Age and the distance to crime

Introduction

Over a century of criminological research has generated a number of empirical regularities, including but not limited to: neighborhoods retain their relative criminogenic ranking over time (ecological stability) (Shaw and McKay, 1931, 1942); males are more violent than females (Boyd, 2000); most offenders age out of crime (Hirschi and Gottfredson, 1983); and the distance traveled to crime is short, particularly for violent crime (Wiles and Costello, 2000). This latter empirical regularity is often referred to as the journey-to-crime literature. This literature dates back at least 80 years (White, 1932), investigating the distance traveled to crime along a number of dimensions.

As outlined by Rengert (2004), the journey-to-crime has a number of components: origin; direction; and distance. However, most of the journey-to-crime literature focuses only on the distance component. This has led Townsley and Sidebottom (2010) to prefer the term ‘distance to crime’, and we follow their lead here unless we are referring to the entire journey-to-crime or its body of literature.

The journey-to-crime literature has generated a number of methodological concerns. As outlined by van Koppen and De Keijser (1997), distance-decay – the primary result that emerges out of the journey-to-crime literature stating that the distance traveled to crime is most often short – may be an aggregate offender population phenomenon that does not emerge at the individual level. They show that if offenders’ distance to crime is always constant and there are more offenders with short distances traveled, a distance-decay pattern emerges in the aggregate. Therefore, if this phenomenon is used to model individual behavior it would be in error. However, Rengert et al. (1999) argued that such aggregations are sound; Levine and Associates (2002) found strong empirical support for the claim by Rengert et al. (1999), but this empirical support has methodological issues of its own (Townsley and Sidebottom, 2010).

Another methodological concern in the journey-to-crime literature is in regard to the origin of the journey. As stated above, Rengert (2004) identifies this as an important component in the journey-to-crime but it is often ignored. Though the trip origin location is often assumed to be the offender’s home, it is quite possible that an offender’s journey begins from another activity node in his/her activity space (Brantingham and Brantingham, 1981, 1993a) such as a bar (Townsley and Sidebottom, 2010; Wiles and Costello, 2000). If this is the case in more than a trivial number of crime trips, the distance to crime based on the offender’s home is likely incorrect. The difficulty in correcting for this concern is that police data do not (typically) have the trip origin location, only the offender’s home address (if known) and the crime location; interviews of offenders with regard to specific crimes would be necessary but then there are interview biases and small sample issues.

Most recently, nesting effects have emerged as a methodological concern in the journey-to-crime literature. ¹ Nesting effects refer to effects that arise because of repeated sampling from the same groups: repeated sampling from within a group tends to generate less variability than repeated sampling across multiple groups. Bichler et al. (2007), Smith et al. (2009), and Townsley and Sidebottom (2010) all investigated this phenomenon and found nesting to be critical in understanding the distance to crime. Specifically, only a small number of prolific offenders display distance-decay in their criminal activities (Townsley and Sidebottom, 2010). Consequently, only a small number of offenders exhibit the assumed aggregate behavior. However, Bichler et al. (2011) support the robust findings of distance-decay put forth by Rengert et al. (1999). As such, the ecological fallacy argument in distance-decay research may be unwarranted. Regardless, the difficulty in confirming or denying distance-decay is that most offenders are not prolific and, hence, may not have enough observations regarding their criminal behavior to confirm or deny a distance-decay propensity when it may, in fact, exist.

In this article, we contribute to the journey-to-crime literature through an examination of what we argue to be another methodological concern that has a nesting effect: the impact of age on the distance to crime. We do not investigate the presence (or lack thereof) of distance-decay or perform any modeling that may require us to address dependent data. Rather, our objective is to determine whether the distance to crime varies by age and across a number of different crime classifications. Using a large incident-based police database, we are able to show the importance of age in the distance to crime and explain previous conflicting results.

Theory and Empirics of Age and the Distance to Crime

Data and methods

Criminal incident data

The data employed in the current analysis are from municipalities (police jurisdictions) in British Columbia, Canada’s western-most province. The entire province of British Columbia has approximately four million residents living in 186 police jurisdictions, with the Royal Canadian Mounted Police (RCMP) being responsible for policing 174 of the 186 police jurisdictions – 67 percent of the provincial population, approximately 2.7 million persons.

The data used in the analyses below are incident-based and extracted from the RCMP Police Information Retrieval System (PIRS). These data cover the period 1 August 2002 through 31 July 2006 and represent the complete set of incidents dealt with by the RCMP. With our ability to combine 48 consecutive months of the PIRS database, any unusual features of any given year or month are ironed out. The entire PIRS database contains approximately five million contacts with the police that involve approximately nine million individuals – offenders, victims, complainants, and witnesses. Of these five million police contacts, approximately 750,000 are common criminal offenses.

From this large incident-based data set, we extract 12 crime classifications that represent the majority of property and violent crimes: homicide; sexual assault; aggravated assault; ⁵ assault; robbery; armed robbery; burglary (residential, commercial, and other); ⁶ theft; theft of motor vehicle; and theft from motor vehicle. This subset of data includes 81,257 crime trips, outlined in Table 1.

OPEN IN VIEWER

Table 1.

Distance to crime by crime type, kilometers, all ages.

Crime type	First quartile	Median	Third quartile	Interquartile range	Count
Homicide	0.00	0.66	4.61	4.61	374
Sexual assault	0.00	0.92	5.13	5.13	5971
Aggravated assault	0.00	0.30	2.52	2.52	8622
Assault	0.00	0.00	1.70	1.70	49,414
Robbery	0.89	2.09	6.49	5.60	856
Armed robbery	0.86	2.51	7.89	7.03	974
Residential burglary	0.35	1.50	4.67	4.32	4872
Commercial burglary	0.71	1.97	5.81	5.10	340
Other burglary	0.62	2.03	4.82	4.19	396
Theft of motor vehicle	0.23	2.21	7.26	7.03	2930
Theft from motor vehicle	0.64	1.78	4.45	3.82	1540
Theft	0.14	1.28	4.71	4.57	4968
All crimes	0.00	0.65	3.51	3.51	81,257

Source: RCMP Police Information Reporting System (PIRS).

Results

In order to be sure there are no peculiarities in our data, we have calculated the distance to crime for all ages aggregated together. These results are shown in Table 1. With regard to the median distance traveled to an offense, our data are consistent with previous research: the distance to crime is short, particularly for violent crime. When all crimes are aggregated together the distance to crime is also quite short, on a par with the violent crime classifications, but this is because violent crime classifications (particularly assault) are highly represented in our data because offenders are more likely to be identified (and, hence, in our database) in violent crimes – by definition, they interact with the victim(s). The only result that may be considered inconsistent with past research regarding this empirical regularity is that robbery, often considered a violent crime, has a median distance to crime more in line with property crime classifications: a median distance to crime that exceeds 2 kilometers for both robbery and armed robbery. However, if one considers the classification of violent property crime put forth by Indermaur (1995), this result falls in line with past research on the distance to crime.

Though not the primary concern of this article, the other statistics are worthy of note. The first quartile, third quartile, and interquartile range all show the same generality that the distance traveled to crime is short, particularly for violent crime. Perhaps most interesting is to note that all ‘pure’ violent crime classifications and ‘all crimes’ have a first quartile value of 0 kilometers. Consequently, at least 25 percent of the distances traveled for these crime classifications are zero, implying that the offenders in these cases are committing their acts of violence in their homes. Moreover, the median distance to crime for assault is also 0 kilometers, stating that at least 50 percent of the distances traveled for assault are 0 kilometers. This result is shown to have a particular age pattern in the results below.

Turning to the relationship between age and the distance to crime, the correlations between single years of age and the respective median distances to crime are shown in Table 2 – parametric and nonparametric results are qualitatively similar so we will not discuss them separately. Similar to Costello and Wiles (2001) and Wiles and Costello (2000) we find that the correlations are statistically significant and negative, statistically significant and positive, and statistically insignificant depending upon the crime classification. In fact, our results are the same for: all crimes, assault, residential burglary, commercial burglary, and theft of motor vehicle. ⁸ The primary difference between our results and those of Costello and Wiles (2001) and Wiles and Costello (2000), aside from our analysis having more crime classifications, is that the magnitudes of our correlation coefficients are greater: the greatest absolute value for Costello and Wiles (2001) and Wiles and Costello (2000) is 0.17, whereas ours is 0.69 for the parametric correlations and 0.79 for the nonparametric correlations. Though bigger is not always better, it is interesting to note the strength of the relationship between age and the median distance to crime. However, correlation coefficients represent the strength of the linear relationship between two variables. And as indicated above, the relationship between age and the median distance to crime is not linear.

OPEN IN VIEWER

Table 2.

Parametric and nonparametric correlations, distance to crime with age, British Columbia, Canada, 1 August 2002 to 31 July 2006.

Crime type	Pearson	Spearman
Homicide	−0.47**	−0.43**
Sexual assault	−0.21*	−0.21
Aggravated assault	−0.60***	−0.72***
Assault	−0.64***	−0.79***
Robbery	–0.15	–0.21
Armed robbery	0.13	0.10
Residential burglary	0.23	0.17
Commercial burglary	–0.14	–0.10
Other burglary	0.31*	0.34**
Theft of motor vehicle	0.33**	0.30*
Theft from motor vehicle	0.15	0.09
Theft	0.21	0.22
All crimes	–0.69***	–0.76***

Note: *indicates statistical significance at the 10 percent level; **indicates statistical significance at the 5 percent level; ***indicates statistical significance at the 1 percent level.

The primary result of our article is shown in Figure 1. Clearly evident from this plot of age and the median distance to crime is that its relationship is far better described as quadratic. Similar to the result shown in Clarke and Eck (2003), the median distance to crime for those youngest in our data (eight years old) is very close to zero and reaches its maximum at the age of 20 (slightly greater than 1.25 kilometers). The rate of decline in the median distance to crime shown in Figure 1 is much faster than shown in Clarke and Eck (2003), returning close to 0 kilometers by the age of 40. Similar to the age–crime curve (Hirschi and Gottfredson, 1983), there exists an age–distance to crime curve, at least in the aggregate.

OPEN IN VIEWER

Figure 1.

The age–distance to crime curve, by single year of age, all crime types aggregated.

Similar plots of age and median distance to crime for violent crime classifications are shown in Figure 2. Immediately obvious is that this relationship, though most often present to some degree, is far from universal. In the case of homicide (Figure 2a), there is no evidence of a relationship between age and the median distance to crime until offenders reach the age of approximately 40. At this age, the median distance to crime becomes incredibly short, most often 0 kilometers. Sexual assault (Figure 2b) has a far more evident quadratic relationship between age and the median distance to crime. Though this relationship is rather volatile after the age of 20, it has a clear downward trend. Aggravated assault (Figure 2c) and assault (Figure 2d) are clearly driving the aggregate results in Figure 1. Both of these crime classifications exhibit sharp increases in the median distance to crime from eight to 20 years, followed by an equally sharp decrease, particularly for assault. In the case of aggravated assault the median distance to crime is 0 or effectively 0 kilometers at and after the age of 40; for assault, this median distance occurs at the age of 25. Moreover, the median distance to crime for assault reaches its maximum of 0.8 kilometers, approximately half of the maximum median distance for aggravated assault. The quadratic relationship between age and the median distance to crime is not particularly evident for robbery (Figure 2e), but is discernible for armed robbery (Figure 2f). Overall, the quadratic relationship between age and the median distance to violent crime is most often evident, but the strength of that relationship clearly varies from crime classification to crime classification.

OPEN IN VIEWER

Figure 2a.

The age–distance to crime curve, by single year of age, homicide.

OPEN IN VIEWER

Figure 2b.

The age–distance to crime curve, by single year of age, sexual assault.

OPEN IN VIEWER

Figure 2c.

The age–distance to crime curve, by single year of age, aggravated assault.

OPEN IN VIEWER

Figure 2d.

The age–distance to crime curve, by single year of age, assault.

OPEN IN VIEWER

Figure 2e.

The age–distance to crime curve, by single year of age, robbery.

OPEN IN VIEWER

Figure 2f.

The age–distance to crime curve, by single year of age, armed robbery.

The age and median distance to crime plots for property crime classifications are shown in Figure 3. Similar to the results for the violent crime classifications, the strength of the quadratic relationship between age and the median distance to crime varies across property crime classifications. The weakest relationships are for commercial burglary (Figure 3b), other burglary (Figure 3c), and theft from motor vehicle (Figure 3e). However, having a weak quadratic relationship between age and the median distance to crime is expected for these crime classifications. Because of the nature of these crimes, their opportunity surfaces are restricted to specific land uses (commercial burglary) or so ubiquitous (theft from motor vehicle) that it is no surprise that age has little to do with the median distances traveled. Though commercial burglary does have the appearance of a quadratic trend (this quadratic trend is confirmed in Table 3), any offender that lives in a residential area, as opposed to mixed land use in a central business district, will necessarily travel further to commit an offense; this is evident in Table 1, as well. As such, if an offender travels to a commercial area, s/he is most likely already away from the watchful eyes of guardians.

OPEN IN VIEWER

Figure 3a.

The age–distance to crime curve, by single year of age, residential burglary.

OPEN IN VIEWER

Figure 3b.

The age–distance to crime curve, by single year of age, commercial burglary.

OPEN IN VIEWER

Figure 3c.

The age–distance to crime curve, by single year of age, other burglary.

OPEN IN VIEWER

Figure 3d.

The age–distance to crime curve, by single year of age, theft of motor vehicle.

OPEN IN VIEWER

Figure 3e.

The age–distance to crime curve, by single year of age, theft from motor vehicle.

OPEN IN VIEWER

Figure 3f.

The age–distance to crime curve, by single year of age, theft.

OPEN IN VIEWER

Table 3.

Parameterization of quadratic trends, distance to crime with age, British Columbia, Canada, 1 August 2002 to 31 July 2006.

Crime type	Age	Age-squared	R2
Homicide	2.94	–2.39	0.65
Sexual assault	2.52	–1.85	0.74
Aggravated assault	2.54	–2.30	0.42
Aggravated assault – truncated	3.15	–2.63	0.76
Assault	1.96	–1.84	0.24
Assault – truncated	2.77	–2.09	0.77
Robbery	2.51	–1.73	0.84
Armed robbery	1.98	–1.13	0.84
Residential burglary	2.02	–1.16	0.89
Commercial burglary	2.28	–1.62	0.67
Other burglary	1.46	–0.67	0.70
Theft of motor vehicle	1.76	–0.91	0.82
Theft from motor vehicle	2.13	–1.29	0.87
Theft	2.34	–1.54	0.87
All crimes	3.05	–2.73	0.63

Note: All parameters are statistically significant at the 5 percent level; all parameters are standardized for comparative purposes.

For residential burglary (Figure 3a), on the other hand, the relationship between age and the median distance to crime clearly follows a quadratic trend, though there is much volatility once offenders reach the age of 30 years. Theft of motor vehicle (Figure 3d), aside from a few outliers, follows a strong quadratic trend, as does theft (Figure 3f). Aside from the variability in the strength of the relationship between age and the median distance to crime, the most notable aspect of the property crime classifications is that they peak in the median distance traveled later in life than for the violent crime classifications: approximately 30 years, rather than 20 years – robbery and armed robbery also peak in later years, but as discussed earlier these crime classifications are more similar to other property crime classifications than violent crime classifications.

Lastly, there is the parameterization of the relationship between age and the median distance to crime, shown in Table 3. As stated in Table 3, all estimated parameters (age and age-squared) for all crime classifications are statistically significant at the 5 percent level, and the estimated parameters reported are the standardized coefficients for ease of comparison across crime classifications. The coefficient of determination, R², is also shown in Table 3, revealing that much of the variation in all the median distances to crime is explained with the quadratic trend. Because assault and aggravated assault consistently have a zero median distance to crime in the older ages, the quadratic trends were also estimated truncating the data series when the median distance to crime becomes zero. This is done because the zero median values for these two violent crime classifications will necessarily bias the age-squared coefficient to be lower in magnitude, making it more difficult to assess the propositions made earlier.

The results shown in Table 3 are expected given the plots shown in Figures 2 and 3. The ‘pure’ violent crime classifications (homicide, sexual assault, aggravated assault, and assault) have the greatest (positive) magnitudes for the estimated parameters representing age, and the greatest (negative) magnitudes for the estimated parameters representing age-squared. This is an expected result because these four violent crime classifications have their median distances to crime peak at earlier ages than the other crime classifications.

Overall, these results support the importance of age factors in the distance to crime. Though not all crime classifications are equally bound to an age–median distance to crime relationship, it is present in most cases. Moreover, the relationship is not linear. The correlation coefficients shown in Table 2 indicate that many crime classifications have negative relationships between age and the median distance to crime. Clearly evident from the graphs, this negative relationship emerges as an average tendency because the median distance to crime peaks at 20 years of age for most of these cases. As such, there are 48 observations that exhibit a negative relationship and only 12 observations that exhibit a positive relationship between age and the median distance to crime; on average, the relationship is negative. Through visual inspection of our data we have been able to show that the often found relationship between age and the distance to crime is not changing (Costello and Wiles, 2001; Wiles and Costello, 2000). Rather, past research that found a positive relationship between age and the distance to crime either had samples that had younger offenders, on average, or analyzed crime classifications with weaker quadratic trends.

Discussion

The journey-to-crime literature has consistently shown that the distance to crime is short, particularly for violent crime. Recent research, however, has questioned this long-standing empirical regularity stating that it is the result of not accounting for nesting effects (Smith et al., 2009; Townsley and Sidebottom, 2010). Essentially, this nesting effect is the improper grouping of different types of offenders (prolific and non-prolific offenders) that leads to the appearance of a distance-decay for the distance to crime. In this article we investigate another nesting effect: the improper grouping of different ages of offenders. Through an investigation of the distance to crime considering individual years of age, we are able to show that on average the distance to crime increases from adolescence to approximately 20 years of age, then decreases thereafter. As such, because we are able to explain the differences found by previous researchers (Costello and Wiles, 2001; Wiles and Costello, 2000), there is no need to invoke a change in this relationship over time.

An obvious question to ask at this point is: how is this relationship between age and the distance to crime explained? We are confident the explanation lays in the spatially expanding routine activities as we age. When we are young (pre-teenage years) we are almost always under the close watch of our guardians. As such, our travels are rarely a great distance from home. However, as we age and move into the teenage years, we increasingly try to move away from the watchful eyes of our guardians. This process is rather natural in the sense that this age is when we strive for independence, but also because our activities are increasingly moved further away from home (Felson and Boba, 2010). One of the fundamental activities for the teenage years (school) is a prime example: as we move from elementary to high school we increasingly have a more dispersed set of friends and, on average, the high school is further from home than the elementary school because of its size (Phillips, 1980). The end result of this spatial expansion of our routine activities is an increase in the distance to crime from adolescence through the teenage years. But this spatial expansion may also emerge for reasons other than avoiding our guardians. As we get older, we may simply be traveling to spend time with friends/boyfriends/girlfriends who do not live close to home. Or, as young adults, when we move out of our parents’ home, we may have to move further away from the best sources of targets for crime.

Regardless of the reasons for this, the spatially expanding nature of our routine activities peaks in early adulthood. At this stage in life it is no longer necessary to escape the watchful eyes of our guardians so we do not need to travel any extra distance to undertake any of our activities. Though the distance traveled to work is going to depend on land use, our occupation, and where we can afford to live, our commuting distances are preferred to be shorter as we age into adulthood (Rouwendal and Rietveld, 1994; Schwanen et al., 2004). The reason for this is simple: efficiency. We do not need to travel further than necessary in our activities (legitimate or illegitimate), so we do not. As such, the distance traveled to crime decreases as we age as well.

On the surface, it appears as though our argument is only consistent with Zipf’s (1949) principle of least effort in adulthood, but this is not the case. This principle must not be thought of as an absolute principle, but a relative principle. In the context of crime, least effort is not simply defined as finding the closest target, but the closest suitable target; suitability is socially defined because for a (relatively brief) portion of our lives we are being watched rather closely. Therefore, in order to find suitable targets we have to travel a little further and this coincides with the spatial expansion of our routine activities – the spatial expansion of our activity space (Brantingham and Brantingham, 1981, 1993a).

Curiously, a similar pattern is found in the research that considers geographic profiling in animal predatory behaviors. Martin et al. (2009) found that the anchor point from which a great white shark hunts from becomes increasingly fixed as the great white shark becomes bigger (older). When the great white sharks are younger their hunting activities are much more sporadic, but become far more predictable as they age. Similarly for human offenders: as we age, our activity space decreases in size to what is necessary – it becomes increasingly fixed. ⁹

Overall, we find strong support for the three propositions stated above. Though not universal, we find strong support for Proposition 1 and Proposition 2: there is an inverted-U shape to the relationship between age and the distance to crime. Support for Proposition 3 (the strength of Proposition 1 and Proposition 2 depends on the opportunity surface for different crime types) is more speculative but this is likely because of the aggregate nature of our analysis. A detailed analysis of individual crime trips considering the home of the offender, the distance traveled, and the land use of the crime site location would provide far more information regarding this last proposition. We leave this as an avenue of future research.

As with any research, ours is not without its limitations. First, because of the size of our data set, we must assume that all geocoding of home locations and crime site locations are correct. Though there is always potential for geocoding error, the nature of geocoding algorithms and the corresponding street networks used for geocoding are constantly improving. Second, as with most research investigating the distance to crime, we assume the trip origin is the residence of the offender – see Costello and Wiles (2001) and Wiles and Costello (2000) for an exception. As outlined by Townsley and Sidebottom (2010), and discussed above, the residence of the offender may not represent the trip origin for the journey-to-crime. Third, we are only considering crime known to the police. This is an obvious and unavoidable limitation common in all similar research. And fourth, our analysis uses people of varying ages, not a group of persons over a long period of time (a longitudinal study).

However, our data span four years and cover 174 police jurisdictions reported by one agency. As such, our analysis is not as sensitive to the time frame of our study because it smoothes out any single aberrant year, if present; having multiple police jurisdictions limits any bias that may be present because of any biases in particular police jurisdictions (we have urban, suburban, and rural police jurisdictions in our data); and having only one reporting agency (the RCMP) limits inconsistencies in the reporting of criminal incidence.

There are a number of directions for future research. First and foremost, our results must be confirmed in other contexts. Our results are consistent with that found in Clarke and Eck (2003), but replication is necessary before any generalizations can be made. Second, the relationship between age and the distance to crime should be investigated separately in different geographic locations. Our analysis considers British Columbia in aggregate, but notable differences may be present between urban, suburban, and rural environments. Third, an investigation into the nesting effects outlined by Smith et al. (2009) and Townsley and Sidebottom (2010) is in order. The presence of distance-decay has shown to be present only for prolific offenders, but it is possible that there is nesting within this result as well: it may only be specific ages of prolific offenders that are driving the results. And fourth, longitudinal research needs to be undertaken in order to see if offenders alter their distance to crime (at an individual level) as they age, and whether this varies by crime type. Clearly, future research is dependent on large data sets that contain information not only on the crime itself, but those involved – notably the offender. The more disaggregation or nesting effects investigated the greater is the need for large data sets to have reliable analyses. Fortunately, such data sets are becoming increasingly available and in use within the criminological community.