The authors wish to thank Eric Beauregard for his feedback on previous versions of this article.
Patrick Michaud, School of Criminology, University of Montreal, C.P. 6128, Succursale Centre-Ville, Montreal, QC, Canada, H3C 3J7. Phone: +1 514-343-6387
E-mail: [email protected]
Purpose. This study empirically quantifies the methodological biases of the journey-to-crime measurement, by investigating the extent to which they affect its ability to estimate the itineraries offenders actually travel during the perpetration of their crimes.
Methods. With the support of police-arrest records, interviews, and web-mapping technologies, the detailed route taken by 98 offenders during 449 theft-related crimes are reconstructed in order to test some of the key assumptions underlying journey-to-crime research.
Results. Police data used to compute home-crime distances have been found to provide satisfactorily accurate crime-location addresses, but poorly accurate offender home-addresses. Several explanations of why the police have problems correctly identifying where an offender is truly residing on the day of a given crime are presented. Even if the offender's residence was the most important node in their crime journey, the actual travel undertaken by offenders was much more complex than the home-crime-home itinerary assumed by the journey-to-crime measurement.
Conclusions. Despite its numerous drawbacks, the traditional journey-to-crime measure is still a valid and useful proxy for the total distance actually traveled by offenders in robbery and “other theft,” but not in burglary and motor-vehicle theft. Implications for criminal mobility research and future studies are discussed.
journey-to-crime, home-to-crime, criminal mobility, environmental criminology, construct validity
The distance between an offender's home and their crime location ― the journey-to-crime ― has been the measure of choice to estimate criminal mobility since the 1930s. Readily available data, consistent results across studies, and numerous implications for criminological theories in general, and for geographic offender-profiling in particular, have certainly contributed to its ubiquity in the offender-mobility literature. Established scientific knowledge about the spatial behaviors of offenders has been ― and still is ― strongly influenced by empirical research examining home-crime distances. Even though these studies have repeatedly addressed most of the drawbacks of the journey-to-crime measurement, they have failed to quantify how these flaws affect its validity and overall usefulness.
The main findings in journey-to-crime research are that most crimes are committed a short distance from the offender's residence, and that the number of crime occurrences decreases the further the offender is from their home — a phenomenon called the distance-decay function (Rossmo, 2000). These results are consistent with the least-effort principle, which postulates that individuals have a propensity to choose the closest opportunity among a pool of equivalently desirable ones (Zipf, 1965). Some studies have found that offenders avoid committing crimes in the immediate proximity of their home, because of the risk of recognition in familiar neighborhoods (e.g., Canter & Larkin, 1993).
In one of the most exhaustive study on the journey-to-crime, Ackerman and Rossmo (2015) found that residence-to-crime distances for 25,154 crimes (regardless of type) ranged from a minimum of 0 km (i.e., a crime committed at home) to a maximum of 47.3 km (M = 10.1 km; Md = 8.5 km; SD = 8.2 km). In general, adult offenders tend to travel more than juvenile offenders to commit their crimes, the latter being more likely to choose a target in the vicinity of their residence (Levine & Lee, 2013). Journey-to-crime distance also varies by crime type, with violent crimes generally occurring closer to the offender's home than is the case with property offenses (Ackerman & Rossmo, 2015; Beauregard et al., 2005). With regard to gender, some studies have reported that men commit their crimes farther from their residence than do women (Nichols, 1980; Rengert, 1975), while other have reported the opposite (e.g., Chainey et al., 2001). Research on the race-distance relationship is more conclusive, with Whites traditionally traveling longer distances to commit their crimes than non-Whites (Canter & Gregory, 1994; Pettiway, 1995; Warren et al., 1998). However, what are these results really telling us about the mobility of offenders?
Construct validity is the extent to which a measure genuinely assesses the concept it is deemed to measure (Strauss & Smith, 2009). To our knowledge, no study has ever evaluated the ability of the journey-to-crime measurement to accurately estimate the real trip undertaken by offenders in the perpetration of their crimes. Its construct validity seems to have just been taken for granted, probably fostered by its apparent face validity (i.e., the fact that “it looks like” a good measurement). However, the construct validity of the journey-to-crime measurement is called into question by the fact that it disregards the real trip origin, the actual itinerary involved, and the true distance covered by offenders.
The journey-to-crime measurement presupposes that all crime journeys start at the offender's home. This is obviously not always the case, but almost no empirical research has been able to provide a reliable estimate of the proportion of offenders truly departing from their residence to commit their crimes. To our knowledge, the only exception is Pettiway (1995), who was able to confirm directly with offenders the true origin of their crime journey, and who found that only a minority started their trip from home. Furthermore, homeless offenders who do not have a fixed home address have no choice but to base their criminal activities from other social-activity locations such as bars and pool halls (Rengert, 1996) or from a relative's or friend's home where they are temporarily staying (Fleisher, 1995).
The validity of the presumption that home is the origin of all crime journeys is also weakened by the fact that most people, including offenders, seem to spend approximately half their day outside their residence (Statistics Canada, 2006; Wikström et al., 2010). Daily-routine activities are not characterized exclusively by round trips from home, but more accurately by a succession of trips and stopovers en route to the many other nodes in one's activity space (e.g., workplace, school, friends’ or family members' dwelling, entertainment location, drug-selling spot, prostitution area). Even if home is the origin in the morning and the destination in the evening within most people’s daily routine, this is not necessarily the case for active offenders. Because the journey-to-crime measurement exclusively considers the offender's residence as the origin of the crime journey, most researchers have started to use other terms, such as home-to-crime (e.g., Snook et al., 2005). Even when a crime journey effectively begins at the offender's home, researchers using a journey-to-crime measure are forced to take for granted the accuracy of the addresses provided. Because secondary data like police arrest records ― the principal source of information in this field of study ― are generally not collected for the purpose of scientific research (Bernasco, 2017), the reliability of information regarding offenders’ home addresses coming from such sources remains to be seen.
Even if the distance between an offender's home and their crime location could provide some indication of the extent of their mobility before the crime, it says little about the real crime journey involved. The real crime journey is the detailed itinerary traveled by an offender from the moment they decide to embark on a crime journey on the day of the crime to the moment they return to a safe location after the perpetration of their crime. This includes all the stopovers and/or detours made over this time span, the geographical extent of target searches, and any movements made during the crime (e.g., change of crime location, also called journey-during-crime) (Bernasco, 2014). By quantifying criminal mobility in terms of a fragmentary measure of home-crime distance that disregards most aspects of a crime journey, journey-to-crime research has been constrained to assume that: 1) the home-crime trip is direct; 2) the search-for-a-target, if any, is carried out during the home-crime trip; 3) there is only one crime location per criminal event; and 4) the journey-to-crime mirrors the journey-after-crime. However, no study has ever validated these assumptions or verified the extent to which they actually reflect the itinerary and true distance traveled by an offender during a crime journey.
The main objective of this study was to empirically test some of the key assumptions of journey-to-crime research that have hitherto been largely taken for granted by scholars. Given the leading influence of journey-to-crime in the field of environmental criminology, it appears essential to make sure the measurement actually estimates what it is supposed to. Ultimately, the question is whether the traditional journey-to-crime measurement is able to provide researchers with accurate information about the true distances traveled by offenders during the commission of their crime. In simple terms, is journey-to-crime a valid indicator of criminal mobility?
This study relies on an original research design applied to a unique sample of theft-related offenders to constitute what we believe to be the most exhaustive investigation of offender mobility to date. We consider that the geographical accuracy of our reconstructed itineraries comes close to the GPS-monitored ones that have been studied by Rossmo et al. (2012), with the benefit of not having the “noise” of the individual's daily travel for non-criminal purposes. Moreover, our study stands out by relying on a significant sample and by having access to the rich personal account of offenders, especially with regard to their mode of transportation, criminal motivation, and target selection. More specifically, this study tested the following hypotheses.
Hypothesis 1: Police arrest records provide researchers with satisfactorily accurate offender home- and crime-location address data.
Hypothesis 2: Home-to-crime corresponds to the itinerary taken in most crimes, because the offender's home is the most important node in a crime journey, the itinerary is usually direct (without any detours and/or stopovers), and the outbound trip (i.e., journey-to-crime) generally mirrors the inbound trip (i.e., journey-after-crime).
Hypothesis 3: Home-crime distance computed from raw police data is a valid proxy of the home-crime distance validated by the offender, and of the true distance traveled by an offender during the pre-crime phase, post-crime phase, and entire crime journey.
The Correctional Service of Canada (CSC) supplied the names, identification numbers, dates of birth, and criminal convictions of the entire population of 3,356 adult males having received a federal sentence of two years or more in the province of Quebec (Canada) between January 1, 2010, and December 31, 2012. A list was drawn up of all 833 offenders having been convicted for at least one theft-related crime, namely robbery (any type), breaking and entering (any type with an intent to steal), motor-vehicle theft, and “other theft” (any type).
Recruitment of participants took place in 2013 in ten distinct correctional institutions located in the province of Quebec. A written invitation for an individual face-to-face meeting was sent to 453 (54.4%) of the 833 offenders. The remaining 380 offenders were not solicited because at the time of recruitment they: 1) had been released on parole and/or had terminated their federal sentence (72.1%); 2) had been transferred to a correctional institution we did not visit, mainly because of its geographical remoteness (21.6%); or 3) were physically or mentally unavailable for a research project (e.g., in court, in hospital, suffering from serious mental health issues) (6.3%). The aim of this one-on-one session was to briefly (i.e., in 10–15 minutes) present the research project, determine the offender’s willingness to participate, and have them sign a consent form if they were willing to participate. Offenders were informed that their participation was totally voluntary and that they would not receive any compensation and/or preferential treatment for their involvement.
Among the 453 called-upon offenders, 172 (38.0%) initially agreed to take part in the study. The others either verbally refused to participate (49.9%) or did not attend the meeting after being invited on two separate occasions, on two separate days (12.1%). Two hypotheses can be formulated to explain such a low participation rate. First, several offenders may have been dissuaded from participating in this non-mandatory research project because of the highly demanding nature of their potential participation (e.g., lengthy duration of interview, unknown number of meetings, intrusive questioning), which did not provide any apparent benefits. Second, despite our explanations and reassurances, some offenders remained suspicious, and reluctant to participate in a study in which their spatial behaviors were scrutinized (e.g., they questioned whether this was an undercover attempt to link them to unresolved crimes).
Before conducting the research interview(s), the correctional file of each participant, including all the available police reports concerning theft-related crimes committed since January 1, 2005, was systematically reviewed and annotated. The decision to restrict our analysis to crimes perpetrated after this date was made to limit the number of criminal events to analyze while ensuring a sufficient pool of “fresh-in-memory” crimes to revisit with offenders. The Quebec provincial police database (Centre de renseignements policiers du Québec) was also consulted to extract, for each theft-related crime in the study, addresses of both the offender's home and the crime location. Even though most of this information was already available in the police reports, it was compiled for comparative purposes and to question offenders in the event of discrepancies.
The preparation phase (i.e., the time between an offender's recruitment and first interview) for this study lasted between several weeks to over a year, and 43 offenders initially interested in participating in the study decided to quit over this period for various reasons. Another 19 offenders were either unreachable because they had completed their federal sentence or had been transferred to a remote correctional institution. Four offenders were also excluded from the study because no police report involving a theft-related crime committed after 2005 was found in their file. Five offenders who were met with during the preparation phase exclusively to help out in testing and clarifying our interview questions and in adapting and standardizing our procedures were not included in the final sample. Finally, three offenders were excluded because of a blatant lack of cooperation during the interview phase of the research project. The final sample thus consisted of 98 offenders.
At the time of admission, the 98 offenders were between 18 and 60 years old (M = 36.2; SD = 10.5) and were serving an average prison sentence of 3.8 years (SD = 1.8). The majority were White (82.7%), chose French as their preferred official language (93.9%), and had a criminal record prior to admission (89.8%). On average, offenders had been previously sentenced on 13.4 occasions (SD = 9.8) and had accumulated 46.0 criminal convictions (SD = 40.2). Independent-sample t-tests and chi-square tests of independence were conducted to compare the 98 “included” offenders to the 735 “excluded” offenders on five different control variables.1 No significant differences were found between the two groups, suggesting that the offenders included in this study could be reasonably considered to be a representative sample of the overall population of theft-related federal convicts imprisoned in Quebec.
A face-to-face interview was conducted with the 98 offenders between October 2013 and June 2014. Offenders were questioned for a total of 2.0 to 11.3 hours (M = 4.5; SD = 1.8) divided into one to five sessions (M = 2.0; SD = 0.9). Interviews were conducted either in French or English by two experienced clinical criminologists (including the first author). Most interviews (80.8%) took place in a correctional setting. The remainder (19.2%) were conducted in the community while offenders were on parole — either in a parole office, in a community-based residential facility (i.e., halfway house), or at the offender's residence.
Interviews were divided into two parts. The first part consisted of a semi-structured interview during which offenders were questioned on different topics, including their childhood, education, employment, substance-abuse history, criminal activities, and geographic behaviors. With the support of a list of their potential home addresses and precise periods during which they were incarcerated versus free in the community (compiled during the preparation phase), offenders were also asked to reconstruct their residential history timeline since 2005. The second part of the interview was dedicated to the revisitation of all the theft-related crimes they had been convicted of since 2005 (and for which a police report was available). The objective was to unravel their detailed itinerary and underlying decisions from the moment they woke up on the day of the crime to the moment they came back to a safe place after its perpetration. Extensive verbatim notes were taken during both parts of the interview.
As a group, the 98 offenders had been convicted for a total of 956 theft-related crimes perpetrated since 2005. A police report was available for 680 (71.1%) of these crimes.2 For each crime supported by a police report, the offender was asked to rate his level of recollection of that particular event and its surrounding circumstances on a scale from 0 (absolutely no recollection) to 6 (perfect or almost perfect recollection).3 Because helping people to sharply recall a personally experienced event has been reported to help them remember its rich spatiotemporal context, even several years later (Burgess et al., 2002), a vignette with key characteristics of the criminal event (produced during the preparation phase) was read out loud by the interviewer and a Google Street View photo of the crime location was shown. Crimes were reviewed successively, in chronological order, from the oldest crime to the most recent one committed by each individual offender. Only the 449 crimes (66.0%) offenders admitted having committed and of which they still had a good recollection (i.e., scale scores of 3 to 6) were kept for further questioning. Each well-recalled crime was then revisited exhaustively, individually, and in chronological order, with the offender. The number of crimes to review ranged from 1 to 18 per offender (M = 4.6; SD = 3.6). This included 162 robberies (by 58 offenders), 198 burglaries (by 49 offenders), 21 motor-vehicle thefts (by 12 offenders), and 68 “other thefts”4 (by 27 offenders).
Independent-sample t-tests and chi-square tests of independence were performed to compare the 449 “included” crimes to the 231 “excluded” crimes on five different control variables.5 Results showed that the most recent and lucrative crimes were overrepresented in the sample of included crimes. Offenders were more inclined to remember robberies (compared to burglaries or “other thefts”), and crimes which led to an expeditious arrest by the police (i.e., the offender is arrested at the crime location or during their flight from the scene of the crime). The proportion of excluded crimes per offender was also higher in those who committed a greater number of offenses.
For each of the 449 crimes included in the study, 10 variables were coded. When it was possible to use a computer during the interview (85.4% of interviews), Google Maps and Google Street View were used for support. When a computer was either not available or not authorized, a detailed street atlas (1:25,000 scale) and Google Street View photo prints of the crime location and offender's presumed home address were used instead.
1) Awakening location. This is the place where the offender claimed to have woken up on the day of the crime. This location was asked specifically to help offenders remember where they were residing on the day of the crime, even though we acknowledge that many offenders may have slept elsewhere. When the offender was living on the street, he was asked to provide the approximate location where he woke up, even if he could not provide an exact address (e.g., under that bridge, in that park, near that street corner).
2) Offender-home location. This is the place where the offender alleged he had his primary home address (PHA) on the day of the crime ― i.e., the place where he generally slept the equivalent of four nights or more during a given week. If the offender had been homeless, sleeping in motels or hotels, temporarily sheltered in community resources, or transiently staying for a few days with people such as family or friends, or was unable to designate a single place where he normally slept four nights or more in a given week, he was considered to have had “no fixed abode” (NFA).
3) Decision location. This is the place that the offender identified as the location at which he made the decision to embark on a crime journey on the day of the crime. In this context, “embarking on a crime journey” means departing from a location for criminal purposes (i.e., subsequent trips had to have been linked directly or indirectly to the criminal event). Even if the decision to commit the crime had been made several days or weeks before its perpetration, the decision location is the place where the offender embarked on a crime journey on the day of the crime that was recorded. The location of another crime could be considered a decision location when the outcome of one crime directly influenced the decision to immediately and suddenly perpetrate a subsequent crime (e.g., the initial crime failed or was disappointing, so another crime was perpetrated right after). If the decision to commit a particular crime was made “on the spot” (e.g., the spontaneous encountering of a criminal opportunity), the address of the crime was recorded as the decision location.
4) Choice of target. When leaving the decision location, did the offender know his precise target and its location (i.e., a planned target), did he need to investigate to find the right target (i.e., a searched target), or was he unaware that he was about to perpetrate a crime (i.e., an opportunistic target)? In crimes with a planned target, the exact crime location is chosen before embarking on a crime journey (e.g., to commit a crime at this specific bank or this particular house), even though the precise routes to get there may still be undetermined. In crimes with a searched target, the offender is also crime-driven when leaving the decision location, but the target is unknown and needs to be found based on specific criteria. In crimes with an opportunistic target, offenders are in movement for non-criminal reasons (e.g., to go shopping, go visit a friend, go to work) until they are confronted with a high-value desirable criminal opportunity. These offenders may have a general predisposition to perpetrate crimes, but when they leave their last location on the day of the offense, they have no specific intention of committing a crime.
5) Distance traveled in the search for a target (if any). When the crime involved a search for a target, the offender was asked to estimate the time (in minutes) he spent searching for a criminal opportunity, along with the mode of transportation (e.g., car, bicycle, walking) used during the process. This information was subsequently used to approximate the distance traveled (in kilometers) during this particular search for a target. For each situation, three simulations of itineraries with the mode of transportation used by the offender were performed in Google Maps. Destinations were chosen randomly and manually directly on the web-mapping platform, in the search area given by the offender, until the overall trip time reached the time estimated. The average distance after three simulations was then used as the approximation of the distance traveled by the offender during the search for a target of this specific crime.6
6) Crime location. This is the place the crime occurred. When an offender was unsure of the exact crime-location address (e.g., in cases of multiple residential burglaries), the crime location recorded by the police was chosen. When the crime-location address was obviously incorrect (e.g., the crime was a bank robbery and the address recorded by the police was a residence) and/or the offender was convinced that the crime location address was erroneous, the correct location was found after consulting other sources, and this address was used instead.
7) Safe-house location. This is the place the offender indicated having sought refuge after committing the crime. To be considered a safe house, the location had to provide the offender with both shelter and a sense of safety for a significant amount of time (e.g., generally more than 30 minutes). The suitability of each safe-house location was assessed on a case-by-case basis along with the offender.
8) Intermediary stops (if any). These are all the transitory places the offender visited between the decision location and the crime location, and between the crime location and the safe-house location. A crime location could be considered an intermediary stop when the offender committed more than one crime in the same crime journey (i.e., from the decision location to the safe-house location).
9) Home-crime distance. For each crime, two variations of the home-crime distance were calculated. The first measure (traditional home-crime distance) was computed from raw police data. The offender's home and crime-location addresses were taken directly from police reports; when one of these was missing or unknown, the Quebec provincial police database was searched for any existing addresses linked to this specific crime. The second measure (offender's validated home-crime distance) was computed from double-checked and cross-validated data. The offender's home address at the time of crime was given by the offender himself and the crime-location address was corroborated by consulting other sources, including the offender. When the offender was NFA on the day of the crime, the address of the awakening location was used instead. Both measures of home-crime distance were calculated in kilometers with Google Maps.
10) Distance traveled during the crime journey. For each crime, three measures of distance traveled by offenders were estimated. First, the distance traveled during the pre-crime phase corresponds to the distance covered by the offender from the decision location to the crime location, including all detours and stopovers, and any searches for a criminal target. Second, the distance traveled during the post-crime phase corresponds to the distance covered from the crime location to the safe-house location (or the arrest location, if the offender was apprehended by the police before reaching a safe house), including all detours and stopovers, and any searches for a criminal target. Finally, the distance traveled over the entire crime journey corresponds to the sum of the distances covered by the offender in the pre-crime phase and in the post-crime phase. Distances were calculated in kilometers with Google Maps, using the mode of transportation indicated by the offender and the fastest route available (without traffic).
The 449 crime journeys reconstructed with offenders yielded a total of 1,743 places and/or stopovers. With the support of various official and non-official sources, it was possible to find the exact address (i.e., civic number, apartment number if any, street name, city, province, and postal code) of 78.9% of these locations. The precise address of public places (e.g., restaurants, schools, stores) was usually easy to find by typing their name in the Google Maps search bar. When an offender did not recall the exact address of a private place, he was asked to provide a nearby landmark (e.g., a bridge, a shopping center, a park) in a constrained zone (e.g., in this neighborhood, near this subway station). Once the landmark was discovered, we entered Google Street View’s immersive mode and followed the offender's directions until we virtually arrived at the exact destination.7
When the offender was unaware of the exact address and was unable to find it with the support of Google Maps or Google Street View, he was asked to provide an approximate location (generally a street corner); this had to be located within 0.25 km² (500 meters by 500 meters) of the real location. Approximate locations accounted for 18.4% of the places and/or stopovers composing crime journeys. When the offender was unable or unwilling to provide a sufficiently precise location, the address was considered missing (2.8% of the places and/or stopovers). When the decision-location, the crime-location, or the safe-house location address was missing, or when it referred to an intermediary stop representing a detour for the offender, the phase of the crime journey to which it belonged (pre-crime and/or post-crime) was excluded from further analysis. However, if the offender was able to confirm that the missing intermediary stop was located “on his way” to the crime scene or to the safe house (i.e., no detour), the place was simply removed from this particular phase of the trip.
An implicit assumption in journey-to-crime research is that addresses derived from police data are satisfactorily accurate. For each criminal event, the accuracy of both the offender's home and crime-location addresses were assessed by comparing those extracted from raw police data to those verified and cross-validated with offenders and other official documents. When the addresses matched, police data was considered to be “accurate.” Spelling errors or incorrect postal codes were not sufficient to classify an address as false or erroneous. To be considered “inaccurate,” the address supplied by police data had to refer to a different geographical location than the one given by the validated address. Results are presented in Table 1.
[INSERT TABLE 1 ABOUT HERE]
In our sample, the accuracy of the offender's home address as supplied by police data was strikingly low, with as much as 50.3% of crimes being recorded with a wrong perpetrator residential address. This low accuracy rate was comparable across the four crime types. Crime-location addresses, however, were overwhelmingly accurate (96.7%) and the accuracy rate was significantly better for burglary than for motor-vehicle theft (97.7% versus 85.7%; χ2 [3] = 9.42, p < .05; Cramer's V = .148). The crime location of three (15.0%) of the 21 motor-vehicle thefts were wrongly recorded as having occurred at the owner's home or vehicle-recovery address rather than the real location of the crime. On average, the street-network distance between the incorrect and the accurate address was 26.9 km (SD = 33.6) for the offender's home (n = 198) and 4.4 km (SD = 3.9) for the crime location (n = 14). Almost half (48.5%) of the inaccurate home addresses and 21.4% of the inaccurate crime locations were in an inaccurate city. Additional verification was undertaken to exclude the possibility of a police coding bias in the recording of addresses.8
The startling discovery that more than half of offenders' home addresses extracted from police data correspond to a place they were not living at on the day of the crime calls for an in-depth examination. During the interview, each time an offender claimed he was not residing at the address recorded by the police on the day of a given crime, he was asked to provide justification for the discrepancy (was this a former residence? If yes, since when did he move from this place? If no, what is this address? etc.).
The three main causes of inaccuracies in offenders’ home addresses were that the police mistakenly: 1) believed the offender was living at the residence location of his mother and/or father (36.2% of inaccurate addresses); 2) used an offender's PHA (i.e., primary home address) that did not correspond to the place he was residing at on the day of the crime, because he had either ceased to live there or not yet started to do so (18.6% of inaccurate addresses); and 3) recorded the address of a correctional institution, given that this was the place the offender was “living” at when they arrested him for this particular crime (17.6% of inaccurate addresses). Among all the inaccurate addresses, only 49.2% had already been a PHA for the offender and 25.1% could have been considered their last valid one (i.e., the last PHA an offender had prior to the current one, excluding incarceration terms and NFA periods). These wrong PHAs were out of date by 374.9 days on average (SD = 472.4), with 23.8% out of date by two years or more. How can the police’s significant difficulty correctly identifying where an offender was truly residing on the day of a given crime be explained? Our analyses have identified five noteworthy problems that can directly influence the accuracy of police data on offenders’ home addresses.
First of all, there is the question of how individual police officers from different forces understand, interpret, and codify primary home addresses (PHAs). The present study relied on the same standardized definition for all 449 crimes in the sample, namely, the place the offender generally sleeps, at the time of the crime, the equivalent of four nights or more during a given week. No interview was conducted with police officers to provide insights into how they precisely determine an offender's PHA, but our preliminary analysis of police reports suggests considerable variability in the way this information is recorded by them.
In some police reports, an offender's name on a rental lease, a residential address on a probation order, or a credible third party pretending to live with the offender was sufficient to settle his PHA, regardless of the number of days he really slept there in a given week. In others, additional evidence was needed to make sure the offender was really living there. In some situations, the recorded PHA was the offender’s “current home address” while in others, it was the offender’s “last known address”; most of the time, no such distinction was made in the police report. When the perpetrator was arrested for a given crime while incarcerated (10.9% of crimes), the offender's PHA was usually recorded as the correctional institution (67.3% of the time), but sometimes as his last known address (24.5% of the time), or as missing or unknown (8.2% of the time). Unknown addresses, missing addresses, and being NFA were often used interchangeably, but sometimes they had different meanings. In most police reports, not having a PHA (or being NFA) was limited to homeless people living on the streets, individuals sleeping in motels or hotels, or individuals temporarily sheltered in community resources. In most cases, the police’s definition of NFA excluded individuals who were house-hopping every few days between their relatives’ and/or friends' homes because they had nowhere else to live. However, our definition did include individuals living in this situation, which is often considered a form of hidden homelessness (see Wright et al. 1998; Rodrigue, 2016). The lack of clarity, consistency, and standardization regarding the definition and the coding of an offender's PHA in police data certainly jeopardized the accuracy of some offenders’ home address in our sample.
More detailed analyses (not shown here) have revealed that 23.6% of inaccurate offender home locations were erroneous at the time of crime, but accurate at the time of arrest, and that 7.0% of wrong addresses corresponded to an offender's future PHA. These results are an indication that the duration of the police investigation can have an impact on the addresses’ accuracy. In our sample, 23.6% of crimes for which the perpetrator was arrested at the crime location or minutes later while fleeing the crime scene were not subjected to that kind of police-investigation bias in the recording of the offender’s home address. In these situations, the perpetrator's place of living at the time of crime was obviously the same as at the time of arrest. However, in the 76.4% of crimes for which a police investigation was needed to identify, find, and/or arrest the perpetrator(s), that person's residential stability over that time period had to be assessed. Results are presented in Table 2.
[INSERT TABLE 2 ABOUT HERE]
The average duration of a police investigation for a theft-related crime ― from the offender's commission of the crime to his arrest ― was 54.9 days (SD = 63.1). Most investigations were generally concluded within days or weeks, but 22.4% had a duration exceeding three months. Due to a violation of the normality assumption, a Kruskal-Wallis non-parametric analysis of variance (ANOVA) was performed on the mean rank duration of the police investigation by crime type, and it revealed significant variation (χ2 [3] = 30.83, p < .001). Police investigations of burglaries (M = 80.9; SD = 73.4) were, on average, almost three times longer than of robberies (M = 29.8; SD = 38.8). As many as 109 criminal events (31.7%) included a residential change by the perpetrator during the police investigation (i.e., a change of PHA or residential status ― domiciliated, NFA, incarcerated ― during that time period). This proportion was comparable across the four crime types. In these situations, where the offender's PHA at the time of crime was different than at the time of arrest, the PHA at the time of arrest was accurately recorded in 46.4% of cases, but this fell to 19.6% for the PHA at the time of crime (in the rest, an inaccurate address was recorded for both times). These results not only suggest that a police-investigation bias exists in the police recording of offenders’ home addresses, but also that perpetrators tend to exhibit a high residential mobility that may complicate the identification of their current PHA by the police.
Offenders were asked to reconstruct their residential-history timeline from the beginning of the study period (January 1, 2005) to their last incarceration date prior to the interview. Depending on the participant, this interval varied from 1208 to 3099 days (M = 2381.7; SD = 355.9). With the support of official documents, each offender was asked to provide the total number of different PHAs they had had during that time span, along with all NFA episodes and prison stays. These numbers were subsequently summed to yield the total number of residential changes made during the study period. Offenders’ residential mobility was estimated by dividing the number of days in the study period by the total number of residential changes. Results are presented in Table 3, along with corresponding accuracy rates for offenders’ home addresses in police data.
[INSERT TABLE 3 ABOUT HERE]
In this sample, the average number of days before a residential change during the study period was 253.7 days (SD = 148.9), with 80.6% of offenders having switched their place of living ― voluntarily or involuntarily ― more frequently than once a year during that interval. On average, offenders had 7.0 different PHAs (SD = 3.5), 1.8 NFA episodes (SD = 2.3), and 3.3 prison stays (SD = 2.8) throughout the period. Half (50.0%) of the offenders declared having used their mother and/or father's home as their PHA for at least one month during the study period. Eighteen (36.7%) of them lived with their parents permanently for some time because they were minors at the beginning of the study period; most of the others had stayed there temporarily following an incarceration sentence, a breakup, and/or financial struggles. Our results suggest that the residential mobility exhibited by offenders affects the police’s ability to correctly identify their place of living. For example, the accuracy rate of offenders’ home addresses in police data was 29.4% for the crimes committed by the most nomadic offenders (i.e., residential change every 4 months or less, on average) and 95.8% for those perpetrated by the most sedentary ones (i.e., residential change every 2 years or more, on average) (χ2 [4] = 30.57, p < .001; Cramer's V = .278).
After having identified and/or arrested the suspect of a crime, the police use different techniques and sources of information, such as police and governmental databases, to locate their residential address. In most jurisdictions, given that the driver's license is a renewable privilege that is associated with an obligation to declare promptly any change of residence (generally under penalty of a fine), it is often considered one of the most up-to-date sources of information regarding an individual’s current home address. However, among the 98 offenders interviewed in this study, 28.6% had never had a driver's license, 43.9% had had a driver's license that was not valid at the time of last arrest, 12.2% had had a valid driver's license with an inaccurate PHA at the time of last arrest, and only 15.3% had had a valid driver's license with an accurate PHA at the time of last arrest. These results not only challenge the utility of the driver's license in corroborating the current home address of an active offender, but also emphasize the importance of verifying the presumably accurate information coming from official sources.
Police regularly questioned suspects to corroborate their place of residence. Almost all (98.0%) offenders in our sample declared having already been questioned about where they were living by the police. Depending on the importance of this information to the overall investigation, this could take many forms, varying from a basic yes or no validation question at the time of arrest (e.g., “Do you still live at this address [e.g., written on their driver's license]?”) to a central theme questioned deeply during a formal police interview. Being the most directly concerned, it is legitimate to consider offenders as a potential source of information regarding where they are living. But to what extent they should they be considered a trustworthy source in that matter?
When questioned about the importance of keeping their true PHA confidential and unknown to the police, most offenders in this study (54.0%) answered that the importance was “crucial” or “high,” and only 25.3% declared that it was “not important at all.” The accuracy rate of offenders’ home addresses in police data was 38.9% for the crimes committed by offenders who declared PHA secrecy to be of “crucial” importance and 63.7% for crimes perpetrated by those who declared that PHA secrecy is “not important at all” (χ2 [4] = 15.69, p < .01; Cramer's V = .214). The majority (61.1%) of offenders admitted to having lied to the police regarding their true home address, with almost half (48.9%) confessing to having been deceitful “always” or “often.” The interviews with offenders revealed four different tactics they used to mislead the police regarding their true home address.
Twenty-six (26.5%) offenders confessed to regularly giving officials (e.g., police, government, court) the residential home address of their mother and/or father even if they were not really residing there. Some explained that they exhibit so much residential instability or are so frequently incarcerated that they prefer to use their parents' home as a fixed and reliable address to receive their important mail, such as social assistance cheques and court orders, rather than risking losing track of it and/or sacrificing time constantly updating their addresses. Others believed that by living (or pretending to live) with a parent, they improved their credibility in the eyes of the authorities when asking for parole or with regard to their ability to respect their conditions during a granted bail or probation. Others explicitly stated that falsely indicating their residence at a parent's home address was a strategy to continue committing crimes with the feeling of being “one step ahead.” As participant #85 illustrated: “Because I always use my mother's home address in all my paperwork, when the police are looking for me, they are coming to her house... and this is a red flag... it means that they suspect me of having done something wrong, so I have to start watching my back, they are on my tail... maybe it is time for me to disappear for some time.” Having a parent to cover up for you if questioned by the police and some personal belongings (clothing, mail) to show as evidence certainly improves the effectiveness of this technique (“My parents have kept my room in the basement ‘as is’ even if I have not lived there for 7 years, so they [i.e., the police] could easily think I currently live there”; participant #67). The fact that more than one quarter of offenders stated that they regularly used this strategy may explain, at least in part, why the most common cause of inaccurate offender home addresses in the sample was the mistaken beliefs of the police that the offender was living at his mother and/or father's home at the time of crime (i.e., 36.2% of inaccurate addresses).
A second method, which nine (9.2%) offenders confessed to frequently employing to protect the secrecy of their true home location, was to falsely claim that they were homeless or NFA. But as noted by participant #9: “You have more chances to be believed [by the police] if they think you are a drug addict and if you commit your crimes in a big city... telling them you are homeless in a small village is not really credible and you may look [even more] suspicious.” Among the 53 crimes for which the offender's home address was missing or undetermined in police data, only 32.1% had in fact been perpetrated by individuals who admitted during the interview to having been NFA on the day of the crime; the rest (67.9%) were committed by offenders who did had a PHA, but one which was, manifestly, unknown to the police.
A third tactic, which eight (8.2%) offenders confessed to having relied on during the study period, is an agreement with a third party, such as a friend or a family member (other than their mother/father), who would pretend that the offender was living with them in exchange for money, drugs, and/or other services (e.g., protection, sex). For example, participant #150 gave $50 per month to a friend to cover for him in case of questioning and to be able to receive his social assistance cheque at his residence (because he was not living there). Another example is participant #124 who was paying a “clean” (i.e., with no criminal record) friend to use his name to rent an apartment and have access to public services (electricity, cable). Among the 23 crimes for which the police inaccurately recorded the home address of a relative, friend, or partner (or ex-partner) as the PHA of the offender (i.e., 11.6% of inaccurate addresses), 8 involved this kind of agreement.
The last strategy, employed by seven (7.1%) of the interviewed offenders, was to have an alternative address they visited regularly to make the people think they were living there, such as a drug house, a stash house, or a fake apartment: “When you are in the drug business, you need to separate the place where you deal your dope and the place where you live... you don't want to have junkies coming anytime for their fix at your real home... Also, if I get busted there by the police or other bandits, at least my family will not be traumatized, and my home will not be damaged” (participant #104). Another offender added: “My [drug dealer] office is a small apartment I rent with no lease and pay for in cash each month... there is some basic furniture, a couch, a television, a small fridge, a mattress on the floor... no stove, no dishes, no decoration... I do not live there, I work there” (participant #72). Inaccuracies in offenders’ home addresses due to the police mistakenly recording that the offender was residing at such an alternative address were present in 6.0% of the crimes analyzed.
Offenders' motivations to keep the police away from their real home location were diverse: preventing the finding of new evidence linking them to unsolved crimes, avoiding being easily arrested at home as soon as they became suspected of another crime, keeping their relatives unaware of their criminal involvement, avoiding attracting the attention of neighbors and/or other criminals. Some offenders even told us that a “burned” (i.e., known to the police) home address was their main reason to move to another place. Offenders exhibiting a propensity to lie about their PHA pointed out the ease with which they could generally provide the police with a misleading home address (“The police's job is to bust criminals, not to check for their home addresses... if they already have everything they need to put you in prison, which is by the way often the case when they arrest you, they won't really care about where you [really] live”; participant #99), the difficulty of proving that someone is truly living or not living at a given location (“When I told the police I live somewhere and they don't believe me, I'm saying to myself, well, prove it!”; participant #163), and the few dissuasive consequences of being caught lying about it (“The worst thing that could happen if I get caught lying about that [i.e., offender's home address] is maybe a new criminal charge? I don't even know if the police could do that? Anyway, I don't care... With all the criminal stuff I have at home [e.g., fake gun, illegal drugs, dirty money, stolen goods], do you think I want them to come and have a look?”; participant #60). Hence, to what extent do the police diligently verify the validity and accuracy of the addresses coming from these different sources?
Police validated the offender's home address “on-site” in only 32.7% of the 449 crimes in the sample. This includes all the situations in which the police followed the suspect or surveilled his whereabouts thoroughly enough to presume his home location (15.5% of crimes), arrested him at a place they considered to be his residence (12.6% of crimes), and/or conducted a house-search warrant at this location (18.6% of crimes).9 Police on-site verification of offenders' PHA during the investigation was significantly more prevalent in robberies (38.9%) than in motor-vehicle thefts (9.5%), but comparable in burglaries (27.8%) and “other thefts” (22.1%) (χ2 [3] = 12.78, p < .01; Cramer's V = .169).
These results suggest that in most theft-related crimes, the police did not confirm in situ the offender's PHA, either because: a) they did not know where the offender was residing, generally after having verified he was no longer living at his last known address; b) the offender was in prison at the time of identification and/or arrest; or c) they did not consider the offender's home to contain relevant new evidence for their investigation, so they took for granted the accuracy of the offender’s last known address (e.g., from a police database, a driver's license) or were satisfied with a less rigorous verification approach (e.g., cross-referencing this address with other sources of information, such as the offender). Even when the offender's home address was physically validated by the police, the corresponding location was not a guarantee of accuracy, with 19.1% of addresses still referring to an inaccurate offender PHA at the time of crime.10 According to several interviewed offenders, the police’s willingness to locate an offender’s home address is highly related to the gravity of the crime in question: “If the police really want to find your home address, they will find it... This is all about the importance you have in their eyes... If they suspected you of robbing banks with a loaded gun, trust me, they will put the resources and find where you live... but for minor crimes [or low-profile criminals], forget it, they don't have the time and the money for that” (participant #10).
Another implicit assumption in journey-to-crime research is that home-to-crime corresponds to the itinerary taken by offenders in most crimes. This presupposes that: a) the offender's home is the most important node in a crime journey; b) the crime journey is direct, without any detours and/or stopovers; and c) the outbound crime journey (i.e., journey-to-crime) largely corresponds to the inbound crime journey (i.e., journey-after-crime).
To what extent is the offender's home an important place in a crime journey? Descriptive statistics summarizing the places where offenders have woken up on the day of the crime, took the decision to commit the crime, and sought refuge after the crime are presented in Table 4. Unsurprisingly, home was the offender's awakening location on the day of most crimes (75.2%). The remainder were crimes perpetrated by people considered NFA (14.4%) or by domiciled offenders who slept elsewhere the night before the crime (10.4%). The proportion of perpetrators starting their day at home was comparable across crime types.
[INSERT TABLE 4 ABOUT HERE]
The offender's home was the decision location in less than half (49.0%) of the theft-related crimes in our sample and this proportion was also similar across crime types. Other important places at which offenders decided to start their crime journey included a friend’s or a family member's residence (14.9% of crimes), a bar or a drug spot (10.0% of crimes), or a prior crime location (6.0% of crimes). Purely opportunistic crimes (i.e., the decision to commit the crime is taken at the crime location), which represent 11.1% of all the theft-related crimes in this study, were significantly more prevalent in “other theft” than in burglary (20.6% vs. 8.6%; χ2 [3] = 11.87, p < .01; Cramer's V = .163). In these haphazard crime situations (n = 50), the offender's initial purpose for traveling was to visit a friend or a relative (24.0%), go shopping (18.0%), buy drugs and/or party in bars (16.0%), head back home (14.0%), or go to work (10.0%). The remainder (18.0%) were typical situations of “manufactured serendipity” (Jacobs, 2010) in which the offender had created a facilitating environment for criminal opportunities (e.g., by wandering the streets, lingering in criminogenic areas, taking a walk with no particular motivation).
Home was the safe-house location for 40.2% of the crimes in our sample. Around one quarter of crime journeys ended abruptly before the offender was able to reach a safe house, either because they were caught red-handed at the crime location (6.7% of crimes) or because they were arrested by the police mid-flight (16.9% of crimes). Such situations were more frequent in “other theft” than in robbery (37.3% vs. 16.1%; χ2 [3] = 12.16, p < .01; Cramer's V = .165). When the offender succeeded in reaching a safe house, home was the place of choice for all crime types except robbery, for which a friend or a family member's residence was slightly preferred. The offender's home was neither a decision location nor a safe-house location in as much as 40.6% of the crimes in our sample.
To what extent is the crime journey a direct trip? In our sample, only 32.3% of the crime journeys could be considered to have been realized without any detour or stopover from the decision location, to the crime location, to the safe-house location (or to the arrest location, if the offender was intercepted by the police before reaching a safe house). This proportion was comparable across crime types. That means that in most crime journeys, offenders deviated from their linear path to and/or from the crime in order to accomplish another task, such as the search for a criminal target, the perpetration of another crime, or any other actions they deemed necessary. Descriptive statistics about the nature of the detours and stopovers made by offenders during their crime itineraries are presented in Table 5.
[INSERT TABLE 5 ABOUT HERE]
The search for a target was a fundamental preliminary step in 36.3% of the theft-related crimes in our sample. Criminal foraging to find the right target was much more prevalent in burglary (53.0%) than in robbery (24.7%) or “other theft” (14.7%) (χ2 [3] = 47.15, p < .001; Cramer's V = .324). However, searching time and its translation into effective distance traveled were comparable across crime types. On average, offenders spent 27.9 minutes (SD = 27.4) or traveled 8.7 km (SD = 10.2) searching for a target. Committing another crime immediately before and/or after the index crime was also an important offender stopover found in 25.6% of the crime journeys. In such situations, the number of other crimes perpetrated ranged from 1 to 6 (M = 1.5; SD = 1.0), with 51.4% of them having also required a distinct search for a target. Other types of intermediary stops were also made by offenders in 40.5% of the crime journeys, in which case the number of stopovers varied from 1 to 6 (M = 1.5; SD = 0.9).
Finally, to what extent does the outbound crime journey (i.e., journey-to-crime) mirror the inbound crime journey (i.e., journey-after-crime)? When exclusively considering crime itineraries in which the offender was able to reach a safe house without being arrested by the police (n = 344), 58.0% started and ended at the same location (i.e., the decision and the safe-house locations were the same), but only 17.2% had the exact same itinerary in the pre-crime and post-crime phases (after taking into account detours and stopovers).
Our results highlight several differences in the itineraries undertaken by offenders through the pre-crime and the post-crime phases. During the journey-to-crime, an offender could have diverged from their route, in order to search for a target (36.7% of crimes11), to commit another crime (13.8% of crimes), to meet with co-offenders (e.g., make a plan, carpool to crime) (6.9% of crimes), to gear up for the crime (e.g., buy and/or organize equipment, find transportation) (6.2% of crimes), to scout out the crime surroundings (2.4% of crimes), and/or for other reasons (4.5% of crimes). During the journey-after-crime, an offender could get arrested by the police (23.6% of crimes) or make a stopover before arriving at the safe house, in order to search for another target (6.7% of crimes), commit another crime (14.0% of crimes), switch their mode of transportation (e.g., change or abandon the getaway car or stolen vehicle, call or wait for a lift) (9.8% of crimes), sell or stash the stolen goods (e.g., professional fence, pawnshops) (8.9% of crimes), buy drugs and/or get intoxicated (e.g., bar, drug house, pusher) (7.6% of crimes), and/or diverge from their route for other reasons (10.0% of crimes).
This brings us to our final point, which is to assess the extent to which the traditional measure of journey-to-crime is a valid proxy of: a) the offender's validated home-crime distance; b) the distance traveled by an offender in the pre-crime phase; c) the distance traveled by an offender in the post-crime phase; and d) the total distance traveled by an offender over the entire crime journey.
What is the percentage of crimes in which the offender has in fact taken the direct home-crime and/or crime-home itinerary as assumed by the traditional measure of journey-to-crime? Results by crime type and crime phase are presented in Table 6. First, only 21.8% of all crimes were perpetrated by an offender whose pre-crime itinerary was a direct trip from home (i.e., without any detours or stopovers). This percentage was significantly lower for burglary (15.7%) and robbery (21.0%) than for “other theft” (39.7%) (χ2 [3] = 17.78, p < .001; Cramer's V = .199). Second, only 19.6% of all crimes were perpetrated by an offender whose post-crime itinerary was a direct trip toward home (i.e., without any detours or stopovers). This time, the percentage was significantly higher for burglary (25.3%) than for “other theft” (7.4%) (χ2 [3] = 11.86, p < .01; Cramer's V = .163). Finally, only 7.3% of all crimes were committed by an offender whose entire crime itinerary was a direct round trip (i.e., without any detours or stopovers) from his residence. The proportion was significantly higher for robbery (9.9%) than for “other theft” (0.0%) (χ2 [3] = 9.02, p < .05; Cramer's V = .142). If one accepts that the offender's home and crime-location addresses recorded in police data were accurate (as assumed by the traditional measure of journey-to-crime), as few as 3.9% of all crimes were committed by an offender whose entire crime journey matched the home-crime-home direct itinerary.
[INSERT TABLE 6 ABOUT HERE]
What is the impact of the mostly misleading itineraries presumed by the traditional measure of journey-to-crime on estimates of the distances really traveled by offenders? Results of a Kruskal-Wallis non-parametric analyses of variance (ANOVA) on the mean ranks by crime type for five different measures of distance are presented in Table 7. Home-crime distance according to police data was significantly longer for “other theft” (M = 31.0; SD = 27.9) than for robbery (M = 16.9; SD = 20.1) and burglary (M = 18.7; SD = 22.3) (χ2 [3] = 10.10, p < .05). No matter the crime type, the home-crime distance according to police data tended to overestimate the offender's validated home-crime distance. No significant differences were found across the four crime types regarding these home-crime distances. Generally, the traditional measure of journey-to-crime also tended to overestimate the distances actually traveled by offenders during the pre-crime and post-crime phases, no matter the crime type. On average, the total distance traveled by offenders during the entire crime journey was 31.1 km (SD = 36.4) for robbery, 33.0 km (SD = 39.7) for burglary, 46.8 km (SD = 50.7) for motor-vehicle theft, and 47.2 km (SD = 49.1) for “other theft”. The mode of transportation used by the offender during their crime journey was usually a motor vehicle that was not stolen (61.7% of crimes), followed by walking (38.8% of crimes), a stolen motor vehicle (10.5% of crimes), a taxi (7.1% of crimes), a bicycle (5.3% of crimes), and/or public transportation (3.1% of crimes). In each of the four categories of theft-related crimes, all five measures of distance followed a distance-decay function at the aggregate level, displaying a disproportionate number of short trips as compared to longer trips.
[INSERT TABLE 7 ABOUT HERE]
Finally, to what extent is the traditional measure of journey-to-crime correlated with the offender's validated home-crime distance, and with the distance actually traveled by offenders during the pre-crime phase, post-crime phase, and entire crime journey? Given the nested structure of our data (crimes nested in individuals), multilevel correlations ― also called hierarchical or random-effect correlations ― were performed with R (version 4.1.2) using the “correlations” package from the “easystats suite” (Makowski et al., 2020). This analytical strategy allowed us to account for differences between groups (i.e., each offender) by partializing the group variable entered as a random factor in a mixed linear regression. Results are presented in Table 8.
[INSERT TABLE 8 ABOUT HERE]
In factor analysis, the acceptable level of explained variance for a valid construct is 60.0% (Hair et al., 2010). Below that threshold, the analysis is considered to leave too much common variance unexplained, thus limiting its explanatory capabilities and ultimately questioning its usefulness. We draw upon this literature to establish that to be regarded as a valid construct of criminal mobility, the multilevel correlation coefficient (r) between the traditional measure of journey-to-crime and the different measures of distance should be at least .774 (i.e., 60.0% of explained variance). Results showed that the home-crime distance according to police data and the offender's validated home-crime distance are highly correlated for robbery (r = .95, p < .001) and “other theft” (r = .87, p < .001), but not for burglary (r = .10, n.s.) or motor-vehicle theft (r = .37, n.s.). The construct validity of the traditional measure of journey-to-crime varied drastically across crime types. For robbery, the measure could be considered a valid proxy of the distance traveled by offenders during the pre-crime phase (r = .84, p < .001) and the total distance they traveled during the entire crime journey (r = .82, p < .001). Similar results were observed for “other theft”, with the difference that the home-crime measure was a good proxy of the distance traveled by offenders during the post-crime phase (r = .81, p < .001), but not of the distance they traveled during the pre-crime phase (r = .61, p < .001). However, the traditional measure of journey-to-crime was an extremely tenuous indicator of the distances traveled by offenders for burglary (i.e., with r coefficients ranging from .03 to .04, n.s.) and motor-vehicle theft (i.e., with r coefficients ranging from .06 to .33, n.s.). The offender's validated home-crime distance was a better proxy of the distances traveled by the offender (pre-crime, post-crime, entire crime journey) for all crime types.
This research has addressed an issue that has never been studied before: the extent to which the traditional measure of journey-to-crime is a valid estimation of the trip actually undertaken by offenders during their crime journey. With the support of web-mapping technologies, past crime journeys were reconstructed with theft-related offenders to test some of the key assumptions of journey-to-crime research. As hypothesized, our findings have demonstrated that police data provided researchers with satisfactorily accurate crime location addresses. But contrary to what was hypothesized, such data was characterized by poorly accurate offender home-address information. Police capability to correctly identify where an offender was truly residing on the day of a given crime was limited by issues concerning the definition and standardization of PHA across police forces, police-investigation bias, offenders' high residential mobility, low reliability of police sources about offenders' home addresses, and insufficient police on-site validation of PHAs. Even though offenders’ residences were absent from up to 40% of crime journeys, our results have shown that the offender's home could still be considered the most important node in a crime journey, as hypothesized. However, contrary to what was hypothesized, most crime journeys were not a direct endeavor (i.e., included at least one detour and/or stopover) and the outbound itinerary (i.e., journey-to-crime) rarely mirrored its inbound counterpart (i.e., journey-after-crime).
Our last hypothesis has to be discussed by crime type and crime phase. As hypothesized, the traditional measure of journey-to-crime could be considered a valid indicator of the offender's validated home-crime distance, but only for robbery and “other theft”. This can be explained by the fact that within these crime types, offenders’ home addresses in police data were slightly more accurate, and the fewer misleading addresses were situated closer to their real location than for burglary or motor-vehicle theft. Even if a direct round trip from home was the exact depiction of the criminal itinerary undertaken by offenders in a substantial minority of crimes, the traditional measure of journey-to-crime could still be considered a sufficiently valid indicator of the distance traveled for robbery (pre-crime phase and entire crime journey only) and “other theft” (post-crime phase and entire crime journey only), as postulated. This suggests that within these crime types, the decision and safe-house locations tended to be situated at or near the offender's home address recorded in police data, and detours and stopovers were likely to be clustered around the home-crime trajectory (i.e., a rather “linear” crime journey).
However, contrary to the hypothesis, the traditional measure of journey-to-crime was not a valid proxy of the distances traveled by offenders for burglary and motor-vehicle theft. In burglary, this could be explained, at least in part, by the substantial difference in distance between the real offender's home address and the inaccurate location in police data, the fact that these mostly misleading addresses were part of the offenders’ itinerary in more than half of the crime journeys, and the fact that a large number of such crimes required a search for a target prior to perpetration. In motor-vehicle theft, our interpretation of why the traditional measure of journey-to-crime is not a valid proxy of the distances traveled by offenders is limited by the small sample size (i.e., 21 crimes by 12 offenders). Still, the relatively low accuracy rate of addresses in police data (both for offenders’ home addresses and crime locations) and the proportion of drastically different post-crime trips (as compared to pre-crime trips) probably is part of the explanation.
Our findings have numerous implications for criminal mobility research, environmental criminology, and criminological theory. On the one hand, despite its numerous drawbacks, the traditional journey-to-crime measure could still be useful to estimate the total distance really traveled by offenders committing robbery (as it is capable of explaining 67.2% of common variance) and “other theft” (as it is capable of explaining 74.0% of common variance), and therefore, should continue to be used in such research, hopefully with more nuance regarding its validity and prudence concerning its interpretation. Scholars would possibly want, however, to improve the accuracy of their estimation by relying on sources of information other than police data concerning offenders’ home and crime-location addresses.
On the other hand, our findings show that the traditional journey-to-crime measure is of little value in estimating the total distance really traveled by burglars (explaining less than 1% of common variance) and motor-vehicle thieves (explaining 2.9% of common variance) during their crime journey. Even when using only accurate offender home and crime-location addresses (i.e., offender's validated home-crime distance), the respective percentages of common variance explained, despite being definitely much better, were still below our acceptable construct validity threshold of 60.0% (except for the distance traveled in the pre-crime phase of motor-vehicle theft). Therefore, we recommend that researchers should avoid relying on the journey-to-crime measure to estimate the distance traveled by offenders who commit burglary and motor-vehicle theft. Caution should also be exercised in interpreting the results of past empirical studies which investigated home-crime distances in such crimes, as our findings suggest it gives no indication of the actual distance traveled by offenders. Moreover, we found that the traditional journey-to-crime indicator was not even capable of accurately estimating the true distance between a burglar's or motor-vehicle thief's home and their crime location. Burglary being one of the most studied crimes in journey-to-crime research, these results are particularly alarming and raise questions about the extent to which our collective understanding of criminal mobility is well founded.
Furthermore, the home-crime distances (computed from police data) found in this study were much longer than those presented in other research investigating offender mobility in theft-related crimes (e.g., Ackerman & Rossmo, 2015). One reason could be that the crimes considered here were not geographically bounded to densely populated cities (like most past research), but were, rather, dispersed over a large territory (i.e., the province of Quebec12) that comprises many rural areas. Past research has shown that rural offenders tend to travel more than urban offenders (Chainey, 2021; Townsley et al., 2015). Another reason could be that we have used street-network distances rather than the invariably shorter Euclidean (i.e., as-the-crow-flies) distances used in most past research (according to Ackerman & Rossmo, 2015, street-network distances are from 18% to 39% longer). A third reason could be that using home-crime distance produces noticeably larger figures than using the distance between the awakening and the offense locations, as suggested by the interview data presented by Wiles and Costello (2000).
Our analyses have enabled us to document a new restrictive deterrence strategy employed by persistent offenders to evade detection, identification or apprehension (Gibbs, 1975). This tactic involves deliberate or indirect actions intended to mislead the police about the real location of a primary home address. It could take various forms, such as lying about a home address in official documents, wittingly “forgetting” to update housing information after moving, falsely claiming to live with a conniving third party, or falsely claiming to be homeless. It should be recalled that almost half of the offenders in our sample confessed to having had a propensity to lie or to be deceitful when asked about where they lived by the police. Offenders may provide the wrong home address after a given crime (e.g., during their police interrogation) to prevent the police from finding new criminal evidence that could link them to other crimes (e.g., possession of guns, drugs, stolen goods, etc.) or to protect the location of their household if they plan to return there after their sentence and continue committing crimes (thus evading further detection, identification or apprehension).
The reconstruction of crime journeys with offenders has also allowed us to differentiate strategically driven crimes from opportunistically triggered ones (see Table 9 for a summary by crime type). In our sample, 88.9% of the crimes were committed by crime-led offenders (i.e., those who were driven to commit a crime when embarking on their crime journey) and only 11.1% were perpetrated by opportunistic offenders (i.e., those who were in movement for non-criminal reasons until they were confronted with a high-value desirable criminal opportunity). Therefore, it seems that offenders predominantly travel to commit crime, and rather seldomly commit crime merely while in travel. For theft-related offenses, while daily-routine activities undoubtedly contribute to offenders’ awareness space and knowledge of proximate criminal opportunities (Brantingham & Brantingham, 1981; Cohen & Felson, 1979), the process of committing a crime seems to be overwhelmingly rational and purposive (Cornish & Clarke, 1986). This also suggests that criminal mobility is probably more a crime-specific than an offender-specific phenomenon. Offenders are not strategic or opportunistic by nature. Instead, each of them perpetrates, to different degrees, a mix of strategic and opportunistic crimes. From a criminal-mobility perspective, studying “offenders-in-crime” rather than crimes or offenders separately could allow researchers to make this distinction ― of great theoretical value in criminology ― between strategically driven and opportunistically triggered crimes.
[INSERT TABLE 9 ABOUT HERE]
Finally, the interview data used in this research allowed us to discover that crime journeys were not entirely independent, thus suggesting that there is another level of nesting in the study of crimes. Indeed, we determined that our 98 interviewed offenders (first level of nesting) undertook 334 crime journeys (second level of nesting) which led to the perpetration of a total of 449 official crimes (third level of nesting). In our sample, around one quarter of the crimes (involving 41.8% of the offenders) were characterized by a crime journey involving the commission of more than one crime. Most of the time, these multiple-crime journeys were undertaken by offenders in response to a failed or disappointing initial crime or because they had planned to offend during a given period of time and/or until they reached a certain amount of reward (e.g., goods, money). However, this proportion is likely to be underestimated, given that our interviewed offenders were told not to give information about unofficial crimes while reconstructing their crime journey. At the same time, our sampling procedure probably contributed to an overestimation of the clustering of official crimes within individual offenders, so the impact of this new intermediate level of nesting remains to be more accurately estimated. This finding is particularly relevant for researchers studying near-repeat victimization in burglary or motor-vehicle theft, which are crimes that do not always have a direct witness to confirm the exact moment of the crime. In these situations, because the date of the crime is recorded as an interval (from this date/time to this date/time), researchers generally used the “committed from” date/time (Chainey, 2021). Without an accurate date/time, how do we ascertain that these proximate targets have not been victimized during the same crime journey by the same offender?
This study has some limitations worth mentioning. To begin with, our sample tends to overrepresent the most prolific offenders (e.g., federally sentenced, high recidivism rates) and their well-recalled official crimes (e.g., recent, lucrative, having led to a quick arrest). Although our overall sample size was fair (98 offenders, 449 crimes), its subdivision into crime types has limited the number of observations, particularly for motor-vehicle theft (12 offenders, 21 crimes), thus increasing the margin of error and limiting the statistical power of our analysis. Even though we regularly refer to the “offender's validated” or “actual” crime journey throughout the article, we do not presume that we were able to unravel the real itinerary taken by each offender in each crime, because some features of the crime journey inevitably had to be surmised. For example, the distance traveled during the search for a criminal target was estimated by presuming that the offender's time estimation was accurate, and that the speed of the journey was constant and faithful to the speed of the mode of transportation established in Google Maps. Also, we assumed that offenders would always take the fastest route available between two points, without traffic or any other unforeseeable circumstances. Finally, given that most offenders could not be considered a trustworthy source by the police regarding their home address (see section 4.1.4), how could we expect them to be a reliable source for the researcher willing to reconstruct their entire crime journey? Because we had limited ways of corroborating offenders’ self-reported itineraries, we relied on other strategies to minimize deception and maximize the accuracy of the information provided. For example, we stuck to crimes the offenders remembered acutely and acknowledged having committed, and for which they had already been convicted. We also told them to avoid giving details about sensitive locations (e.g., unofficial crimes, an active drug house, the residence of an unarrested co-offender), used Google Maps and Google Street View to facilitate recall and improve the precision of addresses, asked validation questions during the interview (e.g., the estimated travel time in the pre-crime phase and in the post-crime phase), and invited the offenders to provide explanations when discrepancies were found (e.g., when the travel time given by the Google Maps Route Planner function was considerably different than their own estimation).
In addition to estimating the validity of the traditional journey-to-crime measure, this study has provided researchers with an innovative and promising ― albeit work-intensive ― research procedure to finally allow the analysis of something closer to the real travel undertaken by offenders. Web-mapping technologies such as Google Maps offer great advantages over the typical data-gathering instruments used by environmental criminologists such as hand-drawn sketch maps and conventional cartographic maps (for a review, see Vandeviver, 2014), and we believe they should be used to a greater extent in criminological research. This study should also be replicated with other samples, in other countries, and with other types of crimes. In particular, the validity of the traditional measure of journey-to-crime remains to be assessed in crimes of interest in geographic offender-profiling, such as murder and sexual assault. While the police will probably want to use more investigative resources to solve such crimes, thus potentially contributing to more accurate addresses (as compared to theft-related crimes), the seriality (implying a notion of time), delayed reporting (e.g., mostly in sex crimes), and/or extended investigations that may characterize some of these crimes could possibly increase the risk of a police-investigation bias regarding the offender's home address (i.e., police data’s tendency to record the address at the time of arrest rather than at the time of crime). Also, journey-to-crime's failure to take into consideration the multiple crime sites typically observed in such crimes (e.g., encounter location, attack location, crime location, victim/body release location; see Rossmo, 2000) may distort the resulting estimation of the overall distance traveled.
To conclude, offenders' willingness to protect the anonymity of their household, intermingled with the police’s inability, reluctance or indifference to conduct on-site validation of offenders' home addresses, calls into question the value of modeling offender mobility solely on the basis of police arrest records. Police data are rarely collected for research purposes, are generally not recorded using a standardized research procedure, and involve “subjects” (i.e., offenders) that may have ― as we found in this study ― strong incentives to be misleading and/or to distort the data. Our findings also pointed toward a less-domocentric and more behaviorally informed approach to the detection and monitoring of individual offenders. Rather than prioritizing crime suspects only on the basis of their presumed home address, the police could benefit from taking also into account where they tend to travel or wander. In the same vein, geographic offender-profiling could become more efficient by determining the suspect’s “socializing area” rather than their most probable area of residence. Finally, GPS-tracking devices should be used to a greater extent, to document the whereabouts of an offender and/or to make sure they truly live at a given home address.
This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.
All authors certify that they have no affiliations with or involvement in any organization or entity with any interest in the subject matter or materials discussed in this manuscript.
Ackerman, J., and Rossmo, D. K. (2015). How far to travel? A multilevel analysis of the residence-to-crime distance. Journal of Quantitative Criminology, 31(2), 237–262. https://doi.org/10.1007/s10940-014-9232-7
Beauregard, E., Proulx, J., and Rossmo, D. K. (2005). Spatial patterns of sex offenders: Theoretical, empirical and practical issues. Aggression and Violent Behavior, 10(5), 579–603. https://doi.org/10.1016/j.avb.2004.12.003
Bernasco, W. (2014). Crime journeys: Patterns of offender mobility. In M. Tonry (Ed.), Oxford Handbooks Online in Criminology and Criminal Justice (pp. 1–19). Oxford University Press. https://doi.org/10.1093/oxfordhb/9780199935383.013.49
Bernasco, W. (2017). Modeling offender decision making with secondary data. In W. Bernasco, J.-L. van Gelder, and H. Elffers (Eds.), The Oxford handbook of offender decision making (pp. 569–586). Oxford University Press. https://doi.org/10.1093/oxfordhb/9780199338801.013.28
Brantingham, P. L. and Brantingham, P. J. (1981). Notes on the geometry of crime. In P. J. Brantingham and P. L. Brantingham (Eds.), Environmental criminology (pp. 27–54). Sage Publications.
Burgess, N., Maguire, E. A., & O'Keefe, J. (2002). The human hippocampus and spatial and episodic memory. Neuron, 35(4), 625–641. https://doi.org/10.1016/S0896-6273(02)00830-9
Canter, D. V., & Gregory, A. (1994). Identifying the residential location of rapists. Journal of the Forensic Science Society, 34, 169–175. https://doi.org/10.1016/S0015-7368(94)72910-8
Canter, D., & Larkin, P. (1993). The environmental range of serial rapists. Journal of Environmental Psychology, 13(1), 63–69. https://doi.org/10.1016/S0272-4944(05)80215-4
Chainey, S. P. (2021). A comparison of burglary near repeat victimization between rural and urban areas using a target-based assessment of criminal opportunity. International Criminal Justice Review, 31(4), 405–419. https://doi.org/10.1177/10575677211041920
Chainey, S., Austin, B., & Holland, F. (2001). London borough of harrow: Insights into the analysis and patterns of offender journeys to crime. InfoTech Enterprises Europe.
Cohen, L. E., and Felson, M. (1979). Social change and crime rate trends: A routine activity approach. American Sociological Review, 44(4), 588–608. https://doi.org/10.2307/2094589
Cornish, D. B., and Clarke, R. V. (1986). Introduction. In D. B. Cornish and R. V. Clarke (Eds.), The reasoning criminal: Rational choice perspectives on offending (pp. 1–16). Springer-Verlag.
Fleisher, M. S. (1995). Beggars and thieves: Lives of urban street criminals. University of Wisconsin Press.
Gibbs, J. P. (1975). Crime, punishment, and deterrence. Elsevier.
Hair, J. F., Black, W. C., Babin, B. J., and Anderson, R. E. (2010). Multivariate data analysis: A global perspective (7th ed.). Pearson Education.
Hinkle, D. E., Wiersma, W., and Jurs, S. G. (1988). Applied statistics for the behavioral sciences (2nd ed.). Houghton Mifflin Company.
Jacobs, B. A. (2010). Serendipity in robbery target selection. British Journal of Criminology, 50(3), 514–529. https://doi.org/10.1093/bjc/azq002
Levine, N., and Lee, P. (2013). Journey-to-crime by gender and age group in Manchester, England. In M. Leitner (Ed.), Crime modeling and mapping using geospatial technologies (Vol. 8, pp. 145–178). Springer.
Makowski, D., Ben-Shachar, M. S., Patil, I., and Lüdecke, D. (2020). Methods and algorithms for correlation analysis in R. Journal of Open Source Software, 5(51), 2306. https://doi.org/10.21105/joss.02306
Nichols, W. W. Jr. (1980). Mental maps, social characteristics, and criminal mobility. In D. E. Georges-Abeyie & K. D. Harries (Eds.), Crime: A spatial perspective (pp. 156–166). Columbia University Press.
Pettiway, L. E. (1995). Copping crack: The travel behavior of crack users. Justice Quarterly, 12(3), 499–524. https://doi.org/10.1080/07418829500096111
Rengert, G. F. (1975). Some effects of being female on criminal spatial behavior. The Pennsylvania Geographer, 13(3), 10–18.
Rengert, G. F. (1996). The geography of illegal drugs. Westview Press.
Rodrigue, S. (2016). Hidden homelessness in Canada. Statistics Canada catalogue no. 75-006-X. Statistics Canada.
Rossmo, D. K. (2000). Geographic profiling. CRC Press.
Rossmo, D. K., Lu, Y., and Fang, T. B. (2012). Spatial-temporal crime paths. In M. A. Andresen and B. J. Kinney (Eds.), Patterns, prevention, and geometry of crime (pp. 16–42). Routledge.
Snook, B., Cullen, R. M., Mokros, A., and Harbort, S. (2005). Serial murderers’ spatial decisions: Factors that influence crime location choice. Journal of Investigative Psychology and Offender Profiling, 2(3), 147–164. https://doi.org/10.1002/jip.35
Statistics Canada (2006). Overview of the time use of Canadians 2005. Statistics Canada catalogue no. 12F0080-XIE. Author.
Statistics Canada (2016). Dictionary of the 2016 census of population: Population center (POPCTR). Statistics Canada catalogue no. 98-301-X. Author.
Strauss, M. E., and Smith, G. T. (2009). Construct validity: Advances in theory and methodology. Annual Review of Clinical Psychology, 5, 1–25. https://doi.org/10.1146/annurev.clinpsy.032408.153639
Townsley, M., Birks, D., Bernasco, W., Ruiter, S., Johnson, S. D., White, G., and Baum, S. (2015). Burglar target selection: A cross-national comparison. Journal of Research in Crime and Delinquency, 52(1), 3–31. https://doi.org/10.1177/0022427814541447
Vandeviver, C. (2014). Applying Google Maps and Google Street View in criminological research. Crime Science, 3(1), 1–16. https://doi.org/10.1186/s40163-014-0013-2
Warren, J., Reboussin, R., Hazelwood, R. R., Cummings, A., Gibbs, N., & Trumbetta, S. (1998). Crime scene and distance correlates of serial rape. Journal of Quantitative Criminology, 14(1), 35–59. https://doi.org/10.1023/A:1023044408529
Wikström, P.-O. H., Ceccato, V., Hardie, B., and Treiber, K. (2010). Activity fields and the dynamics of crime: Advancing knowledge about the role of the environment in crime causation. Journal of Quantitative Criminology, 26(1), 55–87. https://doi.org/10.1007/s10940-009-9083-9
Wiles, P., and Costello, A. (2000). The “road to nowhere”: The evidence for traveling criminals (Vol. 207). Home Office.
Wright, B. R. E., Caspi, A., Moffitt, T. E., and Silva, P. A. (1998). Factors associated with doubled-up housing—A common precursor to homelessness. Social Service Review, 72(1), 92–111. https://doi.org/10.1086/515747
Zipf, G. K. (1965). Human behavior and the principle of least effort: An introduction to human ecology. Facsim. of 1949 ed. Hafner Publishing Company.
Table 1. Accuracy rate of offender's home and crime-location addresses in police data, by crime type
Type of address | N | Accuracy ratea | ||||||||||
---|---|---|---|---|---|---|---|---|---|---|---|---|
All crimes | Robbery | Burglary | MV theft | Other theft | χ2 (3) | |||||||
n | % | n | % | n | % | n | % | n | % | |||
Offender's home | 396b | 197 | 49.7 | 75 | 55.6 | 85 | 47.0 | 5 | 31.3 | 32 | 50.0 | 4.58 |
Crime location | 428 | 414 | 96.7 | 156 | 96.3 | 173 | 97.7c | 18 | 85.7d | 67 | 98.5 | 9.42* |
3Note. MV = motor vehicle.
4a Accuracy rate refers to the percentage of crimes for which the address extracted from raw police data is the same as the verified and cross-validated one. Percentages with different subscripts differ significantly at p < .05 based on Bonferroni adjusted z-tests for independent proportions.
5b There were 53 crimes for which the offender's home address was missing or undetermined in police data.
6*p < .05.
Table 2. Duration of police investigation (in days) and percentage of crimes with a residential change by the perpetrator during that interval, by crime type
Police investigation | Crimes with a police investigationa | |||||||||
---|---|---|---|---|---|---|---|---|---|---|
All crimes (N = 344) | Robbery (n = 136) | Burglary (n = 149) | MV theft (n = 16) | Other theft (n = 43) | ||||||
n | % | n | % | n | % | n | % | n | % | |
Durationb | ||||||||||
Within 24 hours | 25 | 7.3 | 13 | 9.6 | 9 | 6.0 | 1 | 6.3 | 2 | 4.7 |
From 1 day to 1 week | 77 | 22.4 | 38 | 27.9 | 28 | 18.8 | 5 | 31.3 | 6 | 14.0 |
From 1 week to 1 month | 75 | 21.8 | 37 | 18 | 12.1 | 4 | 25.0 | 16 | 37.2 | |
From 1 to 3 months | 90 | 26.2 | 39 | 28.7 | 36 | 24.2 | 2 | 12.5 | 13 | 30.2 |
More than 3 months | 77 | 22.4 | 9 | 6.6 | 58 | 38.9 | 4 | 25.0 | 6 | 14.0 |
M | 54.9 | 29.8d | 80.9e | 47.0 | 47.0 | |||||
Md | 27.3 | 18.5 | 60.0 | 14.7 | 26.0 | |||||
SD | 63.1 | 38.8 | 73.4 | 59.7 | 53.4 | |||||
Residential change by the perpetrator during that intervalc | 109 | 31.7 | 35 | 25.7 | 55 | 36.9 | 6 | 37.5 | 13 | 30.2 |
7 Note. MV = motor vehicle. Extreme values (>= 90th percentile) were recoded to a common value corresponding to their 89th percentile.
8 a Crimes in which the perpetrator was arrested at the crime location or minutes later while fleeing the crime scene were excluded (n = 105).
9 b Number of days between offender's crime commission and arrest. Due to a violation of the normality assumption, a Kruskal-Wallis non-parametric analysis of variance (ANOVA) was performed on the mean rank duration by crime type (χ2 [3] = 30.83, p < .001). Means with different subscripts differ significantly at p < .05 based on Bonferroni corrected post-hoc tests.
10 c Perpetrator's “place of living” at the time of crime is different than at the time of arrest. Includes any change of primary home address or residential status (domiciled, NFA, incarcerated) during that time period. Chi-square test for independence among crime types is non-significant (χ2 [3] = 4.40, p = .222) and percentages do not differ significantly at p < .05 based on Bonferroni adjusted z-tests for independent proportions.
T
able 3. Offenders’ residential mobility during the study period and its effect on the accuracy rate of offenders’ home address in police data
Perpetrator-level (N = 98) | Crime-level (N = 396b) | ||||
---|---|---|---|---|---|
Average number of days before a residential change during the study perioda | N | % | Number of crimes committed | Correct PHA in police datac | |
n | % | ||||
Less than 4 months | 14 | 14.3 | 51 | 15 | 29.4d |
From 4 to 6 months | 30 | 30.6 | 165 | 82 | 49.7d |
From 6 to 12 months | 35 | 35.7 | 107 | 49 | 45.8d |
From 1 to 2 years | 11 | 11.2 | 49 | 28 | 57.1d |
More than 2 years | 8 | 8.2 | 24 | 23 | 95.8e |
M | 253.7 | ||||
SD | 148.9 |
11a Number of days in the study period (range = 1208-3099; M = 2381.7; SD = 355.9) divided by the total number of residential changes by the perpetrator during that interval (range = 1-30; M = 12.3; SD = 6.6). The latter corresponds to the sum of all different PHAs (range = 1-16; M = 7.0; SD = 3.5), NFA periods (range = 0-8; M = 1.8; SD = 2.3), and prison stays (range = 0-11; M = 3.3; SD = 2.8).
12b There were 53 crimes for which the offender's home address was missing or undetermined in police data.
13c Accuracy rate refers to the percentage of crimes for which the address extracted from raw police data is the same as the verified and cross-validated one. Chi-square test for independence among the five categories is significant (χ2 [4] = 30.57, p < .001; Cramer's V = .278) and percentages with different subscripts differ significantly at p < .05 based on Bonferroni adjusted z-tests for independent proportions.
Table 4. Offenders’ awakening location, decision location, and safe-house location, by crime type
Type of place | All crimes | Robbery | Burglary | MV theft | Other theft | χ2 (3) | |||||
---|---|---|---|---|---|---|---|---|---|---|---|
n | % | n | % | n | % | n | % | n | % | ||
Awakening location (n = 444) | |||||||||||
Offender's home | 334 | 75.2 | 114 | 70.8 | 157 | 80.5 | 15 | 71.4 | 48 | 71.6 | 5.24 |
Friend’s or family members' dwelling | 53 | 11.9 | 17 | 10.6 | 24 | 12.3 | 2 | 9.5 | 10 | 14.9 | 1.00 |
Motel or hotel | 21 | 4.7 | 9 | 5.6 | 9 | 4.6 | 1 | 4.8 | 2 | 3.0 | 0.72 |
Therapy center, community shelter, or hospital | 13 | 2.9 | 9 | 5.6a | 1 | 0.5b | ― | ― | 3 | 4.5 | 9.22* |
Other location | 23 | 5.2 | 12 | 7.5 | 4 | 2.1b | 3 | 14.3a | 4 | 6.0 | 9.21* |
Decision location (n = 449) | |||||||||||
Offender's home | 220 | 49.0 | 73 | 45.1 | 102 | 51.5 | 10 | 47.6 | 35 | 51.5 | 1.69 |
Friend’s or family members' dwelling | 67 | 14.9 | 24 | 14.8 | 32 | 16.2 | 1 | 4.8 | 10 | 14.7 | 1.95 |
Actual crime location (opportunistic crime) | 50 | 11.1 | 14 | 8.6 | 17 | 8.6b | 5 | 23.8 | 14 | 20.6a | 11.87** |
Prior crime location | 27 | 6.0 | 16 | 9.9a | 6 | 3.0b | 1 | 4.8 | 4 | 5.9 | 7.46 |
Bar or drug spot | 45 | 10.0 | 20 | 12.3 | 22 | 11.1 | 1 | 4.8 | 2 | 2.9 | 5.66 |
Other location | 40 | 8.9 | 15 | 9.3 | 19 | 9.6 | 3 | 14.3 | 3 | 4.4 | 2.58 |
Safe-house location (n = 445) | |||||||||||
Offender's home | 179 | 40.2 | 54 | 33.5 | 91 | 46.4 | 8 | 38.1 | 26 | 38.8 | 6.23 |
Friend’s or family members' dwelling | 105 | 23.6 | 57 | 35.4a | 33 | 16.8b | 4 | 19.0 | 11 | 16.4 | 19.57*** |
Motel or hotel | 19 | 4.3 | 9 | 5.6 | 5 | 2.6 | 1 | 4.8 | 4 | 6.0 | 2.59 |
Other location | 37 | 8.3 | 15 | 9.3 | 18 | 9.2 | 3 | 14.3 | 1 | 1.5 | 5.48 |
Unable to reach a safe house (arrested by police) | 105 | 23.6 | 26 | 16.1a | 49 | 25.0 | 5 | 23.8 | 25 | 37.3b | 12.16** |
14Note. MV = motor vehicle. Chi-square test for independence was conducted separately for each type of place among the four crime types. Percentages with different subscripts differ significantly at p < .05 based on Bonferroni adjusted z-tests for independent proportions.
15*p < .05. **p < .01. ***p < .001.
Table 5. Type of detour or stopover during the crime journey, by crime phase and crime type
Type of detour or stopover by crime phase | All crimes (N = 449) | Robbery (n = 162) | Burglary (n = 198) | MV theft (n = 21) | Other theft (n = 68) | χ2 (3) | |||||
---|---|---|---|---|---|---|---|---|---|---|---|
n | % | n | % | n | % | n | % | n | % | ||
Search for a target (index crime) | |||||||||||
Pre-crime | 163 | 36.3 | 40 | 24.7a | 105 | 53.0b | 8 | 38.1 | 10 | 14.7a | 47.15*** |
Commit another crime | |||||||||||
Pre-crime | 62 | 13.8 | 24 | 14.8 | 29 | 14.6 | 1 | 4.8 | 8 | 11.8 | 1.94 |
Post-crime | 64 | 14.3 | 22 | 13.6 | 23 | 11.6 | 4 | 19.0 | 15 | 22.1 | 4.97 |
Entire crime journey | 115 | 25.6 | 45 | 27.8 | 45 | 22.7 | 5 | 23.8 | 20 | 29.4 | 1.82 |
Other | |||||||||||
Pre-crime | 69 | 15.4 | 36 | 22.2a | 25 | 12.6 | 3 | 14.3 | 5 | 7.4b | 10.37* |
Meet with co-offenders | 31 | 6.9 | 19 | 11.7 | 10 | 5.1 | 1 | 4.8 | 1 | 1.5 | 10.20* |
Gear up/buy equipment/find transport | 28 | 6.2 | 17 | 10.5 | 9 | 4.5 | ― | ― | 2 | 2.9 | 8.65* |
Scout out the crime surroundings | 11 | 2.4 | 4 | 2.5 | 6 | 3.0 | ― | ― | 1 | 1.5 | 1.08 |
Other | 20 | 4.5 | 8 | 4.9 | 7 | 3.5 | 2 | 9.5 | 3 | 4.4 | 1.75 |
Post-crime | 148 | 33.0 | 54 | 33.3a | 53 | 26.8a | 15 | 71.4b | 26 | 38.2a | 18.37*** |
Switch mode of transportation | 44 | 9.8 | 19 | 11.7a | 12 | 6.1a | 10 | 47.6b | 3 | 4.4a | 40.03*** |
Sell or stash stolen goods | 40 | 8.9 | 5 | 3.1a | 15 | 7.6a | 2 | 9.5 | 18 | 26.5b | 33.06*** |
Buy drugs/get intoxicated | 34 | 7.6 | 18 | 11.1a | 16 | 8.1 | ― | ― | ― | ―b | 10.26* |
Other | 45 | 10.0 | 19 | 11.7 | 18 | 9.1 | 3 | 14.3 | 5 | 7.4 | 1.670 |
Entire crime journey | 182 | 40.5 | 68 | 42.0 | 70 | 35.4b | 15 | 71.4a | 29 | 42.6 | 10.79* |
Any (at least one of the above categories) | |||||||||||
Pre-crime | 227 | 50.6 | 75 | 46.3a | 124 | 62.6b | 9 | 42.9 | 19 | 27.9a | 27.13*** |
Post-crime | 184 | 41.0 | 69 | 42.6 | 66 | 33.3b | 15 | 71.4a | 34 | 50.0 | 15.30** |
Entire crime journey | 304 | 67.7 | 106 | 65.4 | 139 | 70.2 | 17 | 81.0 | 42 | 61.8 | 3.73 |
16Note. MV = motor vehicle. Chi-square test for independence was conducted separately for each type of stopover among the four crime types. Percentages with different subscripts differ significantly at p < .05 based on Bonferroni adjusted z-tests for independent proportions.
17*p < .05. **p < .01. ***p < .001.
Table 6. Percentage (%) of crimes in which the offender took a direct home-crime and/or crime-home itinerary, by crime type and crime phase
Type of itinerary by crime phase | N | Matching ratea | ||||||||||
---|---|---|---|---|---|---|---|---|---|---|---|---|
All crimes | Robbery | Burglary | MV theft | Other theft | χ2 (3) | |||||||
n | % | n | % | n | % | n | % | n | % | |||
Pre-crime phase | ||||||||||||
Offender's home → Crimeb | 449 | 98 | 21.8 | 34 | 21.0d | 31 | 15.7d | 6 | 28.6 | 27 | 39.7e | 17.78*** |
Offender's home → Crimec | 386 | 53 | 13.7 | 20 | 14.8 | 14 | 8.2d | 2 | 11.8 | 17 | 26.6e | 13.42** |
Post-crime phase | ||||||||||||
Crime → Offender's homeb | 449 | 88 | 19.6 | 31 | 19.1 | 50 | 25.3e | 2 | 9.5 | 5 | 7.4d | 11.86** |
Crime → Offender's homec | 383 | 35 | 9.1 | 16 | 11.9 | 16 | 9.5 | ― | ― | 3 | 4.8 | 4.39 |
Entire crime journey | ||||||||||||
Offender's home → Crime → Offender's homeb | 449 | 33 | 7.3 | 16 | 9.9e | 17 | 8.6 | ― | ― | ― | ―d | 9.02* |
Offender's home → Crime → Offender's homec | 383 | 15 | 3.9 | 8 | 5.9 | 7 | 4.2 | ― | ― | ― | ― | 4.74 |
18Note. MV = motor vehicle.
19a Matching rate refers to the percentage of crimes in which the offender took the home-crime and/or crime-home itinerary. Percentages with different subscripts differ significantly at p < .05 based on Bonferroni adjusted z-tests for independent proportions.
20b Direct trip without any detours or stopovers.
21c Direct trip without any detours or stopovers with an accurate offender's home and crime-location addresses in police data.
22*p < .05. **p < .01. ***p < .001
Table 7. Kruskal-Wallis non-parametric analyses of variance (ANOVA) on the mean ranks by crime type for five different measures of distance
Measure | Distance in km | χ2 (3) | ||||||||||
---|---|---|---|---|---|---|---|---|---|---|---|---|
All crimes | Robbery | Burglary | MV theft | Other theft | ||||||||
N | M [Md] | SD | M [Md] | SD | M [Md] | SD | M [Md] | SD | M [Md] | SD | ||
Traditional home-crime distance | 396 | 20.1 [9.7] | 22.9 | 16.9a [7.6] | 20.1 | 18.7a [8.1] | 22.3 | 18.9 [14.5] | 18.6 | 31.0b [21.5] | 27.9 | 10.10* |
Offender's validated home-crime distance | 448 | 13.6 [6.6] | 15.7 | 12.9 [6.6] | 14.8 | 12.2 [6.2] | 14.5 | 11.7 [7.4] | 12.2 | 20.1 [10.5] | 19.8 | 6.28 |
Distance traveled in the pre-crime phase | 449 | 16.0 [5.6] | 20.1 | 14.1 [5.4] | 18.4 | 16.8 [5.5] | 20.6 | 12.7 [8.8] | 15.9 | 19.4 [5.9] | 23.2 | 1.27 |
Distance traveled in the post-crime phase | 441 | 14.4 [6.0] | 17.3 | 14.2 [8.3] | 16.3 | 12.4 [4.4] | 16.0 | 21.1 [14.9] | 19.7 | 19.2 [7.5] | 21.4 | 5.50 |
Distance traveled in the entire crime journey | 441 | 35.0 [17.0] | 41.0 | 31.1 [17.6] | 36.4 | 33.0 [12.9] | 39.7 | 46.8 [20.3] | 50.7 | 47.2 [24.0] | 49.1 | 3.34 |
1Note. MV = motor vehicle. For each measure of distance, extreme values (>= 90th percentile) were recoded to a common value corresponding to their 89th percentile. Mean with different subscripts differ significantly at p < .05 based on Bonferroni corrected post-hoc tests.
2*p < .05.
Table 8. Multilevel correlations between both measures of home-crime distance (i.e., traditional and offender's validated) and the distance traveled by offenders during the pre-crime phase, post-crime phase, and entire crime journey.
Measure (in km) | N Crimes | N Offenders | Multilevel correlation coefficient (r) | ||||
---|---|---|---|---|---|---|---|
All crimes | Robbery | Burglary | MV theft | Other theft | |||
Traditional home-crime distance and... | |||||||
Traditional home-crime distance | ― | ― | ― | ― | ― | ― | ― |
Offender's validated home-crime distance | 395 | 95 | .51*** | .95a*** | .10b | .37b | .87c*** |
Distance traveled in the pre-crime phase | 396 | 95 | .29*** | .84a*** | .03b,c | .33b,c,d | .61c,d*** |
Distance traveled in the post-crime phase | 389 | 95 | .39*** | .71a,c*** | .04b | .06b | .81a,c*** |
Distance traveled in the entire crime journey | 389 | 95 | .40*** | .82a,c*** | .04b | .17b | .86a,c*** |
Offender's validated home-crime distance and... | |||||||
Traditional home-crime distance | 395 | 95 | .51*** | .95a*** | .10b | .37b | .87c*** |
Offender's validated home-crime distance | ― | ― | ― | ― | ― | ― | ― |
Distance traveled in the pre-crime phase | 448 | 98 | .62*** | .86a*** | .41b*** | .95a*** | .64c*** |
Distance traveled in the post-crime phase | 440 | 98 | .78*** | .79a,b,c*** | .76a,b*** | .31d | .87b,c*** |
Distance traveled in the entire crime journey | 440 | 98 | .81*** | .86a,c*** | .69b*** | .53b* | .92a,c*** |
1Note. MV = motor vehicle. Multilevel correlations were computed with R software (version 4.1.2) using the “correlations” package from the “easystats suite” (Makowski et al., 2020). Correlation coefficient (r) with different subscripts differs significantly at p < .05 based on a z-test on Fisher’s z-transformed correlation coefficients (Hinkel et al., 1988).
2*p < .05. ***p < .001.
Table 9. Summary of the offender's choice of target, by crime type
Choice of target | All crimes (N = 449) | Robbery (n = 162) | Burglary (n = 198) | MV theft (n = 21) | Other theft (n = 68) | χ2 (3) | |||||
---|---|---|---|---|---|---|---|---|---|---|---|
n | % | n | % | n | % | n | % | n | % | ||
Planned | 236 | 52.6 | 108 | 66.7a | 76 | 38.4b | 8 | 38.1 | 44 | 64.7a | 34.67*** |
Searched | 163 | 36.3 | 40 | 24.7a | 105 | 53.0b | 8 | 38.1 | 10 | 14.7a | 47.15*** |
Opportunistic | 50 | 11.1 | 14 | 8.6 | 17 | 8.6b | 5 | 23.8 | 14 | 20.6a | 11.87** |
3Note. MV = motor vehicle. Percentages with different subscripts differ significantly at p < .05 based on Bonferroni adjusted z-tests for independent proportions.
4**p < .01. ***p < .001.