Content uploaded by Ray Bull
Author content
All content in this area was uploaded by Ray Bull on Mar 16, 2015
Content may be subject to copyright.
This article was downloaded by: [86.159.162.88]
On: 18 February 2015, At: 02:46
Publisher: Routledge
Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered
office: Mortimer House, 37-41 Mortimer Street, London W1T 3JH, UK
Click for updates
Psychology, Crime & Law
Publication details, including instructions for authors and
subscription information:
http://www.tandfonline.com/loi/gpcl20
Testing the impact of group offending
on behavioural similarity in serial
robbery
Amy Burrella, Ray Bullb, John Bondc & Gary Herringtond
a Division of Psychology, Birmingham City University, Birmingham,
UK
b School of Law and Criminology, University of Derby, Derby, UK
c Department of Chemistry, University of Leicester, Leicester, UK
d West Midlands Police (Retired), Birmingham, UK
Accepted author version posted online: 17 Dec 2014.Published
online: 19 Jan 2015.
To cite this article: Amy Burrell, Ray Bull, John Bond & Gary Herrington (2015): Testing the impact
of group offending on behavioural similarity in serial robbery, Psychology, Crime & Law, DOI:
10.1080/1068316X.2014.999063
To link to this article: http://dx.doi.org/10.1080/1068316X.2014.999063
PLEASE SCROLL DOWN FOR ARTICLE
Taylor & Francis makes every effort to ensure the accuracy of all the information (the
“Content”) contained in the publications on our platform. However, Taylor & Francis,
our agents, and our licensors make no representations or warranties whatsoever as to
the accuracy, completeness, or suitability for any purpose of the Content. Any opinions
and views expressed in this publication are the opinions and views of the authors,
and are not the views of or endorsed by Taylor & Francis. The accuracy of the Content
should not be relied upon and should be independently verified with primary sources
of information. Taylor and Francis shall not be liable for any losses, actions, claims,
proceedings, demands, costs, expenses, damages, and other liabilities whatsoever or
howsoever caused arising directly or indirectly in connection with, in relation to or arising
out of the use of the Content.
This article may be used for research, teaching, and private study purposes. Any
substantial or systematic reproduction, redistribution, reselling, loan, sub-licensing,
systematic supply, or distribution in any form to anyone is expressly forbidden. Terms &
Testing the impact of group offending on behavioural similarity
in serial robbery
Amy Burrell
a
*, Ray Bull
b
, John Bond
c
and Gary Herrington
d
a
Division of Psychology, Birmingham City University, Birmingham, UK;
b
School of Law and
Criminology, University of Derby, Derby, UK;
c
Department of Chemistry, University of Leicester,
Leicester, UK;
d
West Midlands Police (Retired), Birmingham, UK
(Received 22 July 2013; accepted 11 December 2014)
Behavioural case linkage assumes that offenders behave in a similar way across their
crimes. However, group offending could impact on behavioural similarity. This study
uses robbery data from two police forces to test this by comparing the behavioural
similarity of pairs of lone offences (LL), pairs of group offences (GG) and pairs of
offences where one crime was committed alone and the other in a group (GL).
Behavioural similarity was measured using Jaccard’s coefficients. Kruskal–Wallis tests
were used to examine differences between the three categories within the linked
samples. No statistically significant differences were found for linked GG compared to
linked LL pairs. However, differences emerged between GL and the other categories
for some behaviours (especially control) suggesting caution should be applied when
linking group and lone offences committed by the same perpetrator. Differences
between linked and unlinked pairs were assessed using receiver operating character-
istic. The results suggest it is possible to distinguish between linked and unlinked pairs
based on behaviour especially within the GG and LL categories. There were, however,
fewer significant findings for the GL sample, suggesting there may be issues linking
crimes where the offender commits one crime as part of a group and the other alone.
Keywords: case linkage; behavioural similarity; group offending; serial; robbery
Introduction
In the UK, robbery is defined as the theft of property with the threat or use of force
against a person. This includes where the victim resists or where anyone is assaulted, or if
the victim feels the offender might use force due to their language or actions. Where force
is targeted at the property (as in snatching a handbag, wallet or mobile phone) rather than
the person, the offence is classified as theft from the person rather than robbery (Home
Office, 2012).
Group crimes are offences committed by two or more offenders against one or more
victims. The prevalence of group offending varies by type of offence (Alarid, Burton, &
Hochstetler, 2009; Deakin, Smithson, Spencer, & Medina-Ariza, 2007; Erikson, 1971;
Hindelang, 1976; Hochstetler, 2001; Weerman, 2003) and is more common in predatory
street crimes such as robbery compared to other offence types such as sex offences and
fraud (van Mastrigt & Farrington, 2009). In fact, the majority of robberies are committed
by groups (e.g., Kapardis, 1988; Walsh, 1986); a phenomenon that is not surprising given
*Corresponding author. Email: amy.burrell@bcu.ac.uk
Psychology, Crime & Law, 2015
http://dx.doi.org/10.1080/1068316X.2014.999063
© 2015 Taylor & Francis
Downloaded by [86.159.162.88] at 02:46 18 February 2015
that robbery benefits from a division of labour more than other offence types (van
Mastrigt & Farrington, 2009). Research on group offending has highlighted a number of
characteristics that differ between group and lone offending.
Characteristics of group and lone offending
Group offences are more likely than lone offences to involve some level of planning
(Alarid et al., 2009). This makes sense for crime types where individual members of the
group will be assigned roles (e.g., commercial robbery). Even in more spontaneous
crimes (e.g., personal robbery) the offenders may need to discuss, however briefly, the
method of approach. This is in contrast to the lone offender who only needs to consider
his/her own actions to commit the crime.
Group offenders’victims tend to be younger than victims of lone offenders (e.g.,
Lloyd & Walmsley, 1989; Morgan, Brittain, & Welch, 2012). Groups have been found to
target lone victims –for example, Porter and Alison (2004) reported that 87% of group
rapes (194 out of 223) were against a lone victim –but groups are also more likely to
attack multiple victims than a lone offender (Alarid et al., 2009; Hauffe & Porter, 2009).
The latter perhaps is not surprising as a group allows victims to be controlled more easily.
However, Alarid et al. (2009) found no significant differences between how groups and
lone offenders selected victims indicating that there are likely to be other factors also
influencing victim selection. For example, offenders may respond to a spontaneous
opportunity or the offence could be targeted against a particular person (e.g., as a means
of debt collecting or gang-related).
The group context encourages violence (Morgan et al., 2012) and group offenders
commit more violent offences than do lone offenders (Conway & McCord, 1995, cited in
Conway & McCord, 2002). Group offences are more likely to involve physical violence
than lone offences (Alarid et al., 2009; Porter & Alison, 2006a,2006b; Woodhams,
Gillett, & Grant, 2007), and young offenders are more likely to behave violently (e.g.,
punching and kicking) towards the victim(s) when committing a crime with others than
when offending alone (Conway & McCord, 1995, cited in Conway & McCord, 2002).
Furthermore, group offences are more likely to involve multiple acts of violence during
the event (e.g., Hauffe & Porter, 2009). With regard to injury, Alarid et al. (2009) reported
that the probability of (robbery) victims receiving a slight injury was comparable across
group and lone offences, but that group offences were associated with all of the serious
injuries sustained by victims in their sample.
Group offenders are less likely to use weapons than lone offenders (Lloyd &
Walmsley, 1989) suggesting there are different methods of controlling victims. Group
offenders, on the one hand, have strength in numbers which can be used to control the
victim (Porter & Alison, 2006b), if only through intimidation rather than physical
violence. It may not, therefore, be surprising to learn that victims (of rape) are less likely
to resist against a group of offenders (Hauffe & Porter, 2009). The lone offender, on the
other hand, is more likely to need a weapon to achieve the same level of control, and as
such, the weapon could be a substitute for an accomplice (Alarid et al., 2009).
Group offending and case linkage
The theoretical assumptions for case linkage are (1) behavioural consistency –that offenders
behave consistently across their crime series and (2) behavioural distinctiveness –that
2A. Burrell et al.
Downloaded by [86.159.162.88] at 02:46 18 February 2015
offenders are sufficiently heterogeneous from each other for series committed by different
offenders to be separated from one another. Offenders must therefore commit crime in a
consistent but distinctive manner in order for case linkage to be feasible (Santtila, Junkkila, &
Sandnabba, 2005). The impact of group dynamics on behavioural consistency and
distinctiveness is as yet untested in the case linkage literature. However, research on co-
offending has found that group offences are more likely to be planned (Alarid et al., 2009), to
target multiple victims (Hauffe & Porter, 2009), and be more violent than lone offences
(Porter & Alison, 2006a,2006b; Woodhams, Gillett, et al., 2007). This suggests that the
crimes an offender commits with a group might differ from those they commit alone,
potentially reducing behavioural similarity between cases and thus making their crimes more
difficult to link.
Methodology
Sample
The data for this study were extracted from police records for solved personal robbery
offences for two UK police forces: Northamptonshire (the third most rural police force)
and West Midlands (the second most urban police force; Bond, 2012).
The Northamptonshire data-set comprised 160 offences committed by 80 offenders
between 1 January 2005 and 31 December 2007. Seventy-four offenders were male, and
five were female (the gender was recorded as unknown for one offender). The offenders
were aged between 10 and 44 years with an average age of 18 at the time of the offence.
Females were a little older than males (mean = 23, range = 12–44 compared to mean =
28, range = 10–41 for males). Over 70% (n= 55) of the offenders were recorded as being
White (including four out of five of the females), 13 were recorded as Black and 12
(including one female) of mixed heritage.
The West Midlands data-set comprised 554 offences committed by 277 offenders
between 1 April 2007 and 30 September 2008. The majority of offenders were male
(n= 258 or 93%). The offenders were aged between 11 and 45 years with an average age
of 19 at the time of the offence. Females were slightly younger than males (mean = 16,
range = 12–24 years compared to mean = 19, range = 11–45 years for males). Almost
half of the offenders (n= 138) were recorded as being from a Black background
(including nine females). Just under 30% were White (n= 78; of which eight were
female) and 15% (n= 42) were recorded as Asian. Less than 1% (n= 2) were recorded as
being from a mixed or other minority background. Ethnicity was unknown in 17 (6%) of
cases (of which two were female).
1
Procedure
Selecting linked pairs
All offenders who had committed two or more recorded offences in the respective time
frames were identified; 135 in Northamptonshire and 438 in the West Midlands. The two
most recent offences for each offender were used to create a linked offence pair [this
mirrors the approach used by other case linkage researchers (e.g., Woodhams & Toye,
2007)]. However, there were some cases where the two most recent offences could not be
used for fear of compromising the independence of the data-sets. This is because the
Home Office Counting Rules (Home Office, 2012) state that a separate crime should be
recorded by the police for each victim rather than each incident, and so a single incident
Psychology, Crime & Law 3
Downloaded by [86.159.162.88] at 02:46 18 February 2015
can result in multiple offences being recorded if there is more than one victim. There
were cases in both data-sets where the date, time and location of offences were identical
and the modus operandi information suggested that the two most recent offences were
actually part of the same incident. To include such pairs in the analysis would falsely
inflate the level of similarity in linked pairs. Therefore, data for 19 offenders from
Northamptonshire and 70 from West Midlands were removed from the analysis.
In a similar vein, a further 21 offenders were omitted in Northamptonshire and 91
from West Midlands because one or both of the offences associated with the offender
already appeared in the respective data-sets as part of the crime series of another offender
(i.e., their co-offender) and so the inclusion of the pair would again compromise the
independence of the sample. A further 15 offenders were excluded from the Northamp-
tonshire data-set due to missing data about their offences.
The final linked samples contained the two most recent offences (that were not part of
the same incident) for 80 offenders from Northamptonshire and 277 offenders from the
West Midlands.
Selecting unlinked pairs
The current research mirrors previous case linkage research utilising an unlinked sample
with the same number of pairs as the linked sample (e.g., Markson, Woodhams, & Bond,
2010; Tonkin, Grant, & Bond, 2008; Tonkin, Woodhams, Bull, Bond, & Palmer, 2011).
The unlinked pairs were generated using the =RAND() function in Microsoft Excel to
randomly reorder the rows within each linked sample. The unlinked pairs were created
using rows 1 and 2 as unlinked pair 1, rows 3 and 4 as unlinked pair 2, and so on. The
data were checked manually to ensure that all the unlinked pairs were indeed unlinked
and two linked crimes were not randomly reassigned back together.
Identifying group and lone offences
The data for both police forces included a variable relating to the number of defendants/
offenders involved in the crime. However, it is likely that this information under-
represents group offending as there were cases where only one offender in a group was
identified. Therefore, for the purposes of the current research, group and lone offences
were identified by the first author using the modus operandi information.
In Northamptonshire of the 160 robberies, 104 (65%) were committed by groups and
56 (35%) by lone offenders. The ratio of group versus lone offending was similar in the
West Midlands, with 68% of robberies (377 out of 554 cases) identified as group crimes
and 32% as lone offences (177 out of 554 cases).
Crime pairs were split into the three categories for analysis: (1) crime pairs where the
offender committed both offences as part of a group (labelled GG), (2) crime pairs where
both offences were committed by the same lone offender (labelled LL) and (3) crime
pairs where the offender committed one offence as a part of a group and one alone
(labelled GL). Table 1 shows how many linked and unlinked pairs fell into each category
for the two police forces.
Identifying crime behaviour
The linked pairs were identified based on the criminal history of individual offenders and
the classification of group/lone was based on whether the offender in question committed
their offence as part of a group or not. It was not possible to isolate which actions or
4A. Burrell et al.
Downloaded by [86.159.162.88] at 02:46 18 February 2015
behaviours were committed by which offenders (e.g., who determined the timing/location
of the robbery or who committed violent acts), and so the behaviours identified in group
robberies were associated with the offence rather than with an individual offender.
Therefore, the research focuses on behaviours associated with group robberies compared
to behaviours associated with lone robberies rather than considering the roles or actions
of any individual group member.
A description of how the offence was reported to have been committed (i.e., the
modus operandi) was included in the police records. Content analysis of these
descriptions was conducted and a checklist of dichotomously coded behaviour variables
created. Binary coding, where 1 denoted the presence of a behaviour and 0 the absence of
a behaviour, was used because previous research has indicated that more complex coding
methods are difficult to apply to police data in a reliable way (Canter & Heritage, 1990).
Two people independently coded the modus operandi data into dichotomous variables
(for 10% of the overall samples) and their level of agreement was assessed using kappa.
A total of 15 modus operandi behavioural variables, each of which had a very good
overall inter-coder reliability score (κ= 0.95; range = 0.81–1.00 for individual beha-
viours), were selected for inclusion in this study. These variables were combined with
other variables extracted from the recorded crime data (e.g., time of day, day of week,
property stolen, the distance between offences) to form a final ‘behaviour’checklist of 48
behaviours (see Appendix).
Behavioural domain formation
Individual offence behaviours can be arranged into clusters, each thought to serve a
different purpose in the offence (Tonkin et al., 2008). For example, weapon use and
threatening language are both examples of how to seek to control victims during an
offence. Thus, the behaviours were grouped into behavioural clusters or domains for
analysis.
Atarget selection domain was developed that was formed of 16 variables relating to
the day of week and time of day of the offence, whether the offender was known to the
victim and whether the victim was at a cashpoint at the time of the offence. The control
domain included 15 variables relating to weapon use (e.g., whether a weapon was present
during the offence), violent actions, offender commands, and whether the victim and/or
offender were alone or part of a group when the offence occurred. The property domain
contained 14 types of property stolen plus whether any property was returned to the
victim by the robber(s) during or following the offence.
Temporal proximity (i.e., the number of days between offences) and inter-crime
distance (calculated using Pythagoras’theorem on the six-digit geographic coordinates
Table 1. Frequency of GG, LL and GL pairs in the four samples.
Northamptonshire West Midlands
Pair consists of Linked N(%) Unlinked N(%) Linked N(%) Unlinked N(%)
Two group offences (GG) 38 (47.5) 34 (42.5) 165 (59.6) 130 (46.9)
Two lone offences (LL) 14 (17.5) 10 (12.5) 65 (23.5) 30 (10.8)
One group/one lone (GL) 28 (35.0) 36 (45.0) 47 (17.5) 117 (42.2)
Total 80 (100) 80 (100) 277 (100) 277 (100)
Psychology, Crime & Law 5
Downloaded by [86.159.162.88] at 02:46 18 February 2015
for each offence) were also included in the analysis. These were included because they
have proved to be useful predictors of linkage in previous research (e.g., Tonkin, Santtila,
& Bull, 2012). Furthermore, an analyst survey (Burrell & Bull, 2011) revealed that the
majority of analysts (15 out of 18) use spatial and temporal behaviours to support linkage
decisions. It is therefore important to test the usefulness of these variables.
Measuring behavioural similarity
The similarity of pairs across each behavioural domain was measured using Jaccard’s
coefficients. These do not take joint non-occurrences into account (Porter & Alison, 2004,
2006a; Real & Vargas, 1996) and therefore the level of similarity does not increase if the
behaviour is not reported to have occurred within an offence pair (Woodhams, Grant, &
Price, 2007). This is an important issue when working with police data as the absence of a
behaviour does not necessarily mean that the behaviour did not occur (Porter & Alison,
2004,2006a), but perhaps that it was not reported or was not recorded (Tonkin et al., 2008).
Jaccard’s coefficients are expressed as a value of between 0 and 1, with 0 indicative
of no similarity and 1 denoting perfect similarity. Jaccard’s coefficients were calculated
using the Statistical Package for the Social Sciences (SPSS) version 18.0© (IBM
Corporation, NY, USA). SPSS calculates the similarity of pairs of offences based on the
binary coding of behaviours input into the analysis producing a matrix containing the
Jaccard’s coefficients for all possible pairings of offences in the data-set. Jaccard’s
coefficient matrices were produced for each behavioural domain (i.e., target selection,
control and property). The relevant Jaccard’s coefficients for each domain were then
manually extracted from each matrix for each linked pair in the LL, GG and GL samples
(i.e., the Jaccard’s coefficients for target selection for LL pair 1, the Jaccard’s coefficients
for target selection for LL pair 2, etc.). This was repeated for the unlinked pairs. All other
Jaccard’s coefficients were excluded from the analyses. The Jaccard’s coefficients for
each domain plus the variables temporal proximity and inter-crime distance formed the
data-set for the next stage of analysis.
Comparing behavioural similarity of GG, LL and GL linked pairs
In the first phase of the study, we explored how behavioural similarity might be
influenced by group offending by comparing the average Jaccard’s scores, temporal
proximities and inter-crime distances of the three categories (GG, LL and GL). The data
were not normally distributed (contact the first author for details of the Kolmogorov–
Smirnov outcomes) and were considered to be independent. The dependent variable
(group/lone) has three categories, therefore, a Kruskal–Wallis test [a non-parametric
version of the analysis of variance (ANOVA)] was performed to assess whether there
were any statistically significant differences between the three categories. As with the
one-way ANOVA, the Kruskal–Wallis test can determine if there is a difference between
categories but does not identify where differences lie. Therefore, post hoc tests are
needed, in this case the Mann–Whitney Utest. To allay concerns about increasing the risk
of Type I errors (i.e., identifying a significant difference where there is not one) through
repeated Mann–Whitney Utests (Field, 2005), the Bonferroni correction is used to adjust
the critical value for significance. This is achieved by dividing the critical value (0.05) by
the number of tests conducted (in this case three). This means any pvalue of 0.0167
(i.e., 0.05/3) or below is considered to be significant to the p< 0.05 level for the purposes
of the Mann–Whitney Uanalysis in this case.
6A. Burrell et al.
Downloaded by [86.159.162.88] at 02:46 18 February 2015
Comparing the behavioural similarity of linked versus unlinked pairs
The second phase of the study compared the behavioural similarity of linked and
unlinked pairs in each category (GG, LL and GL) to determine if there are any
significant differences between them. Analysis was conducted on each behavioural
domain independently using receiver operating characteristic (ROC). ROC produces a
measure of discrimination accuracy called theareaunderthecurve(AUC)which,in
this study, indicated how well linked crime pairs can be distinguished from unlinked
crime pairs using Jaccard’s coefficients, temporal proximity and inter-crime distance.
An AUC of 0.5 indicates chance level and an AUC of 1.0 indicates perfect
discrimination, meaning the larger the AUC, the higher the predictive accuracy
(Woodhams, Bull, & Hollin, 2007). In each analysis the test direct was selected
according to how the data were coded (i.e., a larger value indicates a more positive
result for Jaccard’s coefficients and a smaller value indicates a more positive result for
inter-crime distance and temporal proximity). The state variable in all analyses was
linkage status with a value of 1 (denoting linked). AUCs of between 0.5 and 0.7 are
indicative of low levels of accuracy, 0.7 and 0.9 indicate moderate levels of accuracy
and 0.9 and 1.0 high levels (Bennell & Jones, 2005). The ROC analysis was also
conducted using SPSS.
Hypotheses
For the first phase of the study, it is hypothesised that there will be no difference in the
level of behavioural similarity between GG and LL because groups behave in a
homogenous way across offences (Porter & Alison, 2006a) and so will not differ from
lone offenders in terms of behavioural consistency (tested using Kruskal–Wallis).
However, where one offence was committed by the offender on their own and the other
as part of a group (i.e., GL pairs) there will be less behavioural similarity, consistent with
evidence offenders behave differently when they are working alone than when they
offend in a group (e.g., Alarid et al., 2009; Porter & Alison, 2006b).
For the second phase of the study, it is hypothesised that linked pairs in each category
will have higher levels of behavioural similarity than unlinked pairs (tested using ROC
analysis). Such results would provide evidence for the behavioural assumptions of case
linkage.
Results
Before outlining the outcomes of the Kruskal–Wallis tests and ROC analyses, descriptive
statistics are presented which provide an indication of trends.
Descriptive statistics
Since the data were not normally distributed the median rather than mean scores should
be used to compare the behavioural similarity of each domain. Table 2 shows the median
scores for linked and unlinked pairs in the three group/lone categories for each
behavioural domain for Northamptonshire.
These data indicate that linked pairs of offences have shorter inter-crime distances,
fewer days between offences and higher Jaccard’s coefficients for target selection than
Psychology, Crime & Law 7
Downloaded by [86.159.162.88] at 02:46 18 February 2015
unlinked pairs across all categories and for control in the LL group. However, there is a
higher median Jaccard’s coefficients for control in the unlinked GG sample compared to
the linked GG sample. The Jaccard’s coefficients for the property domain are low in all
categories whether the crimes are linked or unlinked.
Focusing in on the linked pairs only, these data suggest that there may be some
differences between categories for some domains. Most notably, GL pairs had larger
inter-crime distances and more days between offences than GG and LL pairs. There were
also notable differences between median scores for the control domain across the three
categories.
Table 3 shows the median Jaccard’s coefficients, temporal proximities and inter-crime
distances for linked and unlinked pairs in the three group/lone categories for each
behavioural domain for the West Midlands.
The overall trends mirror the results for Northamptonshire with linked pairs in all
categories having smaller inter-crime distances, fewer days between offences and larger
median Jaccard’s coefficients for target selection than unlinked pairs in that category.
Furthermore, linked pairs in all categories displayed higher median Jaccard’s coefficients
for control compared their corresponding unlinked pairs. As in Northamptonshire, the
Jaccard’s coefficients for property are low across the board.
With regard to the linked pairs, it can be seen that, in the West Midlands, GG pairs
displayed smaller inter-crime distances than LL and GL pairs. There were differences
between all categories for temporal proximity but it is unclear at this stage whether this
difference is likely to be significant given the overall number of days between offences
was low for all categories. The GL category had lower median similarity scores for target
selection and control domains.
Table 2. Median scores for behavioural domains in Northamptonshire.
Linked Unlinked
Behavioural
domain
All
pairs
Two
group
offences
(GG)
Two
lone
offences
(LL)
One
group/
one
lone
(GL)
All
pairs
Two
group
offences
(GG)
Two
lone
offences
(LL)
One
group/
one
lone
(GL)
Inter-crime
distance
(m)
803.8 788.6 741.6 1169.7 14,198.4 9891.41 14,187.2 15,125.1
Temporal
proximity
(days)
34.5 7 16 87 295.5 283.5 351.0 309.5
Target
selection
.225 .250 .225 .200 .200 .100 .200 .100
Control .250 .286 .429 .000 .167 .333 .268 .000
Property .000 .000 .000 .000 .000 .000 .000 .000
Number of
pairs
80 38 14 28 80 34 10 36
8A. Burrell et al.
Downloaded by [86.159.162.88] at 02:46 18 February 2015
Kruskal–Wallis test
Table 4 reveals the outcomes of the statistical tests for Northamptonshire.
A Kruskal–Wallis test revealed a significant difference in relation to temporal
proximity [χ
2
(2) = 6.304; p= 0.043]. Post hoc tests –Mann–Whitney Utests (with
Bonferroni correction) –showed that there was only one significant difference between
categories, that is between GL and GG (p= 0.014; r= .30).
A Kruskal–Wallis test also revealed a significant difference in relation to the
behavioural similarity of control behaviours [χ
2
(2) = 21.384; p< 0.001]. The post hoc
tests showed the significant differences to be between categories GL and GG (p< 0.001;
r= .43) and GL and LL (p< 0.001; r= .69).
Table 5 reveals the statistical test outcomes for West Midlands.
A Kruskal–Wallis test revealed significant differences between categories in relation
to target selection [χ
2
(2) = 6.342; p= 0.042]. The only significant difference between
categories was between GG and GL; however, the effect size was small (p= 0.017; r=
.16). As in Northamptonshire, the Kruskal–Wallis test also revealed significant
differences between categories for control [χ
2
(2) = 34.043; p< 0.001], and again the
differences were significant between GL and GG (p< 0.001; r= .39) and GL and LL (p<
0.001; r= .46).
ROC analyses
Table 6 shows the ROC results for Northamptonshire.
The results of the ROC analysis indicate that inter-crime distance is the single most
useful behaviour for distinguishing between linked and unlinked pairs of offences. This
was true regardless of the category being tested with an AUC of .940 for the GG/GG
Table 3. Median scores for behavioural domains for West Midlands.
Linked Unlinked
Behavioural
domain
All
pairs
Two
group
offences
(GG)
Two
lone
offences
(LL)
One
group/
one
lone
(GL)
All
pairs
Two
group
offences
(GG)
Two
lone
offences
(LL)
One
group/
one
lone
(GL)
Inter-crime
distance
(m)
608.6 475.5 852.1 893.9 10,356.5 10,387.8 13,244.4 9840.1
Temporal
proximity
(days)
1 0 2 4 150 138 135 181
Target
selection
.500 .500 .500 .333 .000 .000 .000 .200
Control .333 .429 .429 .143 .143 .211 .167 .000
Property .000 .000 .000 .000 .000 .000 .000 .000
Number of
pairs
277 165 65 47 277 130 30 117
Psychology, Crime & Law 9
Downloaded by [86.159.162.88] at 02:46 18 February 2015
Table 4. Statistical test outcomes for Northamptonshire.
Mann–Whitney Upost hoc test
Kruskal Wallis GG vs. LL GG vs. GL LL vs. GL
Behavioural domain χ
2
(df) U(z)rU(z)rU(z)r
Inter-crime distance 3.189 (2) 259.500 (.134) .02 382.500 (1.738) .21 147.500 (1.141) .18
Temporal proximity 6.304 (2)* 234.000 (.670) .09 343.000 (2.470)* .30 145.500 (1.350) .21
Target selection 2.733 (2) 264.000 (.042) .01 417.000 (1.533) .19 151.000 (1.237) .19
Control 21.384 (2)* 207.500 (1.215) .17 269.500 (3.507)* .43 31.500 (4.500)* .69
Property .779 (2) 257.000 (.282) .04 485.000 (.879) .11 184.500 (.681) .06
*p< 0.05.
10 A. Burrell et al.
Downloaded by [86.159.162.88] at 02:46 18 February 2015
Table 5. Statistical test outcomes for West Midlands.
Mann–Whitney Upost hoc test
Kruskal Wallis GG vs. LL GG vs. GL LL vs. GL
Behavioural domain χ
2
(df) U(z)rU(z)rU(z)r
Inter-crime distance 4.584 (2) 4599.000 (1.641) .11 3220.000 (1.744) .12 1469.000 (.346) .03
Temporal proximity 2.489 (2) 5010.000 (.833) .05 3354.500 (1.507) .10 1415.000 (.689) .07
Target selection 6.342 (2)* 4750.000 (1.385) .09 3015.500 (2.383)* .16 1354.500 (1.035) .10
Control 34.043 (2)* 5237.000 (.277) .02 1798.500 (5.634)* .39 701.500 (4.912)* .46
Property 1.254 (2) 5177.500 (1.057) .04 3874.500 (.023) .02 1472.500 (.773) .06
*p< 0.05.
Psychology, Crime & Law 11
Downloaded by [86.159.162.88] at 02:46 18 February 2015
sample, .857 for the LL/LL sample and .895 for the GL/GL sample. Temporal proximity
also performed moderately well with AUCs of at least .700 in each sample. In addition,
target selection achieved an AUC of .683 in the GG/GG sample, and a moderate AUC of
.775 was found for control in the LL/LL sample. None of the other results were
significant.
Table 7 shows the ROC results for the West Midlands.
Table 6. ROC analyses for linked versus unlinked offences in Northamptonshire.
Sample Behavioural domain AUC (SE) 95% confidence interval
Linked GG pairs vs.
unlinked GG pairs
Inter-crime distance
Temporal proximity
.940 (.025)*
.830 (.047)*
.891–.990
.738–.923
Target selection .683 (.063)* .559–.807
Control .477 (.069) .342–.611
Property .478 (.069) .344–.613
Linked LL pairs vs.
unlinked LL pairs
Inter-crime distance
Temporal proximity
.857 (.085)*
.850 (.082)*
.690–1.000
.690–1.000
Target selection .646 (.112) .426–.867
Control .775 (.096)* .586–.964
Property .607 (.116) .380–.834
Linked GL pairs vs.
unlinked GL pairs
Inter-crime distance
Temporal proximity
.895 (.042)*
.701 (.070)*
.812–.977
.564–.839
Target selection .585 (.073) .442–.728
Control .579 (.073) .437–.722
Property .482 (.073) .339–.625
*p< 0.05.
Table 7. ROC analyses for linked versus unlinked offences in the West Midlands.
Sample Behavioural domain AUC (SE) 95% confidence interval
Linked GG pairs vs.
unlinked GG pairs
Inter-crime distance
Temporal proximity
.944 (.013)*
.860 (.022)*
.918–.970
.816–.903
Target selection .765 (.027)* .712–.819
Control .699 (.030)* .640–.758
Property .534 (.034) .468–.600
Linked LL pairs vs.
unlinked LL pairs
Inter-crime distance
Temporal proximity
.927 (.030)*
.838 (.040)*
.868–.987
.759–.917
Target selection .764 (.050)* .666–.852
Control .719 (.054)* .612–.825
Property .613 (.058) .498–.727
Linked GL pairs vs.
unlinked GL pairs
Inter-crime distance
Temporal proximity
.923 (.024)*
.884 (.031)*
.876–.970
.824–.944
Target selection .697 (.051)* .598–.797
Control .588 (.052) .486–.690
Property .547 (.051) .447–.647
*p< 0.05.
12 A. Burrell et al.
Downloaded by [86.159.162.88] at 02:46 18 February 2015
As in Northamptonshire, the single most useful linkage factor to emerge was inter-
crime distance which reached high levels of discrimination accuracy in all samples, with
AUCs of .944, .927 and .923 across the GG/GG, LL/LL and GL/GL samples,
respectively. Again, the second most useful factor to emerge was temporal proximity,
with moderate to high AUCs: .860 for GG/GG, .838 for LL/LL and .884 for the GL/GL
sample. In contrast to Northamptonshire, the AUCs for target selection were significant
and moderate for all of the samples achieving AUCs of .765 for GG/GG, .764 for LL/LL
and .697 for GL/GL. Moderate AUCs of .699 and .719 were also found for control for the
GG/GG and LL/LL samples, respectively. None of the other results were significant.
Discussion
The initial rationale for conducting a group/lone comparison was to examine the impact
of group offending on behavioural similarity. Clearly, should group offending adversely
affect behavioural similarity this could reduce the accuracy of case linkage decisions
based on behavioural evidence.
The Kruskal–Wallis tests revealed there were no statistically significant differences
between the median Jaccard’s coefficients, inter-crime distances and temporal proxim-
ities, for GG pairs compared to LL pairs. This supported the first hypothesis and indicates
that with regard to these factors pairs of group offences displayed similar levels of
behavioural consistency to pairs of lone offences. This is beneficial to case linkage as it
means that it is feasible to link group offences with some accuracy based on behaviour.
Such results could be attributed to behavioural coherence within groups. Research has
demonstrated thematic similarities between offenders committing multiple crimes with
the same co-offenders (Porter & Alison, 2004) with further work indicating that this
behavioural coherence is due to group members copying a leader (Porter & Alison,
2006a). Alarid et al. (2009) reported that if offenders commit a series of robberies in a
short time frame, they are likely to select co-offenders from the same group of associates.
This is supported by Warr (1996) who found some offenders to have small social
networks and likely to select the same co-offenders repeatedly, particularly when they
commit multiple offences within a short time frame. This suggests that co-offending
might involve some behavioural similarity (and therefore the ability to link offences)
provided that the offences are committed in relatively quick succession by the same group
of offenders.
Perhaps even more important than the lack of differences in behavioural similarity
between GG and LL pairs are the results that demonstrate there is some behavioural
similarity across GL pairs. The previous literature suggests that people may well behave
differently when offending in a group to when offending alone (Alarid et al., 2009; Porter
& Alison, 2006b) which would lead to lower levels of behavioural consistency in GL
pairs compared to GG and LL pairs. Although this was true for some behavioural
domains, this study suggests it may be possible to link crimes across group and lone
offences based upon certain behaviours. First, despite apparently divergent median inter-
crime distances between GG, GL and LL pairs, the Kruskal–Wallis tests indicated that
these ‘differences’were not statistically significant in either police force. This suggests
that inter-crime distance is useful when trying to identify crimes committed by the same
person, even when linking across group and lone offences. Thus, a general rule that the
smaller the distances between any two crimes, the more likely they are to be linked,
applies regardless of whether the robberies were committed by a group or a lone offender.
Psychology, Crime & Law 13
Downloaded by [86.159.162.88] at 02:46 18 February 2015
Larger temporal proximities were found in GL pairs (in both police forces) than in
GG and LL pairs, particularly in Northamptonshire, where the difference was statistically
significant. There are a number of potential explanations for this difference. First, it could
be due to variations in decision-making processes in the lead up to the offence, e.g., it is
possible that the offender will be more selective about when they commit an offence
alone. Second, it could be due an artefact of the distribution of dates within the
Northamptonshire data-set which, upon re-examination of the raw data (i.e., the 160
offences), was revealed to be unevenly distributed across local policing areas –i.e.,
offences within each borough tended to be weighted towards either the start or the end of
the time frame examined. In fact, no borough had offences from all three years
represented within their sample. Therefore the anomalous finding may be attributable to
the distribution of date of offence within the data.
GL pairs were also less behaviourally similar than GG and LL pairs in terms of target
selection behaviour in the West Midlands data. There are several possible reasons for this.
First, the target selection domain included variables regarding the day and time of day the
offence was committed, and it is possible that offenders might choose different days or
times of day to commit robbery if they are alone compared to when they are with a group.
Second, they may be more likely to target a group of victims when offending in a group
compared to when they are alone.
There were no differences between GG, GL and LL pairs for property possibly
because this domain had poor levels of behavioural similarity. In fact, the median
Jaccard’s coefficients were 0.000 for all pairs indicating that the property stolen during
the offence is not at all useful for measuring behavioural similarity. This can perhaps be
explained, at least partially, by the fact that property stolen during an offence is one of the
most situation-dependent criminal behaviours (Bennell & Canter, 2002) as it is dependent
on what is available to steal (Wellsmith & Burrell, 2005). This could impact on the
consistency of behaviour across offences. However, because different property is stolen
does not mean the offences are not linked, it could be because victims possess different
types of property. Thus, this behavioural domain should probably not be considered
useful for linkage and excluded from linkage decisions whether the analyst is trying to
link lone or group offences, or both.
The only behavioural domain that emerged as a substantial problem for linking across
group and lone offences was control. The behavioural similarity of GL pairs was low for
the control domain, with median scores of just 0.143 in the West Midlands and 0.000 in
Northamptonshire. This is not surprising given the differences in violent behaviour and
weapon use between group and lone offences reported in the literature. The Kruskal–
Wallis tests revealed significant differences between GG and GL pairs and between LL
and GL pairs in both police forces for control, supporting previous findings that control
behaviours differ between group and lone offences. However, there were no significant
differences between GG and LL pairs which suggest that control is equally useful in
linking group offences together and lone offences together (although the specific
behaviours used may differ). The key finding here is that analysts should not look for a
similarity of control behaviours when seeking to link group offences to lone offences.
Instead, such linkage decisions should be made using other information.
The second phase of the study used ROC analysis to examine whether it was possible
to distinguish between linked and unlinked pairs of crimes using behaviour whilst
controlling for group/lone offending. Inter-crime distance emerged as the most useful
linkage factor for all samples (i.e., GG/GG, LL/LL and GL/GL) in both police forces,
14 A. Burrell et al.
Downloaded by [86.159.162.88] at 02:46 18 February 2015
suggesting that this can be used to distinguish between linked and unlinked crimes at
least when examining within-category trends. The results also suggest that temporal
proximity might prove useful for linking offences as it achieved moderate AUCs in all
samples and these findings were replicated across both police forces. It is noted that these
two behaviours were entered into the analysis as individual behaviours rather than as part
of a behavioural domain or theme. Concerns have been raised about how reliable
individual behaviours are as methods of linking crimes due to concerns about situational
dependency (Bennell & Canter, 2002); however, previous research in this area has failed
to demonstrate that using groups of behaviour is more effective than using individual
behaviours (e.g., Bateman & Salfati, 2007). Furthermore, numerous studies have
demonstrated the usefulness of these two particular behaviours as linkage factors (e.g.,
Tonkin et al., 2011) and so this finding is perhaps not surprising. It is also noted that
neither of these behaviours are subjective and they are easy to accurately calculate based
on information in the crime report (i.e., date of offence and geographical grid references).
This perhaps suggests that part of the success is due to the objective nature of the
variables and the ability to accurately identify their values for each offence pair.
The ROC outcomes also indicated that target selection might be useful for
distinguishing between linked and unlinked crimes but the results here were mixed.
There were moderate AUCs reported across all category samples in the West Midlands
but only for the GG/GG sample in Northamptonshire. It is possible that this difference is
attributable to the nature of the police forces; i.e., Northamptonshire is largely rural
whereas the West Midlands is largely urban. It could therefore be hypothesised that the
lower population density in Northamptonshire means targets are more dispersed and this
might have an impact on how they are chosen, which might, in turn, impact on the ability
to link cases to a common offender/group of offenders. Alternatively, the difference could
be an artefact of the larger sample size available for the West Midlands. Further research
in this area would help to unpick these findings.
The ROC outcomes revealed moderate AUCs for the control domain for the LL/LL
sample in both Northamptonshire and the West Midlands. This suggests that it might be
possible to link offences committed by a lone offender based on their control behaviours.
The GG/GG sample in the West Midlands also displayed a moderate AUC, suggesting
that it might be possible to link group offences in urban areas using control behaviours.
This result was not, however, replicated in Northamptonshire. There are number of
possible reasons for this. First, it may be that rural groups behave differently to urban
groups, although there is no evidence of this within the current data-set as the same types
of behaviours are displayed in both areas. However, there were more detailed data
available in the West Midlands which made it easier to identify behaviours displayed
during the offence. Also, the sample was larger which facilitates data analysis. It is
suggested that these latter points are more likely to explain the difference. With regard to
the GL/GL samples, the AUCs in both police forces were non-significant and indicated
that control behaviours performed only slightly better than chance at linking offences in
GL/GL cases. Obviously the situational context is different in a group versus a lone
offence and so linking a group and a lone offence committed by the same perpetrator
would be anticipated to be more difficult. Add to this the evidence that suggests that
groups behave differently to lone offenders in relation to actions that might be used to
control a victim, i.e., violence and weapon use (e.g., Alarid et al., 2009; Lloyd &
Walmsley, 1989), and this finding is not at all surprising. In fact, research using this data-
set has demonstrated that groups were more likely to physically assault a victim, and lone
Psychology, Crime & Law 15
Downloaded by [86.159.162.88] at 02:46 18 February 2015
offenders were more likely to use weapons (Burrell, 2012), indicating that the difference
in offence context explains the findings here.
Limitations
The samples were sub-samples of all robbery offences which comprised only solved
offences. There have been concerns expressed about the use of solved cases (Bennell &
Canter, 2002). It is possible that offences are solved because they display higher levels of
behavioural similarity than unsolved cases, thus introducing a potential positive bias
boosting similarity scores (Bennell & Jones, 2005; Santtila et al., 2008; Tonkin
et al., 2008).
Also, as Gagnon and LeBlanc (1983, cited in Alarid et al., 2009) found that lone
robbers were less likely to be caught. This suggests that lone offenders might have been
under-represented in the present sample. This is further compounded by Erikson’s(1971)
warning that researchers should beware of the ‘group hazard hypothesis’, which contends
that group offences are more likely to be reported to the police (McGloin, Sullivan,
Piquero, & Bacon, 2008). Combined with evidence that group offenders are more likely
to be known to the police (Hindelang, 1976), this suggests that group offending may have
been over-represented in the current study.
It is also noteworthy that there were large confidence intervals reported for the ROC
outcomes. This suggests that the results cannot be generalised beyond the current
samples. Thus, whilst this research provides a useful indicator as to which behaviours are
most useful for linkage, the work needs to be replicated with a larger sample to refine the
ROC analysis and hopefully reduce the size of the confidence intervals.
Conclusion
The study reinforced the value of inter-crime distance and temporal proximity as useful
linking factors. The findings also indicated it might be possible to use to target selection
and control behaviours to link robberies together, albeit only under certain conditions,
e.g., only in urban areas, or only for group offences.
The study provides evidence for strong behavioural coherence within group offenders
as there was no significant difference between the similarity scores for linked pairs of
offences committed by groups compared to pairs of offences committed by an individual
lone offender (for any behavioural domain) in either police force. There was also some
evidence of behavioural consistency within linked pairs of offences where the offender
committed one crime alone and the other as part of a group, as there were few significant
differences between these pairs and other pairs for many of the behavioural domains.
Differences were only found for temporal proximity in Northamptonshire, target selection
in the West Midlands and control in both police forces. The most noteworthy is control,
where group/lone (GL) pairs had significantly smaller Jaccard’s coefficients than group/
group (GG) and lone/lone (LL) pairs in both police forces. Since, the majority of
behaviours in the control domain relate to violent acts and weapon use this finding is not
that surprising. Thus, this study indicates that differences between control behaviours
need to be carefully considered when seeking to link group and lone robberies by the
same offender to avoid false negatives (i.e., failing to link cases that were committed by a
common offender).
16 A. Burrell et al.
Downloaded by [86.159.162.88] at 02:46 18 February 2015
Acknowledgement
The authors would like to thank West Midlands Police and Northamptonshire Police for providing
the data used in this study.
Disclosure statement
No potential conflict of interest was reported by the authors.
Note
1. Note that the ethnicity of offenders is only included for reference and this study does not
examine the relationship between ethnicity and crime. However, it may be of interest to note that
population estimates from mid-2009 (sourced from the Office for National Statistics) indicated
that White offenders were under-represented and offenders from Black background were over-
represented in the samples from both police forces.
References
Alarid, L. F., Burton, V. S., & Hochstetler, A. L. (2009). Group and solo robberies: Do accomplices
shape criminal form? Journal of Criminal Justice,37,1–9. doi:10.1016/j.jcrimjus.2008.12.001
Bateman, A. L., & Salfati, C. G. (2007). An examination of behavioral consistency using individual
behaviors or groups of behaviors in serial homicide. Behavioral Sciences and the Law,25, 527–
544. doi:10.1002/bsl.742
Bennell, C., & Canter, D. V. (2002). Linking commercial burglaries by modus operandi: Tests using
regression and ROC analysis. Science &Justice,42, 153–164. doi:10.1016/S1355-0306(02)
71820-0
Bennell, C., & Jones, N. J. (2005). Between a ROC and a hard place: A method for linking serial
burglaries by modus operandi. Journal of Investigative Psychology and Offender Profiling,2,
23–41. doi:10.1002/jip.21
Bond, J. (2012, March). Fingerprints forever. An evening with …Dr John Bond (OBE). Leicester:
University of Leicester.
Burrell, A. (2012). Behavioural case linkage in personal robbery (Unpublished PhD thesis).
University of Leicester, Leicester.
Burrell, A., & Bull, R. (2011). A preliminary examination of crime analysts’views and experiences
of comparative case analysis. International Journal of Police Science and Management,13,2–15.
doi:10.1350/ijps.2011.13.1.212
Canter, D., & Heritage, R. (1990). A multivariate model of sexual offences behaviour:
Developments in ‘offender profiling’.Journal of Forensic Psychiatry,1, 185–212.
doi:10.1080/09585189008408469
Conway, K. P., & McCord, J. (2002). A longitudinal examination of the relation between co-
offending with violent accomplices and violent crime. Aggressive Behavior,28,97–108.
doi:10.1002/ab.90011
Deakin, J., Smithson, H., Spencer, J., & Medina-Ariza, J. (2007). Taxing on the streets:
Understanding the methods and process of street robbery. Crime Prevention and Community
Safety,9,52–76. doi:10.1057/palgrave.cpcs.8150033
Erikson, M. L. (1971). The group context of delinquent behavior. Social Problems,19,114
–129.
Field, A. (2005). Discovering statistics using SPSS (2nd ed.). London: SAGE.
Hauffe, S., & Porter, L. (2009). An interpersonal comparison of lone and group rape offences.
Psychology, Crime, & Law,15, 469–491. doi:10.1080/10683160802409339
Hindelang, M. J. (1976). With a little help from their friends: Group participation in reported
delinquent behaviour. The British Journal of Criminology,16, 109–125.
Hochstetler, A. (2001). Opportunities and decisions: Interactional dynamics in robbery and burglary
groups. Criminology,39, 737–764.
Home Office. (2012, April). Home office counting rules for recorded crime: Robbery. Retrieved
September 5, 2012, from http://www.homeoffice.gov.uk/publications/science-research-statistics/
research-statistics/crime-research/counting-rules/count-robbery?view=Binary
Psychology, Crime & Law 17
Downloaded by [86.159.162.88] at 02:46 18 February 2015
Kapardis, A. (1988). One hundred convicted armed robbers in Melbourne: Myths and reality. In D.
Challinger (Ed.), Armed robbery (AIC Seminar Proceedings, No. 26) (pp. 37–49). Canberra:
Australian Institute of Criminology.
Lloyd, C., & Walmsley, R. (1989). Changes in rape offences and sentencing. Home Office
Research Study 105. London: Home Office. Retrieved September 5, 2012, from http://
webarchive.nationalarchives.gov.uk/20100418065544/homeoffice.gov.uk/rds/pdfs05/hors105.pdf
Markson, L., Woodhams, J., & Bond, J. (2010). Linking serial residential burglary: Comparing the
utility of modus operandi behaviours, geographical proximity, and temporal proximity. Journal of
Investigative Psychology and Offender Profiling,7,91–107. doi:10.1002/jip.120
McGloin, J. M., Sullivan, C. J., Piquero, A. R., & Bacon, S. (2008). Investigating the stability of
co-offending and co-offenders among a sample of youthful offenders. Criminology,41, 155–188.
Morgan, L., Brittain, B., & Welch, J. (2012). Multiple perpetrator sexual assault: How does it differ
from assault by a single perpetrator? Journal of Interpersonal Violence,27, 2415–2436.
doi:10.1177/0886260511433514
Porter, L. E., & Alison, L. J. (2004). Behavioural coherence in violent group activity: An
interpersonal model of sexually violent gang behaviour. Aggressive Behavior,30, 449–468.
doi:10.1002/ab.20047
Porter, L. E., & Alison, L. J. (2006a). Behavioural coherence in group robbery: A circumplex model
of offender and victim interactions. Aggressive Behavior,32, 330–342. doi:10.1002/ab.20132
Porter, L. E., & Alison, L. J. (2006b). Examining group rape: A descriptive analysis of offender
and victim behaviour. European Journal of Criminology,3, 357–381. doi:10.1177/147737
0806065586
Real, R., & Vargas, J. M. (1996). The probabilistic basis of Jaccard’s index of similarity. Systematic
Biology, 45, 380–385.
Santtila, P., Junkkila, J., & Sandnabba, N. K. (2005). Behavioural linking of stranger rapes. Journal
of Investigative Psychology and Offender Profiling,2,87–103. doi:10.1002/jip.26
Santtila, P., Pakkanen, T., Zappalà, A., Bosco, D., Valkama, M., & Mokros, A. (2008). Behavioural
crime linking in serial homicide. Psychology, Crime & Law,14, 245–265.
Tonkin, M., Grant, T., & Bond, J. W. (2008). To link or not to link: A test of the case linkage
principles using serial car theft data. Journal of Investigative Psychology and Offender Profiling,
5,59
–77. doi:10.1002/jip.74
Tonkin, M., Santtila, P., & Bull, R. (2012). The linking of burglary crimes using offender
behaviour: Testing research cross-nationally and exploring methodology. Legal and Criminolo-
gical Psychology,17, 276–293. doi:10.1111/j.2044-8333.2010.02007.x
Tonkin, M., Woodhams, J., Bull, R., Bond, J., & Palmer, E. (2011). Linking different types of crime
using geographical and temporal proximity. Criminal Justice and Behavior,38, 1069–1088.
doi:10.1177/0093854811418599
van Mastrigt, S. B., & Farrington, D. P. (2009). Co-offending, age, gender and crime type:
Implications for criminal justice policy. The British Journal of Criminology,49, 552–573.
doi:10.1093/bjc/azp021
Walsh, D. (1986). Heavy business. London: Routledge & Kegan Paul.
Warr, M. (1996). Organization and instigation in delinquent groups. Criminology,34,11–37.
Weerman, F. M. (2003). Co-offending as social exchange: Explaining characteristics of co-
offending. The British Journal of Criminology,43, 398−416.
Wellsmith, M., & Burrell, A. (2005). The influence of purchase price and ownership levels on theft
targets: The example of domestic burglary. The British Journal of Criminology,45, 741–764.
Woodhams, J., Bull, R., & Hollin, C. R. (2007). Case linkage: Identifying crimes committed by the
same offender. In R. N. Kocsis (Ed.), Criminal profiling: International theory, research, and
practice (Chapter 6, pp. 117–133). Totowa, NJ: Humana Press.
Woodhams, J., Gillett, R., & Grant, T. (2007). Understanding the factors that affect the severity of
juvenile sex offenses. Journal of Interpersonal Violence,22, 218–237. doi:10.1177/
0886260506295349
Woodhams, J., Grant, T. D., & Price, A. R. G. (2007). From marine ecology to crime analysis:
Improving the detection of serial sexual offences using a taxonomic similarity measure. Journal
of Investigative Psychology and Offender Profiling,4,17–27. doi:10.1002/jip.55
18 A. Burrell et al.
Downloaded by [86.159.162.88] at 02:46 18 February 2015
Woodhams, J., & Toye, K. (2007). An empirical test of the assumptions of case linkage and
offender profiling with serial commercial robberies. Psychology, Public Policy and Law,13,
59–85. doi:10.1037/1076-8971.13.1.59
Appendix: Behaviour checklist
Behavioural domain Offence behaviour
Target selection Day of week (7 variables)
Time of day (6 variables)
Known offender
Unknown offender
Victim at cashpoint/bank
Control Weapon used
Type of weapon (3 variables)
Group of offenders vs. group of victims
Group of offenders vs. lone victim
Lone offender vs. group of victims
Lone offender vs. lone victim
Offender(s) searches victim/victims property
Violence –physical assault
Weapon threatened
Weapon shown/seen
Offender requests property
Offender demands property
Victim resists –met with threat
Property Type of property stolen (14 variables)
Property returned
Temporal proximity
Inter-crime distance
Psychology, Crime & Law 19
Downloaded by [86.159.162.88] at 02:46 18 February 2015
