Near-repeat burglary patterns in Malmö: Stability and change over time

Introduction

Burglary has been described as a crime that has a powerful impact on its victims (Mawby et al., 1999). Burglary rates have declined slightly (Stepanek, 2011), but in Sweden the rate has remained relatively stable, with approximately 1 percent of surveyed households reporting a residential break-in per year according to the Swedish Crime Survey (BRÅ, 2016). The rate of break-ins in Sweden is somewhat higher in single-family housing than in multi-family housing, and is more common in the larger cities (BRÅ, 2016).

According to Morgan (2001), data reveal that a small number of houses, about 4 percent, accounted for 44 percent of all property crimes. Researchers have come to the conclusion that a between-places relationship must exist (Farrell and Pease, 1993; Polvi et al., 1991). Thus Braga et al. (2012) argue that burglary is not evenly spread across places; instead it is clustered in space and time. It is now well established that past victimization is a strong predictor of future criminal offences (Johnson and Bowers, 2005; Mawby, 2013). Research has highlighted an increased risk for the re-victimization of previously victimized properties, which peaks within 48 hours of the first attempt, and then lasts for nearly one month (Braga et al., 2012). The related concept of near-repeat victimization patterns has received increasing attention since the early 2000s. Near-repeat events are crimes that occur in close proximity in both time and space to an initial crime event of the same type, but without targeting the exact same victim as the first event (Weisel, 2005).

The present study focuses on near-repeat patterns in the city of Malmö, Sweden. This study is among the first to cover the topic in a Scandinavian context and thus represents a contribution to the international body of research in this area. In addition, the present study may contribute to our knowledge of yearly variations in near-repeat patterns because six years of burglary data are examined both in a combined analysis and separately for each year.

Methods, setting and data

The aim of the present study is to quantify the presence of near-repeat patterns of burglary in the city of Malmö, Sweden, and to consider whether there are differences in year-on-year patterns of near repeats.

Data

A complete list was provided of police-recorded home burglary data for the years 2009 to 2014. The list includes incidents relating to both single-family houses and multi-family housing. The data contain geographical information on 7257 burglary events in the form of x and y coordinates, as well as the start date of the incidents. ¹ The geocoding was performed by the police and, although geocoding can be problematic for some crime types (see Gerell, 2016, for a discussion relating to assault data), the reliability of such coding is likely to be better for burglary because each home has a specific address. Of the total number of incidents, 1364 events were attempted burglaries and 5893 were completed offences; 8 cases were excluded owing to the absence of geographical data. In addition, two incidents had been geocoded as lying outside municipal boundaries and were thus also excluded. In this study, the chosen point of comparison was the start date of an incident, and data were also available for end date and reporting date. Because many incidents are not discovered immediately, the start and end dates can be imprecise and represent only an estimation of the time of a burglary. Using the start date will tend to result in more incidents on Fridays and at the beginning of holidays, whereas using the end date results in more incidents on Sunday–Monday and at the ending of holidays. Either would be preferable to the reporting date, which would introduce an additional bias to the data through differences in the time taken to report the crime to the police. An overview of burglary rates per year is presented under Table 1. The burglary count was lowest in 2009, when a total of 923 incidents were reported. The highest number of annual incidents was recorded in 2012, at 1489.

OPEN IN VIEWER

Table 1.

Total burglary rates and the number of attempted vs. completed incidents per year.

Year	2009	2010	2011	2012	2013	2014
Total (N)	925	923	1264	1489	1398	1258
Attempted	175	172	233	266	258	260
Completed	750	751	1031	1223	1140	998

Burglaries are clustered in specific parts of the city, largely correlating with building and street network density (Figure 1). Figure 1 presents kernel density maps of burglary for each year, based on a 500 metre radius calculation performed in ARCGIS10. The figure includes separate frames for buildings, for streets and for two areas of the city that will be discussed later in the paper. There is some notable yearly variation, both in terms of the levels of density that are recorded (min = 89, max = 185), and in terms of specific places, which in several cases vary from mid–high to very high across different years.

OPEN IN VIEWER

Figure 1.

Malmö maps. Top row, Left: Street network (2011). Top row, middle: Buildings (2011). Top row, right: Two neighbourhoods with particular changes in burglary and near-repeat patterns. Middle and bottom row: Kernel density based on 500 metre radius for burglary in Malmö, 2009–14. Density visualized by within-year equal intervals.

Results

The results show the presence of repeat and near-repeat victimization patterns in Malmö municipality. Both in the data for each year and in the full dataset, the most overrepresented space–time range is 1–100 metres and 0–7 days. Interestingly, however, there are substantial year-on-year differences in the size of the observed OE ratios, and in several cases also with regard to which spatiotemporal distance bands were associated with significant levels of overrepresentation.

As regards same-location repeat incidents, very high overrepresentations were identified, although it should be noted that, in the case of multi-family housing, several apartments share the same coordinates. Thus, a same-location repeat might involve two different apartments having been burgled rather than a true repeat burglary of the same home, and both types of incident are included in the values for same-location repeats. Although these shortcomings should be noted, the high OE ratios noted for same location nevertheless shows that the risk of an additional burglary at an address is very high after an initial incident, and that such a risk is present for a full eight weeks.

In general, near repeats are identified within 1–100 metres and up to 42 days, 101–200 metres and up to 28 days, and 201–400 metres and up to 21 days, with a significance level of .001 (see Table 2). The overrepresentation of near repeats at larger spatial bands declines to 7 days. In the first band –1–100 metres and 0–7 days – the immediate space–time difference from an initial event is the highest, with a 183 percent higher risk than would be expected if there was no association between spatial and temporal distributions. As the space–time distance from the initial event increases, the OE ratio decays in a fairly consistent way. Using space and time bands that are twice as large yields substantially similar findings, and thus the results do not appear to be sensitive to the specifications employed here (see Appendix Table A2).

OPEN IN VIEWER

Table 2.

Observed/expected ratios for near-repeat patterns 2009–14 in the city of Malmö.

	0 to 7 days	8 to 14 days	15 to 21 days	22 to 28 days	29 to 35 days	36 to 42 days	43 to 49 days	50 to 56 days	57 to 63 days	64 to 70 days	> 70 days
Same location	9.07**	3.37**	2.27**	2.25**	1.93**	1.74*	1.55*	1.64*	1.38	1.01	0.89
1–100 m	2.83**	1.81**	1.67**	1.37**	1.31*	1.45**	1.13	1.21*	1.21*	1.21*	0.96
101–200 m	2.07**	1.45**	1.37**	1.22*	1.17*	1.24**	1.15*	1.24*	1.10*	1.14*	0.98
201–300 m	1.67**	1.24**	1.20**	1.09*	1.07	1.21**	1.02	1.06	1.14*	1.02	0.99
301–400 m	1.33**	1.21**	1.20**	1.17**	1.12*	1.09*	1.10*	1.03	1.09*	1.19**	0.99
401–500 m	1.22**	1.13*	1.14**	1.13*	1.04	1.07*	1.10*	1.12*	1.03	1.05	0.99
501–600 m	1.24**	1.00	1.05	1.02	1.06	1.05	0.98	0.96	0.98	1.05	1.00
601–700 m	1.20**	1.04	1.09*	1.09*	1.11*	1.03	1.04	1.02	1.02	0.98	1.00
701–800 m	1.08*	1.08*	1.11*	1.08*	1.06*	0.99	1.06	1.01	1.04	0.96	1.00
801–900 m	1.18**	1.10**	1.07*	1.10*	1.05	1.08	1.04	1.02	1.04	1.04	1.00
901–1000 m	1.16**	1.09*	1.09*	1.10**	1.04	1.04	1.07	1.11	1.01	1.03	1.00
> 1000 m	0.99	1.00	1.00	1.00	1.00	1.00	1.00	1.00	1.00	1.00	1.00**

**

p < .001; *p < .05.

Table 3 presents a summary of the year-on-year differences in observed OE near-repeat ratios (see Appendix Tables A2–A7 for yearly near-repeat patterns). As can be noted, the yearly fluctuations are fairly large, and this is an important finding that underscores the conclusion that while the basic finding that near-repeat patterns exist is valid, there is also large variability in the specifics of these patterns. Consider for instance the case of the 301–400 metre distance band and the 0–7 days temporal band, for which Table 3 shows OE values between 1.0 and 1.7. In relation to levels of significance, this means that near-repeat patterns are noted at the 99 percent level in 2009 and 2014, and at the 95 percent level in 2012, but that there are no significant associations in the years 2010, 2011 or 2013 (see Appendix for specifics).

OPEN IN VIEWER

Table 3.

Variation in observed/expected values for the years 2009–14 analysed separately.

	1st week, OE min–max (mean)	2nd week min–max (mean)	3rd weekmin–max (mean)	4th week, min–max (mean)	5th week, min–max (mean)
Same location	4.5–6.7 (5.5)	1.5–2.6 (2.0)	0.6–2.1 (1.2)	0.4–2.7 (1.2)	0.2–1.5 (1.0)
1–100 m	1.8–2.9 (2.3)	1.4–1.9 (1.5)	1.1–1.7 (1.4)	1.0–1.3 (1.1)	0.8–1.3 (1.1)
101–200 m	1.4–2.1 (1.9)	1.1–1.5 (1.4)	1.1–1.4 (1.3)	1.0–1.2 (1.1)	0.8–1.2 (1.1)
201–300 m	1.2–2.2 (1.6)	1.0–1.4 (1.1)	1.0–1.4 (1.1)	0.8–1.1 (1.0)	0.7–1.3 (1.0)
301–400 m	1.0–1.7 (1.3)	1.0–1.3 (1.2)	1.0–1.3 (1.2)	0.9–1.3 (1.1)	0.9–1.2 (1.1)
401–500 m	1.0–1.3 (1.1)	1.0–1.2 (1.1)	1.0–1.2 (1.1)	0.9–1.3 (1.1)	0.8–1.2 (1.0)

Discussion

The present study has consistently identified near-repeat patterns of burglary in all seven datasets tested, and the concept of near repeats thus appears to be of importance for our understanding of burglary in Malmö, Sweden. The strongest near-repeat patterns were noted for short time and space distances, and this finding was stable across all datasets. In particular, repeat incidents at the same location are registered for very high OE values, suggesting that repeat burglary at the same address is of great importance. There are, however, important variations to take into consideration when less stable associations are considered. Substantial differences in relative risk were noted between different years, and in many cases this also resulted in non-significant associations for some years while other years show highly significant results for the same spatiotemporal bands. The full dataset with all six years of burglaries has a great deal more statistical power compared with the one-year datasets and thus paints the most reliable picture, but it is still important to note that this picture may obscure year-on-year differences that could yield a more nuanced understanding of the near-repeat phenomenon. Although some of these differences between years are to be expected as normal fluctuations and/or the result of error, we argue that the most pronounced differences are likely to be related to substantial changes in the spatiotemporal patterns of burglary.

Some cases in point can illustrate why such differences may occur. One distinctive burglary cluster was noted in the Triangeln area in 2010 (see Figure 1, top right). In this year, the construction work on the Triangeln Station was completed, which may have led to higher levels of mobility in the area and possibly to new potential offenders incorporating it into their awareness space. Davies and Bishop (2013) emphasize that a burglary cannot be seen as an isolated incident, but rather is a crime that has to be analysed in the context of the environmental circumstances of a neighbourhood. It is possible that an increased flow of people through the area, and the associated changes in routine activities, may have had an impact on the occurrence of burglary. Researchers explain differences in burglaries and near-repeat levels as being due to seasonal factors and urban background influences, such as available transportation, housing density or social factors (Davies and Bishop, 2013; Johnson et al., 2007). But, although the case for changes in burglary rates in Triangeln area related to the new train station is fairly straightforward, it is less clear how and if this would impact on near-repeat patterns. Possibly a change in the awareness space of offenders took place with the new train station that adapted to new opportunities faster than non-offenders could respond to them, but there is no research to substantiate such a process. It can, however, be noted that the area currently has fences/walls around the residential blocks, which were not there at the time the train station was inaugurated.

Another case in point can be noted for the years 2013 and 2014, when near-repeat patterns with higher effect sizes were noted in the first couple of time bands. Some near-repeat burglaries in these two years may be linked to reconstruction efforts in the neighbourhood of Kroksbäck (highlighted in Figure 1, top right). During the winter of 2013–14 more than 50 burglaries were noted in this neighbourhood alone, although not all were reported to police. This is likely to have been related to the presence of scaffolding, which provided potential burglars with easy access to apartments for a limited time period, thus explaining the surge in burglaries that occurred (Gerell and Hallin, n.d.). If such a relationship between reconstruction and an increase in near-repeat burglary indeed exists, it may be considered as providing support for the flag account of near-repeat patterns (for example, Johnson, 2008). The neighbourhood temporarily exhibited an increased vulnerability to burglary as a result of the improved access to apartments provided by the scaffolding, and the neighbourhood was thus flagged as vulnerable, resulting in an increase in the number of near repeats. Possibly this could also be related to some sort of process where information spreads to offenders regarding such a vulnerability. Potential offenders living in the neighbourhood would be aware of such a vulnerability, and could possibly spread their knowledge further to other potential offenders. Such findings would, however, need to be substantiated with additional research before any firm conclusions can be drawn.

In comparison with other studies (Chainey and da Silva, 2016; Johnson et al., 2007;Townsley et al., 2003; Wells et al., 2012), near-repeat burglary patterns in Malmö are generally quite durable, which adds to the complexity and cross-contextual variation that has previously been identified in the research (Johnson et al., 2007). In their comparison of 10 cities across 5 nations, Johnson et al. (2007) noted that there was some consistency between contexts, but also substantial differences. In the case of Philadelphia, USA, significant near-repeat patterns were identified within 200 metres and 8 weeks, and in the case of Canberra, Australia, patterns were noted within 1200 metres and 4 weeks. The present study yields similar results regarding differences for a within-city comparison over time.

Chainey and da Silva (2016) investigated 19,453 domestic burglary incidents in a four-year time period in Belo Horizonte, Brazil. For four out of five distance bands, they found significant overrepresentations up to three weeks after an initial incident, and for one out of five distance bands there was a significant overrepresentation in the fourth week after initial incident. In the case of Malmö, significant overrepresentations were found across all 10 time bands, which is an interesting difference to note.

Chainey and da Silva (2016) themselves argue that the results are caused by a different housing infrastructure in Belo Horizonte. Compared with western cultures, residents live in high-rise apartment buildings or in ‘irregular self constructed houses’ (Chainey and da Silva, 2016: 1). Furthermore, the lower level of repeats and near repeats can be explained by reference to lower levels of opportunity. Buildings, with the exception of favelas, have inbuilt situational crime prevention factors, such as guards or high perimeter fencing. In favelas, however, social capital, for example in the form of networks, together with proximate living, leads to a stronger community and to collective efficacy (Sampson et al., 1997), which is a restraining factor for burglars (Chainey and da Silva, 2016). The analysis thus emphasizes the power of environmental influences on burglaries.

Although it is viable to explain the presence of, and variations in, near-repeat patterns on the basis of environmental theories of criminology, no firm conclusions can be drawn based on the current study. However, the findings presented here show that future investigations should focus on year-on-year differences to untangle whether, and how, near-repeat patterns may be related to routine activities shaped by the developing structure of a city.

As argued by Haberman and Ratcliffe (2012), analysis of near repeats may be of importance for the development of a predictive technique to forecast future burglaries. Since future burglary risk is highly dependent on past victimization, it is essential to develop preventive methods that focus on previous victims and near repeats (Bowers and Johnson, 2005; Hirschfield et al., 2010; Pease, 1998). One step in this direction has already been taken. In some countries, for example the USA and the UK, police patrol cars work with ‘predictive policing’ methods and ETAS crime forecasting models, which have been claimed to decrease the crime volume by 7.4 percent (Mohler et al., 2015).

Owing to stronger and more stable effect sizes within the first week and 200 metres, as well as within two weeks and 100 metres after an initial burglary, specific strategies focused on this time frame are recommended (Bowers and Johnson, 2005). It may be particularly advisable for the police to act immediately, for instance by using different forms of patrolling or monitoring in combination with attempts at involving both the area’s residents and other key actors. First of all, information regarding security precautions such as closing windows and locking doors when leaving the house can be emphasized when the risk of burglary is likely to be elevated. Residents in neighbourhoods where a lot of burglaries occur can also be notified to keep an eye open and implement neighbourhood watch and asked to emphasize mutual caretaking (Mawby et al., 1999). The police, as a central actor, need to support residents and resident groups such as neighbourhood watch associations by means of information, advice and other forms of support. Warning signs, as well as flyers on how to protect homes, can be distributed. In some countries, for example Germany, the police offer free counselling and discount agreements with insurance companies or banks (loans) in order to support target hardening efforts. Further, since significant near-repeat patterns have been detected at greater distances, it is important to implement follow-up police investigations, in addition to patrolling, in order to make residents feel safer. Communication processes between police departments need to be further improved in order to apprehend suspects more quickly (Dreißigacker et al., 2015). The significant differences between years also highlight a need for the police to gain good knowledge of local communities in order to be able to identify increased risks of burglary and to develop prevention efforts accordingly. If increased vulnerabilities to crime are temporary, such as with the case of scaffolding in the Kroksbäck neighbourhood, it becomes important to act on them quickly in order to be able to prevent crimes related to such vulnerabilities.

Conclusion

The present study contributes to academic knowledge on near repeats primarily in two ways. Firstly, it provides peer reviewed evidence from a new geographical context, Sweden. Secondly, it provides new insights into annual variations in near-repeat patterns that may help inform future research and prevention efforts.

The aim of the study was to examine near-repeat burglary victimization patterns in Malmö. As expected, in all datasets the risk of a near-repeat victimization peaked within the first week and 1–100 metres of a previous break-in. The data revealed that the immediate space–time difference within the first week was 183 percent greater than would be the case in the absence of a pattern of near-repeat victimization.

The analysis of differences between years revealed that, although some core patterns are stable across all years, there are also substantial year-on-year fluctuations. This points to a need for further studies into stability and change in near-repeat patterns, and for more empirical work focused on explaining why such changes may occur. This may also be an important consideration for crime prevention practitioners. Crime prevention should focus on the shortest distances from an initial incident in both time and space because the patterns here are more stable, but additional efforts could also be implemented at greater time–space distances if these are informed by some form of analysis showing that near-repeat patterns may be extended as a result of shifts in routine activities or other changes in the relevant environment.