Using Data to Understand and Reduce Recidivism

There is little disagreement that reducing recidivism will improve the lives of former inmates, reduce costs and reduce the overall crime rate.

The Ministry of Justice (“MOJ”) spent £5.42 billion in 2021/22 on prisons and offenders[i].  The latest prison population figures (30 June 2022) show an increase of 3% on the previous year to 80,659[ii]. 

Recidivism (defined as the rate at which individuals return to prison after release) is reducing, but there is still much room for improvement.  (MoJ reported a reduction in recidivism from 31.6% in 2008/9 to 24.4% in July to September 2020.  In terms of total economic and social costs, re-offending costs for 2019 were estimated at £18.1 billion[iii].)

The use of data and analytics to reduce recidivism has been studied closely, particularly in the US with its large prison population.  The US spends $182 billion annually to lock up one percent of their adult population[iv].  The same source suggests the cost of a single recidivism event exceeds $150,000. (£129k when this was written.)  Initiatives to reduce recidivism are important to the US given the costs involved, which could provide useful insight and ideas to explore for the UK.

The authors of one US study on the use of data and analytics to understand and reduce recidivism[v] state:

“Organisations and processes in Criminal Justice and Corrections must evolve to optimise around emerging analytical or technical capabilities”.

They conclude:

“In the end, analytics offer an important starting point for thinking about how to manage any agency or institution where seemingly small decisions can have significant consequences for individuals and their communities.” 

MoJ and analytics

As part of MoJ’s approach to capitalise on data and analytics, it has created an analytics platform”[vi], using proven open-source tooling that offer significant cost benefits and best-of-breed capabilities. 

Within its Digital Strategy 2025[vii], MoJ’s mission is “Simpler, faster and better services for everyone”, and it states the MoJ must be “driven by data”.

To support reducing recidivism, the strategy has two relevant execution themes:

• Deliver simple, clear, fast services for probation colleagues (so practitioners have the digital tools they need to support the right interventions and can quickly and accurately provide recommendations at sentencing); and
• Deliver digital prison services that replace legacy systems and support rehabilitation (to give safe, fair and decent prisons, to improve experiences for prison staff and to increase educational & rehabilitation opportunities for prisoners).

All data and analytics initiatives need foundations.  I cannot go into great detail in a blog, but briefly these are:

• Understanding of the data held and the data required;
• Organisation – preparing data to support analysis to drive desired outcomes;
• Trust – data must be secure, transparent, well governed, curated and of suitable quality.

There must be a strategy to identify the necessary foundations, the steps to create them and sufficient budget to build and maintain them. 

MoJ is trying to put the right foundations in place, but we will see if they can be realised and built on effectively in the light of pressures on spending. 

Caveat Emptor

There are many articles on the use of predictive analytics and algorithms to calculate offender “risk scores” have been established for some time. 

Data does not lie, but the lens through which it is examined can distort the truth.  A prime example of this is the “Correctional Offender Management Profiling for Alternative Sanctions” (COMPAS).  This was used in over one million offender assessments.  A 2016 study by ProPublica[viii] reported that COMPAS was racially biased, over-estimating black re-offending and underestimating white re-offending.  Such bias must be understood and allowed for.  “Buyers should beware” but that is not to say that such methods should be avoided. 

One study by the National Institute of Justice[ix] showed how such biases can be accounted for and that these methods can improve recidivism forecasting significantly.  It also highlighted another lesson I believe MoJ should embrace.

Data from the state of Georgia was provided to the public via “Open Data” access for the NIJ’s “Forecasting Recidivism Challenge”.  Such data is usually siloed with limited access.  Some of those with access (e.g. the institutions themselves) may not have the resources to analyse the data.  Formal research partnerships can limit diversity of expertise and individuals evaluating the data.  Georgia benefitted from the greater variety of research insights this open access gave.  70 organisations took part in the challenge.  The winning models were significantly better at forecasting recidivism in an unbiased manner.  In implementing its strategy, I suggest the MoJ opens up access to its data so more diverse insights can inform initiatives to reduce recidivism. 

Data analysis can inform more effective, lower cost services BUT users must understand the bias and ethical challenges.  The ethics and equality of analytical tools must be addressed.

Authors

Stephen Etheridge

Solution Architect, Insight Practise, Atos

Anju Saxena

Head of Insight, Atos