Abstract
Keywords
Evidence consistently shows that crime is disproportionately concentrated in a small number of micro-geographic locations, often referred to as ‘hotspots’ (Sherman et al., 1989; Sherman & Weisburd, 1995; Weisburd, 2015, 2018, Weisburd et al., 2009, 2012, 2024). This skewed distribution means that a relatively small set of places accounts for a large share of total crime. Using this “law of crime concentration in place” (Weisburd, 2015), some crimes can be predicted, at least in terms of their spatial and temporal characteristics. In turn, controlled trials have demonstrated that focusing police resources on persistent high-crime locations—particularly by increasing visibility compared with business-as-usual deployment—can be a cost-effective way to reduce crime (Ariel et al., 2016, 2020; Sherman & Weisburd, 1995; Weisburd et al., 2024), at least in the short term before the deterrence effect decays (Sherman, 1990). Research has repeatedly confirmed the effectiveness of hotspots policing as a crime reduction strategy (Braga et al., 2019; Braga & Weisburd, 2022; Sherman & Weisburd, 1995; Smith et al., 2024; Weisburd & Telep, 2014). Thus, visible and capable police presence increases the perceived likelihood of apprehension among potential offenders, thereby discouraging criminal activity (Ariel & Partridge, 2017; Smith et al., 2024).
By contrast, comparatively less is known about crime concentration and the effectiveness of policing strategies in mass transit systems (Ariel, Langton, et al., 2024). Public transportation operates as a complex environment, encompassing multiple components—such as buses, trains, trams, stations, and interchanges—and serving diverse user groups, including both passengers and personnel. The design of transit facilities and the environmental conditions within them can influence both crime levels and the capacity for effective policing (Newton, 2004). Experimental research has produced evidence on policing fixed transit points, such as train stations (Ariel et al., 2020) and bus stops (Ariel & Partridge, 2017), with diverse types of capable guardians (Ariel et al., 2017). Yet, the one area that remains largely absent from this literature is the policing of the moving elements themselves: bus interiors, train carriages, ferries, planes and other mobile spaces. These settings are inherently different from static locations: they involve populations in transit, rapidly shifting crime opportunities, and fluctuating levels of guardianship. Determining where and when to position an officer in such environments – as per the classic hotspot policing approach – can be operationally complex, as illustrated by multi-carriage systems like the London Underground or the New York City Subway, where trains arrive within minutes and passenger flows change constantly. Targeting is further complicated by imprecise crime reporting in these contexts (Ariel, 2011). These factors create both methodological and conceptual challenges for identifying and addressing moving hotspots, underscoring the need for rigorous empirical tests in these environments (Lincoln & Gregory, 2015).
This study examines policing within London’s mobile transportation network, focusing on the impact of high-visibility police patrols on buses travelling through selected high-crime bus segments with disproportionately more bus-related crimes, referred to as ‘hot bus corridors’. A six-week randomised controlled trial was conducted in London, in which 556 hot bus corridors were identified and randomly assigned to either treatment or control conditions, with 50% receiving the intervention. The treatment corridors were patrolled by officers engaged in targeted, high-visibility activities on the buses, while the control corridors continued to receive standard patrols at their routine frequency. The intervention’s impact was assessed both in terms of crime counts and the Cambridge Crime Harm Index (CCHI; Sherman et al., 2016) to capture variations in both the volume and severity of recorded crimes.
The potential implications of this study relate both to practice and theory. Should the results indicate measurable crime reductions associated with targeted patrols in moving hotspots, they may help inform decisions about resource allocation in public transport environments, including the possible refinement of deployment patterns to address particular routes or times. From a theoretical perspective, examining the dynamics of bus-based hotspots may contribute to a more nuanced understanding of crime concentration and the applicability of ecological frameworks to mobile settings (and therefore an extension of Weisburd, 2015). Given that buses traverse varied urban environments, the findings may also provide insights into how deterrent effects operate in contexts where crime opportunities and potential guardianship shift rapidly over time and space.
Literature Review
Crime and Place
Contemporary criminological literature has demonstrated that the spatial distribution of criminal activity follows distinct and highly stable patterns of concentration (Braga et al., 2019; Brantingham & Brantingham, 2013; Sherman et al., 1989; Sherman & Weisburd, 1995; Weisburd, 2015; Weisburd et al., 2012, 2024). Crime consistently clusters in a small proportion of micro-geographic locales, such as street segments or addresses. A landmark analysis of more than 300,000 calls for service to the Minneapolis Police Department, covering over 100,000 locations, found that just over 3 % of addresses generated half of all recorded emergency responses, while the vast majority reported no incidents at all (Sherman et al., 1989). This phenomenon has been conceptualised in two complementary ways. Weisburd (2015) formalised it as the ‘law of crime concentration at place’, establishing that roughly 50% of crime is typically concentrated in 2.1%–6% of street segments and that this distribution remains remarkably consistent across time and settings. Sherman (2007) advanced the related notion of the ‘power few’—later referred to as the ‘felonious few’ (Sherman, 2019)—to describe the disproportionate criminogenic contribution of a small number of locations. Supporting these principles, a systematic review of 44 studies conducted between 1970 and 2015 found that, on average, the top 10% of high-crime places accounted for approximately 63% of all recorded crime when measured by prevalence, and 43% when measured by frequency (Lee et al., 2017), reinforcing the empirical regularity of extreme spatial concentration across diverse jurisdictions and crime types.
Notably, many hotspots remain ‘hot’ for extended periods, sometimes for years. In Seattle, Weisburd, Bushway, Lum, and Yang (2004) applied trajectory modelling to crime incidents across individual street segments, uncovering remarkable stability in concentration patterns over a 14-year period. Half of all recorded crime occurred in just 4% of street segments, and crime levels at these locations remained consistent over time, regardless of whether they were categorised as high, medium, or low-crime areas. A Canadian replication by Curman et al. (2014) in Vancouver, analysing over one million service calls between 1996 and 2001, found similar concentration patterns, with 60% of crime occurring in just 7.8% of street segments and intersections. More recently, Kumar et al. (2024) demonstrated that this persistence also characterises domestic abuse, with a longitudinal analysis of police records over 13 years in London, UK, showing that a small number of micro-locations consistently generated a disproportionate share of incidents. These spatial concentrations are further overlaid by temporal regularities, with crime clustering at certain hours of the day, on particular days of the week, and during specific months (Johnson et al., 2008; Ratcliffe, 2004; Townsley, 2008). This dual pattern has led some scholars to propose that the phenomenon is best described as the ‘law of concentration of crime in place and time’ (Ariel & Partridge, 2017, p. 810; see also Kumar et al., 2024).
At least three prominent theoretical frameworks have been advanced to account for the spatial concentration of crime in particular locations. The first, routine activities theory (Cohen & Felson, 1979), posits that crime occurs when three necessary elements converge in space and time: a motivated offender, a suitable target, and the absence of a capable guardian. The effectiveness of guardianship is contingent upon both the visibility and the authority of the guardian, with diminished or absent guardianship increasing the likelihood of criminal events. Accordingly, crime concentration may emerge in settings characterised by a high density of motivated offenders and attractive targets during periods when guardianship is weakened or absent. The second, rational choice theory (Clarke & Cornish, 1985), conceptualises offending as the outcome of deliberate decision-making processes in which offenders weigh potential rewards against perceived risks. Central to this calculus is the perceived probability of apprehension, supplemented—though typically to a lesser extent—by considerations of the severity and immediacy of potential sanctions (Nagin, 2013; Piquero et al., 2011; Pratt et al., 2017). From this perspective, crime is likely to cluster in areas offering high rewards and low perceived risks, and to diminish in locations where visible and authoritative guardianship elevates the perceived likelihood of detection and arrest.
The third, crime pattern theory (Brantingham & Brantingham, 1993), synthesises elements of both routine activities and rational choice perspectives to explain the distribution of crime across places. Building on earlier work in environmental criminology (Brantingham & Brantingham, 1981), the theory argues that offenders encounter criminal opportunities as they move through their routine activities in familiar areas such as home, work, school, and recreation. The distribution of crime is therefore shaped by the spatial and temporal patterns of offenders’ daily movements (Townsley & Sidebottom, 2010) and by the environmental characteristics of particular locations (Brantingham & Brantingham, 2013). Opportunities present in places that come to offenders’ attention are more likely to be targeted. Although crime pattern theory and routine activities theory share complementary assumptions, they differ in emphasis: crime pattern theory regards places as criminogenic due to their situational context and environmental connectivity, whereas routine activities theory focuses more on the presence or absence of particular types of people (Eck & Weisburd, 2015).
Harmspots
Crime hotspots need not be identified and targeted solely based on the number of incidents recorded within the spatiotemporal constraints of the hotspot. For example, while criminologists typically define hotspots as areas that experience the highest levels of recorded crime, they rarely consider the “ambient population” (Andresen & Jenion, 2010; Malleson & Andresen, 2016), such as the number of pedestrians, residents, or passengers in the area (e.g., Ariel, Langton, et al., 2024). More crucially, the traditional practice of counting crime fails to capture the vastly varying degrees of severity that different offences impose on society. The US Uniform Crime Reporting system classifies crimes into Index Crimes – comprising the most serious offenses such as violent crimes (e.g. murder, rape, robbery, and aggravated assault) and property crimes (e.g. burglary, larceny-theft, motor vehicle theft, and arson) – and Non-Index Crimes, which include all other less serious offenses, such as drug offenses, simple assaults, and fraud.
A more nuanced approach to measuring crime severity utilises weighting systems. For example, Wolfgang et al. (1972) assessed crime severity using surveys developed from the Sellin-Wolfgang Index of Crime Severity (Sellin & Wolfgang, 1964). Respondents ranked crimes by their level of harm, providing a method to quantify the perceived severity of criminal offences and aiding in the development of weighted crime severity scales. A more common practice in contemporary crime harm research is anchored in the criminal justice system’s assessment of offence seriousness as reflected in laws, conviction records, or sentencing practices. Thus, this weighted approach offers a more accurate representation of the social costs of criminal activity compared to raw counts, which can misrepresent the relative significance of distinct criminal activities.
The CCHI (Sherman et al., 2016) is a modern method that has gained traction in both research and practice (van Ruitenburg & Ruiter, 2023). This methodology translates the severity of each offence into a standardised metric based on prison sentences – specifically, the number of days that first-time offenders would have been sentenced to (without mitigating or aggravating circumstances) had they been convicted in court. For crimes typically penalised through monetary fines rather than incarceration, the severity score is calculated by converting the financial penalty into labour hours at minimum wage rates specific to the jurisdiction. This framework acknowledges the fundamental difference in severity between offences, such as a violent assault resulting in grievous bodily harm and a minor retail theft.
By incorporating crime harm scores, law enforcement agencies can move beyond a sole focus on volume to capture the inherent gravity of offences in addition to their frequency (Norton et al., 2018). This allows for enhanced resource allocation to areas – and individuals (see, for example, Lay et al., 2023) – that experience a disproportionate level of crime harm. Moreover, crime harm assists in identifying the ‘power few’ among offenders, victims and locations (Curtis-Ham, 2022; Frydensberg et al., 2019; Ottaro et al., 2024). In the context of spatial crime analysis, high-harm locations can be conceptualised as ‘harmspots’, which are not necessarily the same as traditional count-based hotspots (Norton et al., 2018). This approach not only refines spatial crime analysis but also enhances the effectiveness of resource allocation and strategic planning in policing (Macbeth & Ariel, 2019; Verrey et al., 2023; Weinborn et al., 2017).
From a spatial perspective, measures of crime harm are typically more densely concentrated than counts of criminal incidents (Harinam et al., 2022; Ottaro et al., 2024). In Birmingham, UK, for example, half of all recorded criminal incidents were confined to just 3 % of street segments. In contrast, the same proportion of total crime harm was concentrated within a mere 1 % of street segments (Weinborn et al., 2017). This heightened concentration of harm relative to incident counts has been replicated across multiple contexts and offence types. In Sussex, UK, Norton et al. (2018) examined four specific categories of crime, finding that theft and robbery displayed predictable spatial clustering, while sexual offences exhibited a more dispersed and less predictable distribution. Similar patterns have been observed internationally; for instance, Ng et al. (2023) documented pronounced concentration in crime harm within Hong Kong’s train stations. In the United States, Fenimore (2019) reported findings consistent with Birmingham’s pattern, identifying parallel levels of spatial concentration for both crime counts and crime harm. Harinam et al. (2022) reported comparable results in Toronto, showing that half of all recorded crime severity scores were concentrated in only 7.3% of spatial units, compared to 8.4% for incident counts. They also found temporal stability in the spatial distribution of both harmspots and hotspots, further reinforcing the persistence of these patterns over time.
Crime Analysis in Mass Transit Systems
Crime concentration principles also apply in the specialised environment of public transport systems, where temporal passenger flows and environmental design create distinctive opportunity structures for offending. Spatial clustering within transit networks is equally pronounced. In England and Wales, half of all reported crimes on the rail network occurred in just 5% of stations, with micro-locations such as ticketing areas, entry gates, and concourses accounting for a disproportionate share of incidents (Ariel, 2011).
Temporal concentrations are closely tied to ridership patterns, with peak commuting periods—such as morning and late afternoon rush hours—associated with elevated risks of theft, personal offences, and antisocial behaviour (Ariel, 2011). Environmental and situational conditions further shape these patterns. Ceccato and Uittenbogaard’s (2014) analysis of Stockholm’s metro system showed that passenger flow data, when integrated with crime records, revealed distinct temporal “rhythms” for different offence types, varying by time of day, day of week, and season. Similar patterns have been by Ng et al. (2023), who found that in Hong Kong’s Mass Transit Railway, both crime counts and crime harm clustered disproportionately during high-volume travel periods, while Ottaro et al. (2024) reported analogous findings in Canadian transit hubs, where temporal spikes in offending corresponded with predictable surges in passenger density.
Finally, large stations functioning as multimodal hubs—often integrated with shopping arcades and nighttime economy venues—as high-risk convergence points, echoing Bertolini’s (1999) conceptualisation of transit nodes as spaces of diverse social intersections. This aligns with Brantingham and Brantingham’s (1993, 1995) framework of crime generators and attractors, both of which are prevalent in major transit stations. In Washington, D.C., Irvin-Erickson and La Vigne (2015) found that station characteristics such as isolation, visibility, and surrounding socioeconomic conditions predicted elevated risks for theft and violent crime (see also Ceccato et al., 2024). Newton et al. (2014) demonstrated that crowded conditions provide offenders with natural camouflage, reducing the perceived risk of detection—a dynamic particularly relevant to theft, sexual offences, and harassment. National surveys of women and girls have confirmed that such conditions facilitate unwelcome sexual advances within UK rail systems (Ariel, Ceccato, et al., 2024). These characteristics make the mass transit systems’ crime problem more pronounced.
Deterrence Mechanisms in Hotspot Policing
Hotspots policing, though defined differently across studies and contexts (Weisburd & Telep, 2014), consistently demonstrates significant crime reduction effects in targeted areas compared to ‘business-as-usual’ controls (Braga et al., 2014, 2019; Braga & Weisburd, 2022; Smith et al., 2024; Weisburd et al., 2024; Weisburd & Telep, 2014). At its core, the approach often takes the form of a micro-geographic, problem-oriented strategy that may combine a saturated police presence, community engagement, and technological measures such as CCTV, drones, and facial recognition systems to deter offending in designated locations. The underlying mechanism is partly explained by deterrence theory: potential offenders are less likely to commit crimes when the perceived risk of apprehension is high (Nagin, 2013; Pratt et al., 2006). A visible and authoritative police presence can therefore have a substantial preventive effect. This has been supported by experimental evidence, including Sherman and Weisburd’s (1995) Minneapolis Hot Spots Patrol Experiment and Koper’s (1995) patrol dosage study, both of which demonstrated that targeted and strategically timed patrols can significantly reduce crime in high-risk micro-locations.
Importantly, the deterrent effect of hotspot policing is also evident in mass transit systems. For instance, crime rates in the New York City Subway system have likely decreased due to the deployment of more police officers during specific hours (Chaiken et al., 1974). Ariel and Partridge (2017) conducted a study on ‘hot’ bus stop transit environments, suggesting that targeted patrols in these settings can reduce crime, provided that patrols are randomly conducted during peak hours or high-harm periods. Ariel et al. (2020) conducted an RCT on the London Underground to assess the effectiveness of high-visibility policing on platforms across half of 102 platforms, while the other half received no proactive patrols. The study found a significant reduction in crime on patrol days, particularly in theft and violent crime. These patrols, conducted primarily during the peak hours between 14:00 and 23:00 proved to be an effective deterrent, with evidence of a diffusion of social control benefits to adjacent, non-targeted areas within the stations, even in the absence of direct patrols.
The Present Experiment
Despite robust evidence supporting the effectiveness of high-visibility patrols in fixed hotspots – such as platforms, stations, and street blocks (e.g., Ariel et al., 2020; Braga et al., 2019; Sherman & Weisburd, 1995) – crime reduction in moving hotspots, including buses, trains, and underground carriages, remains largely underexplored. Prior research on place-based policing interventions has focused almost exclusively on stationary locations, consistently showing that visible police presence in high-crime areas can deter offending (e.g., Ratcliffe et al., 2011). Nevertheless, no studies have systematically examined the distinct dynamics and challenges of policing confined, transient spaces where potential offenders and victims are in close proximity for defined periods of time.
This study addresses that gap by evaluating the effects of high-visibility police patrols in “moving hotspots”, specifically, buses travelling through selected high-crime street segments, where offenders and potential victims remain spatially fixed relative to each other despite the vehicle’s mobility. This is not merely a technical extension of traditional hotspot policing research, which has been repeatedly tested in environments with stable geographic boundaries and fixed populations (Braga et al., 2019). Rather, it represents a conceptual shift, focusing on the nuances of crime prevention in transient yet enclosed environments. In such settings, passengers are confined by the vehicle’s layout and journey duration, creating immediate and unavoidable proximity between potential offenders and victims. This proximity can heighten the risk of opportunistic offences, such as theft or assault, in the absence of visible guardianship (Clarke, 1995).
While existing studies have shown that patrols on platforms and in stations can reduce crime (Ariel et al., 2020), it is unknown whether these effects extend to enclosed, mobile environments under experimental conditions. We hypothesise that regular patrols in these settings will reduce crime both by deterring potential offenders and by increasing perceived guardianship among passengers (Cohen & Felson, 1979).
Methods
Setting and Design
We used a randomised control group pretest-posttest design to evaluate the effectiveness of targeted high-visibility police patrols in reducing crime counts and harm on London’s bus network. The experiment was conducted in London, the largest city in the UK, with a population of 9.8 million (World Population Review, 2025). However, this figure does not include the daily influx of around 1.1 million commuters (Reed, 2022) and an average of more than 80,000 tourists per day (Touristplaces, 2025), emphasising the crucial role of public transportation that facilitates more than 3.5 million ridership per day (TfL, 2023, 2024).
Geographically, London’s boroughs are policed by the London Metropolitan Police Service (MPS), with the exception of a small square mile managed by the City of London Police. London has historically experienced higher crime rates than most other regions in England and Wales, with 86 offences per 1,000 people recorded in the year ending June 2021 (Office for National Statistics, 2021). In comparison, the Southwest of England had the lowest crime rate at 61.6 per 1,000 people, while the Northeast recorded the highest at 95.0 per 1,000 people (House of Commons Library, 2022). Despite London’s overall crime rates, public transportation remains relatively safe, with an estimated crime rate of 11.4 crimes per million passenger journeys on the London Underground and 10.8 crimes per million journeys on buses (TfL, 2021). The majority of bus-related crime incidents involve theft and fare evasion, with serious offences such as violent crimes occurring more rarely (Home Office, 2021).
To identify eligible hotspots, baseline data collection was conducted between June 2018 and June 2021, with the intervention period spanning from December 2021 to January 2022. Both the end of the baseline data collection and the intervention period occurred during the COVID-19 pandemic, a time when overall crime rates declined, partly due to a sharp reduction in passenger numbers. However, the same period also saw a rise in public order offences linked to confrontations over face masks. In the 2021/22 performance year, total recorded offences fell by 23.4% compared to 2019/20, with public order offences, sexual offences, and violence against the person declining by 4.5%, 6.7%, and 22.3%, respectively. Although bus ridership in London typically rises during the holiday season because of shopping, social gatherings, and other activities, Transport for London data (London Datastore, 2024) show that bus journeys during the intervention period totalled 95.1 million, down from 144.9 million in the same months before the pandemic. This broader context helps explain the overall reduction in bus-related crime during the intervention, as the lower number of journeys offset the seasonal patterns that usually bring increased passenger volumes and higher crime levels.
The Safer Transport Teams within the MPS are primarily responsible for policing London’s bus network. These teams are fully funded by Transport for London, the government body that manages and oversees public transport services, infrastructure, and road systems across the city. They comprise approximately 1,000 officers who respond to emergency calls and investigate crimes occurring on buses and at transport hubs across 13 designated spatial transport areas in London. Importantly, the Safer Transport Teams operate reactively rather than proactively, lacking a progressive patrol strategy based on high-crime locations. Instead, patrol areas are determined through three primary methods: (1) monthly Tactical Tasking Coordination Group meetings, which recommend locations based on recent spikes in crime counts, (2) requests from Transport for London that highlight problematic locations for anti-social behaviour or areas with high rates of drivers activating panic buttons, and (3) the local knowledge of police supervisors.
Moving Hotspots: ‘Hot Bus Corridors’
Identifying hotspots and eligible units for randomisation is crucial for effective causal analysis (Sherman, 2013). While Sherman et al. (1989) recommended starting with the smallest spatial unit of analysis, this is particularly challenging with moving units like buses, as pinpointing crime locations on moving hotspots is more complex compared to fixed, high-crime areas (Newton, 2004). This problem is exacerbated due to the scarcity of crime events on the London bus network (see discussion in Ariel, Langton, et al., 2024).
To identify eligible and stable hotspots, we analysed three years of MPS crime data related to buses. Reports flagged with the ‘bus flag’ from June 2018 to June 2021 were reviewed to identify trends beyond seasonal spikes. This flag indicates bus involvement but does not necessarily denote transport-related crimes. To refine the dataset, we selected only those crimes occurring on the bus, including incidents during boarding and alighting.
Narrowing crime locations to the smallest spatial unit proved challenging, as street names and postcodes recorded most bus-related crimes, while specific bus routes and precise locations were often documented in free-text fields within the crime reports. Given these constraints, the smallest identifiable spatial unit for a moving bus hotspot was defined as a corridor: high-crime street sections, confined within a postcode, through which buses travel. While these corridors are geographically fixed, the crime events are associated with the buses moving through them. Accordingly, these areas are classified as moving hotspots, reflecting the dynamic nature of crime opportunities occurring within buses as they pass through these corridors.
Using this approach, the initial dataset comprised over 20,344 locations within the MPS jurisdiction, categorised by street name and postcode. This number was refined to 4,242 units by applying CCHI scores, establishing a threshold of 1,000 CCHI units over the 36 months preceding the trial. This selection represents 20% of all units, which account for 80% of the recorded CCHI during the baseline period. From this list, 556 units were randomly selected based on the number of available officers capable of delivering the intervention in half of these units (n = 278). Ultimately, these 556 units represent 2.7% of all corridors and are classified as London’s ‘hot bus corridors
Random Assignment and Partial Blinding
Randomised Allocation and Crime Harm Distribution Across Patrol and No-Patrol Conditions by Spatial-Temporal Units
Participating officers were provided only with the treatment units and not the complete list of corridors, ensuring that they remained blinded to the control areas to prevent any potential treatment contamination (see Ariel et al., 2022). Although the officers were aware that they were participating in an academic evaluation, no individuals outside the study group were informed of all the specific treatment or control locations.
The Intervention
The treatment involved high-visibility patrols conducted by officers from the Safer Transport Teams within the MPS. These officers conducted uniformed patrols during peak hours identified as high-harm periods, with each patrol consisting of two 15-min rides in opposite directions, totalling 1 hour per shift. This approach was integrated into the officers’ routine responsibilities to ensure alignment with their standard operating procedures.
Patrol teams comprised both police constables and Police Community Support Officers (PCSOs), typically patrolling in pairs, visibly engaging with passengers, introducing themselves to drivers, and checking upper decks when applicable. They also encouraged face-mask compliance as part of the three E’s (Explain, Educate, and Encourage) developed during the COVID-19 pandemic. Incentives like overtime pay were offered, particularly for night and weekend shifts. Notably, officers continued their regular policing activities, including crime reporting and stop-and-searches, as necessary.
The control bus corridors did not receive enhanced patrols; instead, they were subject to regular policing operations conducted by Safer Transport Teams and Emergency Response Teams, which maintained standard policing levels without additional patrols. These activities did not include proactive, high-visibility patrols on the buses unless they were part of special operations or in response to crime incidents. Consequently, officers typically had only a few seconds to minutes of presence on the buses during their shifts.
Outcome Measures
The two primary outcome measures were crime count and crime harm, using the counted incidents and CCHI scores during the trial period (December 6, 2021, to January 31, 2022) and the baseline period preceding the experiment (September 6, 2021, to December 5, 2021). CCHI, as noted earlier, assigns harm scores based on the minimum recommended days of imprisonment for each offence, as outlined by the Sentencing Guidelines for England and Wales (Sherman et al., 2016). To ensure accuracy, we excluded police-generated incidents and crime types that occur in the context of police presence, such as assaults on officers (see Bland & Ariel, 2020). This approach ensured that the recorded harm was attributed solely to victims and not inflated by the increased presence of officers.
Output Measures
To ensure compliance, supervisors and representatives from 13 Safer Transport Teams participated in briefings one week before the experiment began, outlining the goals and methodology of this RCT. The briefing emphasised the role of high-visibility patrols and the importance of prevention in policing. Regular updates and feedback sessions were conducted to maintain engagement and address any emerging issues.
Officer tracking was self-reported due to technical difficulties with officers’ GPS-enabled radios and the unavailability of officer travel card data (for limitations of this approach, see Wain and Ariel, 2014). Officers recorded their activities using a basic Microsoft form following each patrol, a method that had been tested in advance for simplicity and accuracy. Furthermore, regular meetings were held with supervisors to maintain engagement and address logistical challenges. These meetings proved to be more effective than emails or meetings that involved only senior leaders, thereby replicating earlier findings on tracking by De Brito and Ariel (2017). They provided valuable feedback and updates on patrol activities, aimed at keeping patrolling officers motivated, offering insight into patrol outputs, and ensuring that the treatment units received the appropriate levels of patrols.
Analytical Procedures
We first present the data descriptively, highlighting differences in crime harm and count before and after the intervention across both the treatment and control groups. Additionally, we investigate whether crime harm and crime count were concentrated at specific hours or days to identify any temporal patterns. We also analyse these patterns separately for the treatment and control groups to determine if there were notable differences in the distribution of crime harm or count across different shifts.
We first conducted a series of t-tests to determine if there were statistically significant differences between the treatment and control groups at baseline. The randomisation produced similar groups in terms of crime harm scores (p = 0.618) and crime counts (p = 0.777).
Next, given the data distribution, we employed generalised linear models (GLMs) to assess the effect of the intervention while controlling for the baseline period. Specifically, we used a GLM with a negative binomial distribution and a logarithmic link for the crime count variable, providing a more accurate estimation of the impact of increased patrols on crime volume. For the crime harm variable, we utilised a GLM with a normal distribution and an identity link. 1
Results
Unadjusted Results
Three Months Pre-treatment Crime Harm Scores and Total Crime Count
Six-Week Trial-Period Crime Harm Scores and Total Crime Count
Moreover, peak times and days during the baseline period are consistent across both treatment and control areas, with the highest crime periods occurring from Monday to Friday between14:00 and 22:00. These peak times correspond with the bus network’s busiest travel periods, likely contributing to higher crime harm and count due to increased passenger volume and greater opportunities for criminal activity.
Following the intervention, patrol visits in the treatment areas were associated with a reduction in both crime harm and crime count compared to the control areas. Specifically, crime harm in the control areas increased from 4,678.0 to 6,478.0 (+38.5%) during the trial period, while it decreased from 5,609.5 to 1,734 (−69.1%) in the treatment areas.
Marked decreases were observed in the pretest and post-test comparisons during peak days and times. Total harm fell from 3,485 to 1,626 (−53.3%) from Monday to Friday between 14:00 to 22:00, and from 2,083.5 to 93.0 (−95.5%) from Monday to Friday between 06:00 and 14:00. In contrast, the control group exhibited increases in two of the three peak days and times during the post-intervention period compared to the pre-intervention period. The total harm score increased from 3,468.0 to 3,842.0 (+10.8%) from Monday to Friday between 14:00 and 22:00, and from 837 to 2268 (+171%) from Monday to Friday between 06:00 and 14:00. Interestingly, total crime harm surged from 4.0 to 368.0 (+9100%) from Monday to Friday between 22:00 and 06:00.
In terms of crime counts, we observed an increase from 114 to 152 incidents (+33.3%). However, this increase was more pronounced in the control group, where the total crime count rose by 78.7%, increasing from 108 to 193 recorded events. This contrast in trends between crime harm and crime count highlights the need for complementary metrics in measuring crime (Harinam et al., 2022). While unadjusted crime harm scores decreased in the post-treatment period compared to the pre-treatment period in the hotspots, the overall volume of crime has increased.
GLM Models
Generalised Linear Model (GLM) Effect Estimates of Increased Patrols on Crime Harm and Crime Counts (N = 556): Coefficients, Standard Error, Z-Statistics, and 95% Confidence Intervals (CI)
+p ≤ 0.10; *p ≤ 0.07; **p ≤ 0.01; ***p ≤ 0.001.
Treatment Fidelity
While we adhered to the intention-to-treat model, it is important to report that the delivery of treatment was not uniform. Figure 1 illustrates the planned patrol rates compared to the actual patrols conducted during the experiment. Importantly, only one basic command unit, Area West, met its targeted patrol quota, exceeding the required number of patrols by more than double. This underscores the challenges of conducting field experiments, where real-life conditions can be unpredictable. Each geographical team operates under unique demands, priorities, and supervisory styles, with varying on-the-ground conditions impacting patrol routes. For example, central London bus corridors, such as Area West and Central North, feature more clustered segments, while outer bus corridors, like South Area, have greater travel distances between segments. Patrol Targets vs. Actual Patrols Achieved by the Teams During the Trial Period Experiment
Additionally, differences in staffing and the impacts of the COVID-19 pandemic – such as team unavailability due to illness or self-isolation – contributed to variations in patrol completion across the hot bus corridors. The randomisation of the hottest bus corridors also led to disparities in patrol requirements. For example, Central West personnel patrolled six hot bus route segments within a small area, completing 1,059 patrols, which far exceeded the required 448. In contrast, South West had 34 corridors, with patrol officers completing only 625 of the 2,208 required patrols.
Additional Analyses
Temporal Distribution of Crime Harm and Crime Count During the Trial Period
Temporal Distribution of Crime Harm and Crime Count Separately for Control and Treatment Groups During the Trial Period
In the treatment group, nearly all crime harm (93%) is concentrated in the Monday to Friday 14:00 to 22:00 shifts, which is 34.7% higher than the concentration observed in the control group. Conversely, during the Monday to Friday 06:00 to 14:00 shifts, crime harm was 29.64% lower in the treatment group compared to the control group. This distinct pattern, where virtually all crime harm in the post-intervention treatment group is concentrated in the afternoon to evening shifts, contrasts with the more even distribution of crime harm observed among the pre-intervention treatment group shifts (see Table 2). This suggests that the intervention was particularly effective in reducing crime harm during the morning hours.
Discussion
Public transportation is a complex system that has a significant impact on public safety (Newton, 2004). Previous studies explored crime trends and identified hotspots within mass transit systems (Ariel et al., 2022, Ariel, Langton, et al., 2024; Ariel & Partridge, 2017), but research focusing on the moving hotspots within these systems remains largely unexamined. We examined the impact of high-visibility police patrols in designated ‘hot bus corridors’ – buses travelling through selected high-bus-related crime street segments. Both crime and volume and crime harm measures were used
The degree of crime volume and harm concentration was substantial. About 80% of the crime harm and approximately 96% of the crime count occurred between Monday and Friday during the experimental period, not at weekends. Within these days, 45% of the total crime harm and 67% of the total crime count were concentrated between 14:00 and 22:00. This concentration in both space and time underscores the validity of the ‘law of concentration of crime in place’ (Weisburd, 2015) ‘and time’ (Ariel & Partridge, 2017, p. 810)
The results of the intervention provide additional evidence of the well-established effectiveness of policing hotspots, but in a novel context – moving hotspots. The treatment, which involved intensified police patrols in designated bus corridors, caused a reduction in both crime harm and counted crime compared to control units that did not receive more police presence on buses. This reduction was most pronounced during the morning rush hours. These results support the notion that the presence of capable guardians on moving hotspots can contribute to effective crime management (Ariel et al., 2016)
Theoretical Implications
Our results align with rational choice theory and deterrence theory, suggesting that the presence of police officers in designated bus corridors increases the perceived risk of apprehension, deterring prospective offenders who might otherwise commit crimes or antisocial behaviour in the confined space of a bus
Furthermore, the patrols were conducted randomly within an eight-hour window, which enhanced the deterrence effect (Braga & Bond, 2008). It can be assumed that, during the trial period, frequent travellers who observed these patrols likely perceived a continuous police presence, a perception that could spread on social media among schoolchildren and other passengers (Braga & Weisburd, 2010). Although we lack direct evidence to corroborate this assumption, the visibility of officers boarding and alighting from buses likely amplifies the perceived police presence within the environment, thereby enhancing the deterrent effect (Gill et al., 2014). We observed the physical manifestation of these choices in the form of reduced crime counts and crime rates. The effectiveness of these patrols, even when conducted infrequently, indicates that a minimal police presence can impact crime rates by altering criminals’ risk calculations and exerting a deterrent effect (Ariel & Partridge, 2017).
Practical Implications
These findings highlight not only the feasibility of identifying crime hotspots on moving buses but also the operational effectiveness of deploying targeted interventions in mobile environments. By demonstrating that crime concentration patterns can be mapped and acted upon in transit vehicles, the study provides a replicable framework for extending similar strategies to other high-density transport settings, such as underground networks, commuter trains, and major transportation hubs (Ceccato, 2014; Newton et al., 2014). This mobility-focused approach offers transit authorities and police forces an additional layer of preventive capability, complementing existing hotspot policing in fixed locations (Ariel et al., 2020; Braga et al., 2019; Weisburd & Telep, 2014).
From an operational standpoint, incorporating moving hotspot analysis into routine policing can optimise resource deployment by aligning patrol schedules with both spatial and temporal crime peaks (Braga & Weisburd, 2022; Sherman, 2013). This allows for more precise targeting of patrols during high-risk routes and times, reducing reliance on static, geographically bound enforcement models. Moreover, the approach can be integrated with technological tools—such as real-time incident reporting—to provide dynamic, data-driven deployment plans that respond to emerging crime trends in near real time (Piza, 2019).
At the policy level, embedding mobile hotspot strategies within broader crime prevention frameworks can enhance the overall resilience of urban transit systems. This includes improving passenger perceptions of safety, reducing fear of crime, and increasing public trust in transit policing (Irvin-Erickson & La Vigne, 2015; La Vigne, 1996). These benefits align with broader urban safety objectives, as safer public transport networks can encourage greater use of transit, reduce reliance on private vehicles, and contribute to environmental and economic sustainability goals (Ceccato & Newton, 2015). Focusing on moving hotspots and peak risk periods enables agencies to allocate limited resources more efficiently, extend the deterrent reach of police presence, and achieve measurable crime reductions across an entire transport network.
Limitations and Future Research
This study has several key limitations. First, using crime reports to assess crime harm may result in an underestimation of crime levels, as many incidents go unreported (HMICFRS, 2019). Additional metrics, such as driver activations and calls for service, were not included, but could offer a more comprehensive assessment of crime harm on the bus network (Ariel & Partridge, 2017; Gill et al., 2014).
Second, our analysis was constrained by the availability of aggregated outcome data covering the entire post-intervention period, which prevented us from examining how the intervention’s effects evolved over shorter timeframes. As a result, we were unable to assess potential temporal patterns such as immediate deterrence effects, delayed impacts, or decay over short periods of time—patterns that have been observed in other hotspot policing and situational crime prevention studies (e.g., Braga et al., 2019; Koper, 1995; Sherman, 2007). Such granularity would not only improve understanding of intervention trajectories but also inform operational decisions about optimal patrol timing, frequency, and duration in similar mobile hotspot contexts.
Third, variations in patrol implementation across areas due to operational constraints, including COVID-19-related absences, were a limitation. Future studies should look at dose-response relationship between police presence and crime volume and crime harm.
Fourth, the exact nature of interventions in patrol areas cannot be fully characterised due to a lack of data on the specific activities conducted by officers during patrols (Sherman, 2013). We have no knowledge of the activities that occurred in these hotspots, nor do we know the degree to which officers uniformly applied crime control strategies on the buses. Future research should consider utilising CCTV or body-worn camera footage to code officers’ interactions with members of the public based on generally acceptable parameters such as procedural justice and professionalism (e.g., Barber & Tankebe, 2024; McCluskey and Uchida, 2023; Nawaz and Tankebe, 2018). Future studies should also include structural and socioeconomic data to assess the contexts in which these interventions are most effective.
Fifth, although the sample size was sufficiently large to ensure adequate statistical power, the intervention was limited to a six-week patrol period, creating a scarcity of treatment exposure that reduced the likelihood of detecting statistically significant effects at the standard threshold (p < 0.05). While our results come quite close to this threshold – .07 vs. .05 – we remain encouraged, as they follow the general literature on hotspot policing. Nevertheless, short trial durations restrict the cumulative “dosage” of deterrence delivered, reducing the chance that any underlying effect will be large enough to rise above normal fluctuations in crime data. In such circumstances, even a genuinely beneficial intervention can appear null in statistical testing simply because the observation window is too short for meaningful patterns to stabilise. Longer, sustained interventions, employing stronger evidence thresholds (e.g., p < .05) would not only increase the volume of treatment but also provide a stronger basis for detecting and attributing measurable changes in crime outcomes, and should also be examined over more typical, non-pandemic timeframes.
Finally, while patrols on buses may reduce crime in targeted areas, offenders could shift activities to nearby streets. Prior research suggests displacement can happen, though less than commonly assumed (Johnson et al., 2014), with possible diffusion of benefits to surrounding areas instead (Clarke & Weisburd, 1994). The problem, however, is that we do not have a viable theory of displacement from moving hotspots (i.e., where will the offenders go next and why?).
Conclusion
This study offers valuable insights into the presence of moving crime hotspots. It demonstrates the effectiveness of high-visibility police patrols in reducing crime volume and crime harm scores on ‘hot bus corridors’ in London. The estimates suggest that, albeit based on a relatively short-term test and only partially delivered as planned, the intervention leads to noticeable crime reduction benefits.
Footnotes
Funding
The authors received no financial support for the research, authorship, and/or publication of this article.
Declaration of Conflicting Interests
The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.