Definition and Measurement of Violence in the Crime Survey for England and Wales: Implications for the Amount and Gendering of Violence

Abstract

The definitions and the methodology used in surveys to measure violence have implications for its estimated volume and gendered distribution. The Crime Survey for England and Wales (CSEW) uses quite a narrow definition of ‘violence against the person’ which excludes crimes which are arguably violent in nature. This article expands the CSEW’s measurement of violence by regarding threats, robbery, sexual violence and mixed violence/property crimes as violence. This results in a shift in the gender distribution of violence, with a higher proportion of violence against women (from 39% to 58%) and by domestic perpetrators (from 29% to 32%). Impacts of violence (injuries and emotional harm) are also affected by the change in definition and disproportionally so for women.

INTRODUCTION

The measurement and reporting of interpersonal violence are contested issues. There are varying definitions within research, policy and practice. These range from definitions restricted to physical violence (Walby et al. 2017) to broader definitions that include non-physical threats and coercion (Kelly 1988; Stark 2007). National data on violence is used in the formation of policies to prevent and respond to violence, thus the accuracy of its measurement and reporting are important. The tools and methods used to collect and report on violence vary by jurisdiction. In England and Wales, the Office for National Statistics (ONS) refers to two sources of data to report on violent crime: a national victimization survey—the Crime Survey for England and Wales (CSEW) and police-recorded crime, with the former regarded as the most accurate national estimate of violent crime.

Victimization surveys are considered to be the gold standard in measuring crime (Tilley and Tseloni 2016) and statistics from the CSEW have been designated as National Statistics by the UK Statistics Authority (ONS 2019). The CSEW (formerly known as the British Crime Survey) started in 1982, running every 2 years until 2001, when it became annual. Crime is defined by CSEW crime categories, designed to align as closely as possible with police-recorded crime. The units of measurement are victims as well as crimes. New questions have been added over time, to account for more types and new forms of crime, and some questions have been expanded to allow more detail in respondent answers, though the original questions and crime categories remain largely the same. This makes the CSEW ‘the most reliable source for populating indicators on the scale of different forms of violence, for cross-national comparisons and change over time’ (Walby et al. 2017: 126). Additionally, in general, definitions of crime categories do not change over time.

In contrast, police-recorded crime is widely regarded to underestimate violent crime, as much violence, particularly domestic and sexual violence, goes unreported to law enforcement agencies (Walby and Allen 2004; Felson and Paré 2005). Though police-recorded crime provides a good measure of police activity, they are not a reliable measure of all crime (Mayhew and Van Dijk 2012; ONS 2019). For example, using the 2019/20 CSEW (Table D10 in ONS 2021b), only 49 per cent of violent crimes came to the attention of the police in 2019. Even when crime is reported, it may be screened out by police recording practices (HMIC 2014; Myhill and Johnson 2016). Inconsistency in police recording practices led to police-recorded crime data being undesignated as a National Statistic in 2014 (ONS 2020). Despite the underreporting and screening of police-recorded crime, police-recorded crime is important for capturing information on offences not included in the CSEW, such as homicide, and for crime against populations that are excluded by the CSEW, including those who are homeless, staying in shelters/refuges, student accommodation, hospitals and prisons (Walby and Myhill 2001; Ariel and Bland 2019). Statistics on violence in other countries faces similar issues. For example, in the United States, police crime reporting has historically been carried out through the Uniform Crime Reporting system using the summary reporting system (SRS) which is now being replaced by the National Incident-Based Reporting System (NIBRS) (FBI 2021). However, it is still based on law enforcement reports and suffers from the same issues of underreporting as police-recorded crime in England and Wales.

The methods used by victimization surveys to report on their data impact the accuracy of the reported estimates of violence. In England and Wales, prior research has shown reporting methods used by the ONS underestimate violence, and disproportionately affect estimates of violence against women and domestic violence (Walby et al. 2014; 2016; Cooper and Obolenskaya 2021; Pullerits and Phoenix 2024). Furthermore, there are key differences in reporting methods and data collection used by victimization surveys across jurisdictions, hampering international comparability (Lynch 2006). These include (but are not limited to) question wording, sampling methods, mode of data collection and post-processing of data (such as weights and count capping). Furthermore, the definition of violence and the methodology used to measure it have implications for the amount of violence and its gendered distribution.

In this article, we adjust the reporting methodology of interpersonal violence in the CSEW used by the ONS to more closely reflect those of another large-scale victimization survey—the National Crime Victimisation Survey (NCVS) and to conceptualize violence broader, in line with the World Health Organisation (WHO) definition of violence (Krug et al. 2002). This has two main benefits. First, the definition is broader than that used by the ONS in the CSEW giving a more comprehensive view of the burden of violence on society. Second, reflecting on the methodological practices used to record offences in the NCVS, which prioritizes personal crimes over property offences, means that more violence can be captured (see also Pullerits and Phoenix 2024). The crimes which we consider violence under discussion here are assault, attempted assault, threat, robbery, sexual assault and mixed violence/property events.

We investigate the implications of using an expanded definition of violence on the subsequent estimates of interpersonal violence and its distribution. Using CSEW data, the article investigates the implications of components of an expanded definition for the number of incidents of violence and for the profile of violence, in particular the proportion of incidents of violence committed towards women, and the proportion of incidents that are committed by domestic relations (rather than acquaintances or strangers). Lastly, we investigate how the expanded definition of violence shows an increased health burden of violence by investigating the number of injuries and the number and proportion of victims that are strongly emotionally impacted. This entails treating the crimes of threat, sexual violence and robbery as violence, which are not currently included in the ONS definition of violence in the CSEW.

This article proceeds as follows. In the next section, we outline and discuss various definitions of violence and specify our preferred measure. We then outline our methodology. The results section shows the analyses of violence from the CSEW when applying a broader definition of violence. We then compare estimated incidents and violence rates from CSEW based on the ONS and expanded definitions. The article concludes with a discussion of the results and suggested implications for policy and reporting of violence in official statistics.

INTERPERSONAL VIOLENCE: DEFINITIONS, MEASUREMENT AND PRIORITIZATION SCHEMES

Definitions of violence

The debate over what constitutes violence is ongoing and multifaceted. Definitions of violence are contested and vary not only across fields, but even within them, with recent calls highlighting the need for a shared conceptualization (Blom et al. 2023). Within the field of criminology, definitions of interpersonal violence range from those constrained to physical violence and threats only (e.g. Walby et al. 2017) to broader definitions which include coercion and control (Kelly 1988; Stark 2007); from definitions that are based on rigid crime categories, such as in police-recorded crimes and crime victimization surveys to those which are based on the characteristics of the act and its impact on the victim. Essentially, as Walby et al. (2017: 31) argue, ‘the definition of violence depends on the location of the boundary between violence and not violence’.

From a theoretical viewpoint, the WHO, places the act of violence and likely harm in the centre of the definition, and includes threatening behaviours and neglect, as well as physical force. The definition is the following:

The intentional use of physical force or power, threatened or actual, against […] another person […] that either results in or has a high likelihood of resulting in injury, death, psychological harm, maldevelopment, or deprivation. (Krug et al. 2002: 1084)

The WHO describes three broad types of violence: interpersonal violence, self-directed violence and collective violence (Krug et al. 2002). We focus on interpersonal violence in this study, which consists of familial and intimate partner violence and community violence (Krug et al. 2002). Self-directed violence includes suicidal behaviour and self-abuse, while collective violence concerns violence by large groups or states and is often for advancing a social, political or economic agenda (Krug et al. 2002). Walby et al. (2017) determined violence to consist of four components, an action which is intended, harmful and non-consensual, stating that forms of measurement which focus on only one component are partial and should be rejected. Uncompleted actions are also discussed. Walby et al. (2017) recommend that threats of violence, aiding, abetting and other uncompleted acts of violence should be treated as violence following the United Nations Office on Drugs and Crime recommendations (UNODC 2015), but labelled as uncompleted. They also considered sexual crime to be violence, but there is no consideration of robbery.

Hamby (2017), taking a psychological viewpoint, puts forward a different list of components necessary to define violence, requiring violence to be intentional, unwanted, non-essential and harmful. ‘Intentional’ and ‘harmful’ elements match with Walby et al.’s (2017) intent and harm; Hamby’s ‘unwanted’ matches with Walby et al.’s ‘consent’. ‘Action’ is taken by Hamby as an integral part of violence, as she distinguishes violence from other forms of action which are non-violent. The only difference between Walby et al.’s (2017) and Hamby’s (2017) definitions is the criteria of being ‘non-essential’ in the latter. An example of this is self-defence, which is considered as essential and therefore not violent. Essentiality might from a criminological point of view distinguish criminal violence from non-criminal violence, as reasonable self-defence will not be prosecuted. Hamby refers to power as one of several possible intentions and therefore not an essential component of violence. Thus, there is considerable agreement between Walby et al.’s definition and Hamby’s.

In this article, we aim to be as inclusive in defining interpersonal violence as possible on conceptual and theoretical bases. We therefore consider assault, robbery, threats and rape and other sexual offences to be violent acts, broadly in line with the definition by the WHO. Due to data limitations, we are not able to capture all non-physical components of coercive control often argued to be forms of violence, nor murder, manslaughter and violence against children, which cannot be identified in an adult population victimization survey. The next section discusses the national reporting of violence based on victimization surveys, namely CSEW and NCVS, which determine the extent of violence that is visible in the population. We then discuss how we could empirically redefine violence from that used in the national reporting in England and Wales, using a broader definition of interpersonal violence outlined above.

Reporting of violence in national victimization surveys

Most victimization surveys use a similar structure and method of collecting information on experiences of crime, by asking a representative sample of the population of their experience of crime over the previous 12 months; this is true of the CSEW as well as other surveys, such as those in Scotland, Northern Ireland, New Zealand, Belgium, Denmark, Italy and Sweden (Mayhew and Van Dijk 2012). The NCVS asks about victimizations in the past 6 months. These victimization surveys are often used in the national reporting on crime in general and violent crime in particular. While the definitions of what constitutes violent crime extracted from these surveys differ across countries, their utilization to define violence can also be characterized as the ‘exemplars approach’, whereby operationalized violence consists of a list of specific criminal offences (Hamby 2017). In this section, we discuss the limitation of the current ‘all violence’ measure used by the ONS in England and Wales and suggest how it can be broadened to better reflect the amount of violence experienced by the population.

The ONS definition of violence means that the reporting of violence from the CSEW is restricted to a set of specific offence codes which are deemed to cross a criminal threshold, and which aim to align with the Home Office Counting Rules (HOCR) for police recorded crime in England and Wales (Home Office 2023). This definition includes physical acts of violence and attempted physical acts: serious wounding, other wounding, common assault and attempted assault. It also includes serious wounding and other wounding where there was a sexual motive. There are, however, offences which are violent in nature but are not part of the official ONS measure of violence.

Despite the name ‘all violence’, the ONS measure, firstly, does not contain all components of ‘violence against the person’ as categorized in the HOCR (such as homicides or death by dangerous driving) due to the scope of a victimization survey and, secondly, it omits certain acts which are violent in nature, such as rape and attempted rape, which are also outside the ‘violence against the person’ definition of the HOCR (Walby et al. 2014). This is also partly because of the face-to-face survey design of the CSEW which means that offences such as rape are underreported in the main face-to-face questionnaire, therefore estimates for these offences from the main questionnaire are deemed inaccurate and publishing these numbers could be misleading (ONS 2021c). This has been critiqued by Walby et al. (2014; 2016) and Cooper and Obolenskaya (2021), who argued that sexual violence should be included, at least within the overall measure of violence, on a conceptual basis.

Secondly, although robbery was included in the ONS violence definition until 2013, it was removed as part of changes to the classification of CSEW offences (implemented in July 2014). Robbery is defined in the CSEW as an incident of theft where ‘force or threats were used to further the theft; more force than was necessary to snatch property away’ (Kantar 2021: 25). This is distinct from burglary or snatch theft, for example, the definition of which does not presume force or threat used on an individual. ONS removed robbery from the violence category due to the distinct nature of the offence which could justify its inclusion as either a violent-type offence or a theft-type offence (ONS 2012; 2014). Given that for some robberies, the primary purpose for committing a robbery is to steal, it could be argued that robbery should be classed as a theft-type offence (ONS 2014). However, due to the use (or threat) of force in robberies, we classify them as a violence-type offence. To capture a broader range of violent offences, and to be more aligned with the WHO definition of violence, it is imperative that we include all offences where physical violence, force or threats occur.

The final omission from the ONS reporting on violence is threats. Threats could take on different meanings and impact victims differently depending on the context in which they are experienced. They could be a one-off event, deployed by a stranger, for example, or a partner in a context of domestic abuse. Threats could also be part of the tactics of coercive behaviour employed by intimate partners, and could occur repeatedly (Graham-Kevan and Archer 2003; Johnson 2008). The ability to exert control over their victim by using threats was found to be a crucial component of coercive control identified in the literature (Hamberger et al. 2017; Johnson and Kelly 2008). In the absence of physical violence, threats have been found to have similar consequences for victims of violence by intimate partners, such as fear and perceived risk of future violence (Lammers et al. 2005; Crossman et al. 2015). However, coercion is a complex phenomenon that is difficult to measure within a survey. Myhill (2015) attempted to measure coercive control in the CSEW and concluded that while some indicators when occurring together (fear, emotional harm and denigration) can be used as proxies for coercive control, more comprehensive measures are needed in victimization surveys that consider a broader variety of controlling behaviours. More recently, Hester et al. (2023) have redesigned the self-completion module on domestic abuse, sexual assault and stalking within the CSEW to better reflect lived experience, coercive control and its impact on victim–survivors. This new module is currently in its testing stages. Including threats in the overall measure of violence, therefore, partly captures coercion. Given the significance of threats, especially in the context of domestic abuse, and in alignment with the aforementioned WHO definition of violence which includes threats of physical force or power (as well as Walby et al. 2017 and Hamby 2017), we incorporate offences of threat (to kill/assault/sexual threats/other threats or intimidation) in our broad measure of violence.

Violence reported by the Bureau of Justice Statistics (BJS) in the United States is also based on a crime-led definition, using data from the NCVS. It consists of aggravated assault, simple assault, rape and sexual assaults and robbery (BJS 2017). These offence categories include attempted acts and threats (BJS 2017). The NCVS follows this definition of violence for data collection and its reporting. Although murder is also included in the more general definition of violence by the BJS, it cannot be measured in a victimization survey and is therefore excluded from the measure based on NCVS data. This definition of violence is far broader than the ONS measure and concurs more closely with the theoretical work of Walby et al. (2017) and Hamby (2017), as well as the WHO definition.

It is important to acknowledge that the CSEW captures threats in a different way to the NCVS. The CSEW includes questions on threats made against the respondent, but not necessarily to the respondent. Threats in the CSEW are also inclusive of threats made to damage property and threats to publish personal information online (Kantar Public 2020). This means that the CSEW can capture a wider range of threats compared to the NCVS, which only counts verbal threats made to the respondent such as threats of death, rape, sexual assault, robbery and assault (BJS 2017). We include in our analysis below all threats, which are likely to result in fear of violence, which is inclusive of threats to kill/assault, sexual threats, other threats or intimidation and threats against others, as well as threats made during another event.

Given the different purposes of recording violence across policy and service areas, victimization surveys should be able to capture a broader experience of interpersonal violence in the population than is currently reflected in national reporting. At the same time, victimization surveys should also allow researchers to constrain the measure of violence for their purpose, such as by harms from violence (with injury/without; with mental health impact/without), by intentionality of it (intentional/accident, etc.) and/or by type of violence (physical/non-physical/sexual, etc.).

Crime prioritization schemes and prevalence of violence in national surveys

Research by Pullerits and Phoenix (2024) identified the impact of priority coding practices in victimization surveys on estimates of violence, showing that the prioritization of some property offences over some violence offences leads to underestimations of physical violence, with disproportionate impacts on violence against women and domestic violence. In the CSEW, a single offence code is assigned to a criminal event in complex instances where multiple offences occur during that event. Certain offences have priority over others, which results in some violent offences being categorized under non-violent offence codes. For example, Table 1 highlights the hierarchy of offences used in the CSEW since 1982. It places both burglary and criminal damage above assault (specifically, above both other wounding and common assault), which means that if the offender maliciously damages or unlawfully (even partially) enters property with intent to commit theft, grievous bodily harm or criminal damage at the same time as assaulting the victim, it is likely that this is coded as criminal damage or burglary (unless the assault amounts to a serious wounding) (ONS 2021c). Given this hierarchy, which in some cases means that violent offences are categorized as non-violent, relying on the offence coding alone can potentially result in underestimation of the amount of violence experienced in the population. This article, therefore, builds on and further extends the crime code hierarchy work undertaken by Pullerits and Phoenix (2024) by reclassifying crimes coded as non-violence (but which have an element of violence in them) as violence, and also examining co-occurrence of threats with other offences. The prioritization scheme used in the NCVS is rather different and prioritizes rape, robbery, assaults (including attempts and threats) above burglary and theft. We prefer the NCVS prioritization scheme for identifying violence over that used in the CSEW as it prioritizes acts of violence over property crime, and we seek to apply it to the CSEW.

Table 1.

Prioritization of offences in descending order used in the CSEW and the NCVS when there are multiple crimes within a crime incident

Rank	CSEW (England and Wales)	NCVS (USA)
1	Arson	Rape
2	Rape or serious wounding	Sexual assault
3	Robbery	Robbery
4	Burglary	Aggravated assault (including threat with weapon)
5	Theft	Simple assault (including threat without weapon)
6	Criminal damage	Burglary
7	Assault (includes other wounding and common assault)	Theft
8	Threats

Source: Bureau of Justice (2017) and Kantar Public (2021).

To conclude from the above discussions, assault, robbery, threats, rape and other sexual offences could be considered to be violent acts. We aim to treat them as such in this article, based both on the conceptual/theoretical basis and the definition of violence used by the US BJS and the definition of violence by the WHO. Further, we build on and extend the crime code hierarchy work undertaken by Pullerits and Phoenix (2024), by reclassifying crimes coded as non-violence, but which have an element of violence in them (either force or threats of force) into our violence measure. Furthermore, we investigate how some dual character offences can be recategorized as violence. In the results section, we explore the implications of these recategorizations for the profile of victims, specifically by gender and relationship to the perpetrator. Given the importance of the concept of harm in the aforementioned definitions of violence, we also examine the implications of the recategorizations on extent of harm experienced by victims.

METHOD

Data

We use the Crime Survey of England and Wales, a nationally representative, household victimization survey which collects data on around 35,000 respondents per year (ONS 2019). This article uses 1 year of data from the 2019/20 survey year (ONS 2021a), which was the most recent data available at the time of analysis.

The CSEW consists of multiple modules which record respondent demographic information and victimization experiences. This article focuses on the main victimization module which includes adult respondents (aged 16 and over) who report experiencing victimization in the previous 12 months. Detailed information is collected about each reported event through a face-to-face interview, which ONS subsequently use to categorize the events into offence codes. Whether the events pass the criminal threshold is decided after the interview by specialist-trained coders. Respondents can record up to six discrete criminal victimization incidents in separate victimization forms (also called ‘victim forms’).

It is important to note that there are some key populations and crimes which are not captured by the survey. As previously mentioned, as a victimization survey, the survey asks the victim for their own account of the experience, therefore homicides cannot be included. Additionally, as the survey samples households, populations living in group residences (e.g. care homes, hospitals, or student halls of residence) or other institutions are not captured by the survey. Walby and Myhill (2001) explain how this exclusion is particularly significant when considering violence against women as victims of domestic or sexual violence are more likely than the general population to be in temporary accommodation (refuges, hostels, temporarily living with family/friends, etc.). Westmarland and Bows (2018) also argue that excluding people in care homes disproportionally affects the inclusion of older people in the survey.

The 2019/20 CSEW reached 33,735 respondents, with 6,900 recorded as a victim of at least one crime. The survey contained 9,200 victim forms, and over 13,000 incidents (as one victim form may contain information about multiple repeated incidents). When weighting to the population, this gave an estimated 10.2 million victims, 13.8 million victim forms and 19 million incidents. This sample is broadly representative of households in England and Wales.

Measurement of violence

In this article, we expand the definition of violence used by ONS (shaded grey in Table 2) in two ways: by including additional offences in our measure of violence and by supplementing offence code classification altogether and use an indicator of force used and/or threatened.

Table 2.

Violent offence codes and type of offence, with the ONS classification of violence shaded grey

To achieve the first, we have derived three new measures of violence:

• Measure 1: Physical violence (including attempted, sexual violence and robbery): ONS published measure of violence (codes 11, 12, 13, 21, 32 and 33) expanded to include robbery and sexual violence (codes 31, 34, 35, 41 and 42).

• Measure 2: Physical violence and threats (including attempted and sexual violence, robbery): ONS published measure of violence expanded to include threats (codes 91, 92, 93 and 94) as well as robbery and sexual violence (above).

• Measure 3: Physical violence and threats, also prioritizing violent crime over property crime and following the BJS prioritization scheme (Pullerits and Phoenix 2024).

Measure 3 includes secondary offences which involve a use of force or threat of force, even if the event is recorded as a non-violent offence due to priority offence coding practices. This analysis thus aims to identify violence that has been ignored by priority offence coding practices. This is achieved by using a flag for experiencing force to identify incidents of violence. We use questions within the survey which record whether ‘force’ was either used (following Pullerits and Phoenix 2024) or threatened during the event (to expand on Pullerits and Phoenix 2024). The question on the use of force is asked twice, once as initial screening and then in the victimization form. ONS have then derived a variable which indicates whether the event involved the use of force or a threat of force to the victim.

Repeat victimizations

Each victim form can record either a ‘single’ or a ‘series’ event. A single event is a one-off occurrence, but a series event is where ‘the same thing, is done under the same circumstances and probably by the same people’ (ONS 2021c). When publishing on incident numbers, the ONS cap the maximum number of incidents at the 98th percentile (Flatley 2017). For the ONS measure of violence, this is around 10 incidents, 5 for sexual offences, 5 for robbery, and 9 for threats (Kantar Public 2021). For the first analysis, both capped and uncapped incident counts are presented to show the full effect of adjusting the violence measure. The second analysis uses uncapped incident counts, to ensure that the gendered distribution of violence is not underestimated (as suggested by Walby et al. 2016).

Type of perpetrator

This article is interested in the gender dimension of violence. Violence by domestic relations is considered to be gendered violence (Stark 2007; Kelly and Johnson 2008; Myhill 2015). Therefore, we disaggregate by victim–perpetrator relationship, categorized into three groups: stranger, acquaintance and domestic. Stranger violence includes incidents of violence where the victim does not have any information about the perpetrator(s). In this case, the victim does not know or has never seen the perpetrator before the incident took place. Acquaintance violence is defined as incidents of violence against the person where the victim knows the perpetrator (or at least one of the perpetrators). This covers any person the victim knows by sight and includes but is not limited to; colleagues and people met through work, neighbours and friends of family members. Domestic violence would include violence against the person that involves partners, ex-partners or any family or household member.

Sex of the victim and perpetrators

Victimization events recorded in the victim form can have one perpetrator or multiple perpetrators. The sex of the victim is recorded as binary: male or female. The sex of the perpetrator can be recorded as male or female, meaning that the single perpetrator or all the multiple perpetrators were the same sex. The ONS-derived sex of the perpetrator variable also includes the category ‘both sexes’. This category includes victim forms where there were multiple perpetrators but some of the perpetrators were male and some were female.

We aim as far as possible to capture all violence experienced in the population as reported in 2019/20 in the CSEW. We used data from the CSEW main (face-to-face) questionnaire, which is a valuable source of data on physical violence and threats. However, certain constraints of the data collected restricted our ability to include all forms of violence within our measure. Non-physical abuse and coercion, for example, are not captured by the face-to-face part of the survey and are therefore excluded from our broad measure of violence. Cooper and Obolenskaya (2021) show an increased prevalence of violence, particularly among women, when data from the CSEW self-completion module on domestic abuse and sexual violence is incorporated within face-to-face victimization data. The self-completion module has a higher disclosure rate (Walby et al. 2014; Cooper and Obolenskaya 2021), it is not designed to capture ‘crime’ and it is therefore broader in scope, and data from the module are not affected by prioritization methods. Our estimates of violence against women and domestic violence, in particular, will therefore likely be underestimates.

Indicators of harm

We used two indicators of harm from the CSEW. The first measures whether physical injury occurred as a result of the event (yes/no). This is asked of any respondent who discloses that the perpetrator used force against them. Respondents who did not experience force were coded as having no physical injury.

The second indicator of harm is whether the respondent experienced emotional reactions from the event, and if so to what extent they were affected. Examples of an emotional reaction provided in the questionnaire include anger, shock, fear, depression, anxiety/panic attacks, loss of confidence, difficulty sleeping, crying, annoyance and/or other. Next respondents are asked ‘how affected were you by the incident?’, which includes ‘a little’, ‘quite a lot’ and ‘very much’ as responses. We categorized this into two groups; those who had no emotional reaction or ‘a little’ affect (0) and those respondents who were quite a lot or very much affected (1), the latter from here on referred to as strongly emotionally affected. In additional analyses (available upon request), we differentiated internalized from externalized emotional reactions (based on Iganski and Lagou 2015). Previous analysis of indicators of harm have included emotional reactions from the CSEW (Blom et al. 2023) and both physical and emotional impacts (Blom et al. 2023). Both physical injury and emotional reaction are recorded at the level of the victim form and could relate to multiple incidents of violence if the victim form relates to a series of event.

ANALYSIS

In this article, we estimate incidents of violence using our expanded measures of violence and compare them to estimates using the ONS official measure of violence. While violence can also be measured in terms of the number of victims (prevalence of violence), in this work we focus on the incidence of violence given the importance of repetition to measure gender differences in violence exposure discussed earlier.

The analysis in this article is divided into two parts. We first estimate the number and rate of incidents of violence in England and Wales in 2019/20 using the ONS measure of violence alongside our measures. We then estimate the number of incidents disaggregated by victim and perpetrator gender, victim–perpetrator relationship (type of perpetrator) and extent of harm. The analysis for this article was undertaken using SPSS. All estimates of incidents are weighted to the general population using individual weights provided by the ONS. The unweighted frequencies are provided for all analyses.

FINDINGS

Incidence rates when expanding the measure of violence

We examine reported violence from the ONS and compare to our new violence measures. A published ONS report shows that in 2019/20 (9 months of 2019 and 3 months of 2020) there was an estimated 1,239,000 incidents of violence to victims aged 16 and over who are residents in households, and an estimated incident rate of 26 violent crimes per 1,000 persons (ONS 2020, Appendix Tables A1 and A2; Table 3).

Table 3.

Estimated number of incidents of violence comparing the ONS measure with our new measures of violence

	Definition of violence
	ONS measure	ONS + sexual offences + robbery (Measure 1)	ONS + sexual offences + robbery + threats (Measure 2)	ONS + sexual offences + robbery + threats + force used/threatened (Measure 3)
Uncapped estimates
Number of incidents 2019/20 (unweighted)	1.41m (891)	1.82m (1,150)	4.91m (3,559)	5.40m (3,912)
Percentage of all incidents 2019/20 (unweighted)	7.5 (6.8)	9.6 (8.8)	26.0 (25.8)	28.5 (28.4)
Capped estimates (98th percentile)
Number of incidents 2019/20 (unweighted)	1.24m (757)	1.54m (921)	3.39m (2,281)	3.70m (2,508)
Percentage of all incidents 2019/20 (unweighted)	6.5 (5.8)	8.1 (7.0)	17.8 (17.3)	19.5 (19.1)

Our analysis of the CSEW shows that using a wider definition of violence, which includes sexual offences, robbery, threats and force, there are estimated 3,704,184 violent incidents in England and Wales (Table 3) for an estimated mid-year 2019 population of 48,059,326 aged 16 and over (ONS 2019), a rate of 77 incidents per 1,000 population. This rate is three times as high as the estimated rate for the ONS measure of violence.

As well as the estimated rate of violence increasing, the expanding definitions of violence showed that a much higher proportion of criminal incidents were violent incidents in the CSEW for 2019/20. The ONS estimates show that 5.8 per cent (capped) to 6.8 per cent (uncapped) of all criminal incidents were violence. Measure 1 shows that by including all sexual offences and robbery as violence this increases to 8.8 per cent (uncapped). Including threats as non-physical violence results in the largest increase in the number of violent incidents from 1.82m (Measure 1, uncapped) to 4.91m (Measure 2, uncapped) and over one-quarter of all criminal incidents reported to the survey that year. Our final measure of violence (Measure 3) identifies 28.5 per cent of all criminal incidents from 2019/20 as violence.

Gender inequalities when expanding the measure of violence

We now turn our attention to the effect that using the expanded measures of violence has on the CSEW estimates and the distribution of violence by sex of the victim and victim–perpetrator relationship. To do this, we use the uncapped values (totals presented in Table 3).

Our analysis of incidents using the ONS measure disaggregated by sex shows more violent incidents experienced by men than women (Figure 1). In terms of the total share of incidents by sex, using the ONS measure, there is a higher proportion of incidents experienced by male victims (61%) than female victims (39%).

Fig. 1

Number of incidents of violence by sex of victim, 2019/20. Source: Authors’ analysis of CSEW (ONS 2021a). Notes: Total unweighted count of incidents of violence is 13,117. The number of incidents used here is uncapped. Measure 1: ONS + sexual offences + robbery; Measure 2: ONS + sexual offences + robbery + threats; Measure 3: ONS + sexual offences + robbery + threats + force used/threatened

Comparing this to Measure 1, with the additional inclusion of sexual offences and robbery, the gap in the number of violent incidents between male and female victims decreases. For Measure 1, there are almost as many violent incidents against women (49%) as men (51%). However, for both Measures 2 and 3, women experience more violent incidents than men. For Measures 2 and 3, women experience 58 per cent of violent incidents compared to 42 per cent for men.

Overall, using the official ONS measure of violence we find that more violence is experienced by men than women. However, as we have broadened the measure of violence to include additional offences (robbery, sexual offences and threats), we find that more violence is experienced by women than men. Additionally, compared to the ONS measure, estimates of violence using Measure 3 are up to 2.8 times higher for men and up to 5.2 times higher for women.

Disaggregating by victim–perpetrator relationship and sex of the victim (Figure 2), shows the implications of changing the measurement of violence on the understanding of who perpetrates violence. For the ONS measure, strangers and acquaintances perpetrated a similar number of violent incidents against men (46%/48%), whereas violence by domestic perpetrators accounted for just 6 per cent of violent incidents. The proportions of incidents remain similar across all measures, for example, Measure 2, shows that male victims still experience the most violence by strangers (50%), and are not often victimized by domestic relations (5%). However, while the proportion of violence against men perpetrated by different perpetrators stays similar for each measure, the overall number of incidents significantly increases.

Fig. 2

Number of incidents of violence by sex of victim and victim–perpetrator relationship, 2019/20. Source: Authors’ analysis of CSEW (ONS 2021a). Notes: Total unweighted count of incidents of violence is 13,117. The number of incidents used here is uncapped. Measure 1: ONS + sexual offences + robbery; Measure 2: ONS + sexual offences + robbery + threats; Measure 3: ONS + sexual offences + robbery + threats + force used/threatened

For female victims, the ONS measure shows that female victims are more commonly victimized by acquaintances (48%), followed by domestic relations (29%). For Measure 1, acquaintance violence accounts for 45 per cent of violent incidents towards women, domestic violence makes up 27 per cent of incidents and stranger violence makes up 29 per cent. Including threats (Measure 2) shows that 43 per cent of incidents against women are by acquaintances, the proportion of incidents by domestic relations increases to 32 per cent and stranger violence makes up 25 per cent of incidents. This is similar for Measure 3.

For women, using the broadest definition of violence (Measure 3) shows that stranger violence increased by as much as five times when compared with the amount estimated using the ONS measure, the amount of violence perpetrated by acquaintances increased by 4.4 times and domestic violence is 5.5 times higher.

Finally, Figure 3 shows the proportions of violent incidents by sex of the victim and sex of the perpetrator(s). Using different measures of violence increases the amount of violence experienced but does not change the patterns of who perpetrates the most violence. Both male and female victims are most likely to be victimized by male perpetrators across all four measures of violence. Comparing the ONS measure to Measure 2, we can see that there are twice as many incidents by male perpetrators against male victims and three times as many incidents by male perpetrators against female victims. For Measure 3, there are 4.7 times as many male-perpetrated incidents against women and twice as many against male victims.

Fig. 3

Number of incidents of violence by sex of victim and perpetrator, 2019/20. Source: Authors analysis of CSEW (ONS 2021a). Notes: Total unweighted count of incidents of violence is 13,117. The number of incidents used here is uncapped. Measure 1: ONS + sexual offences + robbery; Measure 2: ONS + sexual offences + robbery + threats; Measure 3: ONS + sexual offences + robbery + threats + force used/threatened

Interestingly, using Measure 3, our analysis shows that there are around six times as many female-perpetrated violent incidents against female victims as using the ONS measure and only 1.2 times as many incidents perpetrated by women against men. Violence perpetrated by perpetrators in the ‘both sexes’ category was eight times higher for Measure 3 than the ONS measure for both male and female victims.

Injuries and emotional impact when expanding the measure of violence

Around half of ONS violent crime offences (ONS definition) result in physical injury to the respondent (this is estimated to be around 400,000 injuries for 2019/20, shown in Table 4). Using our violence Measure 1, which includes sexual offences and robberies, the proportion of injuries decreases to 40 per cent However, the number of violent incidents that led to injuries increased by 13 per cent to 465,000 injuries. Adding threats to the measure of violence (Measure 3), did not, unsurprisingly, pick up more violence-related injuries, and the proportion of violent incidents resulting in injury decreased to 20 per cent using this measure. Crucially, the number of violent incidents that led to injuries increased to around 530,000 when we included whether force was used/threatened. Hence, our most expanded measurement of violence led to an increase by 29 per cent in the number of injuries compared to the ONS measure. Notably, the number of injuries experienced by women rose more than by men, by 44 per cent and 19 per cent, respectively.

Table 4.

Estimated number of injuries and emotional reactions comparing the ONS measure with our new measures of violence, disaggregated by sex of the victim

Definition	Full sample		Male		Female
	Injury	Strong emotional reaction	Injury	Strong emotional reaction	Injury	Strong emotional reaction
ONS measure	411,271	359,112	248,029	169,230	163,242	189,882
	47.10%	41.10%	47.20%	32.20%	46.80%	54.40%
ONS + sexual offences + robbery (Measure 1)	464,693	511,516	258,767	209,849	205,926	301,668
	40%	44%	41.10%	33.30%	38.60%	56.60%
ONS + sexual offences + robbery + threats (Measure 2)	465,641	1,122,771	258,767	420,281	206,874	702,489
	19.10%	46%	21.80%	35.30%	16.60%	56.20%
ONS + sexual offences + robbery + threats + force used/threatened (Measure 3)	530,486	1,302,831	294,974	476,502	235,512	826,329
	19.70%	48.50%	22.80%	36.90%	16.90%	59.20%

Harms are recorded in the victimization module and are counted at the event level rather than incident level. Therefore, totals will be smaller than Table 3.

Table 4 shows that using the ONS measure of violence there are 359,000 instances of strong emotional impact as a result of violence, which amounts to 41 per cent of all violent incidents. This number increases to 511,000 when sexual offences and robberies are included in the measure. It more than doubles when threats are then added in—to 1,122,000, with a further increase to 1,302,000 when all crimes where force was used or threatened are considered violence. As a result, 49 per cent of all violent instances are associated with strong emotional impact. This is an increase of over 260 per cent from the original ONS measure, with strong gendered effects—a rise of 182 per cent for men and 335 per cent for women.¹

DISCUSSION

There are varying measures of interpersonal violence within research, policy and practice. Surveys are considered a more reliable source of information on crime, and violence, than other sources (such as police data). And, while many victimization surveys include relatively similar questions when asking respondents about victimization experiences, there are some key differences in the measurement and reporting of violence within them. This article demonstrates the impact of expanding the measure of violence by aiming to align it more with the WHO definition of violence and using the measurements of another victimization survey (the NCVS). Specifically, the article reveals the impact that this has on the gendered distribution of violence captured in the CSEW.

The CSEW uses a more restricted definition of violence compared to some other victimization surveys, including the NCVS (BJS 2017; ONS 2022). The measurement of violence reported by the ONS using the CSEW excludes some key offences which have a conceptual basis for being included, which could underestimate the prevalence and incidence of violence in England and Wales (Cooper and Obolenskaya 2021, Walby et al. 2016). To arrive at the broadest measure of violence (Measure 3 in the results section above), we perform the following adjustments to the ONS measure of violence using the CSEW. Firstly, in line with the BJS reporting from the NCVS and the WHO definition of interpersonal violence, we include additional offence categories to the ONS’ own measure of violent crime. Namely, we include all sexual violence, robbery and threats. Secondly, we adjust priority coding practices, by reclassifying incidents which had an element of force in them but were coded by ONS as non-violence. This was done in line with recent research by Pullerits and Phoenix (2024) which highlights the importance of critically engaging with methodological decisions which impact the measurement of violence. Specifically, here the issue of prioritizing offences leads to some violent offences not being recorded where priority is given to some property offences. We have expanded on this research, by comparing these practices and including threats into the measurement.

Our results based on the CSEW show that experience of violence in England and Wales is more prevalent than ONS reported violent crime using their current methodology. By aligning the measure of violence in the CSEW to that of the NCVS and definition of the WHO, we estimate that the total number of capped incidents of violence in England and Wales was 2.9 times higher in 2019/20 than officially reported by ONS. The uncapped difference is even higher. Crucially, not only the rate of violence is higher using our broadest measure of violence, but the distribution of violence shifts, especially violence against women. ONS published figures suggest that in 2019/20 men were more at risk of violence than women (ONS 2020). Our analysis of the ONS measure showed 830,000 violent incidents against men and 580,000 against women. However, the ONS measure of violent crime is likely to underestimate violence against women because the current definition excludes violent offences which women are more often victims of, such as rape, attempted rape and indecent assault (Cooper and Obolenskaya 2021) as well as threats of violence (Hester 2009). Indeed, our findings show that using the broadest measure of violence, the uncapped number of incidents experienced by women in 2019/20 was much higher than by men (2.8m and 2.3m, respectively).

Furthermore, our results show that as the volume of violent offences increases as does the volume of harm experienced by victims of violence. Our broadest measure of violence identifies 29 per cent more injurious violent events than the ONS measure. Additionally, the level of emotional harm is often higher for measures which include threats of violence and force. While the CSEW aims to prioritize more serious events over less serious events (ONS 2018), through both the ordering of the questions and the priority ordering of offences, this demonstrates that some potentially serious violent events can be coded as non-violence, supporting previous research (Pullerits and Phoenix 2024). The lower estimation of harmful consequences of violence could also mean that the burden of violence on healthcare provisions is also underestimated. Therefore, not only is the ONS measure showing a different gendered distribution of violence but also an underestimation of the harmful consequences of violence on victims, especially female victims. Further work on the components of emotional impact is needed but this was out of the scope of this article.

We also consider the boundary between coercion and violence with the inclusion of threats. Some definitions of violence are restricted to physical violence only (e.g. Walby et al. 2017). The ONS measurement of violence is also restricted to physically violent acts (and attempted physically violent acts) such as common assault and excludes threats. The inclusion of coercive acts within a broad measure of violence is especially important when considering gender-based violence and domestic violence, whereby tactics can be used to extend the abuse over time (Kelly 1988; Stark 2007). Coercion in the CSEW is not inclusive of all possible coercive acts. The main victimization data used in this article is limited to threats only, and a redeveloped self-completion module on domestic abuse, sexual assault and stalking to better capture controlling and coercive behaviour is being trialled by the ONS at the time of writing (Hester et al. 2023). This article demonstrates the significance of adopting a more comprehensive definition of violence. Including threats in the measurement of violence, drives the largest change in estimates of violence, increasing the estimated number of incidents by over 3m (from 1.82m to 4.91m). Additionally, counting threats as violence shifts the distribution of violence showing that women are more often victims than men and that the estimated number of incidents perpetrated by domestic partners is particularly impacted. Therefore, where the line is drawn between violence and coercion matters for who is perceived to be the victims of violence (men/women) and who is seen as the perpetrators (strangers or known perpetrators). Overall, we see that including all sexual violence and threats into the measurement of violence has implications specifically for the estimates of violence against women, violence by domestic perpetrators and violence by acquaintances.

There are other offences which could be included in the definition of interpersonal violence, for example, arson, criminal damage, environmental crimes or corporate offences. These additional offences are either captured in Measure 3, if threat or force occurred alongside (criminal damage, arson) or they are not captured in a victimization survey (environmental crimes, corporate offences). Additionally, the data does not allow identification of certain characteristics of incidents which would enable us to apply a more nuanced criteria for identification and description of violent acts. For example, we are not able to establish whether experience of force was a result of an act of crime or a result of an essential act such as self-defence. However, our aim was to establish estimates of violence using the CSEW by applying as broad definition of violence as possible. This work is the first to our knowledge to capture such a range of violent experiences using the CSEW. Self-directed and collective violence are outside the scope of this article and further research could investigate how the definition of these factors shapes our understanding of violence inequalities and harms arising from them.

CONCLUSION

In this article, we explore the different definitions of violence, by incorporating information on all forms of interpersonal violence using the CSEW. The ONS applies a more restricted definition of violent crime to their ‘all violence’ measure which, as we show, produces lower estimates of the extent of violence in England and Wales. This article demonstrates that the measure of violence in CSEW used by ONS could be expanded by including offences which are not currently included within an ‘all violence’ measure (such as sexual offences, robberies and threats), but which would fall under commonly used definitions of violence, such as by the WHO, and which are considered violence by other national victimization surveys such as the NCVS. We also emphasize important methodological decisions which mean that some violent events are currently recorded as non-violent by the ONS. It is imperative to understand how widespread violence is, due to its extensive impact on both the physical and mental well-being of victims, as well as its economic costs to society. To be able to gain a comprehensive understanding of it, it is crucial to encompass all its manifestations, including sexual violence, threats and offences where physical violence or threats of violence might not be the primary motivation behind an incident (e.g. robbery). Our estimates indicate that the number of violent incidents in 2019/20 could be up to four times higher when using expanded definitions of violence, compared to using more restricted measures. These findings carry significant implications for the provision of support services for victims, especially specialist support services for addressing violence against women.

Furthermore, the restricted ONS measure of violence presents a different distribution of violence between groups in the population. This article specifically contradicts some of the ONS findings on the distribution of violence by sex of the victim, sex of the perpetrator and victim–perpetrator relationship (ONS 2020). Walby et al. (2016) shows how capping repeated violent events impacts the gendering of violence. We show that the current narrow definition of violence used by ONS is also gendered. By including offences such as threats, which partly reflect acts of coercion, particularly against women, and including all sexual violence—again, predominantly experienced by women—increases the proportion of violence against adult women (from 39% of all violent incidents to 48%). The data used shapes the understanding of violence inequalities, therefore having a more inclusive measure shows the reality of gendered violence. It would be worth investigating in further research how the distribution of violence, defined in a broader sense, differs across other groups, such as by victim ethnicity or age.

It should be acknowledged that a narrower definition of violent crime is appropriate for ensuring consistency when looking at trends in violence over long periods of time (ONS 2022) and greater consistency with police-recorded data. While recategorization of offences may not affect the overall volume of offences captured in the CSEW (they are recorded in the survey either way, even when categorized as a non-violent event), it does affect the volume and understanding of violence. The data and information recorded in the CSEW influences the understanding of violence inequalities in the UK. By broadening the definition of violence, we show that the chosen measure has implications for understandings who the victims and perpetrators of violence are and the gendered dynamics of violence. We would therefore recommend that as well as continuing to publish estimates of prevalence and incidents of ‘violence against the person’, which broadly align with the definition of violence used for police-recorded crime, the ONS should also publish estimates for the broadest violence measure using the definition which we propose in this article. This would allow for a fuller coverage of experiences of violence in the population by using a measure which includes more violent offences, and which is less affected by the priority ordering.

ACKNOWLEDGEMENTS

We thank the anonymous peer reviewers for their helpful comments. The authors also would like to thank Hannah Manzur for her input on discussions about this work and attendees of the 22nd Annual European Society of Criminology conference where feedback was provided on presentations of early findings.

FUNDING

This work was supported by the UK Prevention Research Partnership [Violence, Health and Society; grant number: MR-VO49879/1], a Consortium funded by the British Heart Foundation, Chief Scientist Office of the Scottish Government Health and Social Care Directorates, Engineering and Physical Sciences Research Council, Economic and Social Research Council, Health and Social Care Research and Development Division (Welsh Government), Medical Research Council, National Institute for Health and Care Research, Natural Environment Research Council, Public Health Agency (Northern Ireland), The Health Foundation, and Wellcome.

Footnotes

1

Previous research has separated emotional reactions into ‘internalized’ (e.g. depression, fear, anxiety) and ‘externalized’ (e.g. annoyance and anger) responses (Iganski and Lagou 2015). Additional analyses (available upon request) showed that the proportion of internalized emotions increases as the definition of violence expands, whereas externalized emotions decrease. Women had a higher proportion of internalized emotions than men and were more affected by this increase.