For the types of crime, this analysis will include variables for whether the crime was an assault, burglary, robbery, motor vehicle theft, larceny, possession of contraband, and vehicular crime.3 As an incident can be linked to multiple offences, these categories are not mutually exclusive. The analysis will also include a variable representing if two sequential offences were not the same.4 Dummy variables representing the type of crime are included to assess the moving home effects impact on the IOD independent of types of crimes committed. Committing different types of crimes in two sequential events may lead to a larger IOD, as the spatial distribution of opportunities may differ between crime types. Several of the analyses presented are replicated with crime specific samples as well (i.e. only examining the correlation between moving and the IOD for burglary, robbery, motor vehicle theft, and larceny). This is done to ensure that the findings are not an artifact of differing opportunity structures for different types of crime in the environment. This is also done because the majority of the prior literature has focused on specific crime types, mainly property crimes such as burglary and robbery, and hence the theoretical models explicated by these different theories of offender location choice may be better represented by those specific crimes.

Time in between offences is the interval in days between the current offence and the previous offence. It has been demonstrated in past work that offences committed closer in time and space are more likely to be perpetrated by the same offender (Johnson et al., 2009), and so this may be evidence that for short intervals in between offences they are more likely to be close in space.

Information on where the crime was committed or the home location of the offender is included in a set of dummy variables representing census tracts within the city region. These were used to proxy neighborhood effects and control for spatially outlying offences or home locations. If an offender committed a crime in an area of the city where few crimes take place, it may be expected that future crimes are unlikely to occur in the area nearby. If an offender lives in an area where few criminal events occur, it may cause them to travel to areas with more abundant opportunities to commit crime. Outside of the census defined city region was also included as another dummy variable for both home and offence locations.5

The analytical strategy for this article will utilize difference in means t-tests and fixed effects regression modeling. While past research has shown environmental and individual characteristics influence the JTC distance (Rengert, 2004; Rossmo, 2000), this is not a main concern of this analysis. By fixing the effects of individuals the model only examines variation within an individual, and so a person becomes their own counterfactual. For the multivariable regression model, the IOD was log transformed to account for the skewed distribution of the metric.6 The continuous independent variables of moving home distance and the days in between offences were also log transformed. This allows easier interpretation of slope coefficients as the percentage change in the IOD given a one percent change in the independent variables. Zero values are truncated to the value of one to allow the log transformation.

4.2 SOURCE OF DATA

Information on an offender’s home address and offence location is taken from a large, northeastern police department, from the years of 2003 to 2008. This city contains approximately 150,000 residents as of the 2000 census, and the urban area encompasses approximately 65 square kilometers. The police department has approximately 600 sworn and civilian personnel, and approximately 8,000 reported UCR offences per year (UCR offences include homicide, robbery, rape, burglary, assault, larceny, auto theft, and arson).7

The department in which records were queried maintains unique identifiers for individual offences that allow offence information to be linked to offenders upon a subsequent arrest, and a separate unique identifier for individual offenders that allows offenders who are arrested multiple times to be identified. For the analysis it is necessary to use arrest data, as the analysis will entail linking offenders who committed multiple crimes, and only information on the offender’s home address can be identified when an offender is arrested.

During the sampling period, this northeastern police department recorded 75,017 separate custodial arrests on 33,704 individuals. The study design dictates that the analyses include only offenders who had more than one arrest during the time period, which eliminates 19,838 arrests and offenders from the sample. Another 14,488 arrests and 3,087 offenders were eliminated because the arrest had no linked offence information, the arrest was not within 30 days of the offence, or the offence or home location could not be geocoded. Arrests not within 30 days were dropped as the home location recorded subsequent to the arrest is less likely to be indicative of where the offender lived when the offence was committed. Geocoding hit rates were very high, with over 95% of the total records for offence and home locations having been successfully geocoded. Dropping cases tended to skew the sample to include more active and serious offenders, as well as younger, Black, and male offenders.8

The final analysis consists of 40,691 offences committed by 10,779 offenders (so an average of 3.8 offences committed by the offender in the sample in which all pertinent information is available). Since the IOD examines the distance between offences, this means the first offence for every offender in the sample does not have an associated IOD, and so the final sample consists of 29,912 measures of the IOD.

5. ANALYSIS

The analysis is organized as follows: 1) description of the frequency offenders’ move and the distance of those moves, 2) description of the different characteristics between moving events and non-moving events, as well as the differences in movers compared to non-movers, and 3) estimation of the effect size moving has on the IOD and whether the effect is statistically significant.

More than half of the offenders in this sample (which spans from 2003 to 2008) moved at least once and lived in more than one house. The most moves within the sample was six, and the maximum number of homes listed was six (as an offender could move back into the same home these statistics are not inherently identical). Such extreme mobility may not be unexpected. Cahill and Vigne (2008) found in a sample of parolees that over half moved within one year of their parole date. Wiles and Costello (2000) report that 45% of burglars and car thieves they interviewed had lived at their current address less than 12 months. Intraurban mobility among minority and lower socioeconomic populations is well documented (Quigley and Weinberg, 1977), and such strata encompass the majority of offenders. Moving frequency within this sample also appears similar to the general population of the United States. For the entire United States over 15% of individuals moved within the last year, and 60% of those moves were within the same county (estimated from the Current Population Survey). For individuals between the ages of 18 to 34, 29% of individuals have moved within the last year. For the decennial census, 46% of the population has moved to a different home within the last 5 years.9 In the general population younger people tend to be the most mobile, and most moves are of a short distance within the urban environment (Simmons, 1968). Inevitably the figures used in this study to count moving under count the frequency of moves, as a move can only be observed if an offender has multiple arrests and those arrests coincide with moving in time (e.g. an offender could live in three different residences within a year, and if they were arrested twice within that same year this data could only observe at a maximum two of those residences). Moving frequency also appears similar to that reported by Bernasco, 2010a.

The number of moves speaks nothing of the distance an offender moves though, which could bear importance to an offender’s awareness space and the effects on IOD. Cahill and Vigne (2008) found that parolees’ median first move from their original residence after being paroled was quite far, over 6 kilometers, and such tended to be in different neighborhoods. Wiles’ and Costello’s (2000) findings slightly differ, and that while offenders tend to be itinerant, they still only gain knowledge of few neighborhoods. In this sample the median length of all moves is approximately 2.3 kilometers, and the interquartile range is from 1.0 to 4.1 kilometers. In comparison to Cahill’s and Vigne’s (2008) sample of parolees, these offenders do not move as far.

Table I: Descriptive Statistics

Offence Descriptive Statistics

t-value

Nonmoving events

(n = 22,290)

Moving Events

(n = 7,622)

Mean (SD)

Mean (SD)

Days in between Events

200.7 (263.9)

622.2 (417.0)

-102.41

**

Dummy Variables

%

%

Different UCR

70.9%

77.4%

-11.02

**

Current Offence

Robbery

2.6%

2.1%

2.41

*

Burglary

3.6%

3.2%

1.78

Larceny

9.1%

9.6%

-1.26

Motor Vehicle Theft

3.7%

2.4%

5.15

**

AssaultA

20.1%

21.7%

-3.01

**

Possessing Contraband

32.1%

32.8%

-1.02

In Vehicle

4.5%

5.4%

-2.84

**

Previous Offence

Robbery

2.1%

2.7%

-3.38

**

Burglary

3.3%

3.7%

-1.80

Larceny

9.4%

11.4%

-5.10

**

Motor Vehicle Theft

4.3%

3.8%

1.79

AssaultA

19.9%

23.9%

-7.41

**

Possessing Contraband

30.2%

31.4%

-1.99

*

In Vehicle

4.6%

4.9%

-0.90

A: Assault is for both Simple and Aggravated Assault

*<.05 p-value **<.01 p-value

Table I presents the descriptive statistics for the differences between moving events (i.e. offences that followed a move) and nonmoving events. The only statistic with an obviously different distribution between moving and nonmoving events is that of the time in between events. With a greater amount of time passing, two events are more likely to be committed while living at different home addresses. Summarizing the descriptive differences between movers (i.e. people who moved at least once in their arrest histories) and nonmovers in text, movers tend to have more arrests than nonmovers. This could be an artifact because those with many arrests are more likely to be observed moving. Movers also tend to be younger, and disproportionately minority compared to non-movers, which has been suggested in past literature to decrease the JTC and limit mobility (Rengert, 2004). When examining the spatial distribution of offences and home locations at the census tract level, the distribution for moving and nonmoving events appear to be similar as well. This is important as it suggests that when offenders move, it is spatially random and hence independent of any spatial or neighborhood effects. Thus moving ones home is not correlated with the location of opportunities, which could induce a spurious correlation between moving and the IOD.

Table II: T-test of mean differences in IOD between nonmoving and moving events

Offence

(did not move n, moved n)

Did not move mean (standard deviation)

Moved Mean (standard deviation)

t-value

All offences (22290 , 7622)

1756 (1658)

2125 (1691)

-16.70

**

Burglary (203, 92)

1343 (1726)

2402 (1990)

-4.65

**

Robbery (86, 37)

1606 (1528)

1786 (1596)

-0.59

Motor Vehicle Theft (269, 90)

2468 (1732)

3150 (1723)

-3.24

**

Larceny (958, 455)

2093 (2030)

2446 (2096)

-3.02

**

*<.05 two-tailed p-value **<.01 two-tailed p-value

Table II reports the mean difference in the distance between sequential criminal offences, comparing the IOD after a person moved to the IOD for when the offender stayed at the same home location. The mean distance for the IOD after an offender moves is larger than when they do not move, with an estimated mean difference of 369 meters. This illustrates that when an offender moves, they commit crimes in different locations. The large mean of the IOD is itself surprising, as the offender as a forager paradigm would suggest offences are likely to be tightly clustered in space. Because of the large IOD it may also make the difference of 369 meters substantively uninteresting, and suggest other factors besides where an offender lives play a larger role in where an offender chooses to commit crime.

Although the universe of locations an offender chooses to commit crimes at for all crimes are theoretically dictated by their awareness and activity space, awareness of a space cannot make legitimate opportunities appear. It may be expected that opportunities for instrumental offences are limited in space compared to those for expressive and/or interpersonal crimes, and so moving may not be expected to have as great an impact on where those offences occur.

To demonstrate the moving home effect presented for all crimes is not an artifact of the differing opportunity structures, in Table II below the all crimes sample are t-tests only utilizing crime specific samples of burglary, robbery, motor vehicle theft, and larceny. In these subsets all other offences are eliminated and then the IOD is calculated for offences committed by the same person. Examining subsets of property offences yields the same results as the global test, with the mean IOD increasing between these offences when one moves. The only evidence to the contrary is IOD between robberies, the mean difference between moving and nonmoving events fail to reach statistically significant mean differences. This may be either because robbery events are more influenced by available opportunities within the urban environment, or because the smaller sample size of robbery events lacks power to identify the relationship. While Table II presents evidence that the moving home effect on the IOD is still present even when examining IOD distances between crime specific types, the effect of moving ones home may be variable between different crime types. For burglary and motor vehicle theft, the mean difference is larger than the all crime data, at 1059 and 682 meters respectively. The mean difference in larceny is similar to that for all offences at 353 meters, and the mean difference for robbery events is smaller than all offences at 180 meters.10

Table II presents evidence consistent with the hypothesis that crimes committed after an offender moves are farther away from past offences than would be expected if the offender stayed at the same home address. But the relationships observed could be spurious due to confounding factors, either in the current offence characteristics, the preceding offence characteristics, or in the individual characteristics of movers and non-movers. This bias could be related to past findings of the correlates of the JTC. If one suspects that a smaller JTC is indicative of decreased mobility, one might expect a similar relationship with those same characteristics and the IOD. Also it has been shown offences clustered in space and time have a higher probability of being committed by the same offender (Johnson et al., 2009), and this could be taken as evidence that offences committed closer in time by the same offender are more likely to be closer in space. This is particularly salient because nonmoving events are more likely to be much closer in time compared to moving events in this sample. To demonstrate that the results presented are not confounded by person, offence, or environmental characteristics, a fixed effects regression model is estimated. The form of this model is expressed in Eq. (1) (Gould, 2001).11

[Equation 1]

YijYj=a+ß1(X1ijX1j)+ß2(X2ijX2j)+ßk(XkijXkj)+(eij+vj)Y_{\text{ij}} - Y_{\overline{j}} = a + ß_{1}\left( X1_{\text{ij}} - X1_{\overline{j}} \right) + ß_{2}\left( X2_{\text{ij}} - X2_{\overline{j}} \right) + ß_{k}\left( Xk_{\text{ij}} - Xk_{\overline{j}} \right) + \left( e_{\text{ij}} + v_{j} \right)

Where YijY_{\text{ij}} is the log of the IOD for observation i within individual j, and YjY_{\overline{j}} is the mean logged IOD for individual j. The independent variables on the right hand side of Equation 1 are scaled in a similar manner, in which the model only includes the deviations from the group level mean to estimate the effect of each variable. The group level error term, vjv_{j}, and the intercept, α, \alpha,\ do not have a unique solution, and so in this model the average of vjv_{j} is constrained to equal zero (Gould, 2001). The parameter eije_{\text{ij}} is a residual error term.12 One can see from this specification that the model only assesses variation within an individual. As such the potential confound that movers are inherently different from non-movers is controlled for in the model. This specification makes no assumptions about the distribution of vjv_{j} since its solution is arbitrary (Gould, 2001). The other error term makes the usual OLS assumptions, such as that they are independent and identically distributed, are not correlated with regressors, and have a homoscedastic variance. For the fixed effects model the reported standard errors are the adjusted Hubert-White sandwich errors.

Besides the logged distance of moving ones home, control variables included in the model are offence specific characteristics such as the log of time in between offences, and dummy variables representing whether the current or previous crime was a robbery, burglary, larceny, motor vehicle theft, assault, possession of contraband, or whether the crime was committed in a vehicle. A dummy variable of whether the offence pair was for different types of crimes were included as well. Dummy variables representing the census tract the previous or current offence took place in, and the current or previous census tract the home address was located in were included as control variables as well.

Table III: Fixed Effects Model Predicting logged IOD (n = 29,912)A

Model 1

Model 2 (Fixed Effects)

Variable

B

SE

B

SE

T-value

Log Distance Moved

.035

.004

.021

.005

4.00

**

Log days between offences

.200

.009

.196

.012

16.10

**

Different Offence TypeB

.242

.012

.244

.018

13.73

**

Current OffenceC

Robbery

.179

.035

.110

.041

2.72

**

Burglary

-.182

.032

-.186

.048

-3.11

**

Larceny

.089

.023

.085

.032

2.61

**

Motor Vehicle Theft

.231

.031

.189

.029

6.47

**

Assault

-.221

.014

-.097

.029

-4.69

**

Possession of Contraband

.080

.012

.056

.015

3.81

**

In Vehicle

.214

.025

.185

.027

6.86

**

Previous Offence

Robbery

.046

.037

.023

.041

0.56

Burglary

-.050

.032

-.038

.047

-0.82

Larceny

.072

.022

.086

.031

2.78

**

Motor Vehicle Theft

.195

.028

.185

.027

6.83

**

Assault

-.244

.014

-.129

.021

-6.26

**

Possession of Contraband

.085

.012

.050

.015

3.35

**

In Vehicle

.169

.026

.140

.027

5.15

**

Constant

2.358

.059

2.53

.128

19.82

**

R2

.12

.09D

*<.05 two-tailed p-value **<.01 two-tailed p-value

A: Coefficients for Census Tract Dummy variables, both previous and current home and offence locations, are included in the model but not printed in the table to save space. The Census tract with the most homes and offences within it was used for the reference category.

B: Different Offence Type determined by UCR crime code, simplified to the top crime per arrest. See footnote 4.

C: All offence types are dummy variables equaling 1 when applicable and 0 otherwise. Offence Categories are not mutually exclusive as an arrest can have multiple offences linked to it.

D:R2 for the fixed effects model explains variance in the IOD subtracted from the group means.

Figure 5: Added Distance to IOD when Moving

<p>Figure_5.JPG</p>

Figure_5.JPG

Caption: This figure represents the multiplicative effect that moving ones home has on the IOD. The x axis of the graph represents the expected IOD without the moving home effect (or if one did not move their home). The y axis represents the distance added if an offender did move compared to that same expected IOD without moving. The solid line represents the distance added if an offender had moved 5000 meters from their previous home, and the dashed line represents if an offender had moved 1000 meters from their previous home. The x-axis is limited to the 97.5th percentile of the original IOD distribution. For any move of less than 1000 meters the slope of the line would decrease and approach 0.

Table III presents two regression models predicting the log transformed IOD. The independent variables, the log of distance moved and log days in between offences are represented as continuous variables in the model, and all other independent variables are represented as dummy variables (i.e. if the characteristic is present in that observation a value of 1 is recorded, and 0 otherwise). Model 1 presents the OLS regression estimates without fixing the effects of individuals, and is presented so one can qualitatively assess the affect the fixed effects model has on parameter estimates. Model 2 presents the results for the fixed effects analysis, which will be the focus of the description.

Model 2 finds that when an offender moves the IOD is greater, even when controlling for individual, offence and neighborhood characteristics. The slope value of .021 means a person who has 1 percent increase in the distance moved results in a .02 percent increase in the IOD controlling for other covariates. Although statistically significant, the effect appears small. Mapping the logarithms back to the original units, an offender with an expected IOD of 1500 meters without moving their home, if they had moved their home 2000 meters, all else equal, would add 260 meters to the expected IOD for a total expected IOD of 1760 meters. Figure 5 is meant to be interpreted in a similar counterfactual manner, with the expected IOD without the moving home effect represented on the x axis, and the y axis representing the added distance for each moving event. The x axis is limited to the upper 97.5th percentile of the IOD distribution. Two moving home distances are represented, either moving 1000 meters (dashed line) or moving 5000 meters (solid line). Even for larger expected IODs without the moving home effect, the added distance with moving ones home does not appear to be particularly substantial.

Referring back to Table III, the effect is also small relative to other estimated effects, particularly the time in between offences. Many models including other covariates for offence characteristics were considered, and the only covariate that had any moderating effect on the distance moved coefficient was the time in between events variable. When the days in between offences grow larger, the offender is more likely to have committed the offence a further distance away from the previous offence location. This finding fits well with the popular analogy of an offender as a forager (Johnson et al., 2009), in that offenders are more likely to victimize targets in a smaller area in a short duration of time. Choosing a fixed effects model does moderate the distance moved effect slightly, but in both models the slope of the relationship is rather small compared to days in between offences.

Other dummy variables for offence characteristics have an effect on the IOD, such as when the offences committed are different (represented by “Different Offence Type”) they are estimated to be committed 24 percent farther apart then when they are the same. This could be due to different distributions of targets or opportunities within the environment. For example many assaults may concentrate around areas of commercial bars, but residential burglaries are less likely to occur due to fewer residences in those same commercialized areas. So when a burglary follows an assault, they are more likely to be farther apart then when an assault follows another assault. Also offences committed in vehicles tend to be farther apart, either in the previous or the current offence, suggesting increased mobility when an offender has access to a car. Potential outlying offence locations, which would inflate the IOD, are also controlled by including a set of dummy variables representing in what census tracts the previous and current offence occurred (estimates are omitted from the table). If an offence occurs in a census tract with few other crimes occurring within it, it would be expected the next offence is unlikely to occur within a short distance. Also included are set of dummy variables representing the current and previous census tract the home locations were located in. This would also proxy neighborhood effects of where an offender lived that could theoretically influence where they committed a crime. Because moving events only differed with the observed characteristics of days in between offences, inclusion of these other dummy variables had no appreciable impact on the moving home effect.

To demonstrate that the effect observed is not solely an artifact of differing opportunities within the environment for different crime types, models for the same specific crimes types used in Table II were estimated (i.e. burglary, robbery, motor vehicle theft, and larceny), controlling for time in between offences. While the smaller sample sizes precluded one from using geographic controls that were included in the all crimes model, as well as precluded one from fixing the effects of individuals, prior analysis suggested the only covariate correlated with moving was the time in between offences. Examining the crime specific models, the effect size of moving ones home increases slightly, although it is still substantively small. The effect sizes range from .026 (for motor vehicle theft) to .086 (for burglary). Although the effects are statistically insignificant in the crime specific models, the effect sizes are consistent with the model using all crimes. This suggests the results using all crimes are not biased due to differing opportunity structures for the different crime types. Given the effect sizes are consistent with the all crimes model, the fact that the moving home effect for the crime specific models fail to reach statistical significance could be plausibly attributed to a lack of power due to much smaller sample sizes.

6. DISCUSSION

The current article demonstrates that when an offender moves they tend to commit crime farther away from their last offence than when they do not move. This research reaffirms past work that the home location has a causal effect on the crime location. The effect is small though, suggesting other structural factors may have a greater influence on where offenders choose to commit their crimes.

The small effect size may stand at odds with some of the contemporary literature on the topic, and as such stands to be further examined. While the distance decay effect has been observed in a variety of studies, recent work has questioned the veracity of this distance decay effect when examining intraoffender behavior (Townsley and Sidebottom, 2010). Other recent work by Bernasco and colleagues finds a consistent effect of the distance to home (or previous homes) in where an offender chooses to commit crime (Bernasco 2010a; Bernasco 2010b; Bernasco and Block, 2009). If the home location does not have a real effect on crime location though, the home effect found in these prior studies could be spurious due to unaccounted for environmental characteristics (such as opportunities).

Although the research found a small effect for the home location, it is difficult to say from the study design what factors do influence where offenders choose to commit crime and the IOD. While finding a small moving home effect one may consider the other models of offender behavior presented, gravity models and/or the offender as a forager, as more valid alternatives. But, the data are not entirely consistent with these models of behavior either.

The offender as a forager model for spatial behavior would suggest that offenders should demonstrate sprees of crimes within close proximity in space and time. For the aggregate of offending behavior in this sample this did not appear to be the case. The median IOD for this sample was 1.8 kilometers, a seemingly much wider spread than would be expected with the foraging paradigm. This could be an artifact for several reasons given the study design. It could be a result of the fact that many of the offences included in this study would not be considered predatory behavior. It could also be an artifact that offences are only observed when an offender is arrested, which tend to be far apart in time. As such it could be only longer term factors affecting decision making are observed. It could also be a result of bias from using police arrest data, as an offender is less likely to return to a location after they are arrested. While these critiques make it difficult to suggest the study is evidence against the offender as a forager model of criminal behavior, this model of offender behavior is not consistent with the data presented in this sample of arrests. In the future examining the space/time clustering of offence patterns for both predatory and non-predatory criminal behavior could validate the offender as a forager hypothesis, as well as utilizing separate sources of information other than arrest records.

This leaves the gravity model of criminal decision making left as a plausible alternative to explain the findings in this study. If opportunities are the main driving force behind where offenders choose to commit crimes, it may not be unexpected that the home location has little impact on the offence location. While the home location may bound an offenders awareness and activity space to one urban environment, barriers to travel within that environment may be trivial. Unfortunately this study cannot determine whether the unexplained variation in the IOD is due to the opportunity structure within the environment or other characteristics.

Future studies may improve on this by incorporating effective measures of potential opportunities in the environment, as it is unlikely that the distribution of opportunities to commit crime within the urban environment is random (Rengert et al., 1999). Future research to identify appropriate measures of opportunities, either from census data or from other non-aggregated sources, would help further knowledge of this relationship. Other complications arise when considering different units of analyses to measure opportunities, and that opportunities may differ greatly between crime types. Much potential work remains to attempt to weave all of these aspects together in explaining offender behavior, as both units of analysis and measures of criminal theories differ greatly among articles in this arena.12

7. LIMITATIONS AND CONCLUSION

The greatest weakness of the current study is that it relies on official arrest records. This can have two effects on my analysis: One is that the home listed when one is arrested is likely to have large measurement error. This error is likely to attenuate any relationship found between the home location and where the offence occurs. The second effect it could have on my analysis is in regards to biasing the spatial location of offences. This spatial bias may impact where the offence takes place, as police resources are concentrated in specific criminogenic places, and with that greater police presence the probability of apprehension may be greater. Being arrested may also present a negative stimulus, and make offenders less prone to travelling to that same area to commit crimes. This policing bias is unlikely to influence where offenders live in spatial relation to those offence locations though, and as such the moving home relationship found in this study is unlikely to be an artifact of the filter of police records.

Another weak aspect of the research presented in this article is the non-equivalent nature of moving and nonmoving events, and my attempt to statistically control for their nonequivalence. While only one observed characteristic, the time in between offences, was obviously different between groups and accounted for some of the moving home effect, it is possible other unobserved characteristics are confounded with moving, and thus the moving home effect could be spurious. While this is the case with the majority of nonexperimental research, the causal relationships that were found are unlikely to be an artifact due to the selection into treatment, moving one’s home. The only suggested relationship in the literature would be the anecdote that a drug user is more likely to move closer to their drug dealer (Rengert, 2004). This selection into moving, if no reasonable targets were nearby past residences, would make the IOD smaller, as an offender would be more likely to move to areas in which they already victimized targets, thus biasing the moving home effect on the IOD downward. Also past JTC research suggesting a buffer region around which an offender does not commit crime in close proximity to their home would also make this relationship unlikely; an offender would increase probability of detection if they interacted on a regular basis with those they victimized (Rossmo, 2000).

This research demonstrates that prolific offenders tend to report a wide variety of home locations upon arrest, and over half of the offenders within the sample reported moving within the six years. This begs the question, what exactly is the construct of the home address we are capturing within these records? Is the home address listed when an offender is arrested really a defined node through which his awareness is spread? Although police arrest records may introduce a lot of measurement error, as likely some offenders have little attachment to their listed address, it appears the home address is a valid predictor of where the offence will occur.

In sum, this research finds that offenders’ tend to commit crimes in locations farther away from past offences when they move compared to when they do not move, controlling for potential confounding factors of the offence, of the neighborhood, and of the offender. These results suggest that the home location does have an impact on where an offender will choose to commit a crime, independent of other factors. The effect is small though, and future research may look to other structural factors, particularly those dictating where opportunities to commit crime are most present in the environment, to better explain where offenders choose to commit their crimes. All future research should explicitly take into consideration the extreme itinerancy of offenders’ when examining arrest histories, and consider those limitations when examining the JTC distance.

FOOTNOTES

1. Figures 1 to 5 were made using the statistical package R (R Development Core Team, 2010).

2. Addresses were geocoded using ArcGis version 9.3 based on a street centerline file for the county which contained the northeastern police department for 2008 (maintained by the State government that the police department is located within). The coordinates were projected in the same system as the street file, using the applicable UTM zone with the 1983 North American Datum. A side offset of 20 feet (approximately 6 meters) was used to geocode the addresses.

3. These crime categories were chosen based on that they are a large proportion of the sample and because of theoretical importance.

4. Since an arrest can be linked to multiple offences, to calculate this statistic an arrest was simplified to the most serious offence.

5. Several reviewers suggested that the analysis include structural covariates. While inclusion of these variables would substantively add to the findings, for simplicity only the dummy variables representing different census tracts were included. This accounts for any variation between census tracts for any observable or unobservable attributes, hence the effects reported in the final analysis would not differ even if the analysis were to include structural covariates at the census tract level.

6. The IOD has a similar distribution to that of the JTC (a strong right skew), and is similar to many studies of distance in-between events for human activity (for an example see Gonzalez et al., 2008). Log base 10 was the transformation used for the IOD (and all other variables that were log transformed).

7. UCR figures obtained from table 18, http://www2.fbi.gov/ucr/cius2009 on 12/20/2010. Law enforcement figures taken from the Census of State and Local Law Enforcement Agencies, Bureau of Justice Statistics. ICPSR study number ICPSR28001-v1.

8. Descriptive statistics of offender and offence characteristics, as well as the census tract information for missing data is available upon request.

9. American Community Survey and 2000 census figures were taken from the online utility American Fact Finder provided by the Census. Specific urls and variable codes are available upon request.

10. Because of the skewed distribution of the IOD, I also examined the t-test of mean differences using the log transformed variable. Inferences are largely unchanged, and for simplicity only the mean differences of the original variables are reported. The only difference in inferences is the t-value for the mean difference in the logged IOD for the larceny specific subset changes from 3.32 to 1.68, so the p-value is only significant at the .10 level instead of .05.

11. This model was calculated using Stata’s xtreg command with the fixed effects sub-command (StataCorp, 2009). Note the degrees of freedom in the model are adjusted for the fact that the group level means need to be estimated (Gould, 2001).

12. The actual model that is estimated is rescaled by the grand means for each variable, YijYj+Y=a+ß1(X1ijX1j+X1)+ß2(X2ijX2j+X2)+ßk(XkijXkj+Xk)+(eijej+vj+e)Y_{\text{ij}} - Y_{\overline{j}} + \overline{Y} = a + ß_{1}\left( X1_{\text{ij}} - X1_{\overline{j}} + \overline{X1} \right) + ß_{2}\left( X2_{\text{ij}} - X2_{\overline{j}} + \overline{X2} \right) + ß_{k}\left( Xk_{\text{ij}} - Xk_{\overline{j}} + \overline{\text{Xk}} \right) + \left( e_{\text{ij}} - e_{\overline{j}} + v_{j} + \overline{e} \right) . Because of this rescaling and because the average of vjv_{j} is constrained to equal zero, the predicted values for the IOD is equal to the average IOD.

12. For instance see the recent article of Bernasco and Block (2011) and compare that to the majority of research examining correlates of crime at larger census aggregations (for one example see Hipp, 2007).

References

Bernasco W (2010a) A sentimental journey to crime: Effects of residential history on crime location choice. Criminology 48:389-416

Bernasco W (2010b) Modeling micro-level crime location choice: Application of the discrete choice framework to crime at places. J Quant Criminol 26:113-138

Bernasco W, Block R (2009) Where offenders choose to attack: A discrete choice model of robberies in Chicago. Criminology 47: 93-130

Bernasco W, Block R (2011) Robberies in Chicago: A block-level analysis of the influence of crime generators, crime attractors, and offender anchor points. J Res Crime Delinq 48: 33-57

Bernasco W, Kooistra T (2010) Effects of residential history on commercial robbers’ crime location choices. Euro J Criminol 7: 251-265

Brantingham PL, Brantingham PJ (1993) Nodes, paths and edges: Considerations on the complexity of crime and the physical environment. J Environ Pyschol 13: 3-28

Brantingham PL, Brantingham PJ (1999) A theoretical model of crime hot spot generation. Stud Crime Crime Prev 8:7-26

Cahill M, LaVigne N (2008) Residential mobility and drug use among parolees in San Diego, California and implications for policy. In Thomas YF, Richardson D, Cheung, I (eds.) Geography and Drug Addiction Springer, Netherlands, pp.85-115

Capone DL, Nichols WW (1976) Urban structure and criminal mobility. Am Behav Sci 20: 199-213

Clare J, Fernandez J, Morgan F (2009) Formal evaluation of the impact of barriers and connectors on residential burglars’ macro-level offending location choices. Aust New Zealand J Criminol 42: 139–158

Cohen LE, Felson M (1979) Social change and crime rate trends: A routine activity approach. Am Sociol Review 44: 588-608

Davies A, Dale A (1996) Locating the stranger rapist. Medicine, Science, Law 36: 146-156

Gonzalez, MC, Hidalgo CA, Barabasi A (2008) Understanding individual human mobility patterns. Nature 453: 779-782

Goodwill AM, Alison LJ (2005) Sequential angulation, spatial dispersion and consistency of distance attack patterns from home in serial murder, rape and burglary. Pyschol, Crime Law 11: 161-176

Gould W (2001) Interpreting the intercept in the fixed-effects model. StataCorp LP. Obtained from http://www.stata.com/support/faqs/stat/xtreg2.html on 12/20/2010

Hesseling RBP (1992) Using data on offender mobility in ecological research. J Quant Criminol 8: 95-112

Hipp JR (2007) Block, tract, and levels of aggregation: Neighborhood structure and crime and disorder as a case in point. Am Sociol Review 72: 659-680

Jacobs BA (2010) Serendipity in robbery target selection. Br J Criminol 50: 514-529

Johnson SD, Bernasco W, Bowers KJ, Elffers H, Ratcliffe J, Rengert G, Townsley M (2007) Space-time patterns of risk: A cross national assessment of residential burglary victimization. J Quant Criminol 23: 201-219

Johnson SD, Summers L, Pease K (2009) Offender as forager? A direct test of the boost account of victimization. J Quant Criminol 25: 181-200

Levine N (2007) CrimeStat: A Spatial Statistics Program for the Analysis of Crime Incident Locations (v 3.1). Ned Levine & Associates, Houston, TX

Pettiway LE (1985) The internal structure of the ghetto and the criminal commute. J Black Studies 16:189-211

Quigley JM, Weinberg DH (1977) Intra-urban residential mobility: A review and synthesis. Int Reg Sci Review 2: 41-66

R Development Core Team (2010) R: A language and environment for statistical computing. R Foundation for Statistical Computing. Vienna, Austria

Ratcliffe JH, Rengert G (2008) Near-repeat patterns in Philadelphia shootings. Security Journal 21: 58-76

Rengert GF (2004) Journey to Crime. In Bruisma GJN, Elffers H, de Keijser JW (eds.), Punishment, Places and Perpetrators: Developments in Criminology and Criminal Justice Research, Willan Press, Portland, OR, pp.169-181

Rengert GF, Piquero AR, Jones PR (1999) Distance decay reexamined. Criminology 37: 427-446

Rengert GF, Wasilchick J (2000) Suburban Burglary: A Tale of Two Suburbs, Charles C. Thomas, Springfield, IL

Rossmo, DK (2000) Geographic profiling, CRC Press, Boca Raton, FL

Shaw CR, McKay HD (1969) Juvenile Delinquency and Urban Areas: A Study of Rates of Delinquency in Relation to Differential Characteristics of Local Communities in Am Cities. Revised Edition, University of Chicago Press, Chicago, IL

Sherman LW, Gartin PR, Buerger ME (1989) Hot spots of predatory crime: Routine activities and the criminology of place. Criminology 27: 27-55

Short MB, D'Orsogna MR, Brantingham PJ, Tita GE (2009) Measuring and Modeling Repeat and Near-Repeat Burglary Effects. J Quant Criminol 25: 325-339

Simmons JW (1968) Changing residence in the city: A review of intraurban mobility. Geographical Rev 58:622-651

Smith TS (1976) Inverse distance variations for the flow of crime in urban areas. Social Forces 54: 802:815

StataCorp. (2009) Stata Statistical Software: Release 11. StataCorp LP, College Station, TX

Tita G, Griffiths E (2005) Traveling to violence: The case for a mobility-based spatial typology of homicide. J Res Crime Delinq 42: 275 - 308

Townsley M, Sidebottom A (2010) All offenders are equal, but some are more equal than others: Variation in journeys to crime between offenders. Criminology 48: 897-917

van Koppen PJ, de Keijser JW (1997) Desisting distance decay: On the aggregation of individual crime trips. Criminology 35: 505-515

Warre J, Reboussin R, Hazelwood RR, Cummings A, Gibbs N, Trumbetta S (1998) Crime scene and distance correlates of serial rape. J Quant Criminol 14: 35-59

Wiles P, Costello A (2000) Road to Nowhere: The Evidence for Travelling Criminals. Great Britain Home Office Research Development and Statistics Directorate Information and Publications Group, United Kingdom

Weisburd D, Bushway S, Lum C, Yang SM (2004) Trajectories of crime at places: A longitudinal study of street segments in the city of Seattle. Criminology 42: 283-322

Comments
0
comment

No comments here