Using Administrative Data in research to define groups and identify trajectories

Evidence indicates that offending behaviours peak in adolescence and decrease in adulthood. This is known as the age-crime curve. It is important to examine offending trajectories because some patterns are associated with poorer outcomes than others. However, evidence on the types of offending trajectories to date in England and Wales are based upon approximately 23,000 men born in the 1950’s. Therefore, the first part of my ADR UK funded research project was to update our understanding of child-adult reoffending patterns using official crime data.

Defining groups for analysis

My project used the linked Police National Computer - National Pupil Database for individuals born between 1 September 1985 and 31 August 1999. The linked data contains the anonymised education and social care records for these individuals (approximately 1.4 million), which have been linked to later offending records up to the end of 2017.

My first task was to use information contained in the Police National Computer on age of offence and offence type to develop a series of variables ready for statistical analyses. For my statistical analyses, I used a technique called ‘latent class analyses’, which is where you try and identify ‘groups’ of individuals based on observed information.

Using existing research and previous work by the Ministry of Justice on prolific offending, I developed the following variables:

• Offence type: which grouped people into those who have committed a violent offence at any point, and those who have only committed non-violent offences
• Age of first conviction or caution: 10-13 years; 14-17 years or 18 years and over
• Age of last offence in the Police National Computer: 10-17 years and 18 years and over
• Offending History: which grouped people by the number of offences committed as a juvenile (10-17 years); as a young adult (18-20 years); and as an adult (21 years and over). At each of these stages, an individual was categorised as being ‘prolific’ if they had committed more offences than the median, or ‘low-density’ if they had not.

Identifying trajectories of offending

Previous evidence has shown the age-crime curve highlights two different patterns or trajectories of reoffending. The first is called ‘life-course persistent’ offending where individuals begin to behave antisocially in childhood and continue into adulthood. The second pattern is known as ‘adolescent-limited’ offending. Here, individuals behave antisocially mostly during adolescence, with a minority continuing to offend into adulthood.

Through my research, I found five distinct offending trajectories:

1. Non-violent adolescent-limited prolific (Adol-Prolific): this group committed higher than average non-violent offences during adolescence only
2. Mixed offence type ‘later’ onset with escalating offence history (Adol-Late): this group committed different types of offences, with the number of offences rising more steeply in adulthood than in adolescence
3. Mixed offence type life-course prolific (LCP): this group committed different offence types at a higher-than-average rate throughout their lives
4. Non-violent adolescent limited low-density (Adol-Low): this group committed a low number of non-violent offences in adolescence only
5. Non-violent adult limited low-density (Adult-Low): this group committed a low number of non-violent offences in adulthood.

What’s next?

In line with existing evidence, I found a ‘life course persistent’ trajectory and ‘adolescent-limited’ trajectories. Moving forward, it would be important to see if we can use different education and social care factors to discriminate between these trajectories. This has the potential to provide deeper insights into how these factors might affect offending patterns. This information could be used to inform criminal justice system responses, given that one group does not offend into adulthood.

My sample contains males and females, but the large number of men has undoubtedly driven my results. There is little consensus about the development of female reoffending patterns over time. Future work on trajectories should utilise female-only samples to ensure they achieve better outcomes in the criminal justice system.

How can administrative data research help?

Recently, the Department for Education (DfE) and MoJ created a large dataset linking together de-identified administrative data on education, social care and crime. The dataset contains lots of important information on every child in mainstream education. It is also cost effective because the data has already been collected.

The recent funding of ADR UK by the Economic and Social Research Council, part of UK Research and Innovation means that it is now possible for researchers like me to use this data. I hope that this will showcase new ways to tackle public health problems and improve public services.

The project described here has the potential to identify previously unknown offending patterns based on ‘real world’ data from the offending population in England and Wales. It may also identify unknown drivers or even preventative factors for offending. Overall, the results of this project will highlight whether administrative data can be used to inform UK offending reduction strategies in a highly efficient and cost-effective way.

Using the linked administrative data described here, we went on to examine childhood educational predictors of the five reoffending trajectories, the results of which can be found in this data insight paper.

Administrative data is a powerful low-cost alternative to expensive and lengthy prospective longitudinal studies. However, using de-identified administrative data for research purposes like this can raise some important ethical and legal questions. That is why I plan to consult with the people affected by this research – such as ex-offenders and young people – during the project. Giving those involved a voice in the research I am undertaking will help shape the project and provide ideas on how to use its results in future.