Research and analysis

Identifying factors associated with changes in charge volumes: a statistical analysis

Published 8 May 2025

Applies to England and Wales

Authors

Cloe Cole, Victoria Richardson, Lucy Flanagan, Michelle Diver, Steve Almond and Andy Feist

Acknowledgements

We are very grateful to Professor Shane Johnson MBE, Director of the Dawes Centre for Future Crime at UCL and to Professor Stephen Machin, Director of the Centre for Economic Performance (CEP) at the LSE for peer reviewing this report and for providing helpful comments on an early draft. We would also like to thank Alex Hicks for his contributions to this study.

Executive summary

Background and aims of the study

Between the years ending 31 March 2014 and 2023, charge volumes in England and Wales fell by 34%. However, the factors that contributed to this fall are not well understood.

This report summarises the findings from a study which sought to identify the key factors driving recent trends in charge volumes and other outcomes in England and Wales. The approach involved applying statistical techniques (panel regression) to recent police force-level data to build a series of crime-type specific models. The work was conducted as part of the criminal justice system (CJS) Demand Insights Project, supported by the Treasury’s Shared Outcomes Fund (SOF).

The current evidence base

The existing research into what factors drive charge volumes comprises a small number of studies which have mainly explored how individual factors, such as police workload or resources, influence case outcomes. Few studies have sought to address the question of what, more broadly, drives changes in charges over time.

In terms of the relationship with police officer numbers, most US studies find additional officers tend to be associated with fewer arrests for victim-based crimes. The suggestion here is that the hiring of additional officers results in increases in crime deterrence, and in doing so results in fewer crimes being available to detect.

There are several studies on police workload and crime outcomes. The most robust studies suggest that additional crime workload suppresses the likelihood of charges, and the effect is felt greatest on less serious offences.

A small number of studies have examined the relationship between investigative resources and case outcomes. The most robust studies suggest that increasing investigator resources can lead to more charges.

The relationship between officer experience and charging is not well-evidenced but has been highlighted in some studies. The suggestion is that a less experienced police workforce suppresses charge volumes.

Although rarely a focus of specific studies, there is some evidence that changes in crime availability – the number of recorded offences available to detect – can influence charge volumes for some, mainly less serious, offences.

Methodology

Data for the study was taken from multiple published sources to build a force-level year panel data set, covering various measures of police workforce and capacity, crime workload, crime availability, and police activity (number of stop and searches).

Fixed effects panel regression models were run for 12 offence types: 4 property offences (shoplifting, burglary, other theft, and criminal damage), 4 violence and sex offences (assault with and without injury, adult rape, and rape of a child under 13), and 4 possession offences (weapons, drugs, drugs trafficking and going equipped to steal). Most models used data covering years ending March 2011 to 2023.

The main outcome of interest was charge volumes. For some crime types, additional outcome measures were used: the number of suspects proceeded against, cautions, and out-of-court resolutions (OOCRs).

Findings

Workforce stability

For all 8 victim-based offence types (property, violence, and sex offences) officer experience was a significant predictor of charge volumes. The more experienced the officer workforce, the higher the charge volumes. Officer experience was generally found not to be a significant predictor of possession offence charge volumes.

Annual officer turnover was also a consistently significant predictor of charge volumes for almost all victim-based offences. An increase in officer turnover was associated with fewer charges; the effect size for officer turnover was less strong than for officer experience. However, turnover more often passed a stricter test of association. Officer turnover was not a significant predictor of charges for possession offences.

Workforce size

The relationship between headcount (either total officers or total workforce) and charge volumes was less consistent across the models than either experience or turnover. Only half the property, violence and sex offence models found headcount to be a significant predictor of charge volumes. Within the possession offences, officer headcount was an important predictor of drugs possession charges. Where significant, an increase in headcount was associated with an increase in charges.

Officers in investigative functions were not found to be a significant predictor of charge volumes. This is likely to be due to the flat trend in investigator numbers during the period.

Crime workload and availability

Crime workload, measured by Crime Severity Index-weighted recorded crime, was a significant predictor of charge volumes for all property offences and 2 possession offences (drugs possession and going equipped). In these offences, an increase in crime workload was consistently associated with fewer charges. It was not possible to test the impact of workload in the violence and sex offence models.

For 4 selected crime types (shoplifting, burglary, assault without injury, and adult rape), a separate set of models were run with recorded crime volumes included as a predictor to assess the relationship between crime availability and charge volumes. Crime availability was consistently found to be positively associated with charge volumes.

Stop and search

For all possession offences, the number of stop and searches undertaken was consistently associated with higher volumes of charges. Although these effects were small, they are additive – an increase in stop and search volumes will lead to simultaneous increases in charges for all 4 possession offences.

Finally, for the possession offences, measures of the underlying level of related offending were included in the models (for example, weapons possession charges might be associated with changes in underlying knife-enabled crime); weapons possession charges were positively associated with the level of hospital admissions for assault with a sharp object.

Discussion

Overall, this study finds that a range of workforce and crime factors influence charge volumes and other related outcomes (such as suspects proceeded against).

Among victim-based crimes, measures of officer experience and officer turnover were both consistently associated with charge volumes; these factors were typically less important for possession offence charges.

The relationship between officer headcount and charge volumes was less consistent for victim-based offences than either officer experience or turnover. Within the possession offences, the relationship between officer headcount and charges was most marked for drugs possession offences.

There are important differences between the findings from this study and the US research on the relationship between officer headcount and case outcomes. US studies typically have found a negative relationship (extra officers are associated with fewer arrests). This may reflect the specific characteristics of police workforce expansion in the US which features heavily in these studies and the more focused nature of the US studies on the relationship between officer numbers and arrests.

Although this analysis did not find investigator numbers to be a significant predictor of charge volumes, the finding needs to be treated cautiously. Other studies point to investigator resources being important in increasing charges.

The consistent negative impact of crime workload on charges for property crimes provides some further support for the ‘overload effect’ hypothesis. This states that the negative impact of higher crime workload falls disproportionately on the investigation of less serious offences, such as property and possession offences, with consequent reductions in charge volumes.

Although this study starts to address an evidence gap, it has limitations; a regression approach is useful for measuring association but cannot identify causal relationships with confidence. Other limitations relate to the small samples for some offences, important gaps in the data (for example, around changing crime complexity and Crown Prosecution Service resourcing) and the impact of changes in crime recording practices during this period.

1. Introduction and background

One important, but seemingly neglected, area of criminological research centres on identifying factors associated with changing volumes of upstream demand on the criminal justice system (CJS), typically crimes charged, or suspects proceeded against. There are several reasons why this is an area of critical policy interest. First, until total charge volumes started to increase in the year ending 31 March 2023, there had been year-on-year falls in charge volumes in England and Wales since the mid-2010s. Between the years ending 31 March 2014 and 2023, charge volumes fell from 604,728 to 397,537, a decrease of 34% (Home Office 2023b). This raises questions about what drove the decline, and what might now be driving the reversal. Secondly, more clarity about the factors that drive changes in charge volumes is important in improving understanding of the future pressures on courts and prisons. Finally, it relates to one key area of police performance, the investigation and detection of crime.

This research was carried out as part of the CJS Demand Insights Programme, supported by the Treasury’s Shared Outcomes Fund (SOF). This Home Office-led work was undertaken with support from key partners such as the Ministry of Justice (MoJ) and the Crown Prosecution Service (CPS). The overarching aim of the Programme was to improve understanding of the relationship between crime demand, policing and CPS on the one hand, and courts and prisons on the other. One paper has already been published from the Programme (Wharfe et al., 2024). This used the conditions of COVID-19 as a natural experiment to consider the impact of changing crime demand on charge volumes.

The specific focus of this study was to use statistical analysis to identify factors associated with changes in charge volumes, using force-level data in England and Wales. While the focus is on charge volumes, the analysis also examines the relationship with the numbers of suspects proceeded against, and other positive outcomes (cautions and out-of-court resolutions). The range of factors of interest for this study – which potentially may influence charge volumes – include changes in police resources or the characteristics of the police workforce; levels of police activity (such as stop and search); levels of demand placed on the police (such as changes in crime demand); and wider socio-demographic factors. The relatively small number of existing studies in this area have tended to focus on the relationship between an individual factor (such as police resources) and case outcomes, rather than explore how combinations of factors might drive changes over time. This study aimed to look more broadly at the range of factors that might influence charge volumes.

Multivariate linear regression using panel data was used to identify which factors are most associated with changes in charge volumes over time, using recent force-level data on charges and other outcomes in England and Wales. Before considering the existing evidence base and going into the detail of the modelling approach, the following are some general observations about recent changes in charge volumes which guided the analysis.

Both total charge volumes and the charge ratio (number of charges divided by volume of recorded crime) have been declining annually since the year ending 31 March 2014 (see Figure 1). There was a slight recovery in the year ending 31 March 2023. This general pattern is common to all Police Force Areas (PFAs) in England and Wales, but the pattern varies both in the baseline (charges per capita by force area), and the rate of change by force area. This statistical analysis mainly looks at 40 of the 43 force areas in England and Wales and examines the degree to which the factors included in the models explain this variation. The reasons for excluding forces are provided in Annex A.

Figure 1: Trends in charge volumes and charge ratio, years ending 31 March 2010 to 2023

Source: Home Office (various dates) Police recorded crime and outcomes open data tables

There are important differences in trends in charge volumes by offence type. For instance, as shown in Figure 2, overall victim-based crimes (crimes with a specific, identifiable victim, such as assault or residential burglary) have seen a more consistent and marked reduction in charge volumes than state-based crimes (such as drugs offences). However, even within victim-based offences, the patterns are far from uniform. While theft offences have typically seen charge volumes fall steadily year on year, the pattern for adult rape (both rape of a male over 16 and rape of a female over 16) is markedly different. Stable levels of charge volumes for adult rape were followed by a precipitous fall in charges in the year ending March 2018 (HMG, 2021). This suggests that while there may be some interest in modelling charge outcomes for all notifiable offences combined, the differences in trends for specific offence categories make it important to model crime types individually. This was the approach taken in this study.

Figure 2: Trends in charge volumes over time: victim-based and state-based offences, years ending 31 March 2010 to 2023

Source: Home Office (various dates) Police recorded crime and outcomes open data tables

One important element of this study was to explore whether stop and search activity was associated with changes in charge volumes for those offences which searches might most likely discover. Comparing data for the year ending 31 March 2023 on stop and search-generated arrests for drug and weapons possession offences with all arrests for drugs and weapons offences suggests that approximately two-thirds of total arrests for these 2 offence types are due to stop and search activity. Charges for other offences – possession of drugs with intent to supply and going equipped to steal – might also be influenced by levels of stop and search activity. Figure 3 shows how stop and search volumes have changed over the last 10 years.

Unlike the steady fall in total charge volumes, stop and search and charges for the 4 possession offences (drugs possession, weapons possession, going equipped to steal, and drugs trafficking) appear to have moved in step with each other, independently of other crime types. Drugs trafficking covers a wider range of offences than possession with intent to supply (for example, it includes importation and production) but charge data for this sub-offence are not available for the full period.

Figure 3: Section 1 and Section 60 stop and search volumes, charge volumes and possession offence charges indexed to year ending March 2011

Source: Home Office (2020; 2023a) for stop and search data and Police recorded crime and outcomes open data tables.

The remainder of the report is structured as follows: Section 2 summarises the relevant research literature; Section 3 summarises the methodology and data sources; Section 4 provides the main findings from the analysis; and Section 5 discusses the findings and summarises the main limitations of the analysis.

2. Literature review

To help develop a long list of relevant predictor variables for the models – that is, those factors that might have some relationship with changes in charge volumes – the existing research literature was reviewed. Previous research into this area falls arguably into 2 groups.

1. Several US studies examining how changes in police resources have impacted on arrests and arrest rates.

2. A more diverse set of studies exploring a variety of factors – workload, investigator resource, officer experience – and their influence in determining investigative outcomes.

US studies generally use arrest volumes or arrest rates as a measure of case clearance, rather than charges or suspects proceeded against. This largely reflects the availability of data on arrests in the US, which are collected by the FBI’s Uniform Crime Reports (UCR). For this study, with its focus on charge volumes, studies on arrests or arrest rate are still considered useful and relevant research evidence.

2.1 Officer headcount

Findings from most US studies that have examined – either directly or indirectly – the relationship between changes in officer numbers and arrests suggest that, overall, additional officers tend to be associated with fewer arrests for victim-based crimes. It is suggested that the hiring of additional officers results in greater crime deterrence (for example, through greater police visibility), and this in turn results in fewer opportunities to detect^{[footnote 1]} crime, leading to a reduction in arrests. This relationship appears plausible for many crime types. However, several studies record that more officers have a suppressive effect on serious sexual offences, which is unlikely to be the result of increased visibility or a deterrent effect. Table 1 summarises some of the key studies in this area.

Although there are similar UK studies (for example, Machin & Marie, 2011, which examined the impact of additional police resources on robbery offences), these do not examine the impact on arrests or charges.

Most of these headcount-arrest studies have used panel regression analysis to examine the association between changes in officer numbers and arrest volumes over several years (for example, Kaplan and Chalfin, 2019). Kaplan and Chalfin concluded that for each officer hired, there were 1.3 fewer arrests per annum. However, as with several other studies, there is some imprecision in the estimates and the authors acknowledge the possibility that arrests per additional officer could increase by up to 0.7 per annum. The authors also conclude that increases in police spending do not lead to appreciably larger prison populations and might even lead to reductions.

Some studies have used the introduction of federal grants for hiring police officers administered by the US Department of Justice Community Oriented Policing Services (COPS) office to test the relationship between extra officers and arrests (for example, Chalfin et al., 2022; Cook et al., 2017; Owens, 2013; and Zhao et al., 2011). These grants began in 1994 as part of the Violent Crime Control Act with the intention of supporting hiring officers with a clear crime prevention and community policing focus. These studies generally point to additional officers having an overall negative effect on arrests. Zhao et al., (2011) is the main exception, while Beck and Holder (2022) find a negative effect but focus on arrests for minor offences.

When examined by crime type, the suppressive effect of additional officers on arrests was more common for victim-based crimes. Chalfin et al., (2022) found that arrests per annum for more serious crimes fell by between 0.97 and 1.6 for each additional police officer hired. However, where crime-type specific effects have been explored, these studies tend to find that arrests for quality-of-life offences, such as drugs offences, increased from hiring more officers. Owens (2013) also found a positive relationship between officer numbers and arrests for drugs offences. Chalfin et al., (2022) found a positive relationship for drugs possession offences and reported similar findings for a variety of other quality-of-life offences (for example, gambling and liquor violations). Beck and Holder (2022) found that, after controlling for other factors, reductions in officer numbers and staffing were associated with fewer arrests for low-level misdemeanour offences.

There was a mixed picture on whether the studies included both area and time fixed effects in their models. In general, area fixed effects were most commonly applied, with some studies using time fixed effects as a part of robustness checks or as the main statistical approach. For example, Chalfin et al., (2022) added further controls to capture changes in sentencing practice, technology and crime recording practices. Other studies focused more on using multilevel modelling techniques when the data was hierarchical.

In summary, most of the US research evidence on the police resource–arrest relationship suggested that, at least for victim-based offences, additional officers tended to be associated with fewer arrests. But the opposite pattern existed for drugs and other quality-of-life offences, where extra officers were associated with increased arrests. In terms of the applicability of these findings to England and Wales, several of these studies assessed the impact of increases in officer numbers from the COPS Hiring Program. This expansionist initiative had a clear focus on hiring officers to undertake community policing and crime prevention roles. The findings may therefore be less applicable to understanding the impact of recent changes in officer numbers in England and Wales. The last 10 years have seen marked falls in officer numbers followed by a rapid increase with the Police Uplift Programme (PUP). And there has been no specific national focus on what roles new uplift officers should perform.

Table 1: Selected US research studies on the relationship between resources and arrest volumes/rates

	Beck and Holder (2022)
Focus	Impact of changes in police funding on low-level offence arrests.
Data and approach	UCR Data 1990 to 2016; panel data model
Crimes	Low-level/‘misdemeanour’ offences, such as drug possession, driving under influence, disorderly conduct, drunkenness, vandalism.
Key findings	Reductions in funding for officers and staffing associated with reduced arrest volumes for low-level misdemeanour offences. Modelling the effects of a community policing model did not show an association with arrests, putting the emphasis on funding reductions as being the most critical factor in reducing arrests.
	Chalfin et al., (2022)
Focus	Impact of a larger force size on crime and arrests, additional analysis of impacts on black and white populations. Uses both COPS data on funding and a longer time series of police employment for larger cities.
Data and approach	1981 to 2018 (Annual Surveys of Government [ASG]) and 1990 to 2018 (COPS analysis); panel data model
Crimes	Homicide, low-level ‘quality-of-life’ offences (drinking, loitering, some drug), more serious and Index offences (murder, rape, robbery, aggravated assault, burglary, grand larceny and motor vehicle theft).
Key findings	Additional officers associated with fewer arrests for serious offences and index offences. But additional officers associated with increased arrests for low-level ‘quality-of-life’ offences (such as drugs possession). The impacts were around double or greater for black populations compared to white populations. Reductions in arrests result from deterrence and crime prevention effects.
	Cook et al., (2017)
Focus	Impact of additional funding (COPS) on number of officers, crime, and arrests.
Data and approach	Data 2005 to 2012; regression discontinuity design
Crimes	‘Part 1’ offences: manslaughter, forcible rape, robbery, and aggravated assault (violent crimes), as well as burglary, larceny, and motor vehicle theft (property crimes). Models also use total crime.
Key findings	Demonstrates that COPS funding did result in agencies successfully hiring more officers. More officers reduced both violent and property crimes and reduced the number of arrests available. A 2% increase in the number of officers associated with around 5% fall in crime, but rates vary by agency. Arrest and crime impacts are similar in scale.
	Jang et al., (2008)
Focus	Impact of broken window enforcement (BWE) on arrests and clear-up rates for a range of different crimes.
Data and approach	Data 1990 to 2004; panel data and multilevel model
Crimes	Robbery, aggravated assault, burglary, larceny, auto theft, violent, property, and Index offences.
Key findings	More officers are associated with lower crime and then arrests. BWE and associated increases in arrests increased clear-up rates for burglary and marginally for vehicle theft but decreased clear-up rates for larceny.
	Kaplan and Chalfin (2019)
Focus	Impact of hiring more police officers on arrests and downstream incarceration rates.
Data and approach	Data 1997 to 2015; ordinary least squares, 2-stage least squares, and panel data models
Crimes	Index offences: Murder, rape, robbery, aggravated assault, burglary, larceny, motor vehicle theft, drug possession and drug sales. Analysis was not broken down by crime type.
Key findings	More police officers may yield a ‘double dividend’ to society by reducing incarceration rates as well as crime rates. Study suggests more officers are likely to deter crime, reduce arrests and downstream prison pressures. Investment in law enforcement is unlikely to markedly increase state prison populations and may even lead to a modest decrease in the number of state prisoners.
	Owens (2013)
Focus	Impact of additional funding (COPS) on arrests.
Data and approach	Data 1990 to 2001; panel data and 2-stage least squares models
Crimes	Index offences: Murder, rape, robbery, assault, burglary, larceny, drugs and car theft.
Key findings	Additional funding did not increase arrests and there was a negative relationship with some serious crimes (although findings not significant at conventional levels). Findings for drugs offences were positive. Concludes more officers reduce crime and the number of arrests, that deterrence is more plausible than an incarceration impact.
	Zhao et al., (2011)
Focus	Impact of additional funding (COPS) on arrests for violence, property, drugs, and other crimes.
Data and approach	Data 1994 to 2000; panel data and multilevel model
Crimes	Violent crimes (murder, rape, robbery, and aggravated assault), property, drug, and other offences.
Key findings	Each additional dollar increased arrests per 100k population: 0.58 for violent crimes, 1.8 for property crimes, 4 for drug offences and by 36 for other offences.

2.2 Investigator resource

The evidence base on the relationship between investigator resource and case outcomes is modest. Several older studies (for example, RAND studies in Greenwood, 1979) looked at investigative resources alongside other solvability factors (for example, case characteristics and information obtained from victims). These studies tended to suggest that case characteristics are the key determinant of solvability and are generally more important than resources.

A study into the relationship between solvability and resources in burglary detections in England and Wales by Coupe (2014) acknowledged the importance of solvability factors but found that resources were still important. Coupe used regression analysis to identify how a range of factors – offence characteristics, the initial police response and resourcing levels – influenced detection outcomes. Coupe used outputs from the regression analysis to generate indices of overall solvability and examined the relationship with indices of overall resource time spent on the case (initial response, first officers at the scene and further investigation). The analysis showed that, overall, case solvability was more important than investigative resources in solving and clearing a case (Coupe, 2014).

Critically, however, Coupe identified that additional resources boosted detection rates regardless of the degree of case solvability, although the impact was most marked for crimes which were deemed highly solvable. Coupe also found that the timing of the application of investigative resource mattered – clearance rates increased when resources were used earlier in the investigative process.

Two quasi-experimental studies have also suggested a positive relationship between investigative resources and crime clearance (Braga and Dusseault, 2018; Braga et al., 2019). Both studies focused on homicide cases. Braga and Dusseault’s (2018) initial study evaluated a problem-oriented policing intervention introduced in Boston in the US. The intervention substantially increased ‘key investigative activities’ such as conducting more forensic testing and collecting more physical evidence from the scene. This led to an improvement in Boston homicide clearance rates compared to those in other US jurisdictions. However, as the intervention was not solely limited to increasing investigative resources, the study authors could not attribute increases in clearances specifically to resources.

The Braga et al., (2019) follow-up study was more focused. They looked specifically at investigative resources and actions and found that investigative resources, crime scene activity and subsequent investigative actions all increased the likelihood of homicide case outcomes (in this case, arrests) while controlling for other factors. Additional investigative resources increased the likelihood of getting a homicide cleared by 4.4%. While inherited case characteristics matter, enhanced investigative resources, alongside improved practices, can increase positive homicide investigation outcomes. However, the authors did caution that additional investigative resource may not be enough to achieve a positive outcome if the case is still short of evidential leads. Overall, these findings echo those of Coupe’s for burglary – case characteristics clearly matter, but additional resources can raise solvability. So, on balance, investigator resource should be considered as a potential variable of interest.

2.3 Police workload

A separate strand of research has focused on the relationship between police workload and case outcomes. Roberts and Roberts (2016) set out 2 theoretical approaches to the relationship between police workload and clearances – the ‘production-function’ and ‘overload’ hypotheses. The production-function hypothesis states that police workload is inversely related to the crime clearance rate. As more crimes are allocated to investigating officers, the resource per case diminishes with a consequent impact on clearances (see Liska et al., 1985). The overload hypothesis is a refinement of the production-function. It suggests that impacts from workload increases are not evenly applied across offence types. Rather, the police respond to constrained resources by reducing the effort directed towards less serious, high-volume crimes, thereby protecting resources for the investigation of more serious offences.

In England and Wales, Tilley and Burrows’s 2005 cross-sectional analysis of property offence volumes, clearance rates and officer numbers highlights 2 seemingly contradictory patterns. While police forces with higher crimes per officer were associated with higher numbers of detections per officer, they were also associated with a lower detection rate. Every additional 10 property offences per officer lowered detection rates by 1.5 percentage points. Consequently, the authors suggest that high-crime forces tend to generate more opportunities for officers to solve crimes because they have higher volumes of ‘easy to detect’ offences than equivalent low-crime forces. They also suggest that low detection rates in higher crime forces may be due to the lack of time to investigate more difficult-to-detect cases (so supportive of the production-function hypothesis). While plausible, the analysis does not rule out other interpretations of the relationship.

The findings from US studies on workload clearance rates are mixed. Some studies find little to no relationship between workload and outcomes (for example, Paré et al., 2007) while others have found a positive relationship (for example, Doerner and Doerner, 2012). Roberts and Roberts’ (2016) study arguably represents the most methodologically robust attempt to consider police workload effects on clearances over time. Data on serious violent crimes (homicide, sexual assault, robbery, and aggravated assault) were taken from a record-level incident data set covering 107 police forces in 24 US states in 2007. Each incident’s clearance status was tracked for up to a year to yield daily workload measures for each day of 2007 and 2008. Data on sworn officer numbers were used to capture changes in force resourcing. Using survival analysis, the authors found that police workload measures were significantly negatively related to the clearance hazard rate. The relationship between long-term workload and lower clearances was stronger for less serious offences, providing support for the overload hypothesis.

Wharfe et al., (2024) found further evidence of a crime workload effect – albeit a positive impact on charges from a short-term reduction in crime workload – through their analysis of the impact of changes in crime demand on charges during COVID-19 pandemic lockdowns in England and Wales.

2.4 Officer experience

The relationship between officer experience and the charging of offences is not well-evidenced but has been identified as a factor in some studies.  A Swedish evaluation of a national programme to increase officer numbers concluded that one key reason for the failure to generate an anticipated increase in charge volumes was due to sustained, high levels of inexperience across the workforce (Brå, 2014).  However, the evaluation did not extend to a robust assessment of impact.   

A review of the evidence on what makes an effective investigation identified a small body of research suggesting that, for investigators at least, an officer with greater experience is associated with higher levels of operational effectiveness (McLean et al., 2022).  Investigators, of course, make up just one part of the entire officer workforce, albeit a critical part of the workforce for getting charges in more serious offences. 

The grey literature repeatedly identifies the relationship between low levels of officer experience and low clearance rates. An inspection report by His Majesty’s Inspectorate of Constabulary, Fire and Rescue Services (HMICFRS) on the police response to burglary, robbery and other acquisitive offences stated that ‘the relative inexperience of newly trained officers is also affecting investigations… [M]any newly trained or direct entry detectives carry out high volumes of investigations without any experience in making arrests, building case files or attending court’ (HMICRFS, 2022). A joint inspection by HMICFRS and His Majesty’s Crown Prosecution Inspectorate (HMICFRS and HMCPSI, 2024) on case progression also highlighted the impact of an increasingly inexperienced police workforce on the successful progression of cases through the CJS.

Other studies have examined the relationship between workforce characteristics and the ability of the police to suppress crime. Using a panel vector autoregressive model to examine the relationship between police workforce size, structure, and stability over time for 42 police forces in the UK, Kim et al., (2024) found increased total police workforce ‘churn’^{[footnote 2]} was a significant predictor of higher recorded crime volumes.

2.5 Crime availability

Some studies have, directly or indirectly, highlighted the importance of the size of the pool of available crimes in determining charge volumes (the term ‘crime availability’ is used in this study). Tilley and Burrows’ 2005 cross-sectional study of property crimes in England and Wales found that the more crimes there are for each officer to investigate in a force, the more crimes are detected. This is because higher numbers of recorded crimes yield more opportunities for clearance. With 10 additional crimes per officer, an additional 0.25 detections per officer could be expected.

Research that has explored the relationship between resources and arrests (see Table 1) often implicitly touches on the crime availability relationship (for example, Chalfin et al., 2022; Kaplan and Chalfin, 2019). The inference from these studies is that victim-based crime is suppressed by the recruiting of additional police officers (such as through visible deterrence). Less crime provides fewer opportunities to arrest suspects, implying the existence of some kind of ‘crime availability’ effect.

Wharfe et al., (2024) found good evidence for a temporal crime availability effect on charge volumes during COVID-19 restrictions. Changes in routine behaviour during the most restrictive COVID-19 lockdowns were associated with falls in recorded crime – shoplifting was the most affected offence – accompanied by significant falls in charge volumes for this offence.

2.6 Proactive policing

Few studies were identified that explore how different types of proactive policing operations influence downstream demand. Some of the US resource–arrest studies suggest that increases in quality-of-life arrests associated with increased officer numbers will be facilitated through the greater use of stop, question, frisk (Chalfin et al., 2022). Stop and search has been the subject of many impact evaluations, but most studies have focused on crime reduction effects rather than the impact on crime outcomes (see Petersen et al.,’s 2023 systematic review). A UK study (Braakmann, 2022) not covered in Petersen et al.,’s review, due to the timing of the publication, did examine stop and search and police outcomes and found a small but statistically significant increase in the volume of suspects proceeded against. However, arrests and cautions showed no change.

Yet logic suggests that changes in stop and search activity – and other types of proactive policing – might have some impact on charges, especially for those crimes where the recording and charging of the offence depend on the police discovering the offence through stop and search activity.

2.7 Summary

Most studies which have explored the relationship between increases in officer numbers and arrests – all of which were undertaken in the US – tend to find that increases in officers over time are associated with fewer arrests for victim-based offences. The theory of change here is that additional officers have a suppressive effect on crime (such as through visual deterrence), which in turn reduces crime availability, and ultimately the number of offences that can yield arrests. Some, but not all, of these studies are based on the COPS Hiring Program which had an explicit focus on strengthening community policing and crime prevention.

Several of these studies have highlighted important crime-type variations in the relationship. Additional officers have been found to have a generally negative effect on arrests for victim-based crimes, while for quality-of-life crimes (including drugs offences) extra officers have a positive effect (that is more officers, more arrests for these crime types).

For all these reasons, the size of the overall police workforce was a clear area of interest when building a potential list of predictors of charges.

On workload, the stronger methodological studies suggest that increased crime workload may have a negative effect on case outcomes. The evidence on the role of investigator resources is limited, but robust studies suggests that higher levels of resources are associated with increases in positive case outcomes. And while the evidence on the relationship between levels of officer experience and crime outcomes is limited and not robust, the circumstantial evidence on the importance of this factor also makes it an area of interest. The evidence base on proactive policing and crime outcomes is very limited, but logic suggests this should be considered if only for possession offences. Finally, crime availability, which implicitly is a factor that emerges from the US research on resources and is identified explicitly in several other studies, looks worth considering.

The following sections investigate these factors and their relationship with charges.

3. Methodology

For this study, data was analysed at the crime-type level. Victim-based crimes covered 8 offences: shoplifting, burglary, all other theft, criminal damage, assault with and without injury, adult rape, and rape of a child under 13. Possession offences covered 4 offences: drugs possession, drugs trafficking (possession of drugs with intent to supply), weapons possession, and going equipped to steal.^{[footnote 3]}

3.1 Data

3.1.1 Outcome variables

In total, 5 outcome measures were used across the models. The default outcome measure, tested for all crime types, was the number of crimes charged (charge volumes).^{[footnote 4]} Crimes are charged after an investigation if the police – or CPS, typically for more serious offences – judge there is sufficient evidence for a realistic prospect of conviction. A charge, therefore, represents a higher bar than an arrest. For selected models, additional outcome measures were also used: volume of cautions, a formal out-of-court resolution (OOCR) issued by the police; total volume of OOCRs; volume of suspects proceeded against, as a clearer measure of downstream demand; and charge ratio, the proportion of recorded crime that results in a charge. Annex A provides further details on where these outcomes apply.

Charge volume data on the number of notifiable offences resulting in a charge were extracted from published Home Office statistics. Table 2 presents an overview of the charge volumes data used for each of the 12 crime types modelled. The main advantage of charge volumes data is that they indicate the number of charges in an individual force area. However, as charge volumes count the number offences which are charged rather than the number of suspects (an individual suspect may be charged with multiple crimes), it is a less precise measure of actual demand on courts. For some offence types, proceedings may be a more appropriate measure. The high percentage (40% in the year ending March 2024) of violence against the person offences which result in a ‘charge for an alternate offence’, makes the link between crime volumes and charge volumes for violent offences less clear.

Data on suspects proceeded against are taken from MoJ on the number of suspects received at court (MoJ, 2024a). Although this data is also broken down by force area, the figures will not necessarily relate to equivalent police force generated charges. This is partly because this data will include suspects proceeded against for offences charged by non-territorial forces such as British Transport Police. Additionally, the extent to which non-territorial forces are counted discretely in the MoJ data varies over time.

Table 2: Overview of charge volume data

Crime type	Mean charge volumes (1)	Standard deviation	Time period models run for: Years ending 31 March
Shoplifting	1,950	1,522	2011 to 2023
Total burglary	581	671	2011 to 2023
All other theft	808	924	2011 to 2023
Criminal damage	837	704	2011 to 2023
Assault without injury[footnote 5]	806	812	2011 to 2023
Assault with injury	1,486	1,626	2011 to 2023
Adult rape	37	47	2011 to 2023
Rape of a child under 13	22	21	2011 to 2023
Drugs possession	1,034	1,764	2008 to 2020
Weapons possession	392	600	2008 to 2020
Drugs trafficking	456	525	2008 to 2020
Going equipped to steal	47	74	2008 to 2020

Notes:

1. Annual force level mean values.

Figure 4 shows annual charge volumes for all the offence types included in the modelling. There are 3 offences with relatively small charge volumes: adult rape, rape of a child under 13 and going equipped. The small volumes are likely to impact on the stability of these models so findings for these 3 offences should be treated with caution.

Figure 4: Number of charges: all modelled offence types, 12 months to 31 March 2023

Source: Home Office (2023b)

For the 4 possession offences, the number of individuals cautioned (MoJ data) and the number of offences resulting in an OOCR (HO data) were also considered as alternative outcome measures. The offence group drugs trafficking is broad, covering supply, importation, and production. The most relevant drugs trafficking sub-offence in terms of stop and search is possession of drugs with intent to supply. Unfortunately, Home Office (HO) data on outcomes at the sub-offence code level are not available before year ending 31 March 2016. Data for suspects proceeded against are available at the sub-offence code level for a longer period, and therefore the preferred measure for drugs trafficking is possession of drugs with intent to supply.

Finally, charge ratio – offences charged divided by offences recorded in the equivalent period – was used for one crime type (burglary). Crime recording improvements, which were particularly marked for violence and sex offences during the period covered by the models (HMICFRS, 2020), can artificially depress charge ratios. Charge ratio was used solely for the analysis of total burglary offences (domestic and non-domestic combined), where there was more confidence in the consistency of crime recording during this period.

3.1.2 Predictor variables

A long list of possible predictor variables was created based on factors identified in the evidence review. The full list of all potential predictor variables is given in Table 3 (for victim-based models) and Table 4 (for possession offence models). These tables show average values for each measure across PFA and year, minimum and maximum values, the standard deviation, and the data source.

There were some differences in the predictor variables included in victim-based and possession offence models. Except for drug trafficking, investigator numbers were not expected to be a relevant factor in the possession offence models. Officers in investigator roles are unlikely to be involved in the progression of these offences.

For the victim-based crime models, stop and search activity was not included as a predictor variable as there is no anticipated relationship between stop and search activity and charges for victim-based crimes. Although not listed in Table 3, for some victim-based crimes, recorded offence count was also included in bespoke models to capture the effect of crime availability (discussed in ‘The models’ section below). For example, a count of total recorded shoplifting offences was included in a discrete model of shoplifting charge volumes. These models were not run for possession offences as the offences are much more likely to be determined by police activity.

Table 3: Predictor variables for victim-based offence models (1)

Predictors	Mean value	Minimum value	Maximum value	Standard deviation	Source
Total officers per 100,000 population	198	135	430	44	Published HO workforce data
Officer joiners	183	0	3,711	358	Published HO workforce data
Officer leavers	175	28	2,270	274	Published HO workforce data
Leaver rate (excluding transfers) (%) (2)	6	3	11	1	Published HO workforce data
Police staff headcount	1,898	320	18,136	1,929	Published HO workforce data
Total workforce headcount	4,902	1,456	50,577	6,640	Published HO workforce data
Joiner rate (excluding transfers) (%)	8	0	21	4	Published HO workforce data
Number of officers long-term absent	110	14	1,437	194	Published HO workforce data
Standardised Crime Severity Score (CSS) (3)	11	4	26	4	Office for National Statistics (ONS)
Percentage of officers with 5 or more years’ experience (%)	79	55	99	10	Published, and previously unpublished, HO workforce data
Senior ranks headcount (4)	73	20	759	93	Published HO workforce data
Non-senior ranks headcount (4)	2,931	777	34,220	4,681	Published HO workforce data
Officer turnover rate (5)	12	4	26	5	Published HO workforce data
Officers in investigative roles	509	48	7,530	797	Published HO workforce data from Year end March 2017

Notes:

1. Values are in absolute numbers, except where specified.
2. Leaver rate (excluding transfers) is the total number of police officers leaving during the financial year, as a proportion of the total officers at the start of the year.
3. The ONS Crime Severity Score (CSS) is the proxy measure for crime workload. The CSS weights recorded crime by severity using average sentence lengths/fines and standardises the force values by force area resident population. The measure is used to give an indication of the overall severity of crimes that individual forces were dealing with in a particular year, and the score gives an indication of the overall crime workload of a force.
4. Senior ranks are defined as officers with the rank above sergeant. Non-senior ranks are defined as officers ranked as sergeant or below.
5. Officer turnover is defined as: within any given year, the number of officer joiners plus number of officer leavers, divided by total officer workforce at the end of the year.

Table 4: Predictor variables for the possession offence models (1)

Predictors	Mean	Minimum value	Maximum value	Standard Deviation	Source
Total stop and searches undertaken (Section 1 and Section 60 combined)	19,117	874	597,267	57,720	Published HO data
Hospital admissions for assault with a sharp object (2)	98	0	1,195	168	Unpublished NHS data
Hospital admissions for poisoning by illicit drugs (3)	380	84	1,375	259	Published (England) and unpublished (Wales) NHS data
Police recorded theft from vehicle offences (4)	7,155	451	85,564	10,614	Published HO crime data
Total officers per 100,000 population	198	135	430	44	Published HO workforce data
Percentage of officers with 5 or more years’ experience (%)	79	55	99	10	Published, and previously unpublished, HO workforce data
Percentage of officers with less than 5 years’ experience (%)	21	1	45	10	Published, and previously unpublished, HO workforce data
Officer turnover rate	12	4	26	5	Published HO workforce data
Standardised CSS (5)	11	4	26	4	ONS

Notes:

1. Values are in absolute numbers, except where specified.
2. Variable included in the weapons possession offence model only.
3. Variable included in the drugs possession and drug trafficking offence models only.
4. Variable included in the ‘going equipped’ models only.
5. The ONS CSS is the proxy measure for crime workload. The CSS weights recorded crime by severity using average sentence lengths/fines and standardises the force values by force area resident population. The measure is used to give an indication of the overall severity of crimes individual forces dealt with in a particular year, and the score gives an indication of the overall crime workload of a force.

The Home Office collects force-level data on stop and search volumes (Home Office 2023a). From other data, it is known that stop and search-generated arrests account for a substantial proportion of all drugs offence and weapon offence arrests, so it seems reasonable to assume that stop and search might be a predictor of changes in charge volumes. Combining data on arrests from stop and search by crime type, and total arrests, in the year to 31 March 2023, around 62% of drugs offence arrests and 68% of weapons possession arrests were from stop and search (Home Office 2023a).

For specific possession offences, variables were included to capture related underlying offending in the PFA. For instance, an increase in weapons offence charges might be related to higher levels of underlying knife offending, or charges for going equipped might reflect underlying increases in burglary or car crime.

Data on hospital admissions for assault with a sharp object were included in the weapons possession models, and hospital admissions for episodes with a primary diagnosis of poisoning by illicit drugs were included in the drugs possession models. Police recorded theft from a vehicle were included in the going equipped models.

Some demographic controls were included in the possession offence models (unemployment rate, proportion of young men, concentration of bars, and median house prices). These were selected based on discussions with subject matter experts and after reviewing studies on stop and search and the impact on crime (Tiratelli et al., 2018; and McCandless et al., 2016).

Some data was collected that ultimately could not be included in any models. The CPS provided data on some key elements of their resourcing (for example, staff numbers by type and experience levels). However, it was not possible to break these down geographically.

3.2 Data preparation

From the long list of variables, some were usable in the format in which they were published (such as total officers per 100,000 population) while others needed to be calculated (such as officer turnover). Data for several variables needed to be grouped to maximise the information captured in the models and reduce collinearity between predictors. The cut-off points where data in individual variables were combined were determined by examining distributions of indicators and engaging with subject matter experts about meaningful boundaries. For example, years of officer experience were grouped into ‘under 5 years’ and ‘5 years or more’ in service. Since new officers undergo several years of training and job experience before passing their probation, 5 years seemed a rational cut-off point.

Prior to the models being run, all predictor variables were subject to standard descriptive analysis checks. This included dealing with missing data and potential outliers, checking distributions, and log transforming variables to ensure normal distributions and to ease interpretation of results. Since almost all the data came from published sources, there were few cases of missing data.

To avoid force size impacting the results, all volume measures (including dependent variables) were standardised. Victim-based models were standardised to 100,000 force area resident population, and possession offence models were standardised to 1,000 force area resident population.

3.3 The modelling process

3.3.1 Variable selection

The variable selection process was undertaken for each model individually. To avoid issues with inflated standard errors, correlations between model indicators were checked to ensure each measure selected was genuinely independent. Collinearity between indicators was assessed using an arbitrary but reasonable threshold correlation of 0.7. When 2 indicators were found to be highly correlated, the decision on which indicator to keep in the model was based on univariate regression. The indicator with the most statistically significant relationship with the outcome measure was selected.

All variables that made it past the selection process, and were compatible with the model date range, were inputted into the ‘starting’ model (see below). A backward selection process was then used to iteratively remove variables without a significant relationship with the outcome variable. This process was repeated until all variables in the ‘final’ model had a statistically significant relationship with the outcome measure.

3.3.2 The analytical approach

The main modelling technique was a standard fixed effects panel regression. This approach allowed the modelling to control for (unmeasured) factors that varied across force areas but not over time. Force fixed effects account for omitted characteristics that are difficult to measure with available data. These may include differences in local leadership, force culture, and potentially differences in forces’ operational practices. If force-level fixed effects were not controlled for, these omitted characteristics could have biased the model indicators, or be captured by the error (or residual) term, resulting in less robust models. The force fixed effects were represented by including binary indicators (dummy variables) for each force area.

As part of the general robustness checks, the team also applied a stricter test of association between the outcome and predictor variables. This involved including time fixed effects in the models. Time fixed effects controlled for (unmeasured) factors that varied over time but not by force area, meaning factors that change for all forces at the same point in time; for example, the introduction of new national guidance (such as, on case building, through the issuing of new Director of Public Prosecutions’ Director General’s Guidance (DG6), or national guidance on the use of stop and search). Where any individual model yielded statistically significant predictor variables with both area and time fixed effects included, this increased the degree of confidence that could be placed in statistical results. But equally, models not passing this threshold should not be dismissed. This might simply reflect the limited number of observations on which any model was based, or the consistency of trends of some predictors.

The regression equation used to model the impact of various predictors on outcomes was:

y_it = α_i + χ_itβ + ε_it

Where:

• y_it is the outcome variable for force i and financial year t, that is charge volumes, the number of suspects proceeded against, charge ratio or cautions.

• α_i is the fixed effect (or individual intercept) for force i.

• χ_it is the vector of predictor variables for force i and financial year t, such as measures of police resource and workload.

• β is the estimated relationship between each predictor and the outcome variable.

• ε_it is the error term, the difference between the predicted outcome and the actual outcome.

Where time fixed effects were also included, the regression equation above extended to:

y_it = α_i + μ_t + χ_itβ + ε_it

Where μ_t is the fixed effect for financial year t.

3.3.3 The models

Victim-based and possession offence models were run as separate exercises, but the overall modelling process was similar. The process was set up so that a range of models could be run for different crime types, date ranges and outcome variables.

Table 5 summarises the models, by crime type and outcome measure, which were the focus of this analysis. The default outcome measure was charge volumes, but suspects proceeded against was run as an alternative outcome measure for 4 victim-based offences and for all possession offences. Additionally, a total burglary model was run with charge ratio as the dependent variable. Models were also built for possession offences for individuals cautioned (MoJ data) and for all OOCRs^{[footnote 6]} (Home Office data).

Although not included in Table 5, separate models were also run for a selection of victim-based crime types (shoplifting, burglary, assault without injury, and adult rape) that included recorded offence count, alongside the other police resource and workload measures. This was to test the impact of ‘crime availability’ on charge volumes. This was run as a separate exercise because including recorded offences would introduce endogeneity^{[footnote 7]} to the models. The results of these crime availability models are presented separately in chapter 4, but the main outputs are presented with crime availability excluded.

For each crime type and outcome combination, including crime availability models, a model with force fixed effects was initially run, then the model was re-run with both force and time fixed effects included.

Table 5: Crime type and outcome combinations

Crime type	Crimes resulting in a charge (volumes)	Suspects proceeded against (volumes)	Other
Victim-based offences
Shoplifting	Yes	Yes	None
Total burglary (domestic and non-domestic combined)	Yes	Yes	Charge ratio
All other theft (excluding shoplifting and burglary)	Yes	No	None
Criminal damage	Yes	No	None
Assault without injury	Yes	Yes	None
Assault with injury	Yes	Yes	None
Adult rape	Yes	No	None
Rape of a child under 13	Yes	No	None
Possession offences
Drugs possession	Yes	Yes	Individuals cautioned (MoJ data) and total crimes resulting in an OOCR (HO data)
Weapons possession	Yes	Yes	Individuals cautioned (MoJ data) and total crimes resulting in an OOCR (HO data)
Drugs trafficking (1)	Yes	No	Total crimes resulting in a OOCR (HO data)
Possession of drugs with intent to supply (1)	No	Yes	Individuals cautioned (MoJ data)
Going equipped to steal	Yes	Yes	Individuals cautioned (MoJ data) and total crimes resulting in an OOCR (HO data)

Notes:

1. The offence group ‘drugs trafficking’ is broad, covering supply, importation, and production. The most relevant drugs trafficking sub-offence, in terms of being detected through stop and search, is possession of drugs with intent to supply. Unfortunately, HO data on outcomes at the sub-offence code level are not available before year ending March 2016. However, the suspects proceeded against data is available at the more granular sub-offence code level for a longer period. The proceeded against data is therefore the preferred measure for possession with intent to supply (although the results of the drugs trafficking models are also presented).

The completeness of the data available for the factors relevant to each offence type dictated the time periods modelled. Most victim-based models were run using annual data for the period year ending March 2011 to March 2023. Data for possession offence models were available from year ending March 2008 onwards. COVID-19 lockdowns affected trends in charges for possession offences and stop and search volumes markedly. Therefore for these models it was decided to use data only up to the year ending March 2020. As a robustness check, the drugs possession model was also run up to the year ending March 2023, with dummy variables included for the years affected by COVID-19, to test whether this affected the predictor variables. Including the dummy variables and extending the model up to the year ending March 2023 generated the same significant variables and signs.

Additional models were also run for a shorter period, years ending March 2015 to 2023, to allow for including data on the number of officers in investigator functions. Although the investigator variable was significantly correlated with charge volumes in a univariate model, when included alongside the other predictors and fixed effects, it did not emerge as significant predictors in the final models for any crime type tested. For this reason, the results of these models are not presented in the findings.

3.3.4 Checking model robustness

To test the goodness-of-fit for each model, the R-squared value was used to check how much the predictor variables explained the variation in the outcome variable. During the backward selection process, the R-squared value was monitored to ensure removing a variable did not cause a sudden drop in explained variation.

Standard diagnostic checks were carried out on each model. These typically involved visual checks of the model residuals and examining the difference between predicted and actual values of the outcome variable. This showed how the variation left unexplained in each model was distributed by year, PFA or predictor variable. Checks were conducted to ensure the residuals were normally distributed, and that they did not show any relationship with the final predictors in the model. Any trends found in the residuals implied a problem, such as omitted variable bias.

Additional checks were also carried out to examine if the residuals showed any trend over time and thus provide a more formal justification for the additional inclusion of time fixed effects (as well as a higher bar for including indicators).

4. Findings

This chapter reports the findings from models covering the 8 victim-based and 4 possession offence types. Model results are reported together in 3 groups: victim-based property offences; victim-based violence and sexual offences; and possession offences. The percentage impacts presented in the following tables all relate to a 10% increase in the predictor variable. All variables were log transformed, and coefficients therefore represent elasticities. Most of the results relate to models run with force fixed effects. The hash symbol (#) is used to denote whether the predictor (or an equivalent measure) remained statistically significant once time fixed effects were also included. If an indicator was only significant with time fixed effects included, the result is reported in the table with an added note. Results for each model can be also found in the Annex data tables.

4.1 Victim-based models: property offences

Table 6 presents the findings from the property offence models. For every property offence model (charge volumes, proceedings and charge ratios for all crime types tested), crime workload was a significant negative predictor, and remained so for burglary charge ratios when time fixed effects were included. This suggests that higher levels of severity-weighted recorded crime (‘crime workload’) are significantly associated with lower property offence charges.

Another broadly consistent finding across the 4 property crimes was around the 2 measures of workforce stability: annual officer turnover, and the proportion of officers with over 5 years’ experience.

Levels of officer turnover had a significant negative relationship with charges for the 3 property crime models (it was not significant for shoplifting). A similar relationship was found for the charge ratio and suspects proceeded against in burglary offences. For 2 models, the burglary charge ratio and criminal damage charge volumes, officer turnover also remained significant even with time fixed effects included, giving greater confidence in this finding. Overall, this suggests that high levels of officer turnover have a potentially suppressive effect on outcomes for most property crimes.

The effect of turnover on charges existed independently of officer experience. The percentage of officers with 5 years or more experience was consistently correlated with higher charge volumes and suspects proceeded against (where tested), with significant positive coefficients in all property offence models. This suggests that an increase in the proportion of officers in a force with 5 years or more experience is associated with an increase in charge volumes for property offences. When only allowing for force fixed effects, officer experience had larger coefficients than officer turnover. However, unlike officer turnover, when time fixed effects were included, officer experience did not remain significant in any of the property offence models.

The relationship between workforce and outcomes was less consistent across the 4 property crime types. Additional officer workforce was positively associated with charge volumes for shoplifting and criminal damage. The strongest finding was for shoplifting, where workforce was a significant predictor with time fixed effects included, giving greater confidence in the finding. This suggests that workforce is a particularly important predictor of charge volumes for this offence type.

Table 6: Model results for property crimes

Crime type	Workforce size	Percentage of officers with 5 years or more experience	Officer turnover (4)	Crime workload
Shoplifting
Charge volumes (1)	+7% *** # (5)	+22% ***	NS	-2% **
Suspects proceeded against (2)	+9% ***	+11% ***	NS	-6% ***
Burglary
Charge volumes (1)	NS	+17% ***	-2% ***	-4% ***
Charge ratio (3)	NS	NS	-2% *** #	-5% *** #
Suspects proceeded against (2)	+15% ***	+6% ***	-2% ***	-7% ***
All other theft
Charge volumes (1)	NS	+18% ***	-3% ***	-7% ***
Criminal damage
Charge volumes (1)	+4% *	+11% ***	-2% *** #	-3% ***

Key:

• Percentages depict the associated impact on the outcome variable for every 10% increase in the predictor variable.

• NS = Not statistically significant; * statistically significant at <= 0.05; ** statistically significant at <= 0.01; *** statistically significant at <= 0.001.

• # = Indicator (or equivalent) remains statistically significant when also controlling for time fixed effects.

• All variables are standardised.

Notes:

1. Model based on data covering years ending March 2011 to 2023.
2. Model based on data covering years ending March 2011 to 2020.
3. Model based on data covering years ending March 2012 to 2023.
4. As an additional measure of churn, either leaver rate (excluding transfers) or officer leavers per 100,000 PFA population was also considered in these models. However, since neither were statistically significant in any final models for property offences, they are not reported here.
5. Coefficient result from model including both force and time fixed effects. Measure was not significant when only controlling for force fixed effects.

4.2 Victim-based models: violence and sex offences

Table 7 summarises the main findings for the 4 violence and sex offences modelled. The proportion of officers with 5 years or more experience was a significant, positive predictor in all charge volume models for these 4 offences, and for assault without injury suspects proceeded against. However, as with the property offence models, experience did not remain a significant predictor when time fixed effects were included in the models.

Other measures of workforce stability, officer turnover and leaver rate, were also found to be significantly negatively correlated with charges for both violence offences and for rape of a child under 13 (no significant relationship was found for adult rape). Higher officer turnover rates were associated with fewer charges, although, as with the property offences, the coefficients for this variable were always smaller than the equivalent coefficients for officer experience. However, officer turnover remained significant once time fixed effects were included for both assault with and without injury (and for both charges and suspects proceeded against). This provides greater confidence that this measure is an important predictor of charge and suspect proceeded against volumes for violent crimes.

Results were far less consistent for officer headcount. A positive association was found for assault with injury charge volumes. But no relationship was found for assault without injury or adult rape charge volumes. For rape of a child under 13, a negative relationship was found between headcount and charge volumes. Caution is needed in interpreting the coefficients for the offence of rape of a child under 13. The relatively low charge counts for this offence (see Figure 4) and spikes in recording of historic offences in some years introduce instability into this model which could inflate the values of the coefficients. Once time fixed effects were introduced, no predictors remained significant in the under 13 rape model.

Workforce size measures were positively associated with volumes of suspects proceeded against for assault with and without injury. They remained significant for assault without injury once time fixed effects were included. Workforce was also a significant predictor for assault with injury charge volumes when modelled with time fixed effects.

As the approach for the core set of models was to exclude crime availability, crime workload could not be included in any of the violence and sex offence models either. The 2 measures are highly correlated for these offence types because the workload measure – crime severity weighted crime – is strongly influenced by trends in more serious offences. For these offences, the crime workload measure in effect acts as a proxy for crime availability, which needed to be excluded to avoid endogeneity. Crime availability was tested as a separate exercise (see Table 8 below).

Table 7: Model results for selected violence and sex offences

Crime type	Workforce size	Percentage of officers with 5 or more years’ experience	Officer turnover/leaver rate (3)
Assault with injury
Charge volumes (1)	+7% *** # (4)	+21% ***	-3%***#
Suspects proceeded against (2)	+24% ***	NS	-6% *** # (5)
Assault without injury
Charge volumes (1)	NS	+8%***	-1%*** #
Suspects proceeded against (2)	+8% *** #	+16% ***	-1% *** #
Adult rape
Charge volumes (1)	NS	+19%***	NS
Rape of a child under 13
Charge volumes (1)	-13%***	+39%***	-5%*** (6)

Key:

• percentages depict the associated impact on the outcome variable for every 10% increase in the predictor variable

• NS = Not statistically significant; * statistically significant at <= 0.05; ** statistically significant at <= 0.01; *** statistically significant at <= 0.001

• # = Indicator (or similar) remains statistically significant when controlling for time fixed effects

• all variables are standardised

Notes:

1. Model based on data covering years ending March 2011 to 2023.
2. Model based on data covering years ending March 2011 to 2020.
3. An additional measure of officer turnover, leaver rate (excluding transfers), was occasionally significant for these offences. Details are given in footnotes where this is the case. Full results can be found in the Annex data tables.
4. Coefficient result from time fixed effects model. Measure not significant when only controlling for force fixed effect.
5. Result relates to officer turnover predictor, the most significant measure of turnover in this model. However, leaver rate (excluding transfers) was also significant in this model. See Annex data tables for full results.
6. The significant turnover measure included in the model for rape of a child under 13 offences was the leaver rate (excluding transfers).

4.3 Victim-based models: testing for crime availability

Despite the challenges of modelling for crime availability (the number of recorded offences for the outcome crime type), its relationship to charge volumes was still of interest. Modelling outcome counts with crime counts introduces issues of endogeneity, which can inflate the coefficient for the endogenous variable. For this reason, a separate suite of models was run to test for the effect of crime availability, without this biasing the analysis of the other predictors. Two property crime models (burglary and shoplifting), and 2 violent and sexual offences models (adult rape and assault without injury) were run with a crime availability measure included. This measure was run alongside the other predictor variables and Table 8 presents the results for charges only (full results can be found in the Annex data tables).

The results of the availability models were consistent (Table 8). For every charge volume model run with time fixed effects, the measure of crime availability was a significant predictor of charge volumes. This suggests that while charge volumes can be suppressed by an overall increase in total crime workload, an increase in the ‘supply’ of a specific offence could be associated with an increase in charge volumes for that offence.

Table 8: Model results for victim-based crime availability: selected offences (1)

Crime type	Recorded crime (‘crime availability’ measure)
Shoplifting
Charge volumes	+12% ***#
Burglary
Charge volumes	+10% *** #
Assault without injury
Charge volumes	+2% *** #
Adult rape
Charge volumes	+6% *** # (2)

Key:

• percentages depict the associated impact on the outcome variable for every 10% increase in the predictor variable

• NS = Not statistically significant; * statistically significant at < 0.05; ** statistically significant at <= 0.01; *** statistically significant at <= 0.001

• # = Indicator (or similar) remains statistically significant when controlling for time fixed effects

Notes:

1. All models based on data covering years ending March 2011 to 2023.
2. Coefficient result from time fixed effects model. Measure was not significant when only controlling for force fixed effects.

4.4 Possession offence models

Table 9 presents a summary of the charge volume and suspects proceeded against models for all possession offences (weapons possession, drugs possession, drugs trafficking, and going equipped to steal). Full results can be found in the Annex data tables. It was not possible to obtain charge volume data for possession with intent to supply for the full period as data at the offence sub-group level were only available from year ending March 2016, so the charges model used data on the broader category of ‘drugs trafficking’. For suspects proceeded against, more granular data was available for the whole period and here, the narrower offence sub-categories ‘possession of drugs with intent to supply’ and ‘supply of drugs’ were combined into a single measure. Models which used cautions and total OOCRs as the outcome measure are also presented (Table 10).

Table 9: Model results for possession offences: charges and suspects proceeded against (1)

Crime type	Stop and Searches per 1,000 population	Officers per 100,000 population	Percentage of officers with 5 or more years’ experience	Officer turnover	Crime workload	Measure of underlying offending (2)
Weapons possession
Charge volumes	+1%**	NS	NS	NS	NS	+0.4%**#
Suspects proceeded against	+0.5%***	NS	NS	-1%**#	NS	+0.5%***#
Drugs possession
Charge volumes	+1%***#	+8%***#	NS	NS	-3%***	NS
Suspects proceeded against	+1%***#	+5%*	NS	-1%**#	-4%***#	-2%*#
Drug trafficking/possession with intent (3)
Charge volumes (Drug trafficking)	+1%***	NS	NS	NS	NS	NS
Suspects proceeded against (Possession with intent to supply)	+1%**#	NS	NS	NS	NS	NS
Going equipped to steal
Charge volumes	+2%***	NS	-11%*	NS	-7%**	NS
Suspects proceeded against	+1%***#	+16%***	11%**	NS	-8%***	NS

Key:

• percentages depict the associated impact on the outcome variable for every 10% increase in the predictor variable

• NS = Not statistically significant; * statistically significant at < 0.05; ** statistically significant at <= 0.01; *** statistically significant at <= 0.001

• # = Indicator (or similar) remains statistically significant when controlling for time fixed effects

• all variables are standardised

Notes:

1. Date ranges for possession offence models varied depending on data availability. See Annex data tables for the date ranges specific to each model.
2. The measure of underlying offending used for the weapons possession model was volume of hospital admissions for assault with a sharp object; for drug possession and drug trafficking/possession with intent, hospital admissions for poisoning by illicit drugs; and for going equipped to steal, volume of recorded theft from a vehicle offences.
3. The drug trafficking charge volumes model includes charges for importation and production, as well as supply offences. More granular data just for supplying drugs was not available for the longer time series of data used in the model. The suspects proceeded against model is based solely on sub-offences of possession of drugs with intent to supply or supply of drugs, which is likely to more accurately reflect offences identified through stop and search activity.

The approach to initial variable selection for the possession offence models was similar to the victim-based crime models. However, there was a particular interest in understanding the relationship between stop and search volumes and possession offence crime outcomes. Four predictor variables featuring strongly in the victim-based models were also included in the possession offence models: officer headcount, experience, turnover and crime workload. Officers in investigative functions were also included in bespoke weapon possession, drug trafficking and possession with intent to supply models, as these offences might more typically involve an investigator.

Trends in relevant underlying offending might also influence charge volumes for different possession offences. For weapons possession, the variable selected to capture underlying offending was ‘hospital admissions for assault with a sharp object’ (as a measure of underlying knife crime). For drugs possession and possession with intent to supply, the equivalent measure was ‘hospital finished admission episodes with a primary diagnosis of poisoning by illicit drugs’; and for going equipped, ‘recorded crime for theft from a vehicle’. However, these measures are imperfect. While the ‘hospital admissions for assault with a sharp object’ data is independent of the police, they do not differentiate between street based and domestic offending, and only the former is likely to be influenced by stop and search. Various types of theft offences were considered as the relevant measure of underlying offending for going equipped to steal. Theft from a vehicle was selected through testing for collinearity with other crime variables.

All 4 possession offence models found total stop and search volumes (Section 1 and Section 60 searches combined) to be a statistically significant predictor of charge volumes. The coefficients were small, with a 10% increase in stop and search volumes being associated with increases in charge volumes of between 1% (for weapons possession) and 2% (for going equipped to steal). However, these impacts are additive. In other words, an increase in stop and search would be related to simultaneous increases in all 4 possession offences.

When using suspects proceeded against as the outcome variable, the pattern was similar. Models for all 4 offences found stop and search volumes to be a statistically significant predictor of suspects proceeded against, and the coefficients were broadly similar in magnitude to those found for charge volumes.

The relationship between both officer workforce and crime workload, and possession offence outcomes, was not consistent across possession offences. In line with much of the US evidence on the relationship between officer numbers and arrests, officer headcount was positively associated with the drug possession outcomes. A 10% increase in officer numbers per 100,000 PFA population was associated with an 8% increase in drugs possession charges, and a 5% increase in suspects proceeded against. The finding remained statistically significant when time fixed effects were included in the charges model. For going equipped, officer headcount was also found to be statistically significant, but only for suspects proceeded against. For weapons possession and drugs trafficking and possession with intent to supply outcomes, officer headcount was not found to be a statistically significant predictor.

In terms of the measures of related offending, a statistically significant positive relationship was found for hospital admissions for assault with a sharp object. This was found to be positively associated with increases in charges volumes and suspects proceeded against for weapons possession offences, and remained statistically significant when time fixed effects were added to the models. A 10% increase in hospital admissions was associated with small percentage increases in charges (0.4%) and suspect proceeded against volumes (0.5%). The small negative relationship between hospital admissions for illicit drugs poisonings and suspects proceeded against for drugs possession offences is counterintuitive and no equivalent effect was found for charge volumes. For possession offences, the equivalent concept of crime availability could not be tested due to these offences being largely police detected.

For officer experience, only the possession offence ‘going equipped’ showed any significant association. However, the direction of the association varied with outcome measure, and it is likely that, given the small number of cases, the ‘going equipped’ models are not stable. Officer turnover also had no significant relationship with charge volumes but did have a small negative relationship with suspects proceeded against for weapons possession and drugs possession offences, which remained when time fixed effects were included in the models. There was some evidence of a suppressive crime workload effect on 2 of the possession offences. Higher levels of crime workload (crime severity weighted by recorded crime) were associated with fewer charges and suspects proceeded against for drug possession and going equipped.

Finally, Table 10 reports on the findings for individuals cautioned and the number of crimes that resulted in an OOCR.

Table 10: Model results for possession offences: cautions and total out-of-court resolutions (1)

Crime type	Stop and Searches per 1,000 population	Officers per 100,000 population	Percentage of officers with 5 or more years’ experience	Officer turnover	Crime workload	Measure of underlying offending (2)
Weapons possession
Cautions	NS	+15%***	NS	NS	NS	NS
OOCRs	NS	NS	NS	NS	NS	NS
Drugs possession
Cautions	+1%***#	+12%***#	6%*#	-1%*	-4%***	NS
OOCRs	+3%***#	+5%*	NS	NS	NS	NS
Drug trafficking/possession with intent
Cautions (Possession with intent)	+1%**	+8%*	12%***#	NS	NS	NS
OOCRs (Drugs trafficking)	NS	NS	NS	NS	NS	NS
Going equipped to steal (3)
Cautions	NS	NS	NS	NS	-31%***#	+17%*
OOCRs	NS	NS	NS	NS	-17%***	+16%**

Key:

• percentages depict the associated impact on the outcome variable for every 10% increase in the predictor variable

• NS = Not statistically significant; * statistically significant at < 0.05; ** statistically significant at <= 0.01; *** statistically significant at <= 0.001

• # = Indicator (or similar) remains statistically significant when controlling for time fixed effects

• all variables are standardised

Notes:

1. Date ranges for possession offence models varied depending on the significance of predictors and their available date ranges. See Annex data tables for the date ranges specific to each model.
2. The measure of underlying offending used for the weapons possession model was volume of hospital admissions for assault with a sharp object; for drug possession and drug trafficking/possession with intent, hospital admissions for poisoning by illicit drugs; and for going equipped to steal, volume of recorded theft from a vehicle offences.
3. Given the relatively small numbers of going equipped outcomes, especially in terms of cautions and OOCRs, these results should be treated with caution.

Overall, the cautions and OOCRs models results showed a patchier relationship between the predictor variables compared to charges and suspects proceeded against. Cautions and OOCRs for weapons possession, drugs trafficking and going equipped to steal were not associated with most of the predictor variables. This is likely to be due to the low volumes of cautions or OOCRs given for these 3 offences.

Stop and search volumes were only found to be significantly associated with OOCRs and cautions for drugs possession and drugs possession with intent to supply. Officer numbers were positively associated with drugs possession (cautions and OOCRs), as well as for cautions for weapons possession and possession with intent to supply. Crime workload was negatively associated with cautions and OOCRs for ‘going equipped’, and for drug possession cautions. Going equipped was the only offence type associated with a ‘measure of underlying offending’ (recorded crime for theft from a vehicle, in this case). A 10% increase in theft from a vehicle was associated with around a c16% increase in cautions and OOCRs for ‘going equipped’.

4.5 Model goodness-of-fit: R-squared results

R-squared values provide a measure of how well a model can predict its outcome variable, and how much variation is left unexplained. The Annex data tables give detailed breakdowns of the R-squared values for a subset of final models reported in this chapter. For each model considered, the breakdowns show the level to which the explained variation in outcomes is accounted for by the fixed effects and the significant indicators across various crime types.

First, focusing on force fixed effects models, for both victim-based and possession offence crime types, a significant proportion of the variation could be explained with the force fixed effects alone. Explained variation in outcomes from just force area was greatest for possession offences, with an average of 53%, compared to victim-based offences, with an average of 37%. The marginal improvement of the R-squared values when the significant model predictors were included in the models also varied across offence types. Within victim-based offences, the marginal improvement in R-squared after adding predictors was generally greater for property offences, with 69% additional explained variation on average, compared to violence and sex offences (with an average of 51%). This suggests that, in terms of important factors which impact outcomes captured in these models, there is more missing from the violence and sex offence models than the property models. The next chapter covers what might be behind these issues.

Secondly, focusing on models which also included time fixed effects, the level of explained variation coming from time fixed effects alone was greater for victim-based offences than possession offences. On average, time fixed effects accounted for 47% of the variation in victim-based outcomes, and 18% of the variation in possession offence outcomes. When combining both force and time fixed effects, a considerable amount of variation in outcomes for both victim-based and possession offences could be modelled through the fixed effects alone (on average, 84% for victim-based and 74% for possession offences). This left very little room for the individual predictors to explain the remaining variation.

The marginal improvement from adding indicators to a model may not always equal the difference between the R-squared with significant indicators and the R-squared with just fixed effects added. This is because there may be correlations between some of the fixed effects and the model indicators. This is especially true for time fixed effects, as there is a risk of attributing the influence of any consistently changing indicator to the time variable. While the indicators may appear to make only a small additional contribution to the R-squared value in the time fixed effects models, the fact they appear in the models at all, when so much else is controlled for, gives increased confidence that the significant predictors are genuinely contributing to the explanatory power. It is helpful to consider the range of coefficients (from force fixed effects only models, to force and time fixed effects models) attributed to any indicator that remains in both models.

For most offence types, models of suspects proceeded against had a higher R-squared value than the corresponding charge volumes model.

5. Discussion and conclusions

This study explores factors associated with changes in charge volumes and other case outcomes in England and Wales. It uses a panel regression design at the police force area (PFA) level to examine changes in charge volumes and, selectively, suspects proceeded against, cautions and OOCRs. Most of the models are constructed to cover the period of years ending 31 March 2011 to 2023, which was a period of considerable change in policing and crime in England and Wales.

The analysis can be considered novel in 2 respects. First, while several US studies have examined similar issues, most have focused on the relationship between changes in resources (mainly officer numbers) and police outcomes (mainly arrests). The present study seeks to answer a more general question: what factors drive changes in charges and other positive case outcomes? This remains an area of policy interest given both the importance of planning for future downstream capacity, and the need to better understand the factors have contributed to the recent fall in charges. It is also one of the first studies to examine this area in detail in a UK setting.

Taking the modelled results from the 3 groups of offences together – property offences, violence and sexual offences, and possession offences – several key points emerge. The first is that the relationship between changes within a police workforce and trends in charge volumes is more complicated than being solely a function of changes in officer headcount. The second is that although some common analytical patterns exist, there are marked differences between the factors relevant to victim-based offences and possession offences.

For almost every victim-based crime modelled, measures of officer experience and/or officer turnover were found to be strongly associated with charge volumes (and where modelled, suspects proceeded against). The results suggest the more experienced the officer workforce, the higher the charge volumes for victim-based crimes. There were a range of effect sizes observed across the different offence types, with shoplifting, assault with injury and the 2 sex offences recording the highest coefficients. As a lower severity offence, the large effect size for shoplifting appears to be an anomaly. One possible explanation could be that the recent marked increase in shoplifting charges, at a time when experience is stabilising, has led to a spurious or inflated result.

Time fixed effects were included in the models to control for changes felt across all forces at the same point in time. When these were added, officer experience no longer added sufficient explanatory power in any of the models. However, the context needs to be considered. For much of the period covered by the analysis, levels of officer experience were falling for all forces. By 2023, the percentage of officers in England and Wales (excluding British Transport Police) with less than 5 years’ experience was at 36%, up from 14% in the year ending March 2016. This reflected both the loss of officers without replacement in the first half of the 2010s, and the influx of new officers recruited from late 2019 (Home Office, 2023c).

As the decline in officer experience was relatively consistent across forces, to some degree its impact will be captured by time fixed effects. The true relationship between officer experience and charges is likely to lie somewhere below effect sizes given by models which do not include time fixed effects. The results of the time fixed effect models should therefore not ignore the potential importance of the experience findings, but rather help guide us to considering a more modest effect size for this predictor.

Independent of levels of officer experience, annual rates of officer turnover also appear to be closely associated with charge volumes for most victim-based offences (and suspects proceeded against for weapons and drugs possession). Where significant, high levels of turnover consistently had a negative relationship with both charges and suspects proceeded against. Although the effect size was smaller than that of officer experience, officer turnover was more likely to remain statistically significant once time fixed effects were included. The consistency of this finding, especially for the offence types of burglary, criminal damage, assault with and without injury (where turnover was significant with both force and time fixed effects, and for all outcomes tested) makes officer turnover a key factor of interest.

Taking the findings for officer experience and officer turnover together, they seem to be capturing different elements of the relationship between changes in the workforce and charges. Change in officer experience represents the long-term, cumulative impact of recent recruitment exercises and earlier headcount losses on overall organisational experience. Officer turnover represents the degree of in-year disruption experienced by a force in terms of stretched supervision, and a consequential impact on investigative response performance. Overall, the influence of workforce stability on outcomes is one of the most consistent findings from this study.

In terms of the core relationship between changes in officer headcount and police outcomes, US studies in this area have typically found a negative relationship. That is, increases in officer numbers were associated with fewer arrests for victim-based crime. This general pattern is not reflected in this analysis. Only charges for rape of a child under 13 years old recorded a negative relationship with headcount, and the small numbers in this model mean that this finding needs to be treated cautiously. For 4 of the 8 victim-based offence types, no relationship was found between charge volumes and officer headcount. And for the remainder of the charge volume models, a statistically significant positive relationship was found. For almost all crime types where workforce numbers showed a positive effect, the coefficients were smaller than for experience. Models for only 3 victim-based offences – shoplifting, assault with injury (charges) and assault without injury (suspects proceeded against) – remained significant when time fixed effects were added in.

Several factors might explain the difference between these results and US studies in relation to victim-based offences. First, the analysis in this paper covered an extended period of falls in officer numbers, followed by a relatively short, intense period of increasing numbers in the final few years. Most of the US studies examine the impact of an expansion of officer numbers through the COPS Hiring Program (CHP). Different impacts may well exist depending on whether forces are shedding or gaining officers.

Secondly, one feature of the CHP was that hired officers were intended to have an explicit focus on crime prevention and community policing. If deployed in this way, the extra officers might indeed be more able to suppress crime and, in doing so, reduce arrests and downstream demands (by reducing the available pool of offences to detect). No equivalent operational focus was placed on the officers recruited during the PUP in England and Wales. Similarly, the loss of officers up to the year ending 31 March 2019 was not targeted on shedding staff from specific roles.

Thirdly, the models in this study were designed to identify any factors that influence charge volumes. In contrast, most of the US studies were seeking to understand a narrower relationship, between police resources and arrests. To replicate the workforce numbers focus of the US studies, several models were run which looked solely at the relationship between officer numbers and charge volumes for various crime types. That is, other factors (workload, experience and so on) were excluded from these ‘headcount only’ models.^{[footnote 8]} Using this approach, the results were more closely aligned with those from the US studies. These simplified models found that all victim-based offences showed a negative relationship between officer headcount and charge volumes (that is, more officers were associated with fewer charges).

Where this study’s findings most closely align with the US evidence is on drugs possession offences. Several US studies (for example, Chalfin et al., 2022; Owens, 2013) found officer numbers were positively associated with more arrests for drugs offences. This was part of a wider pattern found to exist with arrests for what are typically called ‘quality-of-life’ offences. In this analysis, a positive, statistically significant relationship was found between officer numbers and both drugs possession charges, and suspects proceeded against, and for going equipped suspects proceeded against. However, the relationship did not hold for charges for weapons possession, drugs trafficking, possession with intent to supply, or going equipped to steal.

To model the impact of crime workload on outcomes, the ONS CSS, which weights recorded crime by severity using average sentence lengths, was used. This indicated the overall severity of crimes a force was dealing with in a particular year. Assuming that crimes higher in the severity index are more resource intensive to solve, the score gives an indication of the overall crime workload of a force, and the prioritisation decisions they may have been facing. Although it was not possible to test the relationship between overall workload and outcomes for violent and sexual offences, charges and suspects proceeded against for all property offences were found to be negatively associated with overall crime workload. This suggests that an increase in a force’s crime workload could lead to suppressed charges and suspects proceeded against for property offences. A suppressive workload effect was also found for charges and suspects proceeded against for the 2 possession offences (drugs possession and going equipped).

In contrast, crime availability reflects the fluctuations in recorded volumes for each specific offence type modelled. The suggestion that an increase in the pool of total offences will lead to increased detectable offences, and lead to higher charge volumes, has been identified in the wider literature. For 4 offences where models were run with a crime availability variable included, all showed significant positive effects (even when controlling for time fixed effects). The findings from this analysis – alongside other studies which have examined natural experiments in terms of changes in crime availability (for example, Wharfe et al., 2024) – provide support for the importance of crime availability in influencing charge volumes.

However, trying to assess the influence of crime availability alongside workload did present analytical challenges for some models. As workload is a severity-adjusted measure of total crime availability, it will typically be collinear for offences with high severity and increasing volumes. Because of this, workload was excluded from the sexual and violence offence models. This means it was only possible to make a partial assessment of the existence of an ‘overload’ effect from changes in crime workload. The property crime analysis does provide supporting evidence that increased workload has a suppressive effect on charges for less serious offences. But it was not possible to test the complementary element of the overload hypothesis – the absence of a workload effect on charges for more serious offences – in these models.

A confounding factor to consider in this analysis is the impact of improvements in crime recording. The period covered by the models coincides with the HMICFRS rolling programme of Crime Data Integrity Audits. This will have inflated crime workload patterns during this period, partly because sex and violence offences were previously the most under-recorded (see HMICFRS, 2020). Nevertheless, some of the increase in crime workload during this period will have been real.

This study also suggests that decisions around police operational activity can influence charge volumes for possession offences. Perhaps not unexpectedly, the possession offence models show that as stop and search volumes increase, the predicted numbers of crimes charged (and suspects proceeded against) increase for all 4 offences. The effects are small, with impacts for individual offences ranging from 0.6% (weapons) to 2% (going equipped), for a 10% increase in stop and search activity. This is perhaps unsurprising given the vast majority of stop and search do not typically result in the discovery of a possession offence. However, these small effects are amplified as the coefficient impacts are additive for each of the 4 possession offences. In other words, an increase in stop and search volumes will lead to a predicted simultaneous increase in charges for all drugs possession, weapons possession, drugs possession with intent to supply, and going equipped.

Weapons possession charges appear to be affected by the level of related offending. Changes in the underlying level of knife crime, as measured by hospital admissions for assault with a sharp object, are associated with higher charge volumes and suspects proceeded against (although the effects here are also small).

Finally, there was one predictor which, perhaps surprisingly, did not appear to explain any of the variation in outcomes seen across forces. A subset of victim-based crime models was run to incorporate Home Office Police Workforce data on the number of officers in investigator functions, on a shorter time series where this data was available (as at 31 March 2017 to 2023). These models consistently failed to identify investigator numbers as a significant predictor of charge volumes. However, this finding needs to be interpreted with caution. During the period when charges for victim-based crimes have been falling at their fastest rate, investigator numbers remained comparatively stable.

While relatively flat, officers in investigator functions were, however, below what police forces believe they should be to operate effectively during this period. HMICFRS identified a detective or investigator ‘gap’ between available resources and investigative demand since 2016 (HMICFRS, 2016). Between the years ending March 2017 and 2023, HO Workforce statistics^{[footnote 9]} indicate that officers in investigator functions increased by around 12% (up 2,500 officers). Given this relatively stable but depleted picture of investigator resource, it is perhaps unsurprising that measures of investigators do not register as a significant predictor in these models. But this should not, of course, preclude the potential downstream impact on charge volumes of any future expansion in investigators. The nature of changes in investigator capacity during the short period modelled may have been relatively unimportant for charge volumes, but this may not be the case going forward.

Considering these findings all together reveals a more nuanced picture of what might drive trends in charge volumes and suspects proceeded against, and how this varies for different types of crime. It also yields some insights into how demand on the downstream CJS might change in response to developments within the police workforce, or changes in levels of crime demand. Although estimated effect sizes for each predictor varied by crime type, there was a degree of consistency in the significance and direction of the findings for officer turnover, officer experience, workload, and crime availability for victim-based crimes, and for police activity for possession offences. This suggests the identification of several key factors that go some way to describing the drivers of the decline in charge volumes in recent years.

This analysis has sought to maximise the value of published data on crime outcomes and potential factors associated with changes in charge volumes, and other CJS outcomes to help identify key factors that influence charge volumes. While several factors have potentially been identified, it is important to acknowledge the limitations of the analysis.

A general point being inferred, but not assumed, is the causal relationships between the predictors included in the final models, and changes in charge volumes and suspects proceeded against. Relatively modestly sized, annual, force-level data sets are being drawn upon to undertake this analysis, and the estimated effect sizes can only ever represent a level of association with the outcome variables, not a causal effect. However, despite these sample size constraints, both force area and time fixed effects have been applied to these models to strengthen confidence in identifying potential causal relationships. Applying these together means a relatively high bar has been set, given the size of the data sets on which the models are built. For some variables in some models, that high bar is met. But equally, findings that do not meet this threshold should not be rejected.

Additionally, the outcomes of interest all followed an almost consistent downward trend with very little year-on-year variation. Consistent trends were also seen in many of the predictor variables – for example, there was a consistent decline in officer experience and an increase in officer turnover over time. Without time fixed effects, the size of some coefficients may have been inflated by this underlying trend. Equally, by including time fixed effects, the consistent downward trend in outcomes is easily explained in the models by the time dummy variables. The true relationship between the predictors and outcomes therefore is likely to lie somewhere between the effect sizes given by models which do not include time fixed effects, and those that do.

It is useful to reflect on the level of explained variation in outcomes, measured by the model R-squared values, coming from the fixed effects in these models. The amount of explained variation added by the predictors was substantial across the variety of crime types tested in the force fixed effects models. However, a substantial proportion of variation could also be explained by the fixed effects, both force and time, alone. In the case of force fixed effects, this suggests that what individual forces do – or do not do – in terms of how they investigate and prosecute crime appears to be important, alongside factors which might be determined solely by the geographical characteristics of the area covered by each police force.

Finally, even with the inclusion of both PFA and time fixed effects, there was, on average, around 20% of variation in charge volumes left unexplained, with more typically left unexplained for the more serious offence types (violence and sex offences). Although these models have tried to maximise the value of existing data on factors which are thought to be influential in determining charge volumes, there are still some major gaps in the model data.

For example, while the police charge a high proportion of notifiable offences, charging authority for more serious offences, and all domestic abuse and hate crime flagged offences regardless of severity, sits with the CPS. Data on the subset of offences charged by the CPS, analysed by CPS area, with relevant CPS area resourcing data as a predictor, would have improved these models. These omissions may explain why the R-squared values for violence and sex offences are lower than those for property offence models. It remains a limitation of the analysis that CPS factors were unable to be accounted for in the modelling. Other important but omitted predictors include the changing nature of crime ‘difficulty’, independent of crime seriousness.

So, while these models are useful in improving understanding of what factors are associated with changes in charges and suspects proceeded against over time, they only constitute a first step. There is much more to be done.

Annex A

Principal data sources

• Home Office police recorded crime and outcomes data (Home Office 2023b and other years) were used to extract recorded offences, charges and out-of-court resolutions (OOCRs)

• Ministry of Justice (MoJ) suspects proceeded against and cautions data was taken from MoJ Criminal Justice Statistics Quarterly (2024a, 2024b and earlier years)

• police resource data came from the police workforce statistics (Home Office 2023c and earlier years)

• stop and search data came from Home Office Police powers and procedures: Stop and search and arrests bulletins (Home Office 2023a and other years)

• external demographic data for the possession models came from Office for National Statistics (ONS) websites (NOMIS, house price statistics, area population estimates)

• the NHS provided an underlying offending variable on hospital admissions for assault with a sharp object to the Home Office as suppressed rolling totals; hospital admissions data are from the NHS Hospital Episode Statistics, which contain volumes of admissions recorded by patient home address and cause code and, when published, are defined as National Statistics; the hospital admissions data analysed in this report use unpublished management information which has not yet undergone the formal badging process, but is produced using similar principles and methodology; the hospital admissions with a primary diagnosis of poisoning by illicit drugs for England were taken from the NHS Digital website and the Welsh data was provided by Digital Health and Care Wales, following an ad hoc request

These data sources were used to create a full measures table (comprising all potential dependent and independent variables). Police recorded crime and outcomes data include measures of recorded offences, charge volumes, and other positive outcomes, all of which can be split by crime type. These measures were predominantly used to calculate all the dependent variables.

Dependent variables

The main interest is on charges because these outcomes will, potentially, lead to further downstream impacts as suspects move through the CJS to courts, prison, and probation.

a. Charges

Unlike US studies, where the preferred measure of case clearance in the number of arrests, this study’s preferred measure is charges. Currently published data in England and Wales on arrests, while available annually and broken down by force, provide much less granularity in terms of offence types. Furthermore, by no means all arrestees ended up being charged. And the growing use of so-called Voluntary Attendance as an alternative further weakens arrests as a consistent and robust measure of clearance over time.

Data on the number of notifiable^{[footnote 10]} offences charged by police force area (PFA) were taken from published annual data on the number of charges recorded (Home Office, 2023). Of the 43 force areas, 3 were excluded from all the charges models: Greater Manchester Police (missing data between 2022 and 2023), one non-Metropolitan force (delays in recording outcomes in one year led to higher-than-expected charges volumes in a subsequent year); and City of London police, due to its small size. British Transport Police was also excluded.

Most models focused on charge volumes as an outcome as it was less sensitive to changes in recording practices; however, in some scenarios, it was possible to model with charge ratios. Crime types known to have consistent recording (such as burglary), or shorter time frames where recording was more consistent, could be modelled with charge ratios as an outcome.

For the possession offence of drug trafficking, the published charges data do not include the detailed offence codes to allow data to be excluded by the sub-offence categories of drug trafficking, including exportation, importation and production.

Charge data used for drug trafficking crimes cover all offences in this group and will therefore have a diluted association with some variables, such as stop and search volumes (but see below).

b. Suspects proceeded against

For a selection of crime types, data on the number of suspects proceeded against were used as an alternative outcome measure. Suspects proceeded against capture the number of individual suspects listed at court due to a charge (unlike charges which relate to the number of crimes charged). This data is collected by MoJ and are fully independent of charge data (which are provided to the HO by police forces).^{[footnote 11]} This data better captures the actual pressures on the downstream CJS in terms of individuals who will need processing. However, there are some issues with how this data breaks down by force level. As well as the 43 territorial forces, MoJ data include a category labelled ‘special, miscellaneous and unknown police forces’. This is likely to include charges from forces such as British Transport Police although the precise composition of this group does not appear to be consistent over time.

Data on the number of suspects proceeded against by force area were extracted from MoJ published data by using the detailed Home Office offence codes which MoJ has included as part of the ‘Outcomes by Offence’ data set (MoJ, 2024a). Having these codes meant it was possible to remove the non-relevant sub-offences within drug trafficking, which would not have been relevant in terms of a possession offence and which would not be picked up by police stop and search activity (for example, production, exportation and importation). As a result, it was possible to generate a model covering a longer time period for the sub-offences of ‘possession of drugs with intent to supply’ and ‘supply of drugs’.

c. Cautions and OOCRs

Since a large proportion of offenders caught in possession of drugs receive an OOCR, a decision was made to include these crime outcomes for the possession offences.

OOCR data was taken from the published annual data on the number of OOCRs recorded (Home Office, 2023). The data is based on the total number of out-of-court formal and informal disposals recorded. OOCRs have changed over the years. Prior to April 2013, they comprised cannabis warnings, cautions and Penalty Notices for Disorder (PNDs). Subsequently, community resolutions were included. The total figures by year relate to the number of crimes that receive an OOCR, meaning as with charges, they are not person based. It is not possible to break down the drug trafficking offence into the more detailed sub-offence categories. In the years modelled, few drug trafficking offences received an OOCR.

An alternative OOCR measure was cautions data published by MoJ. MoJ analysts extract this data from the Police National Computer. MoJ cautions data relate to individuals in receipt of cautions (not offences cautioned).

MoJ cautions data for drug trafficking cannot be fully disaggregated into the more granular sub-categories of supply of drugs and possession with intent to supply.

Predictor variables

a. Police workforce

Data on the police workforce were mainly taken from annual HO published statistics on the police workforce. These are published as accredited official statistics. In the preliminary data collection exercise, various alternate measures of police workforce were initially considered. These included, for example, total officer numbers, and combined officers plus police staff. Individual measures of police workforce were all correlated with each other, so only one could be included in each model.

b. Investigator resources

Data on the number of investigators were taken from published data on the number of officers working in investigative functions as at 31 March each year, published as part of the HO Police Workforce Statistics bulletin. This data is based on the primary role of workers. This will not reflect all of the roles that they fulfil in their job. Some functions may appear to be under- or over-represented in the data. For example, if 100 officers spend just 20% of their time in function A, this will be counted as zero officers working in that area.

Police workers are categorised based on their primary role, that is, the role in which they spend most of their time. If an officer spends 60% of their time in role A and 40% in role B, they would only be categorised as working in role A. While there may be some data quality issues here, they probably give the best published measure of forces’ investigator capacity. The data only go back to the year ending March 2017.

c. Crime workload

Two alternative measures of crime workload were considered: one using notifiable offences weighted by estimated resource cost; and the other weighted by crime seriousness using average sentence lengths (the ONS Crime Severity Index). Although resource cost estimates would be the preferred measure, as this would more accurately reflect the cost to the police of dealing with an offence, the available resource estimates at the time came from data originally collated in the early 2000s and have been only partially updated. ONS Crime Severity Index data was therefore used as the preferred proxy for crime workload.

d. Officer experience

This measure was taken from published workforce stats and previously unpublished police workforce census data, with the following caveats:

• includes the 43 territorial police forces in England and Wales, excluding the British Transport Police

• includes police officers only

• Cheshire were unable to provide data as at 31 March 2010

• the unpublished data from 2010 to 2015 have not undergone quality assurance checks or verification by police forces; the total police officer headcount may differ to the published total police officer headcount in the ‘Police Workforce, England and Wales’, statistical series; some forces may have provided slightly fewer or more officers by length of service compared to the verified total headcount

• several forces are unable to provide data on length of service in the police and can only provide length of service in current force or post; for some forces, length of service may be undercounted, as previous posts in other police forces are not accounted for; this is particularly likely to affect senior ranks; nationally, the data will show too few people in longer service categories, and too many in shorter ones; however, given that constables make up the majority of the workforce, the effect of this is likely to be relatively small

• the data rely on the start date used being the same across all forces and years; data may not always be comparable in different years or forces if different start dates are used; however, it is not expected that this would have a large effect

• unpaid long-term absence and career breaks may be included by some forces; however, most do exclude them; data may therefore, not always be perfectly comparable between forces, although this should only affect a few cases and should not have a major impact on length of service in most cases

Alongside measures of officer experience – typically the proportion of officers with ‘less than 5 years’ or ‘5 years or more’ service – other measures of workforce stability were tested. For example, total staff (officers and staff) turnover rate, officer turnover rate, and, separately, the proportion of the force made up of leavers or joiners in any given year.

e. Police activity

One of the few national data sets on police activity covers the use of stop and search. Annual force-level data on stop and search have been collected since 2007. Stop and search under both the more commonplace S1 (section 1 of the Police and Criminal Evidence Act 1984) and S60 (Section 60 of the Criminal Justice and Public Order Act 1994) were combined into a single measure of stop and search activity. Data was standardised per 1,000 of the population within the PFA. This variable was only used in the possession offence models. Information on data quality for the police activity collection can be found at Police powers and procedures: Stop and search, arrests and mental health detentions, England and Wales, year ending 31 March 2024 - GOV.UK)

f. Complementary offending

For each of the possession offences, one or more data sets that might capture complementary offending associated with the offence were used. The theory is that change in charges for possession offences could be influenced by underlying related offending rather than a change in the main predictor variables. For drugs offences, annual NHS data on admissions to hospital for poisoning by illicit drugs was used as a proxy for underlying illegal drug use in a PFA. The data for England is from the NHS Digital ‘Statistics on drug misuse for England’ (2020). The data for Wales was provided by Digital Health and Care, Wales extracted from the Patient Episode Database for Wales.

As a measure of the underlying level of weapons use, hospital admissions for assault with a sharp object were used (suppressed rolling totals). This has the advantage of being independent of police recording of knife-related offences. Its main disadvantage is that it relates to all knife injuries, not simply those that take place in a public space. Hospital admissions data is from the NHS Hospital Episode Statistics, which contain volumes of admissions recorded by patient home address and cause code and, when published, are defined as National Statistics. The hospital admissions data analysed in this report use unpublished management information which has not yet undergone the formal badging process but is produced using similar principles and methodology.

Similarly, for going equipped, data on the number of acquisitive crimes (theft from a vehicle), as measured by police recorded crime, was used as a proxy for the underlying level of acquisitive crime associated with this possession offence.

Possession offence: external control variables

For the possession offences models, several external demographic variables were used to act as controls in the models.

1. Unemployment rate

Data on claimant count by sex and age were taken from ONS’s NOMIS website. The rate was calculated as claimant count as a proportion of claimant stock within each PFA.

2. Average median house price

Data on median house prices by middle layer super output area were downloaded from the ONS’s website. Data was aggregated to the PFA and the average median house price was calculated.

3. Young people/young men rate

Data on young people and young men – who are a subset of young people – were taken from ONS’s middle layer super output area mid-year population estimates. According to the published stop and search statistics (Home Office, 2023) young people aged 18 to 24 were most likely to be stopped and searched by the police and so a similar age range was used (ONS uses an age range of 15 to 24 in their population estimates). The rate of young men was calculated by dividing the number of 15 to 24 year old young men by the total population within a PFA. Young people aged 15 to 24 included young women, and to create a rate this was divided by the total population within a PFA.

4. Bars per hectare rate

This data was from ONS’s NOMIS website and comprised the total number of bars, pubs and clubs within a PFA divided by the number of hectares for a PFA.

References

Beck B and Holder E (2022) ‘Proximity and Police Departments: Spatial Dynamics of the Misdemeanor Arrest Decline’, Socius: volume 8 (viewed on 10 January 2025)

Braakmann N (2022) ‘Does stop and search reduce crime? Evidence from street-level data and a surge in operations following a high-profile crime’, Journal of the Royal Statistical Society: Series A (Statistics in Society Volume 185, Issue 3, July 2022, Pages 1370-1397 (viewed on 10 January 2025)

Brå (2014) ‘Why did increased police numbers not improve the clearance rate?’ The Swedish National Council for Crime Prevention. 2014_Why did increased police numbers.pdf (bra.se)

Braga AA and Dusseault D (2018) ‘Can Homicide Detectives Improve Homicide Clearance Rates?’ Crime & Delinquency: volume 64(3), pages 283-315 (viewed on 10 January 2025)

Braga AA, Turchan B and Barao L (2019) ‘The Influence of Investigative Resources on Homicide Clearances’, Journal of Quantitative Criminology: volume 35, pages 337-364 (viewed on 10 January 2025)

Braga AA and Weisburd D (2006) ‘Critic Problem-oriented policing: the disconnection between principles and practice’, in Weisburd D and Braga AA (eds.) Police Innovation: Contrasting Perspectives. Cambridge: Cambridge University Press (Cambridge Studies in Criminology), pages 133-152

Chalfin A, Hansen B, Weisburst EK and Williams MC Jr (2022) ‘Police Force Size and Civilian Race’, American Economic Review: Insights: volume 4(2), pages 139-158 (viewed on 10 January 2025)

Cook P, Kapustin, M, Ludwig J and Miller D (2017) ‘The Effects of COPS Office Funding on Sworn Force Levels, Crime, and Arrests Evidence from a Regression Discontinuity Design’ (viewed on 10 January 2025)

Coupe RT (2014) ‘Evaluating the effects of resources and solvability on burglary detection’, Policing and Society: volume 26(5), pages 563-587 (viewed on 10 January 2025)

Doerner WM and Doerner WG (2012) ‘Police accreditation and clearance rates’, Policing: An International Journal: volume 35(1), pages 6-24 (viewed on 10 January 2025)

Greenwood PW (1979) ‘The RAND Criminal Investigation Study: Its Findings and Impacts to Date’, RAND Corporation (viewed on 10 January 2025)

HMICFRS (2016) ‘PEEL: police effectiveness 2016’ (viewed on 10 January 2025)

HMICFRS (2020) ‘State of Policing: The Annual Assessment of Policing in England and Wales 2019: Part 2’ (viewed on 10 January 2025)

HMICFRS (2022) ‘The police response to burglary, robbery and other acquisitive crime - Finding time for crime’ (viewed on 10 January 2025)

HMICFRS (2023) ‘Police performance: Getting a grip’. PEEL Spotlight Report (viewed on 10 January 2025)

HMICFRS and HMCPSI (2024) ‘Joint case building by the police and Crown Prosecution Service’, Interim findings from phase 1 of a joint inspection by HMCPSI and HMICFRS of case building by the police and Crown Prosecution Service

Home Office (2020) ‘Stop and Search open data tables, year ending 31 March 2007 to year ending 31 March 2020’ (viewed on 10 January 2025)

Home Office (2023a) ‘Stop and Search open data tables, year ending 31 March 2021 to year ending 31 March 2023’ (viewed on 10 January 2025)

Home Office (2023b) ‘Recorded crime and outcomes data, year ending March 2023’ (viewed on 10 January 2025)

Home Office (2023c) ‘Police workforce England and Wales statistics’ (viewed on 10 January 2025)

Jang H, Hoover LT and Lawton BA (2008) ‘Effect of broken windows enforcement on clearance rates’, Journal of Criminal Justice: volume 36(6), pages 529-538 (viewed on 10 January 2025)

‌Kaplan J and Chalfin A (2019) ‘More Cops, fewer prisoners?’ Criminology and Public Policy: volume 18, pages 171-200 (viewed on 10 January 2025)

Kim E, Bowers K, Birks D and Johnson SD (2024) ‘Size isn’t everything: Understanding the relationship between police workforce and crime problems’, Journal of Criminal Justice: volume 95 (viewed on 10 January 2025)

Liska AE, Chamlin MB and Reed MD (1985) ‘Testing the Economic Production and Conflict Models of Crime Control’, Social Forces: volume 64(1), pages 119-138 (viewed on 10 January 2025)

Machin S and Marie O (2011) ‘Crime and Police Resources: the Street Crime Initiative’, Journal of the European Economic Association: volume 9(4), 1 August 2011, pages 678-701 (viewed on 10 January 2025)

McCandless R, Feist A, Allan J and Morgan N (2016) ‘Do initiatives involving substantial increases in stop and search reduce crime? Assessing the impact of Operation BLUNT 2’. London: Home Office (viewed on 10 January 2025)

McLean F, Meakins A and White E (2022) ‘Conducting effective investigations: Rapid evidence assessment’, College of Policing (viewed on 10 January 2025)

MoJ (2024a) ‘Criminal Justice System statistics quarterly: June 2023’. Ministry of Justice (viewed on 10 January 2025)

MoJ (2024b) ‘Criminal Justice System statistics quarterly: December 2023’. Ministry of Justice (viewed on 10 January 2025)

Owens EG (2013) ‘Cops and Cuffs’, in Cook, PJ, Machin S, Marie O and Mastrobuoni G, eds, Lessons from the Economics of Crime: What Reduces Offending? Cambridge, MA: MIT (viewed on 10 January 2025)

Paré PP, Felson RB and Ouimet M (2007) ‘Community Variation in Crime Clearance: A Multilevel Analysis with Comments on Assessing Police Performance’, Journal of Quantitative Criminology: volume 23(3), pages 243-258 (viewed on 10 January 2025)

Petersen K, Weisburd D, Fay S, Eggins E and Mazerolle L (2023) ‘Police stops to reduce crime: A systematic review and meta-analysis’, Campbell Systematic Reviews: volume 19(1), e1302 (viewed on 10 January 2025)

Roberts A and Roberts JM (2016) ‘Crime Clearance and Temporal Variation in Police Investigative Workload: Evidence from National Incident-Based Reporting System (NIBRS) Data’, Journal of Quantitative Criminology: volume 32(4), pages 651-674 (viewed on 10 January 2025)

‌Tilley N and Burrows J (2005) ‘An overview of attrition patterns’. Online Report 45/05. London: Home Office (viewed on 10 January 2025)

Tiratelli M, Quinton P and Bradford B (2018) ‘Does Stop and Search Deter Crime? Evidence From Ten Years of London-wide Data’, The British Journal of Criminology: Volume 58(5), September 2018, pages 1212-1231 (viewed on 10 January 2025)

Wharfe H, Furlong S, Feist A and Dunnett A (2024) ‘The impact of COVID-19 lockdowns on crime demand and charge volumes’. Home Office (viewed on 10 January 2025)

Zhao JS, Zhang Y and Thurman Q (2011) ‘Can Additional Resources Lead to Higher Levels of Productivity (Arrests) in Police Agencies?’ Criminal Justice Review: volume 36(2), pages 165-182 (viewed on 10 January 2025)

1. Cases ‘detected’, ‘solved’ and ‘cleared’ are terms used interchangeably throughout this chapter to reflect the language used in each report, but broadly have a similar definition as a case where a suspect was, as a minimum, identified and arrested. ↩

2. Churn was defined as a measure of turnover, to incorporate the effects of both leavers and joiners as a proportion of the overall size of the police force. ↩

3. Defined as a person not in his place of abode, in possession of any article for use in the course or connection with any burglary, theft or cheat (Theft Act, 1968) ↩

4. The label used in the published statistical bulletins is ‘charge/summons’. In this report the term ‘charges’ is used throughout. ↩

5. Excluding stalking, harassment, and malicious communications due to changes in the counting rules for these 3 offences at different points for the period covered by the modelling. ↩

6. The out-of-court resolutions outcome measure was the sum of recorded crimes receiving cautions, Community Resolutions (from 2013), Cannabis Warnings and Penalty Notices for Disorder. ↩

7. Endogeneity occurs when a predictor variable is correlated with the error term of the model. In this case, the endogeneity would be introduced through simultaneity bias. ↩

8. Headcount-only model results are reported in the Annex data tables, under tab ‘Victim-based headcount models’. ↩

9. Figure derived from the total FTE within investigator functions. ↩

10. Offences which the police must inform the Home Office by completing a crime report for statistical purposes. ↩

11. Statistics on prosecutions, convictions and sentencing are either derived from the LIBRA case management system, which holds the magistrates’ courts records, or the Crown Court’s CREST system (or Xhibit from March 2019) which holds the trial and sentencing data, and Common Platform data. We are very grateful to MoJ for providing full crime type level data for force area back to 2010. ↩