ADR UK Research Fellows: Ministry of Justice & Department for Education linked dataset

This project aims to explore the following research questions:

1. How do the frequency and different types (temporary or permanent) of school exclusions relate to offending trajectories? 
2. Are there differences in the offending trajectories of excluded school pupils based on geographical location?  
3. Are there differences in offending severity and offending rates between those who remain in mainstream education after permanent school exclusion and those in a Pupil Referral Unit?  

The methodology used in this study:  

Research question one uses group-based trajectory modelling - a method that identifies groups of individuals who follow similar patterns over time - to examine the different paths pupils take after being excluded from school. This analysis includes various factors, such as socio-demographic details and school-related variables (e.g. attendance, the number and type of exclusions, the duration of exclusions, and academic performance). The analysis also includes information about contact with the criminal justice system (e.g. the age at which contact is first made, types of offences, most severe penalties, number of convictions and cautions, and sentencing outcomes).

Research question two uses ordinal logistic regression - a statistical method used for predicting outcomes with an ordered range - to explore whether school exclusion outcomes vary significantly across different areas in England. This method looks at how different factors predict an individual’s level of involvement with the criminal justice system. It includes for example: types of school exclusions (temporary or permanent); exclusion rates and duration; socio-demographic factors; academic performance; and offence-related data (e.g. the age at which offending began; the type of offence). This involvement can range from no contact to receiving a caution, non-custodial sentence, or custodial sentence.

Research question three uses binary logistic regression - a technique that models the relationship between two categories - to compare outcomes for pupils who remain in mainstream education after being excluded and those sent to Pupil Referral Units. This analysis determines whether these groups are more likely to engage in violent or non-violent offences (where violent offences are defined as violence against a person, and non-violent offences include all other crimes). The model also accounts for other factors, such as the most serious penalty received, the number of convictions, and socio-demographic variables considered in the earlier analyses.

Publications

• Blog: The pathways of excluded school pupils into (and out of) the criminal justice system, February 2025

This project aims to explore the following research questions:

1. What socio-emotional characteristics are exhibited by children in early childhood who engage in offending during adolescence? 
2. How do school difficulties (e.g., absenteeism, exclusions) and poor attainment influence the relationship between socio-emotional development and adolescent offending? 
3. How do unfavourable characteristics within schools and local authorities interact with socio-emotional development to shape offending behaviour? Unfavourable characteristics may include inequalities in local authority expenditure on children’s wellbeing, youth crime rates, etc.

The methodology used in this study:

This project uses data from the early years foundation stage profile for three groups of pupils, aged four-five, from the 2006/2007, 2007/2008, and 2008/2009 school years for reception. This data is matched with their educational outcomes, such as absences, school exclusions, and key stage achievement, as well as any cautions or sentences for offences between ages ten and 18 from the Police National Computer database. Information about schools and local authorities derives from external sources like Get Information about Schools (GIAS) and the Department for Education’s local authority interactive tool, and is linked to the pupils' data at both the school and local authority level.

The project will use a variety of methods, which may include:

Factor analysis - a technique that identifies underlying patterns or “factors” from a set of variables - is used to determine how many socio-emotional domains can be derived from the early years foundation stage profile data. These factors are then used to analyse the impact of multiple variables at once, to predict the likelihood of engaging in offending behaviour during adolescence.

Mediation analysis - a method that examines how one variable indirectly affects another through an intermediate factor (or mediator) - is used to estimate how socio-emotional factors might indirectly influence adolescent offending through mediators, such as educational performance and school-related issues.

Finally, multi-level modelling - a technique used to analyse data that is grouped at more than one level (e.g. students within schools) - is used to explore how socio-emotional development and contextual factors (such as the characteristics of schools and local authorities) interact to influence adolescent offending behaviour. This approach allows us to examine both individual and broader school or local authority-level effects at the same time.

Publications

• Blog: How does early socio-emotional development in childhood relate to later offending?, June 2025
• Impact case study: Socio-emotional characteristics in early childhood and offending behaviour in adolescence, January 2026
• Working paper: Early socio-emotional skills and adolescent offending: evidence from administrative data, July 2026

This project aims to explore the following research questions:

1. Is exposure to alternative provision associated with young people’s offending, compared to young people with similar characteristics in mainstream schools?
2. What factors are associated with offending, and what factors can be considered protective (i.e. reducing the likelihood of offending) for young people exposed to an alternative school  intervention? Does this vary based on the ‘type’ of alternative school  setting and duration in this setting? 
3. Do specific ‘types’ of alternative school settings reduce the likelihood of offending/re-offending for the sample?  
4. Is duration in alternative school settings  associated with reducing the likelihood of offending/re-offending?  

The methodology used in this study:

This project uses a quasi-experimental approach, comparing a treatment group (young people who attended alternative provision) with a comparison group (young people with similar characteristics who did not attend alternative provision and remained in mainstream schools). This comparison focuses on individuals born between 1993 and 2000.  

The method will test different approaches (e.g., marginal structural modelling utilising inverse probability weighting) to establish counterfactual populations to test outcomes for children/young people in alternative provision against those in mainstream schools. Time-to-event analyses will also examine time-to-offending for the two groups.  

A typology of alternative school settings will be established to classify them based on the type of programme they offer (e.g. therapeutic, vocational, online). This allows for an analysis of how different types of alternative school settings  may impact youth offending outcomes and point towards the key enablers of offending and protective factors for children and young people in alternative schools.

Publications

• Blog: Alternative provision and offending: Are there any connections?, July 2025

This project aims to explore the following research questions: 

1. What are the long-term effects of school absenteeism on crime at different stages of life?  
2. Are the effects of school absenteeism on crime moderated by social-demographic and community-level characteristics, such as ethnicity, sex, and economic background?  

The methodology used in this study: 

The MoJ-DfE linked dataset offers a robust and large-scale examination of how absenteeism affects crime over time. School absence measures are obtained from the National Pupil Database, while longitudinal indicators of individual involvement in crime are sourced from the Police National Computer. This study applies quantitative research methods designed for analysing longitudinal data.  

Descriptive and exploratory statistical analyses are used to identify trends in absenteeism and subsequent crime across different demographic groups and over time. This provides an overview of how absenteeism and crime patterns vary by group and context. 

Multivariate regression analysis is applied to test whether absenteeism predicts crime at various ages, while controlling for socio-demographic and contextual variables. This includes binary logistic regression, to distinguish between offenders and non-offenders, and ordinal logistic regression, to predict varying levels of criminal involvement (e.g. minor versus more severe offences). 

Survival analysis examines the time intervals between episodes of absenteeism and subsequent involvement in criminal behaviour. The Cox proportional hazards model is used to assess the relationship between absenteeism and the likelihood of committing a crime. This model controls for potential confounding factors, such as demographics, previous criminal records, and community context. 

Structural equation modelling, including mediated growth models, is employed to track how changes in absenteeism over time influence changes in criminal behaviour. This modelling technique allows for the examination of both direct and indirect relationships, where demographic and contextual factors may act as moderators. This approach helps identify the conditions under which absenteeism affects criminal behaviour and highlights which factors intensify or reduce this risk. 

Publications

• Blog: School absences, deprivation, and crime involvement – or the egg, the farmer, and the chicken, February 2025
• Data Insight: From school absences to crime involvement, August 2025
• Data Explained: Exploring the dynamics of school absenteeism and crime, February 2026
• Article: Unstructured time as a pathway from disadvantage exposure to crime: The role of time away from school, Journal of Criminal Justice, August 2026
• Article: Unstructured spare time and crime: Toward an integrative model, Crime Science, March 2026

This project aims to explore the following research questions:

1. What are the educational factors associated with violent reoffending?
2. Can the Oxford risk of Recidivism (OxRec) risk assessment tool be employed to successfully estimate risk of violent reoffending among individuals released from prison in England and Wales? 
3. Do educational factors improve the OxRec tool’s ability to predict violent reoffending?

The methodology used in this study:  

The study is using a prison discharge dataset to identify a cohort of people born between 31 August 1985 and 31 August 2000, who were released from prison between January 2008 and December 2019. Information about their past offences and whether they committed violent offences again after being released is gathered from the Police National Computer database and cross-checked with the prison population dataset.

Using these release dates, each prisoner can be followed until they either commit a violent crime again or until the last available data (December 2021). Their educational background can also be reviewed using the National Pupil Database, which is already connected to crime records.

Using time-to-event analysis – a way of analysing time elapsed before an event – the project studies how long it takes for violent reoffending to occur. First, it looks at whether educational factors influence the likelihood of violent reoffending research question 1). It then checks if the OxRec tool, which predicts reoffending within one or two years from prison discharge, can be applied to prisoners in England and Wales (research question 2). Finally, it evaluates whether adding the educational factors from research question 1 can improve the OxRec tool's accuracy (research question 3).

Publications

• Blog: Rethinking how we assess the risk of violent reoffending, August 2025