Introduction

When violent offenders collaborate and engage in criminal activities together, the potential for harm and the escalation of violence can significantly increase (Lantz 2020; McGloin & Piquero, 2009). Group dynamics, competition between individuals, and learned behaviour can lead to more violent assaults (Conway & McCord, 2002). Therefore, when multiple individuals engage in violent acts, we can reasonably presume that their shared aggression and destructive tendencies can foster a dangerous symbiotic relationship. The collective nature of their actions may exacerbate the violence and harm inflicted on victims and the community, with potentially important ramifications for violence reduction strategies. However, to date, research quantifying the level of harm caused by violent co-offenders compared to individual offenders is relatively sparse. Lantz (2020) and Felson and Lantz (2016) employed differential classifications of violence by injury, but these were limited in sensitivity as they included only three categories (no injury, minor injury, and serious injury). The recent development of more sensitive instruments presents an opportunity to deepen our understanding, a goal this paper seeks to achieve.

There is a robust body of scholarship on the phenomenon of co-offending in general, particularly concerning violent crime. Our focus is primarily on networks, which have been identified as transmissive and exacerbating (Rowan et al., 2022). Numerous theoretical frameworks explain these phenomena, ranging from the classic (social learning theory, rational choice, denial of responsibility) to the contemporary (compounding). The latter framework is of principal interest to us because it remains fresh and relatively untested (Tillyer & Tillyer, 2019), with an especially high potential for violence reduction programmes. The current view is that violent perpetrators may be influenced by focused deterrence strategies that target risk–reward models (Tillyer & Eck, 2011), employing rational choice theory to bring about behavioural changes. However, there are typically more violent offenders than there is capacity to deal with them. Therefore, additions to the arsenal of targeting tools are important considerations and the principal starting point of this study—looking at co-offending may be a viable approach.

This paper presents a novel approach to estimating the exacerbation of crime using a crime harm index as opposed to assigning each crime an equal weight or using an insensitive taxonomy dependent solely on injury. When tallying the number of offences committed by an individual without considering the severity level of each crime committed, there is a risk of miscalculation regarding the degree of risk an offender poses to society. Equally, categorising offenders based on the types of crimes they commit also has the potential to mischaracterise the offender. This is because crime severity serves as a function of both the type of offence and the overall number of offences an individual commits. For example, an offender who has committed multiple common assaults may not be categorised as a high-harm offender (as common assaults score relatively low compared to attempted homicides or weapons offences), even though they may have caused harm multiple times. Therefore, we utilise a crime harm index, taking into account these nuanced yet critical methodological considerations, to assess the compounding effect of co-offending on violence.

This study aims to provide insights into the dynamics of violence among offenders, emphasising the significance and impact of co-offending on violent behaviour. The findings, both in terms of substance and methodology, can contribute to our understanding of violence transmission within co-offending groups, highlighting that violence contagion is directly connected to group dynamics.

Literature Review

Felson (2003) contended that when criminals assemble, they are highly likely to exchange information, skills, and trusted accomplices across criminal networks. These interactions, Felson argued, are likely to result in the commission of a criminal act. In simple terms, collaborating facilitates behaviour that acting alone does not, a premise supported by a range of empirical evidence (e.g. McGloin & Piquero, 2009; Warr, 2002; Lantz & Kim, 2019). Numerous hypotheses and fully developed theories support this phenomenon, and we review this body of work below.

Definitions of Co-Offending

Reiss (1988) described three types of career criminals: those who commit crimes alone, those who commit crimes with others (two or more individuals), and those who commit crimes both alone and with others. These types have been classified as single-offence, co-offence, and mixed-offence criminals (Reiss & Farrington, 1991). To meet the co-offending definition, Reiss emphasised that individuals must jointly commit a crime and be present at the same time (Reiss, 1988). Tremblay (1993), however, noted that this definition excludes the larger group of individuals who may have supported the primary offender(s) before, during, and after the criminal act (see more recently Denley et al., 2024). Lantz and Hutchinson (2015) expand on this definition to include individuals who assist in the commission of an offence by participating both directly and indirectly. Therefore, some definitions of a co-offender include a broader network of accomplices who can be relied upon for the criminal act to occur.

The definition of a co-offender is not just a matter for scholars, and how people may commit crimes in groups of two or more is also an issue for law enforcement. In the UK, A principal offender is defined as the individual who commits the actual crime (Crown Prosecution Service, 2021). When two or more individuals jointly commit a crime, they are referred to as ‘joint principals’. A secondary offender is defined as an individual who aids or encourages the commission of the substantive offence but does not commit the actual crime (Crown Prosecution Service, 2021).

Theoretical Frameworks

For co-offending to occur, individuals must be willing to collaborate, considering the advantages and disadvantages of such engagement, and this information must be accessible to other prospective co-offenders (Weerman, 2003). When evaluating the costs of co-offending, an individual may assess whether similar benefits could be obtained by acting alone, along with the likelihood of apprehension if they are involved with another person; co-offending can thus be a liability, and researchers have found that the higher the number of offenders involved in a crime, the greater the likelihood of apprehension and betrayal (Erikson, 1973). Still, this decision to co-offend, with whom and with how many accomplices reflect a rational choice (Cornish & Clarke, 2016). The suitability of a co-offender must be calculated too. Tremblay (1993) highlights two key elements when selecting a suitable co-offender: trustworthiness and usefulness; Offenders will either focus on building strong relationships based on trust within smaller networks or opt for more extensive networks of useful co-offenders.

Furthermore, Weerman (2014) described how social selection and co-offending converge when offenders associate with one another based on the availability of criminal opportunities. Weaver and Fraser (2021) offered a slightly different perspective on social selection, arguing that a group’s criminality may be attributed to the socioeconomic disadvantage experienced by its members. They argue that group offending offers members a sense of belonging and respect that they may not achieve otherwise due to their circumstances. This sentiment was echoed elsewhere. In a study involving 298 Canadian inmates, a significant proportion of the sample reported being introduced to crime by an influential individual (Morselli et al., 2006). A Swedish study on youth co-offending involving 22,000 suspects revealed that delinquent behaviour was transmitted across youth networks when members shared common beliefs regarding social injustices, alongside similar feelings of anger and frustration (Sarnecki, 2001).

These theories are fit to explain violent co-offending as well. In the context of aggression, collaboration among co-offenders may encourage each participant to engage in more extreme acts of aggression (Sunstein, 2009:40), forming the basis of the compounding hypothesis. Acting alongside others can lead to a sense of competition among co-offenders, with each individual attempting to be more brutal or ruthless than the others (Hardin, 2002). Consequently, the act of co-offending may result in a violent feedback loop. As each member of the group feels compelled to assert dominance or seek retribution for perceived slights or betrayals, retaliation, vengeance, and group disputes may escalate violence levels (see Walker, 2001 for a historical perspective).

Furthermore, studies in group psychology, including social identity theory and self-categorisation, suggest that co-offending may have a desensitising effect on the individuals involved (Hornsey, 2008). When violent offenders collaborate, they may reinforce and normalise violent behaviour amongst themselves (Baumeister & Butz, 2005; Collings & Magojo, 2003). This normalisation can further erode inhibitions and moral qualms, making individuals and groups more likely to perpetuate violence (Schwartz et al., 2009; see also biological factors in Mazur, 2008).

Violent Co-Offending

While Hodgson (2007), van Mastrigt and Carrington (2013), Stolzenberg and D’Alessio (2008), and Bright et al. (2020) concluded that solo offending represents the most common form of violent offending, a substantial body of empirical research documents the extensive and global prevalence of violent co-offending – especially amongst juveniles. This phenomenon is most prevalent during adolescence and diminishes with age (Andresen & Felson, 2010, 2012; Carrington & van Mastrigt, 2013; Hodgson, 2007; van Mastrigt & Farrington, 2009; Reiss, 1988; Reiss & Farrington, 1991). Stolzenberg and D’Alessio (2008) suggested that young people spend more time in groups than adults, thereby increasing the likelihood of co-offending. However, Zimring and Laqueur (2015) challenged these conclusions, arguing that group involvement among all offenders is significant.

A more nuanced approach considers the type of offence and co-offending prevalence. It seems that co-offending is less prevalent in violent crimes than in property or market crimes. However, extensive evidence suggests that co-offending relationships and network dynamics may lead to an increase in violence. Tracy et al. (2016) examined the transmission of violence within social networks in 16 studies, revealing that personal and social relationships between victims and offenders of violent crime heightened the risk of violent behaviour and victimisation. While this study was limited to weapons-related violence in a domestic setting, comparable findings have emerged in other research on violent crime co-offending and social networks. Studies on gunshot victimisation in two US cities concluded that gang membership and proximity to a gang member increased the likelihood of becoming a gunshot victim (Papachristos et al., 2015a, b). Fox et al. (2021) found that individuals who are exposed to deviant social networks are more prone to becoming victims of homicide or severe assault.

Group Size Matter

Beyond the question of the prevalence of co-offending, collaborating with others has distinguishing features. In a longitudinal study of adolescents, Conway and McCord (2002) found that non-violent offenders who jointly committed their first offence with a violent offender were more likely to commit serious violent crimes in the future. In a study on rape, robbery, and homicide offences, McGloin and Piquero (2009) found that a higher number of accomplices increased the likelihood that the first group offence would be a violent crime. This was the case regardless of whether the offenders had a history of violent crime. Bond and Bushman (2017) found that violent behaviour within adolescent networks is contagious and can propagate through up to four degrees of separation. Therefore, comprehending the extent and reach of this behaviour necessitates an understanding of the size of both social and co-offending networks.

First, studies on both violent and non-violent crime have indicated that the majority of co-offences are committed by two individuals rather than larger groups (Bright et al., 2020; Carrington, 2014; Reiss, 1988; Hochstetler, 2001; Reiss & Farrington, 1991). However, the more offenders an individual associates with, the greater the frequency of violent co-offences committed. McGloin and Piquero (2009) found that larger groups of offenders are more likely to engage in violent acts, particularly in the context of juvenile delinquency. Violent group offences had an average of 3.35 offenders per event, whereas non-violent offences had an average of 2.82 offenders. For each additional offender with whom the individual associates, the expected frequency of violent group offences increased by 9.6% (Incidence Rate Ratio [IRR] = 1.096, p < .05). Logistic regression models predicting whether an individual’s first group offence would be violent indicate that the chances of the first group offence being violent surges by 33% for every additional accomplice present during the event (Odds Ratio = 1.330, p < .01). This relationship remains significant even when controlling for whether the accomplices have a history of prior violence. This is corroborated by studies on firearm ownership (Lantz & Wenger, 2020), kidnap** (Cunningham & Vandiver, 2018), sexual assault (Tillyer & Tillyer, 2019), and shootings (Papachristos et al., 2015b).

It appears that group size impacts perceptions of risks and rewards in deviant behaviour more broadly. In a series of experiments conducted with university students, McGloin and Thomas (2016) demonstrated a positive relationship between group size and anticipated fun or excitement and social inclusion. However, larger groups resulted in diminished perceptions of responsibility and formal sanction risks, suggesting an increased propensity towards deviant behaviour as group size increases. More recently, McGloin et al. (2021) explored the concept of ‘opt-out thresholds’ in criminal behaviour, finding that larger group sizes can lead to a higher likelihood of individuals opting out of participation in criminal activities, indicating that group dynamics can function as a disincentive. Specifically, 68% of subjects who reported an opt-in threshold regarding engaging in fights also exhibited an opt-out threshold.

McGloin et al. (2021) further demonstrated the complex interplay between group size and the decision to participate in or abstain from violent acts, underscoring how group dynamics impact individuals’ inclination to fight in various ways. The study provided a nuanced understanding of how the presence and actions of peers within a group can significantly influence an individual’s decision-making process regarding violent behaviour. Among 1,852 respondents, 62.5% indicated they would not engage in a fight even if others did, while 18.2% would fight regardless of others, and 19.3% would fight if others fought first. Conversely, among those with an intention to fight, a significant portion (68%) stated they would cease fighting once four to five others were involved. This demonstrates a reverse bandwagon effect, where increased group involvement resulted in a decision to stop fighting​​.

Co-Offending Group Dynamics and Their Influence on Patterns of Future Crime

Contemporary studies have explored the characteristics and dynamics of co-offending groups, with a focus on patterns of future behaviour. Meneghini et al. (2023) examined the persistent effects of co-offending on future violent behaviour among offenders involved in organised crime in Italy. Utilising dynamic random-effects probit models, the study analysed how both solo and co-offending past violent behaviour influences the likelihood of future violence. They found that prior violent co-offending has a more substantial effect on future violence than prior violent solo offending. Specifically, the average marginal effect of past involvement in violent co-offending increases the probability of future violent offending by 14.2% (p < .001) compared to offenders who did not engage in violent co-offending. Moreover, engaging in violence multiple times doubles the likelihood of committing violence in the future. This cumulative effect underscores that violence is a persistent behaviour with enduring consequences.

Conway and McCord (2002) also investigated patterns of criminal behaviour among juvenile offenders, particularly focusing on the impact of co-offending on the likelihood of committing violent crimes. They demonstrated that non-violent offenders who commit their first co-offence together with violent accomplices face an increased risk of involvement in subsequent serious violent crime. This finding suggests that exposure to violence in a co-offending context can escalate the severity of criminal behaviour: Offenders whose first co-offence involved violent accomplices subsequently committed a higher proportion of violent crimes than those not exposed to such violent accomplices (F(1234) = 4.93, p = .027). Moreover, exposure to violent accomplices did not heighten a general propensity to crime (as opposed to violence specifically), suggesting that, for some individuals, the impact of co-offending with violent accomplices is specific to violent behaviour​​. However, we do not definitively know whether this relationship is causal; it is plausible that certain individuals may induce others to violence, but it is equally plausible that certain individuals are attracted to violent offenders. Nonetheless, these findings underscore the substantial influence of co-offending, particularly with violent accomplices, on the criminal trajectories of juvenile offenders: the context and companionship in juvenile offences significantly impact the likelihood and severity of future criminal behaviour.

More recently, Bright et al. (2024) conducted a network analysis of co-offending dynamics using relational hyperevent models (RHEM). The study delved into the conditional probability of distinct groups of co-offenders engaging in various crime categories over time, providing insights into the versatility and adaptation of criminal behaviour within co-offending networks. The research revealed that groups of two or more co-offenders are more likely than solo offenders to commit offences involving more than one crime category. This finding suggests a higher level of criminal versatility among co-offenders than solo offenders. Moreover, Bright et al. (2024) illustrated a heterophily effect: the extent to which co-offenders possess varying previous experiences in specific crime categories. A more pronounced heterophily effect was observed for property crimes, while the effect was weakest for violent crimes. This suggests that co-offending groups often comprise both experienced and inexperienced members in property crimes, potentially facilitating the exposure of inexperienced co-offenders to new types of crimes. Nevertheless, the results indicate that offenders are likely to commit offences that involve crime categories previously committed by their co-offending collaborators. Thus, co-offending groups tend to exhibit homogeneous experiences with the violent crime category: it is improbable that an individual without a history of violent offending will engage in co-offending related to violent crimes, which contrasts with other crime categories, such as property crimes. Bright et al.’s (2024) findings support the concept of social contagion, which entails learning new types of crime through exposure to co-offenders.

Crime Severity and Violent Co-Offending

Lantz (2021) illustrated the complex dynamics within co-offending groups and how they impact the severity of violent outcomes, highlighting the importance of considering both the composition of co-offending groups and the nature of the victim–offender relationship in understanding the patterns and severity of violence in criminal incidents. Utilising NIBRS data, Lantz reported that incidents involving co-offenders were (a) 64% more likely to involve the use of a weapon, (b) 31% more likely to result in minor injury, and (c) 79% more likely to result in serious injury, compared to incidents committed by solo offenders. These findings demonstrate that group offending is associated with more serious crimes and the propensity for more harmful victimisation. Moreover, as the size of the co-offending group increases, the likelihood of offence severity also escalates. Each additional co-offender was found to be associated with a 33% increase in the likelihood of weapon use, an 18% increase in the likelihood of minor injury, and a 43% increase in the likelihood of serious injury.

Moreover, Lantz (2021) reports that incidents involving groups with a higher proportion of males were approximately 47% more likely to result in serious injuries compared to those involving female groups. The age composition of co-offending groups is also significant: Young adult offenders were more likely than other age groups to be involved in incidents that result in serious injury to victims. This supports the hypothesis that young adult offenders are more likely than other offenders to engage in significant violence.

The Present Study

In this study, we aim to address the following empirical inquiries:

  1. 1.

    What are the prevalence and frequency of violent crime among solo or group offenders? Addressing this question requires analysing not only the number of incidents but, more crucially, identifying escalatory patterns based on the level of harm inflicted on victims (i.e. the compounding effect).

  2. 2.

    What is the network structure of co-offending groups, and does crime harm (as opposed to crime count) vary across various types of networks? Answering this question would provide insights into the interconnectedness of violent criminals and the extent of these relationships.

  3. 3.

    What is the propensity for repeated offending in co-offending networks of violent offenders? Understanding the conditional probability of further assaults within each study group can inform prevention and crime management policies aimed at addressing violent crime.

Methods

This study is set in Dorset, a county on the south coast of England, under the jurisdiction of Dorset Police. The county is largely rural, with a population of 800,000 residents and over 25 million annual visitors (Dorset Police, 2021).

Data and Procedures

The unit of analysis is violent crime offenders in Dorset. To qualify, individuals need only be identified as offenders in police records, regardless of whether they were charged or convicted. Consistent with Tremblay’s (1993) approach, in crimes involving multiple offenders, the individual need not have been present at the scene of the crime. Due to changes in Dorset’s crime recording systems, a seven-year time series (2015–2022) was the maximum period for which data were available.

Types of violent crime included in this study are largely based on the Home Office Counting Rules for ‘violence against a person offences’ (Home Office, 2021a). The Home Office is the government department responsible for policing in the UK, and the counting rules are national standards for recording crimes in England and Wales. Only two types of violent crime offences were omitted: corporate manslaughter (as the offender is a corporation) and offences involving death or serious injury in the unlawful driving category (as this involves a vehicle and is less about personal violence). Robbery and possession of weapons were included, as both involve some element of violence. The Home Office’s crime recording standards stipulate that only one crime report should be generated for each crime event, regardless of the number of individuals involved. Consequently, we were able to capture co-offending across all violent crimes recorded by Dorset Police, with each offence and offender assigned a single unique ID. In this dataset, there were 75,097 unique offences committed by 41,591 unique offenders.

Estimating Crime and Harm

Not all crimes carry equal levels of harm. Therefore, a weighting system is necessary to measure the severity of a crime (Andersen & Mueller-Johnson, 2018; Ashby, 2018; Curtis-Ham & Walton, 2017; Ignatans & Pease, 2016; Kärrholm et al., 2020; Mitchell, 2019; Ratcliffe, 2015; Sherman et al., 2016). However, measuring harm poses a challenge as it can impact a wide array of individuals (such as victims, parents, friends, and family) and groups (including justice systems, communities, and society). This complexity becomes particularly pertinent when assessing co-offending, as multiple individuals participate in the commission of the crime (Andresen & Felson, 2010).

Traditional crime counting methods treat each incident as carrying equal weight, resulting in a limited understanding of crime issues (Sherman et al., 2016). Instead of tallying the frequency of various types of crimes, the Cambridge Crime Harm Index (CCHI) aims to quantify the harm that crimes inflict on society. It does not, however, consider aggravating or mitigating circumstances in individual cases. The underlying principle of the CCHI is that a more sensitive measure of crime takes into account both the frequency of crime (i.e., how many times it occurred) as well as the level of harm the crime category caused (i.e., severity). For example, a burglary or serious assault inflicts more harm on society than a minor assault, as evidenced by the fact that the punishment for burglary is more severe (i.e. the prison sentence is longer). This approach has limitations (see Bland et al., 2022 for a discussion), but the key point lies in the differentiation of severity.

Formally, the CCHI assigns ‘harm scores’, quantified in days, to various types of crimes. These scores are based on the minimum length of prison sentence each type of crime would typically incur under English law for first-time offenders, without aggravating or mitigating circumstances. The longer the minimum sentence, the higher the score assigned. For instance, a crime that typically results in a 10-year prison sentence would be considered 10 times more harmful than a crime that typically results in a one-year sentence. The total harm caused by crime in a particular area or time period can then be calculated by adding up the harm scores of all recorded crimes.

Thus, the CCHI provides a more nuanced and meaningful assessment of crime’s impact, enabling policymakers and law enforcement to allocate resources more effectively and efficiently to address the most harmful crimes. Additionally, it provides a deeper understanding of violence trends over time and within specific locations and individuals (see Bland & Ariel, 2015, 2020; Fenimore, 2019; Mitchell, 2019; Hiltz et al., 2020; Loewenstein et al., 2023). We use this method of violent co-offending networks in the present study.

Statistical Techniques

We leverage a combination of statistics on crime count and crime harm to discern patterns of violent solo crime versus co-offending crime in Dorset. To assess the differences in crime counts and crime harm, we employ independent sample t-tests, as the data distribution justifies this approach.

Subsequently, we employ conditional probability to gauge the likelihood of future offences given prior offences. Specifically, we calculate an offender’s likelihood of committing the next sequential offence based on their commission of a prior offence.

Finally, we employ social network analysis to discern patterns of violent offending in Dorset. We construct a one-mode ‘offender-to-offender’ network based on the unique offences each offender was involved in. Each of the 41,591 offenders was identified based on their unique ID. A link existed between offenders if they participated in the same violent offence together. We calculate the network density, degree of centrality, and distance of the co-offending network. Network density measures the overall connectedness of the actors within a network and is represented by a coefficient ranging from 0 to 1, with a coefficient close to 1 indicating that offenders often co-offend together. Degree centrality measures the number of unique ties an actor has within their respective network, indicating how many unique co-offenders an actor has. Distance measures the shortest path between two nodes in a network.

Data Limitations

This study employs violent crime data recorded by the Dorset Police. It does not include data held by other agencies, such as the National Health Service, which treats victims of violent crime. Furthermore, it is unlikely that this study represents all violent crimes committed in Dorset during the period of study, as most crimes are not reported to the police (Ariel & Bland, 2019). This limitation is not unique to this study (Sutherland et al., 2017; Ariel & Bland, 2019). Moreover, a national data integrity audit in 2020 reported that Dorset Police failed to record over 1,900 violent crimes a year (HMICFRS, 2020). Thus, it cannot be assumed that all violent crimes have been captured within the period of study in Dorset.

Results

Descriptive Statistics

There were 41,591 distinct offenders involved in 75,097 violent crime events during the study period, spanning 71 months, according to the Dorset Police. Overall, there were 31,118 (41%) acts of violence without injury, followed by 24,375 (32%) acts of violence with injury (32%), 16,979 (23%) incidents of stalking and harassment, 1,706 (2%) cases of possession of weapons, 887 (1.5%) instances of robbery, and 32 (0.5%) homicides. Violence with injury accounted for 8,165,958.5 (80.3%) of the overall crime harm based on the crime harm index, while robbery, violence without injury, stalking and harassment, homicide, and possession of weapons accounted for 532,900 (5.3%), 508,506 (5%), 377,913 (3.7%), 301,855 (3%), and 280,908 (2.8%) of the overall crime harm, respectively. Statistics with the offence as the unit of analysis, with a break down of solo versus non-solo offences, are presented in Table 1.

Table 1 Solo offender vs. multiple offenders recorded in the crime file
Full size table

Male offenders represent 69% (n = 28,770) of all violent offenders, with female offenders representing 31% (n = 12,812); we have no statistics on non-binary offenders. Importantly, males, on average, are responsible for 35% more harm than females; an average harm score of 128.1 (SD = 342.6) and 83.7 (SD = 240), respectively (not shown in the table). Offenders who were white British committed 74% (n = 65,552) of all violent crimes. However, due to the proportion of missing information (17%, n = 14,515), the analysis of ethnicity was discontinued at this point. Regarding age, offenders aged 16–20 had the highest average harm per person, with a score of 155.1 (SD = 376.4), followed by offenders aged 21–25 (mean = 142.1, SD = 354.5), and offenders aged 26–30 (mean = 127.0, SD = 325.2). The pattern of reduced involvement in violence with age follows previous conceptualisations (see Loeber & Farrington, 2012), but this is the first time the pattern is shown empirically using crime harm analysis.

Prolific Offending

We utilise conditional probability to understand prolific offences over time. Figure 1 demonstrates an increasing likelihood of future offences following a prior offence. In this context, violent crime offenders have a 39% likelihood of committing a second offence following the commission of their first offence. Following the second offence, the likelihood of committing a third offence increases to 55%. There is a consistent increase in the likelihood of future offences thereafter, with a slight decrease after the commission of the 17th violent offence. This trend fluctuates when offenders commit their 35th or subsequent consecutive offence, as there is a diminishing pool of offenders with such an extensive criminal history. These results provide an empirical basis for categorising an offender as prolific if they have committed three or more offences (see further statistical justification in Bailey et al., 2020). Using this threshold, prolific offenders represent 21% of Dorset’s violent offender population (n = 8,837). More notably, prolific offenders were responsible for 55% (48,139) of all counted crimes and 53% (5,428,041 CHI) of the total crime harm. This finding is consistent in every year in the data (Figs. 2), with the exception of 2015–2016.

Fig. 1
figure 1

All violent offenders – probability distribution of future violent crimes

Full size image
Fig. 2
figure 2

(a) Prolific offenders – proportion of offender crime count by year. (b) Prolific offenders – proportion of offender crime harm by year

Full size image

Among the three offender categories, 47% (n = 4,163), 44% (n = 3,888), and 9% (n = 786) of prolific offenders were mixed offenders (those who sometimes commit offences with and sometimes without accomplices), strictly solo offenders, and strictly co-offenders, respectively. Among prolific offenders, mixed offenders are active in 57% of all crimes and are responsible for 66% of all crime-related harm. Among non-prolific offenders, solo offending is the most common categorisation. Regarding escalation, Fig. 3 presents the average escalation of crime harm for each successive crime committed by a prolific offender. While the average crime rate fluctuates for each successive offence, the trend line demonstrates a steady decrease. As such, it appears that the harm committed by prolific offenders lessens with each successive offence.

Fig. 3
figure 3

Prolific offender – average crime harm for successive offence

Full size image

Comparing Violent Solo and Co-Offenders

Among the entire cohort of violent offenders, solo offenders represented 68% (n = 28,418) of all offenders, while mixed offenders and co-offenders represented 14% (n = 5,552) and 18% (n = 7,621) of all offenders, respectively. Importantly, solo offenders accounted for 53% (n = 47,148) of all counted crimes but 42% (4,242,964.5 CCHI) of all crime harm. In comparison, mixed offenders and co-offenders were jointly responsible for 47% (n = 41,086) of all counted crimes and 58% (5,925,076 CCHI) of all crime harm.

Figure 4a and b present the total and average crime count and crime harm for solo offenders, mixed offenders, and co-offenders in each fiscal year of study. Mixed offenders and co-offenders (combined group offenders) committed 21–33% more violent crimes, on average, than solo offenders each year. As such, this combined group, which is numerically smaller than solo offenders, consistently commits more crimes on average. A similar pattern is observable for crime harm, as mixed offenders and co-offenders account for 45–60% more crime harm than solo offenders each year.

Fig. 4
figure 4

(a) Total and average crime count for offender categories by year. (b) Total and average crime harm for offender categories by year

Full size image

Notably, we see that offences with solo offenders yield a significantly lower crime harm score than offences committed in groups, as well as a lower number of charges in each incident (Table 2).

Table 2 Solo vs. more than one offender recorded in the crime file: harm analysis
Full size table

Based on analyses of the ‘felonious few’ (Sherman, 2019; see also Sherman, 2007) – offenders responsible for a disproportionate ratio of crime counts or crime harm out of all offenders in the dataset – we find greater concentrations in a smaller number of mixed offenders and co-offenders than among solo offenders. While 20% of solo offenders were responsible for 46% of counted crimes, 20% of mixed offenders and co-offenders accounted for 55% of counted crimes. Similarly, 20% of solo offenders were responsible for 70% of crime harm, while 20% of mixed offenders and co-offenders accounted for 78% of crime harm.

The likelihood of becoming a future offender is higher among the combined non-solo offenders group than among solo offenders (see Fig. 5). The data suggest that while the likelihood of committing a second violent offence is 31% among solo offenders, it rises to 57% in the combined offender group. Solo offenders do not reach a comparable probability (> 57%) until the fifth sequential offence, where the likelihood is 60%. Moreover, the likelihood of offence among combined group offenders is consistently higher than that of solo offenders in the first 14 offences. The probability of future offences is less sporadic for combined group offenders than for solo offenders.

Fig. 5
figure 5

Combined non-solo and solo offender – probability of future violent offending

Full size image

Figure 6 presents the average harm for both solo and combined offenders for each sequential violent crime. As is evident, offenders in the combined group exhibited a higher average crime harm score for the majority of sequential offences compared to solo offenders. Nevertheless, both groups demonstrate a consistent decrease in crime harm for sequential offences based on the trendlines, with fluctuations present at specific points.

Fig. 6
figure 6

Combined and solo offender – average escalation in CCHI

Full size image

Co-Offender Networks of Violent Crime

Table 3 provides social network statistics for the violent co-offending network in Dorset. The network appears to be notably diffused, with a network density of less than 0.01. This suggests that the majority of Dorset’s violent crime offenders do not co-offend or, at the very least, co-offend with a small number of co-offenders. This is reflected in both the degree of centrality and distance, where offenders have 0.67 co-offenders on average and are at least 12.1 nodes removed from one another. Moreover, 28,418 violent crime offenders are isolated, meaning they were solo offenders. Furthermore, 17.3% (7,179) of offenders possessed a degree of centrality of 1, indicating they had a single co-offender, while 6.3% (2,620), 3.4% (1,401), 1.8% (762), and 1.1% (477) of offenders had a degree centrality of 2, 3, 4, and 5, respectively. Only 0.3% (145) of offenders had 10 or more co-offenders.

Table 3 Descriptive Network Statistics
Full size table

Finally, Fig. 7 illustrates Pareto distributions for offenders, depicting the degree of centrality, counted crimes, and crime harm. Notably, 20% of offenders account for 83% of co-offenders and are responsible for 53% and 78% of the counted crimes and crime harm, respectively. However, these variables are not mutually exclusive, as 43% and 43% of the top 20% of degree centrality offenders are also in the top 20% of crime count and crime harm offenders, respectively. As such, the offenders with the highest number of co-offenders committed the most violent crimes and generated the most crime harm. Among the top 100 offenders for crime count and crime harm, 75% and 83%, respectively, had at least one co-offender at some point in their criminal careers. Among the top 1,000 offenders for crime count and crime harm, 62.7% and 62.1%, respectively, had at least one co-offender at some point in their criminal careers. Consequently, offenders with the highest crime counts and crime harm scores were rarely solo offenders; they frequently collaborated with other violent crime offenders. Indeed, while solo offenders averaged 1.7 crimes, with an average crime harm score of 90, offenders who had one or more co-offenders averaged 3.1 crimes, with an average crime harm score of 144.2.

Fig. 7
figure 7

Power curves – degree centrality, crime count, and CCHI

Full size image

Discussion

This study seeks to understand violent behaviour patterns in co-offending networks by exploring the differences across various dimensions. We employ official statistics, despite acknowledging the recognised concerns regarding external validity associated with these data (Ariel & Bland, 2019), employing a population-level dataset of violent crimes in Dorset, UK. We contribute to the body of evidence by applying a crime harm index analysis rather than a crime count approach, which provides insight into the dynamics of violent group offending.

Previous studies have encountered obstacles when attempting to address the distinctive nature of offence severity due to the methods used to measure severity. Initially, studies exploring the potential compounding effect of offences committed in groups of two or more often relied on qualitative assessments, generalised assessments (such as mob mentality or terrorism), or a combination of both. Lantz (2020, 2021) used broad violence categories to suggest that larger groups were prone to committing serious crimes, yet quantifying this disparity proved unfeasible. Until very recently, a nuanced approach to estimating variations in harm across offenders was absent. This gap has been filled with the introduction of crime harm or crime severity indices.

Overall, the findings support the presence of a compounding effect of co-offending. Violent offenders in Dorset with more than one connection to another violent offender (those with a degree centrality score of 2 or higher) were identified to have a higher total and average harm score (31% more harmful on average). The higher the number of connections, the higher the average crime count and crime harm.

Overall, solo violent offenders account for less crime and crime harm compared to offenders who jointly commit crimes with others. Aligned with the findings of previous research in the UK, Canada, and the US, the evidence suggests that solo offending is a predominant career type (Bright et al., 2020; van Mastrigt & Carrington, 2019; Hodgson, 2007; Stolzenberg and D’Alessandro 2008), but prolific offenders were responsible for 55% of all counted crimes and 53% of all crime-related harm. Combined groups of offenders were responsible for 47% of all counted crimes and 58% of all crime-related harm. These figures are consistent on a year-to-year basis.

The literature further demonstrates that exposure to violence through co-offending or network relationships may increase the risk of violent behaviour (Tracy et al., 2016; Papachristos et al., 2015a; Fox et al., 2021; Bailey et al., 2020; Conway and McCord 2020; McGloin & Piquero, 2009; Bond & Bushman, 2017). Our data suggest that the combined co-offender group was responsible for 45–60% more crime harm, on average, compared to solo offenders across each year of study, and that offenders in the combined group were more likely to commit a second offence (57%), than solo offenders (31%). Collectively, the evidence suggests an increased propensity for recidivism and more harmful instances of re-offending when violent crime is committed in groups. The greater the frequency of re-offending recorded, the higher the likelihood of another violent offence being documented, implying that conditional probability offers robust yet crude forecasting opportunities.

Policy Implications

The emphasis on co-offenders is not only theoretical but also pragmatic. Conway and McCord (2002), Andresen and Felson (2010), and Lantz (2021) all concluded that intervention strategies should focus on co-offenders to effectively reduce recidivism. Police forces in the United Kingdom face mounting pressure to reduce violent crime. The Home Secretary has recently pledged over £130 million to combat serious violence, murder, and knife crime (Home Office, 2021b). Additionally, the UK government has encouraged a multi-agency public health approach following the 2018 publication of the Serious Violence Strategy (HM Government, 2019). Combating violent crime is undeniably a clear priority. However, smaller police forces that do not receive the same level of financial investment as larger counterparts face significant obstacles. Therefore, directing attention towards co-offending networks appears to offer a potentially fruitful path forward for police forces with limited financial resources (Hodgson, 2007; Carrington & van Mastrigt, 2013). A comprehensive understanding of the behavioural patterns within co-offending networks can lead to a substantial reduction in serious violence and victimisation, as targeting certain key nodes in the network may have a ripple effect on their accomplices (Ariel et al., 2019; Bright et al., 2020). Preventative deterrence messaging to prolific offenders and their co-offenders in the US has proven effective, reducing crime by 21% among those directly targeted and 15% across the total network (Ariel et al., 2019). This aligns with Bailey et al.’s (2020) findings, which suggested that a knife crime prevention strategy in the Thames Valley, UK, but more formal evaluations in the UK context are needed.

Furthermore, the UK College of Policing provides the Approved Professional Practise (APP) guidelines for police forces in England and Wales to identify and manage the most serious sexual offenders and violent offenders. The majority of management occurs within multi-agency public protection settings and relies on prior convictions. However, the criteria for classifying an individual as a potentially dangerous person (PDP) are somewhat more subjective. There is no specific legislation to define these individuals; instead, it is based on the reasonable grounds that they are likely to re-offend in the future (College of Policing, 2022). However, the PDP process should include offenders based on cumulative harm. If co-offending was also taken into consideration, then a criterion for selecting offenders to be managed can potentially lead to further cumulative crime harm reduction. However, more research is needed to validate this approach under controlled settings, where existing mechanisms to manage offenders should be prioritised for high-harm, violent offenders with the propensity to co-offend with others.

Limitations

Due to bias in reporting, not all violent incidents are reported to the police. Crimes are often underreported due to a fear of retaliation, a lack of trust in the police, cultural barriers, and a belief that the police will not take appropriate action. These unreported incidents constitute the ‘dark figures’, or unknown instances, of violence. Consequently, relying solely on police records provides an incomplete representation of the true prevalence of violence. We advocate for studies to utilise additional data sources to expound on the compounding effect, though still consider implementing a crime harm index that is sensitive not only to the number of documented incidents but also to their severity.

Furthermore, even when incidents are reported, inconsistencies may arise in police documentation and classification of violent incidents. Discrepancies between jurisdictions can occur regarding what constitutes a violent crime and the criteria for recording an incident as such. This inconsistency in data collection practices can result in wide variations in the recorded statistics of violence, making it challenging to compare or generalise findings across regions or time periods (Ariel & Bland, 2019).

Finally, the findings of this study pertain only to one police force in the UK and may be unique to Dorset. Therefore, replication of the findings would be necessary to determine whether similar levels of harm are observed in the prolific and combined offender groups across different contexts.