Introduction The victim-offender overlap is an important phenomenon in criminology (Jennings et al, 2012). The research supporting the existence of the overlap is undisputable and it is arguably one of the most significant facts in criminology (Bottoms and Costello, 2010). Current research has neglected critical areas and answers are needed about how victimisation and offending co-occur and how to identify those victim-offenders who are most harmed (Bottoms and Costello, 2010). Furthermore, there is limited knowledge about what effect interventions such as out of court disposals (OOCD) have on the overlap, or if there is potential to build a triage tool or algorithm to identify the most harmed in future.

Research questions These will focus on four areas victimisation – including types, frequency and harm, the victim-offender overlap, the impact on police reported victimisation by an OOCD, and finally if from criminal and victimisation history prior to the intervention date can outcomes post be predicted.

1). What does victimisation look like in low-level offenders when explored through police records in terms of prevalence, frequency, types, and harm? 2). What are the patterns and relationships between victimisation and offending in this sample? 3). What is the impact of an out of court disposal that aimed to be as effective at reducing offending as sending individuals to court on victimisation? 4). Can victimisation be predicted, and can who is most at-risk of becoming victim-offenders be predicted?

Methods This PhD thesis used the police records of offending and victimisation from the sample of low-level offenders taking part in the Turning Point Project. Which was a randomised control trial (RCT) comparing sending low level offenders through court processing against an OOCD. Victimisation and offending data were collected from police data systems (CRIMES, Police National Computer, and ICIS), matched manually using name and date of birth. Before being analysed in R, basic descriptive statistics, correlations, and odds ratios were used for the first two parts of the analysis. Results from the RCT were analysed using chi square, effect sizes and survival analysis. The final section of the thesis used cox’s regression and binomial logistic regression to examine the impact of pre randomisation variables.

Results The victim-offender overlap was found to be extensive with 63% of the sample reporting a form of victimisation. Victimisation experiences and involvement in offending varied throughout the sample. Violence was most the most prevalent form of both victimisation and offending, caused the most victimisation harm, and had the largest overlap between victimisation and offending. The analysis of harm indicated these low-level offenders reported victimisations that equalled a total 82,180.5 harm points on the Cambridge Crime Harm Index. Using a harm score allowed five different groupings for victimisation to be created, based on the total harm and total number of victimisations suffered. Combining victimisation and offending in this sample showed some complex patterns, and while the two were clearly related this was not a simple positive correlation.

The results of the RCT showed no effect of the intervention on male low-level offenders for either prevalence, frequency, survival, or harm for victimisation. However, a significant backfire effect on all measures was seen for female low-level offenders. Further research concludes this effect is most likely attributable to the significantly higher victimisation occurring prior to randomisation. Finally, the results of the regression analysis indicated key variables associated with increased risk, although the models used here produced high rates of false negatives. Victimisation is more likely to occur if the individual is still involved in offending and key predictive variables differ between victim only, offender only, non-involved and victim-offenders. With victim-offenders tending to be younger, be involved in offending or victimisation prior.

Discussion Consistent with prior research low-level offenders show a substantial overlap, indicating that low-level can be experiencing problematic and concerning levels of victimisation. While the precise mechanism cannot be discerned from this study, it is proposed that understanding both the individual propensity and the environmental exposure is important. This provides some suggestions for beneficial interventions and how to target victim-offenders effectively. While the results here did not produce a clear case for the benefits of OOCD, the results indicated for male low-level offenders the OOCD was “as good as” preventing victimisation as court processing. This mirrored the findings for offending for the OOCD, suggesting that inventions that have null effects on offending are likely to have the same on victimisation. The picture for female low-level offenders is more complex, and while it is likely related to the initially higher levels further investigation would be advisable. Finally, while the models used here produced high rates of false negatives and were limited in their explanatory power, they did highlight key variables and groups to focus on. Indicating this may be an approach to explore further in future.

Policy implications This research suggests six key considerations for policy: 1). Given the amount of victimisation present in low level offenders any policy aimed at low level offenders needs to be written with the explicit understanding that there will be high levels of victimisation present. 2). Prevention of violence is a key policy that should be taken from this thesis. Violence was the most prevalent form of both victimisation and offending and caused the most harm from victimisation. 3). Issues are not distributed equally throughout, and resources should be targeted to those suffering or causing the most harm. Using number alongside harm may provide a context that allows better targeting of resources. 4). Any intervention research into preventing offending needs to include a measure of victimisation alongside that of offending, and vice versa. Without these important effects may be being missed, and policy decisions are not being made based on the best evidence. 5). Due to the link between victimisation and offending in those where cooccurring issues are identified, interventions should aim to approach both simultaneously. 6). Victimisation, offending and becoming a victim-offending appear to be outcomes that could to some degree be predicted through algorithms or machine learning. Therefore, policy should consider utilising this approach to improve the accuracy of decisions.

Conclusion The study reiterates the importance of the victim-offender overlap and indicates even among low-level offenders the overlap can be extensive and problematic. The results here present important findings on several aspects including the first known analysis of victimisation from a RCT aimed at prevention of offending. The potential to prevent future harm from the policy implications outlined in this study are potentially vast, and the approaching victimisation and offending simultaneously could produce wide ranging benefits. The victim-offender overlap should be the centre of future policy and research.