PSR Lament

Over the years we've covered this subject a number of times and once again it's become the focus of attention with a recent HMI report revealing that:-  

• Less than half of all inspected court reports were deemed to be sufficiently analytical and personalised to the individual, supporting the court’s decision making.

• There were notable differences in quality between the types of court report. Oral reports met the overall quality judgement in only four out of 10 cases, short format reports in half of the cases, and standard delivery reports in more than six out of 10 cases. 

• Court reports for those individuals from a Black, Asian or minority ethnicity background were less likely to be deemed sufficiently analytical and personalised, supporting the court’s decision making.

Right from the introduction the HMI report proudly confirms:-

The provision of pre-sentence reports (PSRs) has been a key part of the Probation Service’s work since its very earliest days, providing advice and information to help judges and magistrates decide upon the appropriate sentences for those appearing before the courts. The importance of this work has been highlighted as follows: 

‘it can be argued that the provision of reports for the courts is – in some ways – the most significant task. Pre-sentence reports (PSRs) are the primary point of contact for sentencers, who are the main customers for probation work.’ (Mair, 2016).

so it might come as some surprise to many that over the last 20 years or so all manner of interventions and policies have conspired to downgrade the task, marginalise it within the criminal justice system and make the task as difficult and unwieldy as possible. Once a key part of every field Probation Officers role and often informed by their extensive knowledge of a person's offending history, the task is now performed by dedicated court-based staff with limited time and no prior knowledge beyond that possibly gleaned from crap IT systems such as OASys.

Oh and this again as outlined in the HMI report put a tin hat on the whole thing:-

Over the last decade, under the Transforming Summary Justice and Better Case Management efficiency programmes, and in line with long-standing efforts to reduce the length of the court process, there has been a move away from the more thorough and detailed standard delivery reports (with their typical turnaround of up to 15 working days) and towards fast delivery reports, first towards oral reports (which can often be delivered on the day of request or within 24 hours) and later to short format written reports (turnaround time of around five working days). Following concerns about the numbers of people being sentenced without any form of PSR, there has most recently been a focus on reversing this downward trend, particularly for cases where a community order was most likely; in 2014, 85 per cent of community orders had involved a PSR, which had reduced to 45 per cent by 2019. Research shows that, at least for community sentences, completion of the sentence is more likely when a PSR has been requested and provided than where it has not (Ministry of Justice, 2023).  

Very much late in the day it now seems that dots are being belatedly joined up and realisation dawning that much of the mess we now find ourselves in is a direct result of having so successfully done away with a vital element of probation's work in informing sensible sentencing decisions. But it's all too late as the damage has been done by the absurdity of moving skilled PO staff into prisons under OMiC, thus breaking down the local knowledge base and de-skilling PO staff in the field who now struggle with the whole report writing concept and task.

In my view this HMI aspiration will not improve the situation one bit:-

The measures set out in the 2021 Target Operating Model for probation services aim to increase both the number of cases that receive PSRs and the quality of those reports, leading to increased sentencer confidence in the probation service and potentially helping to reverse the recent decline in the use of accredited programmes (HM Inspectorate of Probation, 2024). High-quality PSRs also have a positive impact in terms of supporting well-informed, analytical, and personalised post-sentence assessment and sentence planning, which is the starting point of the well-established and recognised ASPIRE model for case supervision.

Probation practitioners working in courts now use the Effective Proposal Framework (EPF) digital tool to ensure that ‘interventions recommendations and licence conditions address risk and need, are in line with policy and sentencing guidelines and supports consistency of practice, proportionality and the reduction of bias’ (HM Prison & Probation Service, 2021). The EPF tool contains a list of all available interventions and their eligibility criteria. Court officers can input the details of the individual before the court, such as risk levels, gender, age, geographical location and offence details, and the EPF provides a shortlist of interventions that are suitable and available to that individual.

Just more alphabet soup bloody processes and tools to add to an endless list of them, but failing utterly to understand and get to grips with fundamental flaws that now inhabit all aspects of probation work that have made it part of the problem and not any part of a solution. The new Labour Government promised a wide-ranging 'Probation Review' in their election manifesto and they'd better bloody get on with it before there's nothing much left of a once proud gold standard profession worth saving. 

At this point can I flag up the following quote from a blog from a former Probation Officer now working for Revolving Doors:-

As a former probation officer myself, I can say that probation officers took a real pride in these reports, and sentencers were immensely grateful for them. The other benefit was that if the person received a community order and was supervised by the probation officer that wrote the report, there was already a level of trust and good faith between them which helped set up a purposeful relationship which made meaningful work more likely.

I wholeheartedly agree! The blog continues:-

Failed reforms and increased time pressures

However, reforms aimed at expediting criminal case resolutions and improving efficiency, including initiatives like Transforming Summary Justice, promoted an increase in pre-sentence reports with quicker turnaround times. This shift led to a decline in the use of standard delivery reports, which typically take up to 15 working days, in favour of fast delivery reports, which take around five days, and oral reports, often completed on the same day or within 24 hours.

The problem with oral and fast delivery reports is that the interview happens over a very short period of time: often within the court building, with maybe just 20 minutes to spare and with the officer under pressure either to get back to the court and deliver the oral report, or to get to another person also waiting for their interview.

Under such circumstances it is not surprising that the officer is not able to build a relationship with the interviewee that allows them to extract the same level of information and so the assessment is less detailed and so, in our view, holds less validity.

The irony of this move towards speed is that it has happened in the context of mass court backlogs, meaning cases take longer than ever before. The rush to sentence is in contrast to the glacial progress of cases through the system.

Impact on the revolving door group

In 2022 we published the Probation Inquiry, exploring the experiences of those in the revolving door on probation, including in relation to court reports. Many people reflected on how rushed their interview felt, and how disappointed they were with the process.

However, others reported it as a transformative experience, where they had been able to disclose things they had never felt comfortable to before, including domestic abuse experiences, and others reported the report resulting to them finally getting them the interventions they needed. We think this demonstrates that report interviews need to be done over a good length of time – with opportunity for both parties to build a rapport.

It is worth noting that the Inspectorate also reported that Court reports for those individuals from a Black, Asian or minority ethnicity background were less likely to be deemed sufficiently analytical and personalised, supporting the court’s decision making. It is our belief that short interviews do not allow for nuanced discussions around cultural and racial issues within the interview, disadvantaging those who already face discrimination within the justice system and society more broadly.

We believe the reduction in longform standard delivery pre-sentence reports has significantly contributed to the decline in the use of community orders by sentencers, with these decreasing by more than half (54%) between 2012 and 2022.

Reports no longer offer strong enough analysis as to why a specific community sentence will be so rehabilitative. This is just one of the reasons why we believe that the decline in standard delivery reports has significantly disadvantaged those in the revolving door group, whose offending is driven by unmet health and social needs.

Presenting for sentencing without the opportunity for previous convictions and former failed sentences to be contextualised seems more likely to end in a custodial sentence in our already calamitously overcrowded prisons, rather than with the targeted, rehabilitative non-custodial support our group so desperately need.

Return to more thorough PSRs to break the cycle

The findings of the Inspectorate’s report underscore the critical need for high-quality, detailed pre-sentence reports to support effective sentencing.

The shift towards faster, less comprehensive reports has unfortunately compromised the ability of sentencers to make fully informed decisions, particularly for those with complex needs, such as individuals struggling with substance use or mental health issues.

This trend not only undermines the goal of rehabilitation but also disproportionately impacts those who are already vulnerable, perpetuating the cycle of crisis and crime.

To break this cycle, it is imperative that we advocate for a return to thorough, personalised pre sentence reports that truly inform sentencing decisions and promote more equitable outcomes for all.

Kelly Grehan
Policy Manager

A Reminder How OASys Did For Probation

Thanks yet again to ace researcher and contributor 'Getafix, here is a long but incredibly well-researched article on theConversation.com website into risk assessment and particularly the dreaded OASys so hated by many probation staff. In many people's eyes, including mine,  OASys has been the single most significant factor in probation's demise as a useful endeavour, hence here is the complete article:-   

A ‘black box’ AI system has been influencing criminal justice decisions for over two decades – it’s time to open it up

Justice systems around the world are using artificial intelligence (AI) to assess people with criminal convictions. These AI technologies rely on machine learning algorithms and their key purpose is to predict the risk of reoffending. They influence decisions made by the courts and prisons and by parole and probation officers.

This kind of tech has been an intrinsic part of the UK justice system since 2001. That was the year a risk assessment tool, known as Oasys (Offender Assessment System), was introduced and began taking over certain tasks from probation officers. Yet in over two decades, scientists outside the government have not been permitted access to the data behind Oasys to independently analyse its workings and assess its accuracy – for example, whether the decisions it influences lead to fewer offences or reconvictions.

Lack of transparency affects AI systems generally. Their complex decision-making processes can evolve into a black box – too obscure to unravel without advanced technical knowledge.
Proponents believe that AI algorithms are more objective scientific tools because they are standardised and this helps to reduce human bias in assessments and decision making. This, supporters claim, makes them useful for public protection.

But critics say that a lack of access to the data, as well as other crucial information required for independent evaluation, raises serious questions of accountability and transparency. It also calls into question what kinds of biases exist in a system that uses data from criminal justice institutions, like the police, which research has repeatedly shown is skewed against ethnic minorities.

However, according to the Ministry of Justice, external evaluation poses data protection implications because it would require access to personal data, including protected characteristics such as race, ethnicity and gender (it is against the law to discriminate against someone because of a protected characteristic).

Oasys introduced

When Oasys was introduced in the UK in 2001 it brought with it sweeping changes to how courts and probation services assessed people convicted of crimes. It meant that algorithms would begin having a huge influence in deciding just how much of a “risk” people involved in the justice system posed to society. These people include those convicted of a crime and awaiting punishment, prisoners and parole applicants.

Before Oasys, a probation officer would interview a defendant to try to get to the bottom of their offending and assess whether they were sorry, regretful or potentially dangerous. But post 2001 this traditional client-based casework approach was cut back and the onus was increasingly put on algorithmic predictions.

These machine learning predictions inform a host of decisions, such as: granting bail, outcomes of immigration cases, the kinds of sentences people face (community-based, custodial or suspended), prison security classifications and assignments to rehabilitation programmes. They also help decide the conditions on how people convicted of crimes are supervised in the community and whether or not they can be released early from prison.

Some attempts at more rigorous risk assessments predate Oasys. The Parole Board in England and Wales deployed a re-conviction prediction score in 1976 which estimated the probability of a reconviction within a fixed period of two years on release from prison. Then, in the mid-1980s, a staff member with the Cambridgeshire Probation Service developed a rather simple risk prediction scale to provide more objectivity and consistency about predicting whether probation was an appropriate alternative to a custodial sentence. Both these methods were rather crude in terms of using only a handful of predictors and deploying rather informal statistical methods.

Harnessing computer power

Around this time, Home Office officials noticed the increased interest in the UK and the US authorities for developing predictive algorithms that could harness the efficiencies computers offered. These algorithms would support human opinions with scientific evidence about what factors were predictive of reoffending. The idea was to use scarce resources more effectively while protecting the public from people categorised as being at high risk of reoffending and causing serious harm.

The Home Office commissioned its first statistical predictive tool, which was deployed in 1996 across probation offices in England and Wales. This initial risk tool was called the Offender Group Reconviction Scale (OGRS). The OGRS is an actuarial tool in that it uses statistical methods to assess information about a person’s past (such as criminal history) to predict the risk of any type of reoffending.

The OGRS is still in use today after several revisions. And this simple algorithm has become incorporated into Oasys which has grown to incorporate additional machine learning algorithms. These have developed over time, predicting different types of reoffending. Reoffending is measured as reconviction within two years of release.

Oasys itself is based on the “what works” approach to risk assessment. Supporters of this method say it relies upon “objective evidence” of what is effective in reducing reoffending. “What works” introduced some basic principles of risk assessment and rehabilitation and it gained currency with governments around the world in the 1990s.

Risk factors can include “criminogenic needs” – these are factors in an offender’s life that are directly related to recidivism. Examples include, safe housing, job skills and mental health. The “what works” approach is based on several principles, one of which involves matching appropriate rehabilitation programmes to a person’s criminogenic needs. So, a person convicted of a sex crime, with a history of alcohol abuse, might be given a sentence plan that includes a sex offender treatment programme and drug treatment. This is meant to reduce their likelihood of reoffending. Following Home Office pilot studies between 1999 and 2001, Oasys was rolled out nationally and His Majesty’s Prison and Probation Service (HMPPS) have used the technology widely ever since.

What the algos do – scoring ‘risk’

The Offender Group Reconviction Scale and variations of Oasys are frequently modified and some information about how they work is publicly available. The available information suggests that Oasys is calibrated to predict risk. The algorithms consume the data probation officers obtain during interviews and information in self-assessment questionnaires completed by the person in question. That data is then used to score a set of risk factors (criminogenic needs). According to the designers, scientific studies indicate that these needs are linked to risks of reoffending.

The risk factors include static (unchangeable) things such as criminal history and age. But they also comprise dynamic (changeable) factors. In Oasys, dynamic factors include: accommodation, employability, relationships, lifestyle, drugs misuse, alcohol misuse, thinking and behaviour, and attitudes. Different weights are assigned to different risk factors as some factors are said to have greater or lesser predictive ability.

So what type of data is obtained from the person being risk assessed? Oasys has 12 sections. Two sections concern criminal history and the current offence. The other ten address areas related to needs and risk. Probation officers use discretion in scoring many of the dynamic risk factors.

The person becomes a set of numbers

The probation officer may, for example, judge whether the person has “suitable accommodation”, which could require considering such things as safety, difficulties with neighbours, available amenities and whether the space is overcrowded. The officer will determine whether the person has a drinking problem or if impulsivity is an issue. These judgments can increase the person’s “risk profile”. In other words, a probation officer may consider dynamic risk factors like having no fixed address and having a history of drug abuse, and say that the person poses a higher risk of reoffending.

The algorithms assess the probation officers’ entries and produce numeric risk scores: the person becomes a set of numbers. These numbers are then recombined and placed into low-, medium-, high-, and very high-risk categories. The system may also associate the category with a percentage indicating the proportion of people who reoffended in the past.

However, there is simply no specific guidance on how to translate any of the risk of reoffending scores into actual sentencing decisions. Probation officers conduct the assessments and they form part of the pre-sentence report (PSR) they present to the court along with a recommended intervention. But it is left to the court to determine a sentence, in line with the provisions of the Sentencing Council.

There is no dataset available to us that directly links Oasys predictions to the decisions they are meant to inform. Hence, we cannot know what decision-makers are doing with these scores in practice. The situation is muddier considering that multiple risk tools put out results in different ratings (as in high, medium, or low) for the same individual. That’s because the algorithms are predicting different offence types (general, violent, contact sexual and indecent images). So a person can collect several different ratings. It could be the person is labelled high risk of any reoffending, medium risk of violent offending, and low risk of both sexual offending types. What is a judge to do with these seemingly disparate pieces of data? Probation officers provide some recommendations but the decision is ultimately left to the judge.

Impact on workloads and risk aversion

Another issue is that probation officers have been known to struggle with completing Oasys assessments considering the significant amount of time it takes for each person. In 2006, researchers spoke to 180 probation officers and asked them about their views on Oasys. One probation officer called it “the worst tax form you’ve ever seen”. In a different study, another probation officer said Oasys was an arduous and time-intensive “box-ticking exercise”.

What can also happen is that risk-aversion becomes entrenched in the system due to the fear of getting it wrong. The backlash can be swift and severe if a person assessed as low risk commits a serious offence - there have been many high-profile media scandals that prove this. In a report for the Prison Reform Trust, one long-term prisoner commented:

They repeatedly go on about ‘risk’ but I realised many years ago that this has nothing to do with risk … it’s all about accountability, they want someone to blame should it all go wrong.

The fear of being blamed is not an idle one. A probation officer was reportedly sacked in 2022 for gross misconduct for rating Damien Bendall as medium risk rather than high risk after a conviction for arson. Bendall was released with a suspended sentence. Within three months, he murdered his pregnant partner and three children. Jordan McSweeney, another convicted murderer, was released from prison in 2022 with an assessment of medium risk. Three days later, he raped and brutally killed a young woman walking home alone. A review of the case determined that he had been incorrectly assessed and should instead have been labelled high risk. But unlike in the Bendall case where an individual probation officer was apparently blamed, the chief inspector of probation, Justin Russell, explained:

Probation staff involved were … experiencing unmanageable workloads made worse by high staff vacancy rates – something we have increasingly seen in our local inspections of services. Prison and probation services didn’t communicate effectively about McSweeney’s risks, leaving the Probation Service with an incomplete picture of someone who was likely to reoffend.

‘Bias in, bias out’

Despite its widespread use there has been no independent audit examining the kind of data Oasys relies on to come to its decisions. And that could be a problem - particularly for people from minority ethnic backgrounds. That’s because Oasys, directly and indirectly, incorporates socio-demographic data into its tools.

AI systems, like Oasys, rely on arrest data as proxies for crime when they could in some cases be proxies for racially biased law enforcement (and there are plenty of examples in the UK and around the world of that). Predicting risks of reoffending on the basis of such data raises serious ethical questions. This is because racially biased policing can permeate the data, ultimately biasing predictions and creating the proverbial “bias in, bias out” problem.

In this way, criminal history records open up avenues for labelling and punishing people according to protected characteristics, like race, giving rise to racially biased outcomes. This could mean, for example, a higher percentage of minorities rated in the higher risk groups than non-minorities.

Another source of bias could stem from the way officers “rate” ethnic minorities when answering Oasys-led questions. Probation officers may assess minority ethnic people differently on questions such as, whether they have a temper control problem, are impulsive, hold pro-criminal attitudes, or recognise the impact of their offending on others. Unconscious biases could be at play here resulting from cultural differences in how various ethnic groups perceive these issues. For instance, people from one cultural background may “see” another person with a bad temper whereas that would be seen as acceptable emotional behaviour in another cultural background. Read more: How can black people feel safe and have confidence in policing?

In its review of AI in the justice system in 2022, the justice and home affairs committee of the House of Lords noted that there are “concerns about the dangers of human bias contained in the original data being reflected, and further embedded, in decisions made by algorithms”. And it’s not just the UK where such issues have arisen. The problem of racial bias in justice systems has been noted in various countries where risk assessment algorithms similar to Oasys are deployed.

In the US, the Compas and Pattern algorithms are used widely, and the Level of Service family of tools have been taken up in Australia and Canada. The Compas system, for instance, is an AI algorithm used by US judges to make decisions on granting bail and sentencing. An investigation claimed that the system generated “false positives” for black people and “false negatives” for white people. In other words, it suggested that black people would reoffend when, in reality, they did not and suggested that white people would not reoffend when they actually did. But the developer of the system has challenged these claims. Read more: AI: why installing 'robot judges' in courtrooms is a really bad idea

Studies suggest that such outcomes stem from racially biased decision making embedded in the data which the developers select to represent the risk factors that will determine the algorithm’s predictions. Criminal history data, such as police arrest records, is one example.

Other socio-economic data that developers select to represent risk factors may also be problematic. People will score as being higher risk if they do not have suitable accommodation or are unemployed. In other words, if you are poor or disadvantaged the system is stacked against you. People are also classed as “high risk” for personal circumstances which are sometimes beyond their control. Risk factors include “not having a good relationship with a partner” and “undergoing psychiatric treatment”.

Meanwhile, a report issued by Her Majesty’s Inspectorate of Probation in 2021 alludes to the problem of conscious and unconscious biases which can enter the process via probation officers’ assessments, thereby infecting the outcomes. More transparency could be useful for tracking when and how probation officer discretion has potentially tainted the final assessment, which could have resulted in people being incarcerated unnecessarily or being allocated inappropriate treatment programmes. This could result in flawed risk predictions.

For example, the report states:

It is impossible to be free from bias. How we think about the world and consider risk is intrinsically tied up with our emotions, values and tolerance (or otherwise) of risk challenges.

Social engineering?

Miklos Orban, visiting professor at the University of Surrey School of Law, recently engaged with the Ministry of Justice seeking information on Oasys. One of us (Melissa) spoke with Orban about this and he expressed concerns that the system might be a form of social engineering. He said that governmental officials were eliciting personal and sensitive information from defendants who may think they are making these disclosures to get help or sympathy. But the officers may instead use them for another purpose, such as labelling them with a drinking or drugs problem and then requiring them to go on a suitable treatment programme. He said:

As a convict, you know very little of how risk assessment tools work, and I have my doubts as to how well judges and parole officers understand statistical models like Oasys. And that’s my number one concern.

Not much is known about the accuracy of Oasys in relation to gender and ethnicity either. One available study (though a bit dated as it looked at a sample from 2007) shows the non-violent and violent predictive tools are less accurate with women and minority ethnic people. Meanwhile, Justice, a legal reform organisation, recently cited a lack of research on the accuracy of these tools for women and trans prisoners.

In terms of racial bias, an HM Inspectorate of Prisons’ audit found that an Oasys assessment had not been completed or reviewed in the prior year for almost 20% of black and minority ethnic prisoners. This is a serious issue because further evaluation can help ensure that minority ethnic people are receiving similar treatment or being assigned to helpful programming. It can avoid probation officers simply assuming the risk status of minority ethnic people is unchangeable and thus reduce their chances of early release since Oasys assessments are required to ascertain whether interventions have reduced risks of reoffending.

Researchers with the Inspectorate of Probation encouraged designers of Oasys to expand the ways it can incorporate a person’s personal experiences with discrimination and how it may impact their relationship with the criminal justice system. But, so far, and to the best of our knowledge, this has not been done.

Algorithms affect real people

Oasys results follow a person’s path through the criminal justice system and could influence key decisions from sentencing to parole eligibility. Such serious decisions have huge consequences on peoples’ lives. Yet officials can decline to disclose Oasys results to the defendant in question if they are thought to contain “sensitive information”. They can ask and be shown their completed assessment, but they are not guaranteed to see it.

Even if they are given their scores, defendants and their lawyers face significant hurdles in understanding and challenging their assessments. There is no legal obligation to publish information about the system, although the Ministry of Justice has commendably made certain information public. Still, even if more data were released, defence lawyers may not have the scientific skills to examine the assessments with a sufficiently critical eye.

Some prisoners describe additional challenges. They complain that their risk scores do not reflect how they see themselves. Others believe that their scores contain errors. While some also feel that Oasys mislabels them. In another report compiled by the PRT, one prisoner stated: “Oasys is who I was, not who I am now.” And a man serving a life sentence described the repeated risk assessment when he spoke to a researcher at the University of Birmingham:

I have likened it to a small snowball running downhill. Each turn it picks up more and more snow (inaccurate entries) until eventually you are left with this massive snowball which bears no semblance to the original small ball of snow. In other words, I no longer exist. I have become a construct of their imagination. It is the ultimate act of dehumanisation.

Not all judicial officers are impressed either. When asked about using a risk assessment tool that the state required, a judge in the US said: “Frankly, I pay very little attention to the worksheets. Attorneys argue about them, but I really just look at the guidelines. I also don’t go to psychics.”

There have been relatively few legal challenges to any of the risk assessment algorithms in use across the world. But one case stands as an outlier. In 2018, the Supreme Court of Canada ruled in the case of Ewert v Canada that it was unlawful for the prison system to use a predictive algorithm (not Oasys) on Indigenous inmates.

Ewert was an Indigenous Canadian serving time in prison for murder and attempted murder. He challenged the prison system’s use of an AI tool to assess his risk of recidivism. The problem was the lack of evidence that the particular tool was sufficiently accurate when applied to the Indigenous population in Canada. In other words, the tool had never been tested on Indigenous Canadians.

The court understood that there might be risk-relevant differences between Indigenous and non-Indigenous peoples as to why they commit crimes. But since the algorithms had not been tested on Indigenous people, its accuracy for that population was not known. Therefore, using the tools to assess their risks violated the legal requirement that information about an offender must be accurate before it can be used for decision making. The court also noted that the over-representation of Indigenous people in the Canadian justice system was in part attributable to discriminatory policies.

Individual vs group risk

The feeling that the scores produced by risk assessment algorithms such as Oasys may not be properly personalised or contextualised finds merit when considering how predictive algorithms in general work. They assess people and produce risk scores and this has a longer history in business. The lending industry uses algorithms to assess the creditworthiness of customers. Insurance companies deploy algorithms to generate quotes for car insurance. The insurance algorithms often use driving records, age and gender to determine the likelihood of claiming against the policy.

But an all too common and mistaken assumption is that algorithms can provide a prediction about the specific person. On the contrary, publicly available information shows that the algorithms rely upon statistical groups. What does this mean? As we said earlier, they compare the circumstances and attributes of the person being risk assessed with risk factors and scores associated with criminal justice populations – or groups.

For example, what if “John” is placed in the medium-risk category, which is associated with a reoffending likelihood of 30%? This does not mean there is a 30% chance that John will reoffend. Instead, it means that about 30% of those assigned medium risk are forecasted to reoffend based on the observation that 30% of the medium risk had in the past been reconvicted.

This number cannot be directly assigned to any individual within that medium-risk group. John may, individually, have a 1% chance of reoffending. The scales are not individualised in this way and so John, himself, cannot be assigned specifically with a number. The reason for this is that the predictive factors are not causal in nature. They are correlated, meaning there may be some relationship between the factors and reoffending. Oasys uses male gender as one of the predictive factors of reoffending. But being male does not cause reoffending. The relationship as perceived by Oasys merely suggests that males are more likely to commit crimes than females.

There are important consequences to this. The individual can thereby be seen as being punished, not for what he or she is personally predicted to do. They face imprisonment because of what others – who share a similar risk score – have done. This is why more transparency of predictive algorithms is needed.

But even if we know what the inputs are, the weighting system is often obscure as well. And developers are frequently changing the algorithms for a host of reasons. The purposes may be valid. It could be that predictors of reoffending change over time in connection with societal shifts. Or it could be that new scientific knowledge suggests a modification is necessary.

Nevertheless, we have been unable to discover much about how well the Oasys system, or its components, performs. The Ministry of Justice has, to our knowledge, only released retroactive results. Those statistics cannot inform on the predictive performance of the tool for predictions made today, or for how accurate they are when we relook at the offenders in two years. Frequent retrospective results are needed to provide up to date information on the performance of algorithms.

Independent evaluation

To the best of our knowledge (and to the knowledge of other experts in the field), Oasys has not been independently evaluated. There is a clear need for more information on the effectiveness and accuracy of these tools and their impact on gender, race, disability and other protected characteristics. Without these sources it is not possible to fully understand the prospects and challenges of the system.

We acknowledge that the lack of transparency surrounding Oasys is a common, though not universal, denominator that unites these types of algorithms deployed by justice systems and other sectors across the world. A court case in the state of Wisconsin that challenged the use of a risk assessment tool that the developer claimed was confidential succeeded only to a point.

The defendant, convicted of charges related to a drive-by shooting, claimed that it was unfair to use a tool which used a private algorithm because it prevented him from challenging its scientific credentials. The US court ruled that the government did not have to reveal the underlying algorithm. However, it required authorities to issue warnings when the tool was used.

These warnings included:

• the fact that failure to disclose meant it was not possible to tell how scores were determined
• the algorithms were group-based assessments incapable of individualised predictions
• there could be biases toward minority ethnic people
• the tool had not been tested for use in the state of Wisconsin.

Opening up the black box

Problems such as AI bias and lack of transparency are not peculiar to Oasys. They affect many other data-driven technologies deployed by public sector agencies. In response, UK government agencies, such as the Central Digital and Data Office and the Centre for Data Ethics and Innovation (CDEI) have recognised the need for ethical approaches to algorithm design and implementation and have introduced remedial strategies. A recent example is the Algorithmic Transparency Recording Standard Hub which offers public sector organisations the opportunity to provide information about their algorithms.

A relatively recent report published by the CDEI also discussed bias-limitation measures, such as reducing the significance of things like arrest history as they have been show to be negative proxies for race. A post-prediction remedy in the CDEI report requires practitioners to lower the risk classification allocated to people belonging to a group known to be consistently vulnerable to higher risk AI scores than others.

More generally, researchers and civil society organisations have proposed pre and-post implementation audits to test, detect and resolve AI problems of the kind associated with Oasys. The need for appropriate regulation of AI systems including those deployed for risk assessment has also been recognised by key regulatory bodies in the UK and around the world, such as Ofcom, the Information Commissioner’s Office and the Competition and Markets Authority.

When we put these issues to the MOJ, it said the system had been subject to external review, but it was not specific on the data. It said it has been making data available externally through the Data First programme and that the next dataset to be shared with the programme will be “based on” the Oasys database and released “within 12 months”.

An MOJ spokesperson added: “The Oasys system has been subject to external review and scrutiny by the appropriate bodies. For obvious reasons, granting external access to sensitive offender information is a complex process, which is why we’ve set up Data First which allows accredited researchers to access our information in an ethical and responsible way.”

In the end, we recognise that algorithmic systems are here to stay and we acknowledge the ongoing efforts to reduce problems with accuracy and bias. Better access to, and input from, external experts to evaluate these systems and put forward solutions would be a useful step towards making them fairer. The justice system is vast and complex and technology is needed to manage it. But it is important to remember that there are people behind the numbers.

--oo00oo--

The article generated some interesting responses:-

Not an AI in 2001. The article goes on to refer to an algorithm which is likely correct. An AI is not an algorithm. I’d imagine the one being used by Oasys probably started out like something used by actuaries to calculate risk, the two fields look like there considerable overlap. Also the calculations done to calculate a credit score. I doubt it’s an AI even now..

Nitpicky stuff out of the way, it does illustrate the problems with black box systems and peoples reactions to them. First you can’t easily work out why the results are what they are. Secondly people tend to blindly accept the output even when it clearly diverges from reality, remember when people just followed satnav systems into fields or rivers? Generative systems are already fairly opaque, but it’s going to get worse.

*****
“I have likened it to a small snowball running downhill. Each turn it picks up more and more snow (inaccurate entries) until eventually you are left with this massive snowball which bears no semblance to the original small ball of snow. In other words, I no longer exist. I have become a construct of their imagination. It is the ultimate act of de-humanisation”. The person quoted is maybe not aware just how accurate they are. It used to be my practice to share what their files said with the person I was assessing for a DV or a Sex Offenders group. In every single instance the range of inaccuracies was enormous, from those with insignificant impact to those with monumental impact. Clearly, just like Chinese whispers, the initial mistake is made, repeated by the second worker with another couple added in and so on and so on. 

With AI taking over so much and humans more and more cut out of the equation in gaining information based on relationship (and hopefully a highly skilled interviewer aware of their own inherent biases) God knows where all this might end up. Whilst outcomes for this AI system have not been analysed outcomes for the What Works Approach (imported here into Western Australia) and for Sex Offender Treatment programs have both been analysed and found to be significantly flawed with positive outcomes seriously over-estimated. We still use these patently flawed systems in Western Australia. 

When I was first employed by the (in)Justice system I was one of about 15 Senior Programmes Officers delivering these awful programs with 1 worker only focusing on prisoners education and job seeking stuff. It needed to be the other way around. And any money saved on the fairly useless work we were doing (and I was a very highly skilled and experienced practitioner) would have been better spent on dentists, tattoo removals, and secure accommodation post release. What would AI make of that I wonder?

*****
I have no doubt that OASys has many flaws and would benefit from rigorous academic assessment of its accuracy and biases, but it remains a risk assessment tool and is not AI. Now, if a computer was tasked with analysing videos of defendants in court and generating assessments based on their behaviour, that would be dangerous reliance on AI. On the other hand, analysing OASys outcomes and suggesting improvements in the algorithm would be, if properly managed and evaluated, a positive use of AI.

Lets Re-invent the Wheel

“what about this wheel thingy? It sounds a terribly interesting project.” “Ah,” said the marketing girl, “well, we’re having a little difficulty there.” “Difficulty?” exclaimed Ford. “Difficulty? What do you mean, difficulty? It’s the single simplest machine in the entire Universe!” The marketing girl soured him with a look “All right, Mr. Wiseguy,” she said, “you’re so clever, you tell us what colour it should be.”

Douglas Adams, The Restaurant at the End of the Universe

--oo00oo--

Right from the beginning of this endeavour I have written extensively on the importance of the PSR, 

The end of the PSR is Nigh

Death of the PSR

Folly of PSR Demise

When Is a PSR Not a PSR?

March of the FDR

Would You Adam and Eve It!

PSR Special

PSR At Last Lamented

MoJ Admits PSR Problem

but it seems as if those clever people in charge at the MoJ seem to take ages to join up the dots:- 

The impact of oral and fast delivery pre-sentence reports (PSRs) on the completion of court orders

1.1 Background 

A pre-sentence report is advice given to the court following the facts of the case, expert risks and needs assessments, including an independent sentencing proposal and additional relevant information. They must be as objective as possible and exist to assist the judiciary with sentencing. 

The number of pre-sentence reports written in England and Wales has decreased in recent years – from 211,494 in 2010 to 103,004 in 2019.This was an area of concern in the 2020 white paper, A Smarter Approach to Sentencing, which stated that “The purpose of a pre-sentence report (PSR) is to facilitate the administration of justice, and to reduce an offender’s likelihood of reoffending and to protect the public and/or victim(s) from further harm. A PSR does this by assisting the court to determine the most suitable method of sentencing an offender (Criminal Justice Act 2003, section 158)”. 

The paper proposed that “further work is undertaken to build the evidence base on the impact that a PSR has […] on offender outcomes, sentencing behaviour and the efficient administration of justice”. Following this, a PSR pilot began in March 2021. Separately, this report examines the impact of a PSR oral or fast delivery report on the reason that a court order (or sentence) terminated. This analysis specifically looks at adults sentenced to a Community Order or Suspended Sentence Order (with requirements) in England and Wales in 2016. 

1.2 Key findings 

The overall results show statistically significant evidence that those who received a PSR oral or PSR fast delivery in 2016 were more likely to successfully complete their court order, compared with a group of similar offenders who did not receive a PSR. In particular: 

• 80% of those who received a PSR fast delivery in 2016 successfully completed their court order. This is higher than for similar offenders who did not receive any type of PSR (73%). 

• 73% of those who received a PSR oral in 2016 successfully completed their court order. 

This is higher than for similar offenders who did not receive any type of PSR (68%). These impacts on completion of court orders are based on estimates of what would happen if instead of receiving a PSR oral or fast delivery, the case had not received a PSR. This analysis includes PSRs prepared for both magistrate and crown courts. 

The results of this analysis should not be directly compared to termination outcomes in other analyses or to figures such as national averages due to the use of a time-bounded sample, and the propensity score matching (PSM) technique used to ensure otherwise dissimilar groups were comparable. In addition, the results for PSR fast delivery and PSR orals should not be directly compared. 

The results of this analysis should not be regarded as definitive; it is intended to provide initial evidence of the impact of PSRs and so only looks at the reason that a court order terminates as an outcome. The limitations of this study are discussed in section 5.1 alongside suggestions for areas that could be valuable to explore in greater depth. However, these findings increase the evidence on the effectiveness of PSRs and therefore it can be recommended that oral and fast delivery reports are requested and delivered to increase successful termination of court orders.

4. Results 

The analysis found statistically significant evidence that an offender sentenced in 2016 that received a PSR fast delivery or PSR oral was more likely to successfully complete their court order and less likely to terminate early for failure to comply with requirements or conviction between 2016-2019, compared with a group of similar offenders which did not receive a PSR – see the main results in Appendix E. 

4.1 PSR Fast Delivery 

• ~80% of those who received a PSR fast delivery in 2016 successfully completed their court order. This is significantly higher than the comparison group (7 percentage points higher) which comprised similar offenders to those who received a PSR fast delivery but did not receive a PSR. 

• ~11% of those who received a PSR fast delivery in 2016 terminated their court order early for conviction of a further offence. This is significantly lower than the comparison group (4 percentage points lower) which comprised similar offenders to those who received a PSR fast delivery but did not receive a PSR. 

• ~9% of those who received a PSR fast delivery in 2016 terminated their court order early for failure to comply with requirements. This is significantly lower than the comparison group (3 percentage points lower) which comprised similar offenders to those who received a PSR fast delivery but did not receive a PSR. 

4.2 PSR Oral 

• ~73% of those who received an oral PSR in 2016 successfully completed their court order. This is significantly higher than the comparison group (5 percentage points higher) which comprised similar offenders who did not received a PSR. 

• ~15% of those who received an oral PSR in 2016 terminated their court order early for conviction of a further offence. This is significantly lower than the comparison group (2 percentage points lower) which comprised similar offenders to those who received a PSR oral but did not receive a PSR. 

• ~13% of those who received an oral PSR in 2016 terminated their court order early for failure to comply with requirements. This is significantly lower than the comparison group (2 percentage points lower) which comprised similar offenders to those who received an oral PSR but did not receive a PSR. 

Please note the analysis of PSR oral and PSR fast delivery should not be directly compared as the matched comparison groups contain different offenders with different characteristics. Therefore, it cannot be concluded that fast delivery reports were 7% more likely to terminate successfully than oral reports, although the relative efficacy of fast delivery and oral PSRs is of interest. 

5. Discussion 

These findings demonstrate that the requesting of a PSR oral or PSR fast delivery is related to the successful completion of a court order, and therefore it is recommended that these reports are requested and delivered. 

Additionally, these findings support the wider work across the MoJ to increase the instances of advice being given to court in the form of a PSR. 

This study does not consider the influence of a PSR report on the sentence handed down. As PSRs typically propose a sentence, the extent to which, if at all, the advice of the PSR is implemented and further work to explore whether these findings are impacted by the PSR advice being followed would be useful. 

While in Section 3.3 Assumptions, it was considered that further data linking (e.g. to extracts from OASys and the PNC14) might reduce sample size and bias the dataset towards more complex cases, this assumption could be tested. Making further “presentencing” variables available for review/matching would enable the goodness of match between the treatment and control groups to be verified or improved, and facilitate further analysis of the relationship between sentence completion and sentence characteristics, and the impact of a PSR on these characteristics. 

Additionally, whilst the groups receiving PSR oral and PSR fast were different in some ways, in other characteristics, such as gender, they were similarly distributed. It would be useful to consider whether the PSR report has the same statistical significance across all different groups. 

Research on the quality of a pre-sentence report was undertaken in 2020, which questioned whether PSRs provided sufficiently analytical and personalised (to the service user) advice, in order to aid with the court’s decision making (HM Inspectorate of Probation, 2020). The report considered “standard” (these can take up to 15 working days to complete, and are used for serious cases involving high complexity or serious sexual or violent crime), “short format” (fast delivery) and oral reports, and found that 97% of standard delivery reports, 82% of the short format, and 65% of the oral reports were sufficiently analytical and personalised to the service user (although it is noted that only 30 (4%) of the reports available were standard delivery reports; 535 (67%) were oral and 237 (30%) were short format). This report was broadly supportive of the HMPPS reforms promoting the use of PSRs, and indicated that fast delivery PSRs might offer a necessary increase in analytical depth relative to oral PSRs. However, as noted before, the dataset used in this study contains no information on the quality of each report, and the oral and fast delivery cohorts considered by this study are not directly comparable. This suggestion would therefore require further investigation, e.g. by matching cohorts between PSR types. 

This study considers only the reason for court order termination. While the successful completion of a court order is desirable and indicates no reoffending during the period of the sentence, it is not a direct proxy for reoffending (arguably the most desirable outcome) or for other outcomes that might be considered markers of “success”. The evidence base for PSR reports would be augmented by investigation into the relationship between a case that receives a PSR in court and further outcomes related to the offender, for example whether they attended their initial meeting with a probation officer, reoffended (following the expiry of the court order), had suitable accommodation or employment post-sentence.

Is Training Good Enough?

There may be silence from Napo, but interestingly there were 4,309 hits to this blog yesterday as news broke in relation to the damning report on the Damian Bendall case, confirming it to be the 'go to' place in times of stress and drama. Before that, we had been discussing various aspects of training, an issue which of course features heavily in the case. I found these contributions to be of particular note:-   

I don’t agree with pushing these long service PSOs through shortened PO training just to get completions. There should be one route for probation officer training. Degree based university study with on the job placement training. Those with and without existing degrees should be eligible and relevant work experience should be part of the eligibility criteria. It’s not difficult to see why we have PSO, PQiPS and POs that can’t do the job and short cuts to qualification doesn’t help anybody. Qualified in name only in turn impacts on the rest of us who are too busy already.

******
I know there is concern about the qualification now after some managers including at senior level have, since reunification, gained the PO qualification by a different route without doing the day job of a PO ie directly managing their own case load, completing OASys, delivering interventions etc. So in effect a way of qualifying as a PO seems to have been created just for some managers who did not have the qualification but this was different from the qualifying route the rest of us have to follow. Several SPOs who had previously worked as PSOs and continued in their manager role whilst doing it were successful. Little information seems available on this but there seem to be several different routes now to gaining the PO qualification, including some sort of fast track.

******
Proper practice! SFO after SFO attributable to the lofty PO grade. The attitudes displayed here are the reason that Probation will not survive and no one takes it seriously. Many talk about Probation culture and mourn it but this culture appears to be built on self satisfaction and a belief that the PO is somehow better than the rest of humanity whilst bewailing the fact that others have more to give and its the case with most PSO's that they have more life experience as opposed to I have attended Uni thus I am superior.

******
That’s not the point really. The moaning stems from a desire to maintain standards. The current probation service is driven by process and systems, whilst the targets support this shift. Unfortunately, there is no evidence that processes and systems, no matter how well the targets are met, have any impact on offending behaviour. In a sense this highlights a problem at the heart of the organisation. Targets have become an end in themselves largely because they simplify what is in effect a complex area of work. As this continues the type of staff and managers required to meet the demands has shifted. 

I completed the CQSW, and have to agree with an earlier comment that it was garbage. What wasn’t however was my practice teacher and the high standard of education I was afforded. Both of these introduced me to new ideas and different ways of viewing the world. They were both demanding and made me appreciate the complexity of human life. And with that the complexity of criminal behaviours. I woke up to how society targets the disenfranchised and policed the poor. The depth of this education matched the demands of the work when I started my first job in 1987. Complex behaviour requires the ability to process complex ideas and formulate hypotheses that you test and adjust as you move forward.

Shortly before I retired I read a number of PSRs. Gone was the human messiness of life. Authors tried to simplify something so hard to grasp with pat phrases and jargon that lacked both evidence and rigour. To a one they were shallow and relied on puerile vacuity as though that could explain anything. The job I left championed process over practice. There are many reasons why this happened. But one is undoubtedly that it removes the need to address complex human behaviours. Behaviours that are difficult to judge, understand and almost impossible to change without using a high level of intelligence to understand what it was that lead to an offence occurring. Which of course means you don’t need highly trained staff, and can get by with people who only need to understand processes, time scales and systems. Not all POs were intellectual powerhouses but some were and they were a delight to work alongside. Not all PSOs were dullards but some were. And they certainly weren’t a delight to work alongside.

******
The current training provides staff with what they require for the current service. It’s crap but then they don’t need anything else. You can’t blame for applying. If I was starting out again I would probably think it was great. But I worked with students who were bright, clever, and socially aware. That was encouraged. This isn’t the job for those people anymore. Not because people are more stupid, but because the current job doesn’t require that level of intelligence. To be honest 2 GSEs is probably enough to fulfil the role. I suspect a well trained chimp can press buttons, scroll a screen and complete an OASys on time.

******
I am currently on the second cohort of this and have mixed opinions. It’s a good progression route for those who do not have the existing degree qualification. However, the pilot was designed to capture the skills and experience of the existing PSO. Like any role the skills and experience vary greatly. I’ve found myself fortunate to have a wealth of experience and knowledge and found the process fairly straightforward. However, there are many others struggling. Some who worked in TTG teams who have never experienced case management or others just joining to then apply via the progression route. The push for POs is ongoing and unlikely to stop, however how do we back fill competent and experienced PSOs holding 50 plus caseloads? The progression route should be available but you can’t pull a cork out of one hole to fix another and expect the boat not to sink.

******

In agreement with so much that has been written about the dire state of Probation in today's blog. Probation has so many irredeemable flaws that it's difficult for me to contribute to the Blog in any way constructively. So just one small point - the online training that was made absolutely mandatory by Senior Management is almost entirely worthless. As anyone who has completed the training will know there is usually an online training pass mark of 80 per cent. This sounds fine until you realise that overworked frontline Probation Workers are often unaware of the exact nature of the 20 per cent they may have got wrong in the quiz at the end of the training - this means that the online training is virtually useless. Imagine being operated on by a surgeon who has only 80 per cent of knowledge in his or her field of expertise and who doesn't actually know for sure what s/he does or doesn't know... terrifying.

******
It's the knowledge and understanding of the person that comes from the relationship between the PO and the client that provides the most valuable tool in assessing risk.

Assessing risk, as with everything else in probation has become a process, and because it's become a process it becomes easy to delegate parts of that process to anyone capable of following that process regardless of experience or training.

As long as probation continues along the road of processing its case load rather then engaging with it, the number of SFOs will continue rise, and blame for them happening will continue to be attached to (often undeservedly) those at the lower end of the conveyor belt.

Report Highlights Structural Failings

Press release issued today:-

Independent serious further offence review of Damien Bendall

Background

Damien Bendall murdered Terri Harris (aged 35), John Paul Bennett (aged 13), Lacey Bennett (aged 11) and Connie Gent (aged 11). He also raped Lacey. These crimes took place in September 2021 in Killamarsh, Derbyshire. He pleaded guilty in December 2022 and was later sentenced to a whole-life prison term.

Bendall was on probation when he committed these offences. The Lord Chancellor and Secretary of State for Justice the Rt Hon Dominic Raab asked HM Chief Inspector of Probation Justin Russell to conduct an independent review into this case. This review was completed in January 2022 and can now be published following the completion of the criminal proceedings.

Statement

HM Chief Inspector of Probation Justin Russell has made this statement:

“This was a deeply concerning case. The Probation Service’s assessment and management of Bendall at every stage, from initial court report to his supervision in the community, was of an unacceptable standard and fell far below what was required.

“Bendall had previously committed violent offences. His records show a former partner had made allegations of domestic abuse against him and a police child sexual exploitation unit had made enquiries about him with the Probation Service. Probation practitioners should take account of this sort of intelligence when assessing potential risks of serious harm. But this does not appear to have happened in this case.

“Bendall committed arson in May 2020. A member of the probation service’s court team interviewed him in June 2021 in order to prepare a report with sentencing options for the judge. The report author noted Bendall was suitable for a curfew requirement at the home of Terri Harris. They came to this wholly inappropriate conclusion without speaking to Ms Harris, visiting the property, conducting domestic abuse enquiries, or taking into account past domestic abuse claims.

“The court report author assessed Bendall as posing a medium risk of serious harm to the public and posing a low risk of serious harm to partners and children. We do not agree with this risk assessment; they under-estimated the risks Bendall posed and this had serious consequences.

“Probation managers and practitioners took the risk assessment from the court report as a given, and missed several opportunities to scrutinise and change it. If Bendall had been assessed as presenting a higher risk of serious harm – which would have been appropriate – it is unlikely a curfew order would have been deemed suitable and he would have been assigned to more experienced and confident probation officers.

“Instead, Bendall’s case was transferred to the East Midlands in the summer of 2021, and he was supervised by insufficiently qualified and experienced probation practitioners. The safety of Ms Harris and her children was not given due consideration. This was especially troubling as Bendall had started drinking alcohol and smoking cannabis again, which is likely to have increased the risk of serious harm.

“Probation services must strike the right balance between protecting the public and supporting individuals to move towards crime-free lives. Sadly, in this case, the balance was out of kilter.

“In January 2022, we published a separate thematic report on electronically monitored curfews which questioned why domestic abuse and child safeguarding enquiries are not mandatory before court ordered curfews. The Bendall case demonstrates clearly why these checks are so important.

“This review also highlights common issues that we have found in previous and recent inspections of probation services: the lack of qualified probation officers and managers with too many responsibilities to provide effective oversight for less experienced staff.

“The Probation Service must tackle these workforce issues. Probation practitioners must have the right knowledge, skills and experience to manage their assigned cases – and appropriate support and oversight from managers.

“We want to see probation practitioners and managers scrutinising case files and past criminal behaviour properly and developing a deeper understanding of the people they manage. We did not see enough ‘professional curiosity’ in this case – Bendall’s words and assertions were often taken at face value. Probation practitioners should be interrogating and verifying claims to build up a complete picture of the individual.

“As a result of this review, I made 17 recommendations for improvement to the Ministry of Justice, HM Prison and Probation Service and His Majesty’s Courts and Tribunals Service. They have accepted all these recommendations and responded with an action plan for implementing them. While this is welcome, over the past year in our local and national probation inspections we have continued to raise deep concerns about the quality of probation practice we find more generally in relation to the assessment and management of risk of harm. This is a subject I have raised repeated concerns about since becoming Chief Inspector. It is vital that this time lessons are learnt from this awful case.

--oo00oo--

Extracts from the full report:-

1. Foreword 

In September 2021, Damien Bendall was charged with the murders of Connie Gent (aged 11), Lacey Bennett (aged 11), John Paul Bennett (aged 13) and Terri Harris (aged 35), who was pregnant, and with raping Lacey. These shocking crimes have devastated families, friends, and the local community in Killamarsh, Derbyshire and beyond. In December 2022, the courts imposed a whole life sentence. 

Damien Bendall was on probation when he committed these crimes. The Probation Service typically conducts a Serious Further Offence (SFO) review when an individual on probation commits a serious violent or sexual offence. However, in this case, the Secretary of State for Justice asked me, as Chief Inspector of Probation, to conduct an independent SFO review into the Probation Service’s management of Damien Bendall. 

This report sets out the findings of that independent review. My inspectors found that the Probation Service’s assessment and management of Mr Bendall at each stage of the process from initial court report to his supervision in the community were of an unacceptable standard and fell far below what was required. 

Vital information about the serious risks posed by Mr Bendall to those he lived with, and the public, was not included in the Probation Service’s report and recommendations to the judge when he was sentenced for an arson offence in June 2021. As a result, he was sentenced to an entirely inappropriate curfew condition to reside with Ms Harris and her children. This was then compounded by a failure to allocate his case to an appropriately experienced and trained probation officer who could have managed him at the higher risk of serious harm level his past history certainly warranted. Several opportunities to correct these mistakes and amend his risk of harm classification and reallocate Mr Bendall’s supervision to an appropriate practitioner were missed in the period from June to September 2021. 

Inspectors found successive probation practitioners missed opportunities to ensure vital information known about Damien Bendall was included in assessments and plans to manage and address the risk of serious harm he posed to both women and children. Practitioners did not carry out safeguarding enquiries when he was sentenced for his most recent offence of arson. The impact of unmanageable workloads at both the probation practitioner and senior probation officer levels resulted in reduced oversight of new or struggling staff, frequent role changes and sickness absence. This made consistency and continuity of practice challenging. In this case, there was an increasing reliance on unqualified and trainee staff to manage workloads; this contributed to emerging factors linked to risk of harm not being recognised and escalated appropriately. 

This is a deeply concerning case that raises serious issues around the Probation Service’s assessment and management of risks of harm. This is a subject that has been of repeated concern to us in our local inspections and on which I have commented in my annual reports and in relation to other SFOs, 1 including that of Joseph McCann, 2 on which we reported in 2020. 

As a result of our findings, we make 17 recommendations for improvements to His Majesty’s Prison and Probation Service, His Majesty’s Courts and Tribunals Service and the Ministry of Justice regarding safeguarding and risk assessment practice and procedures, which I expect the service to respond to as a matter of urgency. It is vital that key lessons are learned from this awful case. 

Justin Russell

HM Chief Inspector of Probation

5. Executive summary 

Inspectors found that, at every stage of probation involvement, from the pre-sentence report provided to the court on 08 June 2021 to the commission of the SFOs in September 2021, the Probation Service’s assessment and supervision of DB fell well below the necessary standard. A failure to assign the correct risk of harm level to DB (which should have been ‘high’ risk of serious harm given his past history) meant that the court was missing vital information when reaching its sentencing decision. It is possible that, had a holistic assessment been provided to court (including his pattern of offending against Asian men, use of callous and organised violence against prison staff, an analysis of previous noncompliance and the most recent high risk of serious harm assessments), an immediate, rather than suspended, prison sentence might have been imposed. 

As it was, the court imposed a suspended prison sentence, which included an entirely inappropriate curfew condition to reside with Ms Harris and her children. The case was then allocated for community supervision to an inexperienced and inappropriate practitioner. 

There were then subsequent failures by supervising managers and new practitioners to adequately read the case and amend the initial, incorrect ‘medium risk of serious harm’ to ‘high risk of serious harm’. 

Had DB’s risk of serious harm to the public and children been correctly assessed as high, and had his risk of serious harm to partners been correctly assessed as medium, the court may not have curfewed him to an address with Ms Harris and her children. He would have been allocated to an experienced probation practitioner. This would have led to enforced weekly face-to-face appointments and improved communication with partner agencies, and assertions lacking evidence would not have been relied upon and repeated in future assessments. 

In sections 7 to 11 of this report, we analyse the management of DB during his two most recent sentences, the first a prison sentence with probation licence supervision imposed on 29 January 2017 and the second a suspended sentence order managed in the community imposed on 09 June 2021. In this summary we focus on our key lines of enquiry and summarise why, in our view, the following deficiencies occurred. 

Process for recommending curfew requirements 

The Criminal Justice Act 2003 requires that, ‘before making a relevant order imposing a curfew requirement, the court must obtain and consider information about the place proposed to be specified in the order (including information as to the attitude of persons likely to be affected by the enforced presence there of the offender)’. The current court process requires that court officers undertake domestic safeguarding enquiries ‘in order to assess risk of harm and suitability for sentencing options in all offences involving domestic abuse’,  which DB’s index offence did not. 

HM Inspectorate of Probation recently published a thematic inspection on electronic monitoring, including its use for curfews. In this report, we recommended that HMPPS: 

• mandate the requirement to make domestic abuse and safeguarding checks before recommending a sentence or release on electronically monitored curfew 

• work with the police and children’s social care at a national level to ensure that probation practitioners in every region are provided with domestic abuse and safeguarding checks in a timely manner.  

Before DB’s sentencing for arson, the court officer did not carry out domestic abuse enquiries on the address, find out whether Ms Harris’s children were known to children’s services or speak directly to Ms Harris to ensure she consented to her home being used as a curfew address by DB. The Sentencing Act 2020 requires courts to have sight of this information before imposing a curfew order. However, it appears that courts do not have a mechanism to ensure this information is seen in every case. In this instance, important checks were not carried out and the court proceeded to issue a curfew order without them. 

Child safeguarding 

Inspectors found that probation practitioners in this case based their risk of harm assessments on whether DB had convictions against children or for domestic abuse, or if children’s services were involved with the family. These are highly relevant factors, but probation practitioners should delve deeper to explore the broader attitudes and behaviours of the person under supervision, including their impact on the children in their lives. DB did not have a history of offending against children. However, we found that insufficient consideration was given to whether his racist, manipulative and controlling attitudes and his violent and unpredictable behaviour would have a negative impact on the wellbeing and safety of children. 

We did not find evidence of sufficient professional curiosity about the nature and level of the role he played in the lives of the children of his partners. 

Intelligence was available to the Probation Service from Wiltshire police’s child sexual exploitation team regarding DB’s risk of serious sexual harm to girls. However, this information was not explored or recorded sufficiently to inform the risk of serious harm assessment and plans to keep children safe. 

The probation practitioner who prepared the court report following DB’s arson conviction took his account and version of events in relation to his offending and circumstances at face value. This included DB’s assertion that he played an important part in taking care of Lacey and John Paul Bennett. This information was not checked with their mother. There were no checks to find out if children’s services were currently working with the family or had previously done so. Most egregiously, the report stated that DB was ‘suitable’ for a curfew at Terri Harris’s address. When considering a curfew in the home of children, the attitudes of the people in that home8 and the best interests of the child should be given weight. 

At the start of the most recent order, in June 2021, there was again a failure to be professionally curious about the children living with DB. To probation practitioners, DB presented himself as a father figure to the children of Terri Harris and this was accepted without challenge. No contact was made with the children’s parents. When DB admitted to using drugs and alcohol this was not escalated to a manager and a children’s safeguarding referral was not completed. We found that the risk of serious harm to children was inaccurately assessed and seriously underestimated. 

It is our view that there should be a section of the offender assessment system (OASys) that solely considers the wellbeing and safety of the children – actual and potential – in the life of the person on probation. This would separate children from assessments of broader familial and intimate relationships, and specific prompts should be used to facilitate a more rigorous and defensible assessment of the impact on a child’s ability to thrive. 

Domestic abuse 

During previous orders, DB’s relationships with his mother and grandmother were not explored appropriately. Probation practitioners did not demonstrate sufficient professional curiosity, did not conduct safeguarding enquires, and took information from DB, again, at face value. 

Inspectors found that key information on risk from prison and from DB’s ex-partner and her current partner was not given due consideration and was not recorded appropriately. The impact of this failure was significant, as successive probation practitioners did not recognise that DB posed a risk of serious harm within relationships. 

Probation at court appeared to take DB’s word without verification. The author of the court report noted that a curfew would be ‘suitable’; they did this without undertaking safeguarding enquiries on the address or communicating with the owner/lead tenant of the property. This loophole in the mandated checks required before a curfew recommendation needs addressing urgently. 

Probation practitioners should have explored DB’s relationship with Ms Harris in greater depth, including whether he was coercively controlling her. DB was open about the fact that he had very limited income and that Ms Harris was paying for his accommodation, bills and food. Inspectors conclude that contact with Ms Harris by the Probation Service before sentencing, and at key assessment stages and when there was evidence of increasing risk, would have been appropriate.

Inspectors found that the risk of serious harm to known adults, including partners, was underestimated. There was no focus on safeguarding in this case and, as a result, DB was sentenced to an inappropriate curfew requirement that may have exacerbated the risk of harm to Ms Harris and her children. 

Fast delivery report 

The use of a short format report in this case, rather than a standard delivery report, was incorrect. Mr Bendall’s criminal history was complex and as such met the threshold for a suitable adjournment period to allow for a thorough read of his case to inform the completion of a more detailed report. This case met HMPPS’s own criteria for a standard delivery report as ‘additional assessment, professional discussion and multiple enquiries [were] required to aid risk assessment’ and ‘liaison where medical report [was] unavailable on the day’.

Senior probation officer workload 

Inspectors found that high workloads and staff shortages in the Swindon office impacted on the ability of probation practitioners to undertake high-quality work. Inspectors heard that this was a long-standing issue that they had experienced since the changes introduced with Transforming Rehabilitation. 

HM Inspectorate of Probation has often found that the span of line management control for senior probation officers (SPOs) is concerning. SPOs increasingly deal with complex staffing and human resources issues, for which some feel unequipped. This also reduces the time they have available to provide effective professional oversight of the work of the practitioners they line manage with individual cases. HM Inspectorate of Probation has previously found that SPOs do not have enough time to supervise all members of their teams to the standard they would wish, and when they do hold supervision sessions, there is often a focus on managing volumes of work rather than improving quality. This case highlighted this issue on two specific occasions. 

Firstly, there was insufficient oversight of a member of the probation court team, which led to a poor-quality fast delivery report being presented to the court. This was due to SPO sickness and a lack of resources to cover the absence. 

Secondly, SPO3, who managed the probation practitioner responsible for DB after sentencing from June 2021, was unable to engage with the case fully. SPO3 managed a large number of staff. She directly managed 16, but when covering for colleagues she had oversight of up to 30 PQiPs. This is far in excess of the line management span recommended by HMPPS, of 10 full-time equivalent posts for SPOs. This prevented her reading DB’s case at the allocation stage and from providing the necessary oversight. 

Inspectors found that the SPOs were also not given meaningful, regular and effective supervision and support. 

Professional qualification in probation and probation services officer training and oversight

The probation practitioners who managed DB from June to September 2021 were inexperienced, unqualified and had insufficient support to understand and recognise the risks and needs in the case. We conclude that they should not have been exposed to cases such as DB at this stage in their careers. Following the unification of probation services, new guidance on allocations has been published, and this is welcomed. This guidance sets out clearly that ‘some case allocation decisions will rely on the judgement of the operational manager to decide whether a case is suitable to be managed by a probation officer or a probation services officer (PSO). This decision will be based on individual circumstances of the case, and the skills, ability and experience of the individual officers.’ 

Inspectors heard concerns about the efficacy of online training, especially for key learning on domestic abuse and child safeguarding, from all grades of staff, not just professional qualification in probation (PQiP) and PSO staff. There had been an understandable reliance on this method during the period of Covid-19 restrictions; however, some staff noted that prior to the pandemic there had been a trend towards self-reliant e-learning and development. Practitioners said that such self-selective training and development suffered when staff spent their hours ‘firefighting’ with excessive caseloads. DB’s case was one of 10 being managed by a staff member who had yet to complete basic safeguarding training.

6. Recommendations 

We have directed the recommendations to HMMPS and the Ministry of Justice to ensure national learning. HMPPS should: 

Court work and curfew requirements 

1. ensure that domestic abuse enquiries are carried out on everyone sentenced so that accurate risk assessments can be made and safe proposals are made in court reports 

2. ensure that child safeguarding enquiries are made in all cases where the person being sentenced lives with, is responsible for, has access to, or is likely to have a negative impact on the wellbeing or safety of a child 

3. develop a mechanism and reliable processes with relevant agencies to allow sufficient safeguarding enquiries to be completed, to verify information and therefore reduce reliance on self-disclosure 

4. ensure that sufficient safeguarding enquiries with relevant agencies are always carried out before finding a curfew requirement suitable, and that policy/practice guidance clarifies that assessment of suitability post-sentence should be ongoing. 

5. quality-assure risk assessments and proposals to the courts for accuracy and suitability 

6. introduce a process to contact relevant adult residents of the proposed curfew address and obtain their prior consent to a curfew condition at their address to assess whether the address is suitable for an electronically monitored curfew 

7. ensure that court reports provide a sufficient analysis of the person’s circumstances, including analysis of risk of harm, to provide safe sentencing options. 

Child safeguarding 

8. include a specific section in OASys that is dedicated to assessing and planning for the safety of children, and ensure that the nature of contact and impact the person on probation has in the life of the child have been considered on both current and future children in the person’s life 

9. ensure that the impact on children’s safety and wellbeing is sufficiently considered in every case. 

Risk management plans 

10. ensure that probation practitioners contact partners, family or other key adults in the lives of the person under supervision to determine and discuss their inclusion in risk management plans. 

Training and support 

11. consider the suitability and efficacy of online training, particularly on domestic abuse, child safeguarding and other key training required to correctly assess and robustly manage risk of serious harm 

12. ensure that each PQiP has access to a mentor who has at least two years’ experience as a qualified probation practitioner. 

13. dedicate time for probation practitioners to engage in reflective discussions with colleagues and the line manager regarding cases. 

Allocation practice 

14. ensure that NDelius entries for ‘management oversight – allocation’ include evidence that the manager has considered the complexity of the case and the capabilities and capacity of the probation practitioner receiving the case. 

Oversight of SPOs 

15. review and monitor SPO workloads to ensure that sufficient line management and management oversight of case work can be provided effectively 

16. review the line management responsibilities and supervision of SPOs responsible for PQiPs to ensure the standard of PQiP management and oversight is appropriately robust, including the suitability of the cases allocated to them. 

Ministry of Justice should: 

17. amend legislation to be more prescriptive of the information that should be obtained and considered by the court, to assure themselves of the safety of other household members at a proposed curfew address before they impose an electronically monitored curfew. 

Until this can be actioned HMCTS should issue guidance to court staff requiring them to satisfy themselves that relevant checks have been undertaken by the probation courts team.