Abstract

In the course of an police investigation, police in England and Wales can use “pre-charge bail” (also referred to as “police bail”) following arrest to place broad restrictions on suspects, such as forbidding them from visiting certain areas or forcing them to reside at a given address, with the aim of protecting victims or ensuring rigorous investigations. Following press criticism, pre-charge bail was reformed in April 2017, introducing broader oversight and time limits, to discourage perceived overuse. The impact of these reforms have not been studied in detail, and there have been concerns that they may have gone too far, placing victims at risk (notably vulnerable victims or victims of domestic abuse). We use data from the Metropolitan Police Service to examine the impact of these bail use and domestic revictimisation, using the sudden change created by the legislation as a natural experiment using regression discontinuity. We find that while the reforms did lead to large reductions in the use of pre-charge bail and modest increases in domestic revictimisation. This strongly suggests pre-charge bail use does reduce revictimisation, but not consistently, and that better application or targeting may be able to reduce police bail use while eliminating revictimisation risks.

Introduction

Pre-charge bail, also known as police bail, is a policing power introduced by the Police and Criminal Evidence Act 1984 (“PACE”). It enables police officers to impose restrictions on suspects which have been arrested, but not yet charged (eg, the investigation has not yet been completed). The suspect will be released from police custody and required to return at a later date while evidence is gathered. Notably, conditions can be imposed by police during this period if they are deemed necessary to prevent someone from failing to surrender to custody, prevent further offending, prevent someone from interfering with witnesses, or otherwise obstructing the course of justice.

A government review (Furlong et al., 2021) summarises the main purposes of pre-charge bail as:

• The protection of victims and witnesses, primarily linked to conditions applied to PCB such as no contact with the victim.

• Investigative management, allowing investigations to progress to obtain evidence.

• Suspect management, including reducing the risk of re-offending.

Whether police bail has a meaningful impact on these three goals is less clear: data on application of bail and bail conditions is not readily available, and as such its impact on compliance and revictimisation is not clearly understood, with a general lack of rigorous evidence.

It should be noted this paper is focused on pre-charge bail in England and Wales - this differs both legally and practically from bail after the point of charge, of bail as introduced in other legal systems, such as in Scotland or the United States.

Following concerns around the misuse of police bail, the Police and Crime Act 2017 introduced broad reforms from April 2017 aimed at curbing perceived unjust, unnecessary lengthy periods on bail1. These included strict time limits and expanded internal and external oversight, and introduced a new status of “Released Under Investigation” (RUI) for suspects while the investigation concludes.

While there is little rigorous data or evidence on the impact of these reforms, there is a broad consensus that they have not had the intended effect. Notably, there have been concerns that the reduction in the use of bail and conditions may lead to suspects being able to continue offending while the investigation is ongoing. This has led to calls for reform, which are expected to be introduced in the Police, Crime, Sentencing and Courts Bill 2021 which is currently being debated in parliament.

In this paper, we use custody and victimisation data from the Metropolitan Police Service to examine the impact of these reforms on bail use and revictimisation for victims of domestic abuse, using the 2017 reforms as a natural experiment, examining two hypotheses:

1. The April 2017 Reforms led to a reduction in bail use for suspected domestic offenders

2. Reduction in use of bail led to an increase in revictimisation for domestic victims

History of Police Pre-charge Bail

The Police and Criminal Evidence Act 1984 (“PACE”) was a wide ranging piece of legislation that aimed to reform much of the policing and criminal justice system in England and Wales, reviewing key police powers to investigate, arrest suspects, and seize evidence. Key to these powers is the use of bail: where an investigation is ongoing but has not concluded, be it during the investigation or judicial proceedings, suspects can be released “on bail”, introducing a legal requirement to return (or “answer bail”) at a certain time and place, and where appropriate, impose conditions - for example, this may include not engaging or visiting the address of vulnerable victims, or where a suspect is believed likely to escape or re-offend, a curfew.

Bail can be broadly differentiated into police or pre-charge bail (which takes place while an investigation is ongoing, and before a suspect is formally “charged”) and court bail, which occurs when a suspect has been sent to court or the judicial process is ongoing. These operate under different legal frameworks: police bail is given and enforced entirely by policing, and has no judicial punishment, while court bail is (after the initial court date) set by magistrates or judges, and can lead to additional judicial penalties if breached). Conditions can be imposed in either, and may vary from one to the next - for instance, police may enforce a curfew while the investigation is ongoing, that would then be removed by a magistrate upon first hearing at court.

Perceived overuse of police bail use was often criticised, especially for lengthy policing investigations which did not lead to a court case, but these criticisms came to a head in the mid-2010s. Significant, organised pressure began to be established in early 2011, with a media campaign against police bail launched funded by The Guardian Media Group and Telegraph Media Group highlighting the case of journalists kept on bail for over a year (Mediapoint & Twitter, 2017). By 2014, the BBC investigation found found that “40 out of 44 police forces in England, Wales and Northern Ireland show that there are currently 5,480 people who have been on bail for more than six months”, with the longest up to 3 and half years (BBC, 2014). In 2015, a series of high profile subjects such as Jim Davidson and Paul Gambaccini were arrested on suspicion of sexual offences as part of Operation Yewtree, kept on bail for months, and not charged, leading to highly public testimonies to government.

Following an initial consultation in 2014, the the Police and Crime Act was signed into law in 2017, introducing “a presumption against pre-charge bail unless necessary and proportionate and clear statutory timescales and processes for the initial imposition and extension of bail, including the introduction of judicial oversight for the extension of pre-charge bail beyond three months” (Unknown et al., 2021). Writing in the Sun, one columnist welcomed the reforms as a “common sense change” (Dathan, 2017).

Pre-charge Bail since 2017 and future reform

The reforms came into effect on the 3rd of April across England and Wales, introduced a new status for suspects where an investigation was ongoing but bail was not perceived as necessary: while no longer legally on bail, they would now be “Released Under Investigation”.

This created an overnight shock to policing behaviour, described by one researcher as “staggering”: while 67,838 were released on police bail in London in 2016/17, by the following year that had shrunk to 9,881, with 46,674 “Released Under Investigation” (Cape, 2019).

While a suspect who is “RUI’d” is not under any legal restrictions, they remain a suspect: far from accelerating police investigations, some suspected it was instead lengthening them. By 2019, the Law Society started a campaign against perceived abuses of RUI, highlighting that “unlike pre-charge bail it has no time limits or conditions. This can leave the accused and alleged victims in limbo with no updates on their case for an unlimited time.” Beyond this legal limbo, the Law Society highlighted the risk to the public from the removal of bail conditions, with “tens of thousands of suspects including some accused of violent crimes are being released by police without bail restrictions, leaving victims and the public increasingly at risk” (Bowcott, 2019). This perception of increased risk, especially for the most vulnerable victims, was highlighted by the Domestic Abuse Commissioner, saying “perpetrators of domestic abuse have not been subject to bail conditions which meant they could continue to intimidate their victims and cause untold anxiety, fear and concern” (Oppenheim, 2021).

The government began a consultation around the review of pre-charge bail in February 2020, and in March 2021, the Police, Crime, Sentencing and Courts Bill 2021 saw its first reading, stating “The Bill will encourage the police to impose strict conditions on bail more frequently” and “new pre-charge bail time periods for suspects which will be more balanced and proportionate”, noting that the consultation had “raised concerns that bail is not always being used when appropriate, including to prevent individuals from committing an offence whilst on bail or interfering with victims and witnesses”2. At time of writing, the legislation is being debated in Parliament.

Literature Review

While there is an extensive body of research around UK policing, evidence on pre-charge bail is rare, especially in regards to its impact on suspect behaviour.

What research exists is focused on small datasets, and focuses only on those suspects who were bailed (rather than comparing to the non-bailed suspects). Hucklesby (2001) undertook a small study of one force in the North of England, using survey responses from 38 custody officers and custody records from 281 bail decisions. They found that while the power had reduced the need to retain suspects in custody, there was broad variance in how the power was used between units, enabling “informal practices” to shape the process. Custody conditions were used frequently, with 61% being required to reside at a given address, and 52% having some restrictions on movement (Hucklesby, 2001). They highlight that given this variation in decision making and lack of accountability, some uses of bail may be open to legal challenge under the Human Rights Act, while also noting practical challenges: as breaching bail is not punished with a formal judicial sanction, officers noted they at times felt disincentivised to enforce given conditions.

The legal concern is restated by Cape & Edwards, 2010, who highlight the long term, unsupervised use of the power by individual officers. In a 2015 pamphlet, Cape would go onto justify the abolition of the power given these objections (Cape, 2015).

In response to government proposals for proposed reform in 2015, the College of Policing (the professional body for the police in England and Wales, established in 2012) began a pilot study based in 6 forces, attempting to understand the potential impacts of the legislation by introducing internal oversight and restrictions on length. While the government pushed ahead with the reforms before the conclusion of the pilot, “overtaking the purpose of the study”, Hillier & Kodz, 2012 raise concerns around how reforms may impact length of investigations, particularly around the availability of forensic analysis - this concern is also shared by Horsman & King, 2018 around digital forensics. Neither examine re-offending or victimisation.

In response to government consultation on police bail reform, Hucklesby, 2020 conducted a wider data-driven exercise into two forces bail data, analysing 14,173 pre-charge bail records and interviews. They note that evidence on the effectiveness of bail conditions “does not exist”, noting the difficulties in accessing data.

Despite this lack of evidence, several researchers highlight that bail conditions create a perception of security amongst victims: in a 2020 report by the criminal justice inspectorate HMICFRS (2020), based on interviews and poling, found that suspects released under investigation left victims “without reassurance and protection that bail conditions can provide” (Unknown et al., 2020). This concern is echoed by a report for End Violence Against Women, where Learmonth, 2020 notes “Victim-survivors’ safety is doubly impacted, both by the reduction of the police use of pre-charge bail and by the fact that conditional pre-charge bail is used as affirmation influencing victim-survivor protection in the wider safeguarding system”.

Furlong et al., 2021 conducted a review of the available evidence for government: they highlight the lack of meaningful data or rigorous research, while noting that decreased use of bail had led to various criticisms, particularly around weakened investigative management, less information for suspects regarding the progress of their case, and weaker protections for victims and witnesses. They highlight the reduction in bail has in part been reversed, with the preceding two years seeing an 81% increase in the use of PCB, from approximately 84,200 to 153,500, in England and Wales. Nonetheless, they conclude that there is “very limited robust evidence directly associated with the effectiveness of PCB”.

This review highlights some broad themes around the use of pre-charge police bail since 1994: despite legislative reforms and concerns around consistency and legacy of its use, it has remained a relatively constant and common part of the policing investigative process, including the use of conditions. There is widespread perception that the application of these bail, and conditions, enhances the safety of victims, and prevents further offending. Despite this, the lack of meaningful data means there is no rigorous evidence around the impact of police bail on victim safety or offending.

Data and Methods

Data

This paper uses two data sources provided by the Metropolitan Police Service, for the period of April 2016 to 2018 (a year to either side of the reform implementation on the 3rd of April 2017).

Custody Data

Our dataset contains a record of all individuals arrested, for a total of 341,191 records, as well as whether the arrest was listed as for a domestic offence (though not the exact nature of the offence), and whether the individual was subject to police bail.

Victimisation Data

Our dataset contains details of victims crimes reported to the MPS where the crime was flagged as domestic - a total of 350,124 victim records. For each, we create a unique identifier by combining surname and date of birth, and create a measure of revictimisation, measured as whether the victim reports a second offence to police within 30 days of the first report.

Method

To examine the impact of legislation, we use the impact of the legislation as a natural experiment, using a regression discontinuity or regression kink design, exploiting the shift caused caused on either side of “threshold” on the 3rd of April, examining the impact on both:

• Use of pre-charge police bail, for domestic and non-domestic detainees in custody

• Revictimisation rates (eg, likelihood of reporting a second domestic offence within 30 days of a prior report) for domestic victims

Results

Impact on Bail Use

Examining bail use on either side of the legislative change, a visual observation suggests a stark shift in officer behaviour, not only in response to the legislation change, but potentially in anticipation of it: both domestic and non-domestic cases see a stark rise in police bail use as early as November 2016, and a very sharp increase in the month prior to the legislative change.

Figure : Bail Use in Custody, 30 Day Rolling Mean

This suggests strong seasonal variation, as well as shifts in reaction to the legislation: 20.6% of domestic detainees are bailed over the course of our dataset, and this shifts from 28.6% before the legislation to only 11.1% afterwards, with non-domestic detainees moving from 39.5% to 8.5%. It is notable that the reduction is relatively less impactful among domestic detainees - this suggest the legislation did have the desired effect, with a reduction in broader bail use tempered for perceived most serious offences or the most vulnerable victims.

To examine this impact in detail, we use the regression discontinuity we discussed earlier. Instead of focusing on the individual binary outcome of whether or not any individual is bailed, we instead examine average (mean) bail use across all our domestic detainees, and build a ordinary least squares regression modelling how this varied as a factor of:

1. Our intercept (or constant), accounting for baseline bail use

2. “running_days”, the number of 24 hour days (from 0000 until 2359) since the introduction of the reforms on the 1st of April

3. “threshold”, a binary indicator which is 1 after the introduction of the reforms, and 0 beforehand

4. The interaction effect between the two variables.

Given OLS regression is not well suited to binary outcomes (such as whether or not someone was bailed), we instead aggregate to daily mean bail use, illustrated as below:

$Pr(Bail\ Use)\ = \ constant\ + \ running\ days\ + \ threshold\ + \ runningdays*threshold$

Modelling purely with these parameters, our regression is a good fit to our data, with 69% of the variance in bail use during this period accounted for by our variables (based on our r-squared of 0.687). Our output is available in , suggests a strong, statistically significant (at p<0.001) reduction in police bail use with the implementation of the reforms, indicated by our threshold variable coefficient of -0.1818. In layman’s terms, this shows a reduction in bail use from 29% prior to the legislation to 11% afterwards.

This is illustrated in , which highlights mean daily bail use by day, centered around the reforms at “day 0”, as well as our regression’s predicted bail use line, showing the stark, rapid changed in behaviour centered on the date of the reforms.

Figure : Regression Discontinuity - Proportion of Domestic Detainees Bailed

One limitation of regression discontinuity applied here is extrapolation: we fit a line through the “before” and “after” part of our data. If the impact is not “sharp” and centred on a specific date, this may lead to misleading results, with our model attempting to fit a straight line to a non-linear relationship. To account for this caveat, we repeat our model using triangular kernel weighting, focusing entirely on the 60 day period on either side of the reforms, and over-weighing those observations closest to the reform date. Adding these weights reduces our model fit (as measured by our r-squared of 0.58, in contrast to 0.68 for our full model), suggesting the change was sharp and centered around the reform date, with our initial, “regression kink” design being most appropraite.

As such, we can conclude that the reforms did have a strong impact on bail use, leading to large reductions for domestic suspects.

Impact on Revictimisation

We have shown that the reforms led large reductions in the use of police bail. If our theory is correct - that police bail us a protective measure, reducing further victim harm - we would expect this reduction to lead to large impact on victimisation.

A visual observation of mean re-victimisation on 90 days of either side of the reforms (see ) does suggest some variation, though the data is noisy and any shifts appear minor - there is no clear visual break, in contrast to bail use (though given our 30 day revictimisation time window, a noisier effect would be expected)

Figure : Mean Repeat Victimisation, 90 Days Before and After Bail Reform

To provide a more coherent visual model, we visualise the revictimisation effect per day, baselined to the 2nd of April, by building a regression model and examine each daily coefficient - the coefficients are visualised in , and again, suggest not clear impact around or in the month following implementation.

Figure : Daily Re-victimisation and Errors, Baselined to 2rd of April

To obtain a metric for the impact of the reform, we repeat our regression discontinuity, using the same method as bail. Our model includes

1. Our intercept (or constant), accounting for baseline daily revictimisation

2. “running_days”, the number of 24 hour days (from 0000 until 2359) since the introduction of the reforms on the 3rd of April

3. “threshold”, a binary indicator which is 1 after the introduction of the reforms, and 0 beforehand

4. The interaction effect between the two variables.

Or mathematically illustrated as below:

$Pr(Daily\ Revictimisation)\ = \ constant\ + \ running\ days\ + \ threshold\ + \ runningdays*threshold$

Modelling purely with these parameters, our regression is a poor fit to the data (based on our r-squared of 0.025), our threshold variable does suggest reforms led to statistically significant (at p<0.05) increase in revictimisation. The full table of coefficient is available in , but in layman’s terms, suggest that prior to the reforms, mean domestic revictimisation was 9.26% (based on the intercept coefficient of 0.0926), but increased by to 10.11% after the reforms (with our threshold coefficient 0.0085 being added to our intercept coefficient, for a total of 0.1011).

Our daily variable “running_days”, as well as the interaction effect between that and the threshold variable, are both non-significant (at p<0.05), suggesting that we see no meaningful other shifts in revictimisation over time - there is no longer term shift in revictimisation. The trends are illustrated in .

Figure : Repeat Victimisation Regression Discontinuity - No Weights

Unlike the reduction in bail use, which we would expect to be stark and immediate (given it was mandated by legislated change), impacts on revictimisation are a “secondary” effect - as offenders are released without bail, we predict they will then go on to re-offend over the coming month. As such, we would expect the effect to be “noisier”, though remain centred on the date of reforms.

With this in mind, we re-run our initial model, but apply triangular kernel weighing with a bandwidth of 90 days (meaning the most important observations are the days on either side of the reforms, with the days further away being progressively discounted up to a maximum of 90 days, following which they are disregarded - the model coefficients are highlighted in .

In this model, our threshold variable (the impact of the legislation) is no longer significant (at p<0.05). Given our earlier finding that we do see a significant increase in our unweighted model, this suggests that while an increase in revictimisation does occur during this period, it is not focused around the 3rd of April.

To identify where the nature of this change took place, where it is centred, and how the effect varies over time, we build a model matrix: we rebuild our dataset, but placing our “threshold" date at varying monthly intervals from December 2016, up until August 2017, and rebuild our regression discontinuity for unique threshold date. In addition, we vary the duration (or bandwidth) of our dataset, from a minimum of 30 days to either side of our threshold, to a maximum of 180 (this is as reasonable maximum given the dates of our dataset). For each model, we record the p-value and coefficient of our threshold. To aid in interpretation, this is visualised as a heatmap in .

Figure : Revictimisation Threshold Coefficient Data and Dataset Duration Matrix (rounded to 5dp)

While p-values should be interpreted with caution (as we have conducted a high number of tests without corrections), this serves as a useful visualisation and impact of discontinuities in domestic revictimisation in our dataset, with the smallest p-values suggesting sharp changes around the given threshold date, within the limitations of the given duration. For our smallest day dataset (a bandwidth of 30 days to either side of the threshold), there are two threshold dates where p < 0.05, suggesting a meaningful increase in revictimisation: December 2016, and May 2017 (30 days after the introduction of bail reform). If we extend our bandwidth to 60 days however, we see the effect of December “kink” disappears, given the p-value of 0.27. Conversely, p-values for our May threshold date remain significant (at p <0.05) up until 120 days to either side of the threshold. Re-examining our initial reform threshold date of April, we can see that while the effect is not immediate (given at the 30 day bandwidth p > 0.05 for both the 3rd and 17th of April), it does become clear from the 90 day bandwidth mark, and remains so as bandwidth increases. This strongly suggests a lagged, statistically significant increase in revictimisation, centred around the date of the reforms in April 2017.

Discussion

As we noted in the introduction to this work, two of the stated goals of pre-charge bail are the protection of victims and witnesses, and the management of suspects. While prior research has not established how well (if at all) police bail succeeds in achieving these goals, we hypothesised that by reducing the overall use of bail and bail conditions, the 2017 reforms led to an increase in repeat victimisation, focusing on domestic victims.

We considered first part of our hypothesis - that reforms led to a reduction in bail use for suspected domestic offenders - by using custody data, examining what proportion of domestic suspects were subject to police bail before and after the reforms, using a regression discontinuity design to identify the effect of the “break” in behaviour created by the legislative reforms. We found the reforms had a strong, statistically significant effect on bail use, leading to a reduction in bail use from 29% prior to the legislation to 11% afterwards. In simple terms, of the 32,090 domestic records in the year after the reforms (from April 2017 until April 2018), around 3,500 were placed on police bail, but had pre-reform trends continued, we would have expected around additional 5,800.

We then considered the second stage our hypothesis: did this reduction in bail use lead to an increase in revictimisation for domestic victims. Using a regression discontinuity design, exploring the proportion of victims who report a further crime to police within 30 days of the initial offence, we see a statistically insignificant increase in revictimisation, increasing from 9.26% to around 10.11%. Our dataset shows 174,183 records of victimisation between April 2017 and 2018 - our model suggests that during this period, around 17,600 would have been revictimized within 30 days, in contrast to around 16,130 under pre-reform trends: our model suggests the reforms led to an additional 1,500 domestic victimisations across London between April 2017 and April 2018.

These findings should be caveated due to limitations in our method, driven by the disconnected nature of the available data. We suggest that repeat victimisation is driven by offender “availability” when not on bail, but information on each side of this are held on different systems, with no unique individual identifiers - while this limitation is inherent to how the data is recorded, it limits the validity of our analysis, we bail and victimisation records separately (rather than examining on a crime by crime level). We have relied on the artificial break in behaviour forced by legislation to examine the changes in bail and revictimisation patterns, but these could be correlational rather than causal. A more rigorous approach would benefit from a true experimental approach, such as a Randomised Control Trial, or detailed, person level statistical approaches for causal inference (such as a regression model or matching examining the effect of bail on otherwise similar suspect-victim pairs).

What we have not examined is the nature of this relationship, or the exact mechanism by which it functions. We have shown a sharp reduction in bail use of around 62% (from 29% of offenders to 11%) was linked to an increase in revictimisation of 9% (from 9.3% of victims to 10.11%). This suggests that only a small proportion of offenders who were on bail went on to re-offend - for instance, it may have been possible to reduce bail use by only 40% instead of 60%, but focus more accurately on the offenders most likely to re-offend (potentially through some type of algorithmic forecasting tool, such as the HART model used in Durham (Oswald et al., 2018)), or even a more simple approach relying on offence type or offender history. Future research, especially in the form of a randomised trial, may want to consider subgrouping offenders at an early stage to understand where bail use can be safely reduced.

Conclusion

This research is one of the first rigorous, data-driven approaches to examine the impact of police bail in domestic revictimisation. By using the 2017 legislative reforms as a natural experiment, we have conclusively shown there is a link, and it is highly likely that this is causal - a 62% reduction in use of police bail led to a 9% increase in repeat victimisation.

We have noted that our method is not perfect, and our findings should be suitably caveated - we cannot confidently state exactly how these factors are interlinked, and future research with better data may wish to explore the nature of this relationship. Nonetheless, we have shown police bail does help reduce revictimisation for domestic abuse victims, and future policy changes should consider how to best test the impact of policy changes in advance.

A secondary finding from this work is that the relationship is not “1 for 1” - a very large reduction in bail (of 62%) led to a relatively modest increase in revictimisation (of 9%). Previous research has highlighted justified concerns with the potential over-application of pre-charge bail, and the long-term impact on suspects - the reduction in bail use likely led to meaningful positive impacts, and it is likely that were the use of bail better targeted on suspects most likely to re-offend, future reforms may be able to maximise the benefits while reducing or eliminating potential impacts on victims.

Acknowledgements

Data for this work by the Metropolitan Police.

All analysis was coded in Python using open-source libraries, notably Statsmodels and Matplotlib.

Causal Inference for the Brave and True (Facure, 2021) was critical in letting me finally feel like econometrics was not terrifying.

References

Furlong, S., Richardson, V., & Feist, A. (2021). Pre-charge bail: an overview of the evidence (p. 73) [Techreport]. Home Office.

BBC. (2014). Police bail time limit for suspects considered. https://www.bbc.com/news/uk-29624498

Cape, E. (2019). Police bail without charge – leaving suspects in limbo [Techreport]. Centre for Crime. https://www.crimeandjustice.org.uk/resources/police-bail-without-charge-leaving-suspects-limbo

Cape, E. (2015). What if police bail was abolished? [Techreport]. Howard League for Penal Reform. https://howardleague.org/publications/what-if-police-bail-was-abolished/

Mediapoint, P. A., & Twitter, P. G. (2017). Police bail limited to 28-days from today after campaign launched when journalists spent years in legal limbo. Press Gazette. https://pressgazette.co.uk/police-bail-limited-to-28-days-from-today-after-campaign-launched-when-journalists-spent-years-in-legal-limbo/

Horsman, G., & King, A. (2018). Policing and Crime Act 2017: Changes to pre-charge bail and the impact on digital forensic analysis. Computer Law & Security Review, 34(5), 1139–1143. https://doi.org/10.1016/j.clsr.2018.04.005

Bowcott, O. (2019). Sharp rise in suspects being released without bail conditions. The Guardian. https://www.theguardian.com/law/2019/oct/09/rise-number-suspects-released-under-investigation-without-bail-conditions-police

Dathan, M. (2017). Police bail cut down to just 28 days. The Sun. https://www.thesun.co.uk/news/3239432/police-bail-cut-to-just-28-days/

Hucklesby, A. (2001). Police Bail and the Use of Conditions. Criminal Justice, 1(4), 441–463. https://doi.org/10.1177/1466802501001004004

Police Powers, Pre Charge Bail Government Response (p. 29). (2021). [Techreport]. Home Office. https://assets.publishing.service.gov.uk/government/uploads/system/uploads/attachment_data/file/953715/2021-01-14_Response_to_PCB_consultation__003_.pdf

Hucklesby, A. (2020). Response to the Government consultation on Police Powers: Pre-charge Bail [Techreport].

Oppenheim, M. (2021). Police to implement strict bail conditions on domestic abuse and sexual violence suspects under new measures. The Independent. https://www.independent.co.uk/news/uk/crime/bail-conditions-police-domestic-abuse-sexual-violence-b1787227.html

Hillier, J., & Kodz, J. (2012). The police use of pre-charge bail: An exploratory study [Techreport]. College of Policing.

Facure, M. (2021). Causal Inference for the Brave and True.

Cape, E., & Edwards, R. A. (2010). Police bail without charge: The human rights implications. The Cambridge Law Journal, 69(3), 529–560. https://doi.org/10.1017/S0008197310000796

Oswald, M., Grace, J., Urwin, S., & Barnes, G. (2018). Algorithmic risk assessment policing models: lessons from the Durham HART model and ‘Experimental’ proportionality. Information & Communications Technology Law, 27(2), 223–250. https://doi.org/10.1080/13600834.2018.1458455

Learmonth, S. (2020). A system to keep me safe: an exploratory study of bail use in rape cases [Techreport]. End Violence Against Women. https://www.endviolenceagainstwomen.org.uk/wp-content/uploads/20-July-2018-Executive-Summary-Sarah-Learmonth-A-system-to-keep-me-safe-an-exploratory-study-of-bail-use-in-rape-cases.docx

Pre-charge bail and released under investigation: striking a balance (p. 32). (2020). [Techreport]. Her Majesty’s Inspectorate of Constabulary. https://www.justiceinspectorates.gov.uk/hmicfrs/wp-content/uploads/pre-charge-bail-and-released-under-investigation-striking-a-balance-1.pdf