Factors affecting perceived criminality: evidence from victims of assault

Foreword | Recent research shows that not all assaults described in victimisation surveys are considered to be crimes by the victims. This paper investigates this issue and puts forward findings which have implications for the role surveys play in measuring crime. Using ABS 2005 Personal Safety Survey data, it examines the extent to which surveyed incidents of assault are perceived by victims to be criminal events, aspects of incidents that predict perceptions and any existing variations by sex. Findings show that male victims under 25 years of age are less likely to perceive assault victimisations as crimes, and women were less likely to perceive an event as criminal if the perpetrator is known to them. Incident severity increased the perceived criminality of an incident, but location was predictive of perceptions of criminality for male victims only. The study points to the potential for victimisation surveys to overestimate the extent of violent crime. It is suggested that approaches for reducing violence should acknowledge the importance of victim perceptions, as the way incidents are defined by individuals has a significant bearing on whether they are reported to police and come to the attention of the criminal justice system.

Judy Putt
General Manager, Research

To date, an implicit assumption of victimisation surveys is that all assaults described are considered to be crimes, at least from the victim's perspective. However, as will be discussed in this paper, recent research evidence questions the validity of this assumption. This paper explores this issue and produces findings which have significant policy and practical implications for the role surveys should play in the measurement of crime.

Victimisation surveys have been widely utilised since the 1970s and have provided a valuable alternative source of information about the prevalence and incidence of crime. The public are questioned directly and information is gathered about victimisation events, regardless of whether such events have been reported to police. Further, surveys explore the reasons why events are, or are not, reported. Surveys therefore provide one way of investigating the 'dark figure of crime'. This is defined by Biderman and Reiss (1967: 1) as 'occurrences that by some criteria are called crime, yet that are not registered in the statistics of whatever agency was the source of data being used.'

Surveys generally find that property offences, such as completed break-ins and motor vehicle theft, are reported to police at a high rate (74% and 90% respectively; eg ABS 2006a). Surprisingly, however, violent victimisations such as robbery, assault and sexual assault are less likely to come to police notice. For example, only 32 percent of Australian assaults were reported in 1994 (ABS 1995), 28 percent in 1998 (ABS 1999) and 31 percent in both 2002 (ABS 2003) and 2005 (ABS 2006a). Assault victimisations are of particular interest because they are sufficiently numerous to be amenable to analysis and they have a large reporting gap that requires explanation.

Surveys overcome well-documented limitations of police recorded crime data. They cut across jurisdictional differences in the law and in police practices with respect to offence recording. Many criminologists assume, therefore, that surveys provide the best way of measuring the true incidence of crime (eg Matka 1990), despite certain limitations of their own (as discussed by Skogan 1981). Surveys have been particularly useful in exploring that part of the dark figure of crime arising from victim non-reporting. A number of previous studies have analysed the relationship between victimisation and reporting to police (eg Coumarelos & Allen 1999; Felson, Messner & Hoskin 1999; Kury, Teske & Wuger 1999; Skogan 1994; Tanton, Jones & Lubulwa 2001). For assault, this research has demonstrated that reporting is influenced by a range of offence characteristics that can be broadly placed into three major categories: victim characteristics, victim-offender relationship characteristics and characteristics of the incident, particularly concerning incident seriousness (eg see Carcach 1997; Skogan 1976; 1984; Wolf Harlow 1985, amongst others for a discussion of these issues). Still other researchers have identified additional factors that include past experience(s) with police and also characteristics of the communities where victims live (eg Goudriaan, Lynch & Nieuwbeerta 2004; Goudriaan, Wittebrood & Nieuwbeerta 2006).

A rarely acknowledged assumption of survey research on the dark figure of crime is that the events that victimisation surveys uncover are genuinely criminal. An important study by Ruback, Greenberg and Westcott (1984) made explicit predictions about the importance of the victim labelling an event as a crime when determining if the event would be reported to police. As these authors suggest, '[w]hether the victim labels this a crime depends on two factors: the victim's own definition of a crime and the similarity between the incident and the victim's definition' (Ruback et al 1984: 55). Interestingly, despite the potential significance of this element in accounting for the discrepancy between victimisation as captured by surveys and crime recorded by police, the victim's perception of the criminality of the incidents they describe in surveys has been largely overlooked from a criminological perspective.

There are a number of recent research findings that indicate this oversight may have significant implications for the realist perspective which is of the view that:

  • victim surveys provide the most comprehensive measurement of crime
  • all incidents captured by victim surveys can be labelled as crime.

Instead, it appears that the victim's own perception of the criminality of their experiences may be exerting a significant influence on reporting to police. For example, in an analysis of the 1996 British Crime Survey, Mirrlees-Black (1999) discovered that only 39 percent of female victims of chronic domestic assault described the most recent incident to be a crime. Furthermore, less than half (45%) of respondents who defined themselves as 'victims of domestic violence' also considered the incident to be a crime, with the remainder considering the event to be wrong, but not a crime (34%), just something that happens (15%), or unsure how to classify the event (6%). This trend was also demonstrated by the 2002 National Crime and Safety Survey (ABS 2003), where only 57 percent of assaults were considered to be crimes, and by the 2004-05 British Crime Survey (Coleman, Hird & Povey 2006), where approximately 40 percent of violent crime victimisation was not considered to be a crime by the victims. Even lower percentages of victims considered their experiences to be crimes according to the 2005 Personal Safety Survey (ABS 2006b), with victims perceiving assaults perpetrated by a male to be criminal 37 percent of the time and only 25 percent of the time if the perpetrator was female.

Therefore, considering prior research, the purpose of this study is to explore:

  • the extent to which surveyed incidents of assault are perceived by victims to be criminal events
  • the aspects of incidents that predict whether they will be perceived as assaults
  • what variation, if any, exists between male and female victims' perceptions.

This examination of sex differences was motivated by the mixed, interactive relationship between the sex of the victim and the nature of the victim-offender relationship demonstrated by previous research. Carcach (1997) demonstrated that female (but not male) victims were less likely to report assaults to police if the offender was known. Felson and Paré's (2005) analysis of the United States National Violence Against Women Survey (the equivalent to the Personal Safety Survey (PSS)), found that all victims were less likely to call the police when the offender was known, regardless of sex.

The implications of these results for policy, practice and research will be discussed.


Data specifications

The 2005 PSS sampled 16,413 individuals (11,861 females, 4,552 males). The survey collected demographic details from each respondent, as well as information about any history of childhood abuse, harassment, experiences of partner violence, stalking and violence (as discussed in ABS 2006c). The analysis focused on 1,557 victims' most recent incident of physical assault perpetrated by a male offender that had occurred within the preceding five years. (As a consequence of the way the dataset is structured it was not possible to analyse the most recent incidents of assault, regardless of the sex of the offender.) Of this sample, approximately 44 percent considered the assault incident they described in the survey to be a crime.

Choice of variables and method of analysis

Prior to undertaking the modelling, univariate tests were conducted on all selected parameters to determine if they were predictive of crime perception when considered in isolation. The parameters capturing whether the victims had been repeat victims of violence and whether the victim considered drugs had played some role in the most recent assault were subsequently excluded from the final analysis because they were found to be unrelated to the outcome variable. Standard procedures for conducting stepwise binary logistic regression were followed for all models reported below. The dependent variable in all cases was whether the victim perceived the assault captured by the survey to be a crime (coded 1=yes, 0=no). Consequently, for all parameters, an odds ratio significantly greater than 1.0 suggested that a unit increase in this factor increased the likelihood of perceived crime.


The descriptive statistics for the independent measures involved with this analysis are displayed in Table 1. Three broad categories of independent variables were involved in this modelling process:

  • victim characteristics
  • victim-offender relationship
  • incident characteristics/seriousness.
Table 1 Descriptive statistics of variables
Category Label n Mean SD Min Max
Victim characteristics Victim male 1,557 0.40 0.49 0 1
  Over 24 years 1,557 0.77 0.42 0 1
  Employed 1,216 0.92 0.27 0 1
  Not in workforce 434 1.57 0.82 0 2
  Physically abused child 1,557 0.19 0.39 0 1
Social disadvantage and fear Fearless night walk 1,557 0.50 0.50 0 1
  Unsafe home alone 1,557 0.18 0.38 0 1
  Individual disadvantage 1,557 0.24 0.43 0 1
  Area disadvantage 1,557 0.20 0.40 0 1
Routine activities Drunk regularly 1,557 0.28 0.45 0 1
Victim–offender relationship Current partner 837 0.34 0.48 0 1
  Previous partner 853 0.71 0.96 0 2
  Known non-partner 965 1.29 1.49 0 3
Incident characteristics and seriousness Private dwelling 1,313 0.55 0.50 0 1
  Licensed premises 834 0.59 0.91 0 2
  Injury severity 1,557 0.78 0.76 0 2
  Multiple perpetrators 1,557 0.21 0.41 0 1
  Informal support 1,557 0.86 0.34 0 1

Victim characteristics

Three broad types of victim characteristics were included in the models:

  • aspects of the individual
  • their levels of fear and relative levels of social disadvantage
  • their movements and lifestyle, as captured by their routine activities.

With respect to aspects of the individual, three victim characteristics variables were included in the model. Of these, victim male and over 24 years were both binary (coded 1 when positive). The victim's employment status was entered into the analysis as a categorical variable, with unemployed respondents compared with employed and not in workforce respondents separately.

Two binary measures were included targeting the victim's general levels of fear: fearless night walk, separated those victims who walked alone at night and felt safe, and unsafe home alone, identified those respondents who reported feeling unsafe when home alone at night. An additional two parameters were included to test the impact of disadvantage on perceived crime. The first, individual disadvantage, was a binary variable calculated by comparing victims who, could not raise $2,000 within a week with the rest. The second, area disadvantage, compared victims from the most disadvantaged quintile of areas, as measured by the SEIFA index of relative socioeconomic disadvantage (based on the findings of Morgan et al 2008).

One variable was included in the model to expose elements of the victim's routine activities. This captured the frequency at which the victim got intoxicated (drunk regularly) and was binary coded 1 for those victims who reported getting drunk at least twice a month.

Victim-offender relationship

Given the structure of the dataset, only one relevant victim-offender parameter was utilised. This was dealt with as a categorical variable, with separate comparisons conducted for assaults by strangers compared to current partners, previous partners and known non-partners.

Incident characteristics and seriousness

One specific incident characteristic that could be examined concerned the location of the assault. This was tested as a class variable in the model, with private dwellings and licensed premises both compared separately with assaults that happened at other locations. The final three parameters tested in the model examined the seriousness of the assault. First, injury severity was coded 0 when victims were not physically injured, 1 when they were physically injured but did not see a doctor and 2 when they were injured and a doctor was required. Second, if there was more than one attacker involved, multiple perpetrators was coded 1 (and 0 otherwise). Finally, a binary variable was included to capture whether the victim had sought any form of informal support (with informal support=1 if this was the case).


Predicting perceived crime

Given the mixed research findings discussed previously for the relationship between sex of the victim and reporting (eg Carcach 1997; Felson et al 1999; Felson & Paré 2005), the analytical strategy in this case was to produce separate models for male and female victims in addition to the summary model that analysed the complete set of assault victimisation data. The odds ratios and error bars for odds ratios of parameters retained when modelling the full data set are displayed in Figure 1 and represented by black circles (this presentation style for odds ratios was motivated by the work of Gelman, Pasarica & Dodhia 2002 and Kastellec & Leoni 2007). The patterns of odds ratios and error bars for the female and male models are displayed in Figure 2 and Figure 3 respectively.

Figure 1: Odds ratios and error bars for parameters predicting perceived crime for all data

 Odds ratios and error bars for parameters predicting perceived crime for all data

Source: 2005 Personal Safety Survey (Australian Bureau of Statistics) Confidentialised Unit Record File (CURF) accessed via the Remote Access Data Laboratory (RADL)

Figure 2: Odds ratios and error bars for parameters predicting perceived crime for female-only data

 Odds ratios and error bars for parameters predicting perceived crime for female-only data

Source: 2005 Personal Safety Survey (Australian Bureau of Statistics) Confidentialised Unit Record File (CURF) accessed via the Remote Access Data Laboratory (RADL)

Figure 3: Odds ratios and error bars for parameters predicting perceived crime for male-only data

 Odds ratios and error bars for parameters predicting perceived crime for male-only data

Source: 2005 Personal Safety Survey (Australian Bureau of Statistics) Confidentialised Unit Record File (CURF) accessed via the Remote Access Data Laboratory (RADL)

Victim characteristics

As expected based on previous research, the full model indicated that victims who were male and under 25 years were less likely to perceive assault victimisations to be crimes. Across all models, the area-level social-disadvantage parameter, the fear parameter and the victim's employment status were found to be non-predictive. Individual-level disadvantage was retained in the full model however, with the pattern indicating that highly disadvantaged individuals were increasingly likely to perceive their assault experience to be a crime. Subsequent analysis revealed this was true for female victims only. Within the overall model, the routine activities parameter concerned with excessive alcohol consumption did have an influence on the perception of crime, but the separate models demonstrated that this was specific to male victims, with those who reported getting drunk regularly also being less inclined to perceive their most recent assault as a crime. Alcohol consumption was non-predictive of perceived crime in the female model.

Victim-offender relationship

In the full model, when the victim and the perpetrator were involved in an ongoing relationship, the likelihood that the victim perceived their assault to be criminal was significantly reduced. This was not found when the perpetrator was a previous partner. Large differences in results by sex direct attention to the separate male and female models. For females, victimisation by a current partner or a known non-partner significantly reduced the perception of the criminality of their most recent event. This reduction did not extend to victimisation by ex-partners, although there was some tendency for this to occur. Conversely, the victim-offender parameter made no significant contribution to the male-only model.

Incident characteristics and seriousness

The full model indicated the importance of incident seriousness, with injury severity and informal support seeking both increasing the perceived criminality of the incident. However, significant differences emerged in the sex-specific models that indicated the importance of this class of parameters for predicting male perceptions of crime. First, the location of the assault was predictive of perceptions of criminality for male victims only, with assaults in licensed premises less likely to be perceived to be criminal and the opposite effect found for assaults that occurred in private dwellings. Second, the presence of multiple perpetrators impacted on male victims' perceptions only, potentially reflecting the greater frequency of such situations for males relative to females.

The relationship between perceived crime and reporting

As stated previously, the purpose of this research was to explore the relationship between assault and perceived crime. However, the significance of this novel question for crime prevention policy and the measurement of crime is enhanced by a brief examination of the relationship between perceived crime and reporting to police, as displayed in Table 2. These patterns (which were consistent across male and female victims when analysed separately) demonstrate four important findings. First, approximately one-quarter of the five-year assault cases captured by the 2005 PSS were both considered to be crimes and reported to police. Second, just under half of all these assault cases were neither perceived as crime nor reported. Third, approximately one-sixth of the sample could be classified as the dark figure of crime, as their incidents were crimes according to the victims but not brought to the attention of the police. Fourth, just under one in 10 of the assaults included in this sample were reported despite the victim themselves not considering them to be crimes. When analysed in isolation, perceived crime produced an odds ratio of 7.50 when attempting to account for reporting. Further exploration of this relationship is the focus of a forthcoming work by the same authors.

Table 2 Crime and reporting correlation matrix
Reporting   Perception  
    Not crime Crime Total
n=1,557 cases
Police not told Row % 72.3 27.7  
  Column % 83.5 40.3  
  Total % 46.5 17.9 64.4
Police told Row % 25.8 74.2  
  Column % 16.5 59.7  
  Total % 9.2 26.5 35.6
Total %   55.7 44.3  


Overall, the models performed well when predicting perceived crime for assault incidents perpetrated by a male. Importantly, the models point to both differences and commonalities in the male and female perception of the criminality of these types of assault incidents. For women, relationships with male perpetrators had a significant influence on the perceived criminality of their victimisation, with female victims generally less inclined to perceive incidents to be crimes when the perpetrator was known (provided the perpetrator was not an ex-partner). In contrast, the study indicated that male perceptions of incident criminality are affected by serious incident characteristics or routine activities such as heavy drinking and the location in which the assault incidents occurred. Future research could explore the interactions between these variables to examine whether these results reflect unwillingness by male victims to criminalise incidents in these settings unless victimisation involved multiple offenders. While this study cannot be definitive on the issue, it points to the acceptance of some violent incidents as fights and as things to be left outside of the criminal justice system. To the extent that this conclusion is valid, it emphasises the important role proprietors and venue managers must play in ensuring public venues, particularly licensed premises, are well managed.

Importantly and contrary to the model of Ruback et al (1984), the PSS points to a range of incidents which are not perceived as crimes, but are still reported to police. More research is needed to investigate why these incidents are reported and what services are sought. In the case of violence against women by known offenders, it may be that victims believe on the basis of past experience that a non-criminal incident will escalate unless police are called. Consistent with this perspective and in questioning the conceptualisation of domestic violence under the heading of vulnerable victims, Hanmer (2003: 268) suggests, '[w]hy not conceptualise domestic violence as interpersonal crime or violent crime in which some calls for assistance will not involve a criminal offence?' This position is supported by research evidence that the demonstration of initiative by victims can be preventative, regardless of the police response (eg Felson, Ackerman & Gallagher 2005).

This study points to the potential for victimisation surveys to overestimate the extent of violent crime. Differences between the perceptions of victims and the objective determination of surveys regarding the criminality of victimisation have implications for policy, practice and research. At a practical level, the judgements of victims are of great relevance with respect to their willingness to bring events to the attention of police and to be involved in a prosecution. Policy needs to acknowledge the importance of victim perceptions and to support not only criminal justice approaches, but also alternative avenues for reducing violence. Researchers also need to acknowledge the increasing gap between behaviours labelled as crime in modern surveys and judgements made by individuals who are closest to reported events. Given the variations in victimisation estimates across surveys (as a result of differences in collection methodology), future research in this area should explore the significance of the victim's perceptions of crime across a range of datasets and crime types. Further, given the exploratory nature of this investigation, the research did not examine the relationship between predictor variables and perceived crime for assaults involving female perpetrators, which is an issue that is necessarily left for future research.

The victim's perception of the criminality of their violent incident has been down-played in modern surveys, which adopt an objective, behavioural definition of victimisation and then apply the label 'crime' in the presentation of their results. This approach presents a potential interpretation trap for criminologists and policymakers. At present, Australian survey questions focus on criminal justice responses to violence and the reasons why victims do not report crime to police, rather than the service they are seeking when they do report to police. The PSS provides an exception in this regard because it investigates the victim's own perception of the nature of their incident and the range of other services sought.


The authors wish to acknowledge the Australian Bureau of Statistics for allowing access to the 2005 Personal Safety Survey expanded, confidentialised unit record file via the Remote Access Data Laboratory.


  • Australian Bureau of Statistics (ABS) 2006a. Crime and safety, Australia 2005. ABS cat. no. 4509.0. Canberra: ABS
  • Australian Bureau of Statistics (ABS) 2006b. Personal safety survey, Australia 2005 (reissue). ABS cat. no. 4906.0. Canberra: ABS
  • Australian Bureau of Statistics (ABS) 2006c. Technical manual: personal safety survey, expanded confidentialised unit record file, Australia 2005. ABS cat. no. 4906.0.55.002. Canberra: ABS
  • Australian Bureau of Statistics (ABS) 2003. Crime and safety, Australia 2002. ABS cat. no. 4509.0. Canberra: ABS
  • Australian Bureau of Statistics (ABS) 1999. Crime and safety, Australia 1998. ABS cat. no. 4509.0. Canberra: ABS
  • Australian Bureau of Statistics (ABS) 1995. Crime and safety, Australia 1994. ABS cat. no. 4509.0. Canberra: ABS
  • Biderman AD & Reiss AJJ 1967. On exploring the 'dark figure' of crime. The annals of the American academy of political and social sciences 374: 1-15
  • Carcach C 1997. Reporting crime to the police. Trends & issues in crime and criminal justice no. 68. Canberra: Australian Institute of Criminology. http://www.aic.gov.au/publications/tandi/tandi68.html
  • Coleman K et al 2006. Violent crime overview, homicide and gun crime 2004/2005 (2nd ed): supplementary volume to Crime in England and Wales 2004/2005. London: Home Office, Research, Development and Statistics (Crime Reduction and Community Safety Group)
  • Coumarelos C & Allen J 1999. Predicting women's responses to violence: the 1996 women's safety survey. Crime and justice bulletin: contemporary issues in crime and justice no. 46. Sydney: NSW Bureau of Crime Statistics and Research
  • Felson RB et al 2005. Police intervention and the repeat of domestic assault. Criminology 43(3): 563-588
  • Felson RB et al 1999. The victim-offender relationship and calling the police in assaults. Criminology 37(4): 931-947
  • Felson RB & Paré P 2005. The reporting of domestic violence and sexual assault by nonstrangers to the police. Journal of marriage and family 67(3): 597-610
  • Gelman A et al 2002. Statistical computing and graphs-let's practice what we preach: turning tables into graphs. The American statistician 56(2): 121-130
  • Goudriaan H et al 2004. Reporting to the police in western nations: a theoretical analysis of the effects of social context. Justice quarterly 21(4): 933-969
  • Goudriaan H et al 2006. Neighbourhood characteristics and reporting crime: effects of social cohesion, confidence in police effectiveness and socio-economic disadvantage. British journal of criminology 46: 719-742
  • Hanmer J 2003. Mainstreaming solutions to major problems: reducing repeat domestic violence, in Bullock K & Tilley N (eds), Crime reduction and problem-oriented policing. Cullompten, Devon: Willan Publishing: 252-283
  • Kastellec JP & Leoni EL 2007. Using graphs instead of tables in political science. Perspectives on politics 5(4): 755-763
  • Kury H et al 1999. Reporting of crime to the police in the Federal Republic of Germany: a comparison of the old and the new lands. Justice quarterly 16(1): 123-151
  • Matka E 1990. Uses and abuses of crime statistics. Crime and justice bulletin: contemporary issues in crime and justice no. 11. Sydney: NSW Bureau of Crime Statistics and Research
  • Mirrlees-Black C 1999. Domestic violence: findings from a new British crime survey self-completion questionnaire. London: Home Office
  • Morgan F et al 2008. Schizophrenia and offending: area of residence and the impact of social disorganisation and urbanicity. Trends & issues in crime and criminal justice no. 365. Canberra: Australian Institute of Criminology. http://www.aic.gov.au/publications/tandi2/tandi365.html
  • Ruback RB et al 1984. Social influence and crime-victim decision making. Journal of social issues 40(1): 51-76
  • Skogan WG 1976. Citizen reporting of crime. Criminology 13(4): 535-549
  • Skogan WG 1981. Issues in the measurement of victimization. NCJ-74682. Washington DC: US Department of Justice Bureau of Justice Statistics. http://www.ncjrs.gov/App/Publications/abstract.aspx?ID=74682
  • Skogan WG 1984. Reporting crimes to the police: the status of world research. Journal of research in crime and delinquency 21(2): 113-137
  • Skogan WG 1994. Contacts between police and public: findings from the 1992 British crime survey. London: Home Office
  • Tanton R et al 2001. Modelling crime victimisation and the propensity to report crime to police (methodology advisory committee), June 2001. ABS cat. no. 1352.0.55.040. Canberra: Australian Bureau of Statistics
  • Wolf Harlow C 1985. Reporting crimes to the police. Washington DC: US Department of Justice

About the authors

Dr Joe Clare is a Research Fellow at the Crime Research Centre, University of Western Australia.


Dr Frank Morgan is the Director of the Crime Research Centre, University of Western Australia.


This paper is taken from the report of research commissioned by the Criminology Research Council.