ADR UK Research Fellows: Ministry of Justice and the Department for Education linked datasets fellowships

This project aims to explore the following research questions:

• What are the offending trajectories of individuals born on, or after, 31 August 1985 up to 31 August 1999?
• Which administrative education and social care record data is most helpful in predicting these specific offending trajectories?

The methodology used in this study:

The proposed 12-month project will use a retrospective cohort design. The project will link the individuals identified in the crime dataset, with the care and education datasets:

• National crime records for individuals born on, or after, 31 August 1985 up to and including 31 August 1999 
• Linked to these individuals’ prior educational and social care records.

The project will first use a statistical analysis approach called latent class analysis to identify different trajectories of offending behaviours following a first recorded conviction or caution.

Then the project will adopt a statistical learning approach, to see if it can identify the education and social care predictors of the different types of offending patterns or trajectories. The identification of children and young people at higher risk for persistent offending has the potential to inform early intervention approaches and criminal justice responses to reduce offending and by extension contribute to evidence-based policy making.

Duration: January 2022 - January 2023

Funding: £127,575.80

Publications

Blog: Exploring the impact of education and social care on offending patterns, March 2022

Blog: Developing trajectories of (re)-offending using UK administrative data, March 2023

Data Explained, September 2023

Data Insight: Childhood educational predictors of re-offending trajectories, August 2024

This project aims to explore the following research questions:

• What risk or protective effect does support from child welfare services during childhood offer education engagement and attainment, and involvement in the criminal justice system, when other multi-systemic factors are controlled for?
• Are the outcomes for child educational attainment and engagement, and engagement with the criminal justice system following intervention from child welfare services fair and impartial across local authorities?
• What is the effect of a child’s age at first identification of welfare needs and the duration and degree of subsequent social welfare intervention, on education attainment and engagement with the criminal justice system?

The methodology used in this study:

• Research question one will use multilevel models to explore the effects of different modes of child welfare service interventions (none, child in need, child looked after) on:
  • education attainment (maths and English GCSE level 5 or above)
  • education engagement (number of unauthorised absences, temporary or permanent school exclusion)
  • criminal justice service outcomes (convictions: none, one, two or more).

• The second research question will use the same data sources and knowledge from the research question one, but the effects will be placed in the geographical context of the local authority. Specifically, for this question, income inequality and child welfare service intervention rates for each local authority area will be added as an additional predictor in the model.
• Research question three will build again on the previous research questions and analysis to explore the effects of modes of child welfare service interventions at different ages and stages of development on education and criminal justice outcomes. It will also take account of the influences of multisystemic risk and protective factors.

 

Duration: January 2022 - January 2023

Funding: £121,944

Publications

Blog: How do different types of social care involvement affect children's education and offending outcomes?, March 2022

This project aims to explore the following research questions:

• What proportion of individuals have experience of the looked after system and what are the details of offending?
• Does offending profile vary according to ethnicity and/or legal status and placement type?
• How do care experienced and non-care experienced individuals’ sentence lengths compare for three offence types (actual bodily harm, robbery and possession of an article with blade or point), and do these relationships vary by ethnicity?
• How do factors (offender characteristics, legal status, placement type) impact frequency of offending for care experienced individuals?

The methodology used in this study:

• Individuals with experience of the looked after care system will be identified. Descriptive statistics will enable the offending profiles (offence type, frequency and disposal) of care experienced individuals to be produced and disaggregated by ethnicity and/or legal status and placement type.
• For each offence type, log-transformed custodial sentence length will be considered using logistic regression. Ethnicity and care experience will be included as key variables in these models, while also controlling for gender and court type.
• Where appropriate, multilevel models will be used to account for individuals who have multiple disposals within the offence types.
• Linear regression will be used to measure average frequency of proven offences for care experienced and non-care experienced individuals over a specified time frame, accounting for exposure to periods of imprisonment. Multilinear regression models will then be used to explore the extent to which ethnicity, legal status and placement type impact upon average frequency of proven offences for care experienced individuals.

 

Duration: January 2022 - January 2023

Funding: £118,962.17

Publications

• Blog: Exploring ethnicity, care experience and justice systems involvement, April 2022
• Data Explained, October 2022
• Policy Briefing: Care experience, ethnicity and youth justice involvement - key trends and policy implications, September 2023
• Report: Double Discrimination - Black care-experienced young adults navigating the criminal justice system, September 2023
• Animation: Challenging (In)Justice, October 2023

This project aims to explore the following research questions:

• Are young people with lower or decreasing school performance trajectories more likely to commit any offence after Key Stage 4 (up to age 21 years), as compared to higher or improving trajectories?
• Is school performance trajectory associated with re-offending?
• Does school performance trajectory modify the risk for different types of offending behaviour in vulnerable and marginalised groups? This may include socioeconomically disadvantaged, special educational needs and looked after children.
• Are the identified patterns in school performance trajectories and offending risk consistent across different regions of the United Kingdom?
• Are there discrepancies in gender and ethnicity data between the National Pupil Database and Police National Computer, and how can these be reconciled?
• What is the potential for conducting quasi-experimental trials using the data linkage, to investigate whether interventions such as special educational needs provision mitigate offending risk?

The methodology used in this study:

• The project will use growth mixture modelling to analyse latent school performance trajectories over Key Stages 1, 2 and 4. Regional differences by local authority will be explored. An advisory group comprising young people will be consulted on the optimal number of trajectories to accept in the final model. The resulting trajectories will be contextualised using published national school performance data from the same timeframe.
• The latent school performance trajectories will then be used as an exposure variable in regression models. Outcomes will be offending after Key Stage 4 (ages 17 to 21), and reoffending. Different types of offending will also be examined, using Home Office offence codes such as violence, robbery and sexual offences.
• The project will then use multilevel regression modelling to investigate interactions between school performance trajectory and pupil characteristics in predicting offending risk after Key Stage 4 (ages 17 to 21) while investigating clustering at the local authority level. Pupil characteristics of interest include gender, ethnicity, free school meals eligibility, special education needs status and looked after child status.
• Finally, descriptive comparisons of gender and ethnicity recordings in the National Pupil Database and Police National Computer will be conducted to identify discrepancies.  Descriptive explorations of interventions such as special educational needs provision will also be conducted to scope the feasibility of using these variables for quasi-experimental trials. This will be informed by the findings of a feasibility study funded by ADR UK and being undertaken by Dr Rosie Cornish, University of Bristol, who is a senior advisor on this proposal.

 

Duration: January 2022 - January 2023

Funding: £130,000

Publications

Blog: Exploring educational attainment patterns and criminal offending, May 2022

Blog: Using linked data to evaluate special educational needs provision and offending risk, January 2023

Public engagement case study: Consulting a pre-existing advisory group about a research project on ethnicity data, September 2023

Data Insight: Changes in school performance and involvement in the criminal justice system, April 2024

Data Explained: Discrepancies in gender/sex and ethnicity data between the National Pupil Database and the Police National Computer, May 2024. See also: detailed methods supplement. 

This project aims to explore the following research questions:

• Did school funding programmes that are known to have increased pupil performance, also have the effect of reducing crime, both in the short and the long term?
• What are the potential benefits of this linked dataset for other researchers, policy stakeholders and the public?

The methodology used in this study:

• This project will use estimated regression discontinuity models of crime. Regression discontinuity models are based around the idea that if a policy is implemented on an institution (e.g., school, local authority) according to passing a certain threshold value of a continuous variable, then any ‘jump’ in outcomes observed at this threshold can be interpreted as the causal effect of the policy, subject to identifying assumptions.
• Two education funding policies will be evaluated separately (i.e., not within the same set of models): Excellence in Cities and the Leadership Incentive Grant.
• In both cases the analysis will test for heterogeneity in effects, i.e., whether the programmes’ impact varies depending on pupil, school, and area level characteristics.

 

Duration: December 2021 - December 2022

Funding: £92,474

Publications

Blog: Area-based education policy - what can we learn from past efforts?, March 2022

Data Explained, May 2023

Data Insight: School funding, pupil performance and crime, October 2023