Measuring the dark figure of crime in geographic areas: small area estimation from the Crime Survey for England and Wales

*David Buil-Gil, Department of Criminology, School of Social Sciences, University of Manchester, Humanities Bridgeford Street Building, 176 Oxford Road, Manchester, UK; [email protected]; Juanjo Medina, Department of Criminal Law and Criminology, University of Seville, Seville, Spain; Natalie Shlomo, Social Statistics Department, University of Manchester, Manchester, UK. © The Author(s) 2020. Published by Oxford University Press on behalf of the Centre for Crime and Justice Studies (ISTD). MEASURING THE DARK FIGURE OF CRIME IN GEOGRAPHIC AREAS: SMALL AREA ESTIMATION FROM THE CRIME SURVEY FOR ENGLAND ANDWALES David Buil-Gil, Juanjo Medina and Natalie Shlomo* For decades, criminologists have been aware of the severe consequences of the dark figure of police records for crime prevention strategies. Crime surveys are developed to address the limitations of police statistics as crime data sources, and estimates produced from surveys can mitigate biases in police data. This paper produces small area estimates of crimes unknown to the police at local and neighbourhood levels from the Crime Survey for England and Wales to explore the geographical inequality of the dark figure of crime. The dark figure of crime is larger not only in small cities that are deprived but also in wealthy municipalities. The dark figure is also larger in suburban, low-housing neighbourhoods with large concentrations of unqualified citizens, immigrants and non-Asian minorities. Key Words: crime mapping, crime reporting, divergence, neighbourhood, spatial statistics Introduction For decades, criminologists have been aware of the severe consequences that the dark figure of crime has for designing and evaluating crime prevention policies and, by extension, for citizens’ everyday lives. In 1977, Skogan stated that the dark figure of crime ‘limits the deterrent capability of the criminal justice system, contributes to the misallocation of police resources, renders victims ineligible for public and private benefits, affects insurance costs, and helps shape the police role in society’ (Skogan 1977: 41). These risks have been exacerbated by the generalization of the use of crime mapping techniques, the adoption of place-based policing and the more recent focus on predictive policing (Brantingham 2018). Geocoded police-recorded crimes constitute the basis for all of these. Yet, certain social groups are more likely to report crimes to the police than others, and police forces are more effective in recording crimes in certain areas (Baumer 2002; Hart and Rennison 2003; Goudriaan etal. 2006). The dark figure of police statistics is thus likely to be unequally distributed across geographic areas. This fact has remained as something to be borne in mind and stated as a limitation in many crime mapping studies, but little has been done to account for the geographical inequality of the dark figure of crime. This paper produces small area estimates of the dark figure of crime in England and Wales to serve as basis for future research This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted reuse, distribution, and reproduction in any medium, provided the original work is properly cited. doi:10.1093/bjc/azaa067 BRIT. J. CRIMINOL. (2021) 61, 364–388 Advance Access publication 16 October 2020 364 Downloaded from https://academic.oup.com/bjc/article/61/2/364/5924614 by UNIVERSIDAD DE SEVILLA user on 29 August 2022 aiming to correct for measurement error in crime data by combining police records with survey-based estimates. Crime reporting rates are larger for female victims than males, and elderly citizens are more likely to report crimes than young people (Hart and Rennison 2003; Tarling and Morris 2010). There are also contextual factors that explain why the dark figure of crime may be larger in certain geographic areas. Victims from suburban areas report crimes less frequently than urban and rural residents (Hart and Rennison 2003; Langton etal. 2012), and the neighbourhoods’ economic disadvantage, concentration of immigrants and social cohesion affect crime reporting rates (Goudriaan etal. 2006; Xie and Lauritsen 2012; Berg etal. 2013; Xie and Baumer 2019). The dark figure of crime also varies between crime types (Gove etal. 1985; Tarling and Morris 2010). Although, nowadays, it is well known that police records are more reliable in some areas than others, most crime mapping methodologies geocode offences known to the police and examine their patterns. Advanced techniques are applied to produce crime maps at small spatial scales to enable targeted policing strategies (Weisburd etal. 2012), while potential sources of error arising from the dark figure of crime are usually left touched upon as a limitation or area for future work. Victimization surveys were developed to address the limitations of police statistics as crime data sources (Skogan 1977), and estimates produced from crime surveys can be used to mitigate the sources of measurement error in policedata. However, surveys have their own methodological issues, and measurement error may arise from victims’ non-recall, overestimation or underestimation of situations. Moreover, surveys have limitations to produce crime estimates at the increasingly smaller focus of the criminology of place (Weisburd etal. 2012). Most surveys are designed to record representative samples for large geographical areas, and direct estimates of target parameters are unreliable for most small unplanned areas. This paper makes use of model-based small area estimation (SAE) to produce small area estimates of the dark figure of crime. SAE uses available survey data and auxiliary information to produce reliable estimates of parameters of interest for unplanned areas for which direct estimates are not precise enough (Rao and Molina 2015). In this study, we produce estimates of crimes unknown to the police from the Crime Survey for England and Wales (CSEW), which can then be used to complement existing police statistics. The Crime Mapping and the Dark Figure of Crime section introduces the implications of the dark figure of crime for crime analysis. The Factors Affecting the Geography of the Dark Figure of Crime section discusses the contextual conditions that affect the dark figure of crime. The Data and Methods section presents data and methods. The SAE of Crimes Unknown to the Police section shows model results and estimates. The Discussion and Conclusions section presents conclusions. Crime Mapping and the Dark Figure ofCrime Police records are the main source of data used for crime mapping. However, the likelihood of crimes to be missing in police statistics is related to social conditions unequally distributed across areas (e.g. economically deprived areas and concentration of minorities). The dark figure of police statistics is likely to be larger in some places than others, and crime maps produced from police records may show an unreliable representation MEASURING THE DARK FIGURE OF CRIME IN GEOGRAPHIC AREAS 365 Downloaded from https://academic.oup.com/bjc/article/61/2/364/5924614 by UNIVERSIDAD DE SEVILLA user on 29 August 2022 of crime. Producing estimates of the dark figure of crime from survey data may help mitigate the measurement error in police statistics. Some argue that maps produced from police records and surveys show similar results and assume that police-recorded crimes show true crime levels (Gove etal. 1985). In 1979, Mawby compared police records with other data sources (victimization surveys and self-report studies) in nine areas of Sheffield and concluded that all data showed similar results (Mawby 1979). Bottoms and Wiles (1997) argue that such results must not be overstated and cannot be overgeneralized. They argue that these findings cannot be used to establish a working presumption that police statistics are always valid indicators of crime. Bottoms etal. (1987) conducted a similar study in seven areas of Sheffield and concluded that police data may provide valid crime rates for comparisons across areas with similar housing types and within the same police force jurisdiction. However, police records underestimated crime rates in high-rise areas, and authors warned that caution is necessary when comparing statistics across policeforces. Others show evidence that residents from certain areas are more likely to report crimes to the police than others. Goudriaan etal. (2004) argue that victims of property crimes are more likely to report in countries where police forces are perceived to be competent. Xie (2014) analysed crime reporting trends in large American metropolitan areas and observed that there was a national trend towards higher reporting rates, which was shown in several cities but not in New York. Victims from suburban areas report crimes less often than urban and rural citizens (Hart and Rennison 2003; Langton etal. 2012), and residents from deprived neighbourhoods are less likely to report certain crime types (Goudriaan etal. 2006; Slocum etal. 2010; Xie and Lauritsen 2012; Berg etal. 2013). Baumer (2002) shows that citizens living in deprived neighbourhoods, but also those living in wealthy areas, are less likely to inform the police. Although public reporting is not the unique pathway through which the police become aware of crimes (police can witness crimes, observe cues of crimes and be informed by private law enforcers, and offenders may surrender), it is arguably the main source of data for most crime types and has large impacts on crime rates (Mawby 1979; Brantingham 2018). Maps produced from police statistics may not show accurate visualizations of the geography of crime, and it should always be checked whether other variables are conditioning crime records in geographic areas. Surveys provide key information to complement police statistics. Factors Affecting the Geography of the Dark Figure ofCrime Researchers have found several contextual conditions affecting the dark figure of crime. Factors can be aggregated in variables that affect victims’ reporting rates, unequal police surveillance across areas and differences in counting rules. In England and Wales, the latter was scrutinized and all 43 police forces follow common counting rules (i.e. Home Office Counting Rules for Recorded Crime, National Crime Recording Standard). Thus, we focus on the other factors. We note, however, that a 2014 inspection into police statistics reported a series of practices that need improvement to increase police records’ comparability (HMIC 2014). Economically deprived areas tend to suffer from lower reporting rates than middleclass neighbourhoods (Goudriaan etal. 2006; Slocum etal. 2010; Xie and Lauritsen BUIL-GIL ET AL. 366 Downloaded from https://academic.oup.com/bjc/article/61/2/364/5924614 by UNIVERSIDAD DE SEVILLA user on 29 August 2022 2012; Berg etal. 2013). Social identities in disadvantaged areas may develop legal cynicism that reduces public cooperation with police services and, in certain areas, ‘call the police, or even to cooperate with them, may also be deviant’ (Black 2010: 106). Berg etal. (2013) argue that normative constraints on crime reporting may not exist in middle-class areas where antipolice views play a less important role. Baumer (2002) shows that socio-economic disadvantage affects the likelihood of crime reporting in the case of crime indices dominated by simple assaults, but it does not affect reporting rates for serious crimes (e.g. robbery and aggravated assault). Moreover, while victims of simple assault living in disadvantaged neighbourhoods were less likely to report crimes to the police than elsewhere, the lowest reporting rates were found in the wealthiest areas. Baumer (2002) argues that deprived and wealthy areas are characterized by high levels of social cohesion that help residents cope with minor crimes without the need to contact the police (including taking the matter into own hands). Thus, the relationship between neighbourhood disadvantage and crime reporting would be curvilinear. Research examining the effect of social cohesion on victims’ willingness to report crimes to the police show contrasting results. Some argue that areas characterized by high social cohesion are those where reporting rates are lower due to residents’ higher social resources to cope with crime through social mechanisms alternative to the police (Baumer 2002). As discussed by Black (2010: 7), ‘a citizen is more likely to call the police if he has no one else to help him’. However, Jackson etal. (2013) argue that high collective efficacy measures, defined as shared values and willingness to act to achieve collective goods, are associated to more cooperation with the police. Goudriaan etal. (2006) also show that larger social cohesion measures are related to an increased likelihood of reporting crimes to the police. According to these results, the more cohesive a neighbourhood is, the greater the cooperation with thepolice. Residents living in suburban areas are less likely to report crimes to the police than citizens from urban and rural contexts (Hart and Rennison 2003; Langton etal. 2012). Xie and Baumer (2019) show that crime reporting rates are lower in non-traditional destinations with high concentrations of immigrants, and they argue that this might be due to the poor police effectiveness in assisting immigrant victims and the small social support for immigrants. Moreover, certain demographic characteristics are related to an increased likelihood of victims’ reporting to the police (Hart and Rennison 2003; Tarling and Morris 2010). Areas characterized by larger proportions of these groups may have larger crime reporting rates. For example, reporting rates may be larger in ageing areas but also in neighbourhoods with more married residents. Ethnicity and crime reporting are not always related (Skogan 1977). Berg etal. (2013) found that the most important contextual factor to explain the victims’ likelihood to report crimes to the police is the area crime rate: those who live in areas with high levels of crime are less likely to notify the police. Other research shows no significant effects of crime rates on citizens’ cooperation with police services, while perceived disorder reduces calls for police services (Jackson etal. 2013). The public perceptions about police services, which vary between neighbourhoods, are known to be good predictors of crime reporting rates (Xie 2014). Jackson etal. (2013) show that perceptions of police legitimacy and trust in police fairness are the most important predictors of citizens’ willingness to cooperate with the police. Citizens’ perceptions of police legitimacy are related to their contact with the police, but these are also explained by social identities shaped by neighbourhood conditions (Bradford 2014). MEASURING THE DARK FIGURE OF CRIME IN GEOGRAPHIC AREAS 367 Downloaded from https://academic.oup.com/bjc/article/61/2/364/5924614 by UNIVERSIDAD DE SEVILLA user on 29 August 2022 Finally, policing strategies may affect the proportion of crimes known to the police, both through affecting the victims’ willingness to report and the police probability to witness incidents. Braga (2007) reveals that place-based policing strategies reduce calls for services in treatment places relative to control areas. Schnebly (2008) shows that police notification is higher in cities with more police officers trained in community-oriented policing, but victims are less likely to call the police in cities with more community-oriented officers. Research analysing the effect of stop and search on crime reporting shows conflicting results: ‘[t]he presence of police undertaking stop and searches may increase the opportunity for victims to report crimes, as well as increasing so-called discovery crimes. But stop and searches, if poorly handled, may discourage cooperation in the short and long term, and possibly reduce reporting rates’ (McCandless et al. 2016: 37). Targeting stop and search practices on selected areas might increase crime figures. However, overpolicing areas and targeting stop and search practices in specific locations may also alienate residents from the police and decrease the willingness to cooperate with the police (Jackson etal. 2013). Previous research about contextual predictors of the dark figure of crime is necessary when selecting area-level covariates in SAE. We will examine available data about these variables to produce estimates of crimes unknown to the police. Data andMethods First, we present the survey used to produce estimates of crimes unknown to the police. Second, we introduce SAE methods. Third, optimal covariates are selected to estimate SAE models. Fourth, we apply different SAE approaches and select the approach with the highest performance with respect to the reliability and goodness-of-fit criteria to analyse and map the dark figure ofcrime. Data The CSEW is used to produce small area estimates of crimes unknown to the police. It is an annual victimization survey conducted since 1982. The sampling design consists of a multistage stratified random sample by which a sole randomly selected adult (aged 16 and over) from a randomly selected household is asked about instances where the respondent (household in some cases) had been a victim of a crime in the last 12months. The questionnaire includes questions about perceived safety and perceptions about the police among others. The main part of the questionnaire is completed face-to-face in participants’ houses, but some questions (alcohol and drugs use and domestic abuse) are administered with computer-assisted personal interviewing. Some modules are asked to a sample of 10-to-15-year-old respondents, but these are not used here. Aspecial licence to the survey’s secure access was needed to obtain information about respondents’ low-level geographies (Office for National Statistics 2018). Survey series from 2011–12 to 2016–17 are used in this research. Respondents’ sample sizes are n=46, 031 in 2011–12, n=34, 880 in 2012–13, n=35, 371 in 2013–14, n=33, 350 in 2014–15, n=35, 324 in 2015–16 and n=35, 420 in 2016 –17. Participants are asked about their personal victimization for a list of crimes, which range from misdemeanour-type offences (e.g. theft and criminal damage) to major BUIL-GIL ET AL. 368 Downloaded from https://academic.oup.com/bjc/article/61/2/364/5924614 by UNIVERSIDAD DE SEVILLA user on 29 August 2022 felonies (e.g. sexual assault and burglary). In case of a positive answer, respondents are asked about the details of each victimization, with a cap of five incidents per person. Although this cap allows cross-sectional comparisons, it reduces information in an arbitrary way and is being reviewed by survey administrators. This is not expected to have a large impact in our analyses since only 2.1 per cent of victims registered by the CSEW reported more than five crimes in 2011–12, 2.1 per cent in 2012–13, 1.7 per cent in 2013–14, 1.7 per cent in 2014–15, 1.0 per cent in 2015–16 and 1.5 per cent in 2016–17. In this research, we analyse recorded crimes. Sample sizes of recorded crimes are n=14, 758 in 2011–12, n=10, 296 in 2012–13, n=9, 282 in 2013–14, n=8, 259 in 2014 – 15, n=10, 594 in 2015–16 and n=11, 352 in 2016–17. Each victim of each crime is asked ‘Did the police come to know about the matter?’, which is used to estimate the percentage of crimes unknown to the police. At a national level, the percentage of crimes unknown to the police remains stable around 60 per cent (Figure1). Among those who answer that the police know about the incident, the most common pathway through which the police become aware of offences is by the victim’s report (around 64 per cent of crimes), followed by a report by another person (around 32 per cent), which appears to decrease slightly over time. The percentage of crimes known to the police by another way (the police were there or found out by another way) is small (Table1). The dark figure of crime appears to be quite stable at a national level. Nevertheless, based on this information, we cannot rule out the possibility that the proportion of crimes known to police varies across years in certain areas but not others and, hence, we need small area estimates. Table2 shows the percentage of crimes unknown to the police by victims’ characteristics, their relationship to offenders and crime types. The dark figure of crime tends to be lower among female victims, persons living in low-income households, unemployed or economically inactive victims and married respondents. These differences are statistically significant in some years but not others. The percentage of crimes unknown to the police by victims’ ethnicity and age show inconsistent results across years due to small samples in certain categories. The percentage of unknown crimes is larger when the offender is a stranger. The crime type which is most likely to be known to the police is theft of motor vehicle, but a small dark figure is also observed for burglary. The Figure 1. Percentage of crimes known and unknown to the police (unweighted). Source: CSEW. MEASURING THE DARK FIGURE OF CRIME IN GEOGRAPHIC AREAS 369 Downloaded from https://academic.oup.com/bjc/article/61/2/364/5924614 by UNIVERSIDAD DE SEVILLA user on 29 August 2022 largest proportions of crimes unknown to the police are observed for robbery, criminal damage and threat/intimidation. Differences based on the victim’s relationship to the offender and the crime type are statistically significant in all editions. Table3 shows descriptive analyses of the percentage of crimes unknown to the police by respondents’ area of residence. Crimes suffered by victims from rural areas are generally more likely to be known to the police than those in urban areas. The dark figure tends to be larger among respondents living in inner parts of cities. These differences are only statistically significant in some years and after merging all years. When comparing the dark figure of crime by deciles of the Indices of Multiple Deprivation, we observe that the dark figure of crime tends to be larger in the least deprived areas, but this is inconsistent across years and is only statistically significant in some years. The Index of Multiple Deprivation classifies small areas in England and Wales by their relative deprivation. It is constructed from seven domains (income, employment, education, health, crime, barriers to housing and services and living environment), and it is generally used as a measure of multidimensional deprivation. Small area estimates are produced for local authority districts (LADs) and middle layer super output areas (MSOAs). LADs represent local governments and MSOAs are small areas designed to improve the reporting of statistical information. LADs have an average of 168,000 citizens according to estimates from 2016: a maximum of 1,128,077 in Birmingham and a minimum of 2,331 in the Isles of Scilly. Greater London has 33 LADs. Each MSOA contains between 5,000 and 15,000 residents (on average, 7,200), and between 2,000 and 6,000 households. There are 7,201 MSOAs and 348 LADs in England and Wales. Producing estimates at smaller geographical scales would allow for more precise spatial analyses of the dark figure of crime. However, the main arealevel SAE techniques require that the assumption of normal distribution of the direct estimates is met at the target spatial scale. Such an assumption is only met when aggregating CSEW victimization data at the LAD level or at the MSOA level after merging more than five editions together, while data aggregates at smaller scales suffer from zero-inflated data and distributions skewed towards zero. New SAE methods are being developed in agricultural research to deal with zero-inflated data and skewed distributions (Dreassi etal. 2014), but further research is needed before these can be applied in criminology. Instead, the aggregation of data in LADs and MSOAs allows for using extensively researched methods within social science research. Table 1 Descriptive statistics of how the police come to know about crimes (unweighted) 2011–12 2012–13 2013–14 2014–15 2015–16 2016–17 Police told by respondent f3,445 2,352 2,147 1,935 2,004 1,902 %63.2 63.2 65.0 63.4 64.0 65.5 Police told by another person f1,759 1,214 1,032 974 978 872 %32.3 32.6 31.2 31.9 31.3 30.0 Police were there f127 79 65 65 84 58 %2.3 2.1 2.0 2.1 2.7 2.0 Police found out by another way f121 76 60 77 65 70 %2.2 2.0 1.8 2.5 2.1 2.4 Total 5,452 3,721 3,304 3,051 3,131 2,902 Source: CSEW. BUIL-GIL ET AL. 370 Downloaded from https://academic.oup.com/bjc/article/61/2/364/5924614 by UNIVERSIDAD DE SEVILLA user on 29 August 2022 Table 2 Descriptive statistics of crimes unknown to the police by victim’s characteristics, relationship to offender and crime type (unweighted) 2011–12 2012–13 2013–14 2014–15 2015–16 2016 –17 2011–17 Total F8,874 6,205 5,669 4,868 5,038 4,753 35,407 %61.0% 61.5% 62.3% 60.6% 60.8% 61.3% 61.2% Gender Male 62.2% 62.6% 63.5% 61.6% 62.1% 62.3% 62.4% Female 59.9% 60.5% 61.1% 59.7% 59.8% 60.4% 60.2% χ2 test 7.2** 4.4* 5.4* 3.0+4.2* 2.9+27.6*** Ethnicity White 60.8% 61.5% 62.3% 60.7% 60.5% 61.1% 61.2% Black 61.9% 62.9% 58.8% 53.8% 68.4% 63.5% 61.7% Asian 63.3% 60.0% 61.6% 61.0% 59.4% 60.2% 61.1% Other 61.9% 62.9% 67.6% 64.1% 68.0% 67.0% 65.0% χ2 test 2.2 1.0 3.8 5.6 11.6** 3.8 8.1* Age 16–29 62.5% 63.6% 63.7% 63.1% 59.3% 59.5% 62.2% 30–49 60.2% 60.2% 61.6% 59.5% 59.4% 61.1% 60.3% 50–64 62.2% 62.3% 61.7% 60.6% 65.1% 62.9% 62.4% 65 or older 58.7% 59.9% 63.2% 59.7% 59.8% 61.4% 60.3% χ2 test 10.5* 9.3* 3.3 6.5+20.7*** 4.3 24.4*** Marital status Never married 62.8% 62.7% 62.0% 61.7% 60.4% 61.0% 61.9% Married 60.2% 60.8% 61.6% 59.6% 61.4% 61.8% 60.8% Other 59.2% 60.2% 64.1% 60.8% 60.8% 60.8% 60.8% χ2 test 13.2** 4.8+3.8 2.8 0.7 0.7 6.9* Employment status Employed 62.0% 62.6% 62.9% 61.4% 61.6% 62.0% 62.1% Unemployed 61.6% 59.2% 62.8% 61.3% 64.6% 57.0% 61.1% Economically inactive 59.0% 59.4% 60.9% 58.8% 59.1% 60.3% 59.5% χ2 test 11.8** 9.9** 3.1 4.5 6.3* 4.0 33.5*** Household income Under £20,000 60.0% 59.8% 62.0% 61.3% 59.7% 60.3% 60.4% £20,000–£49,999 63.0% 63.6% 62.9% 61.2% 61.9% 61.2% 62.4% £50,000 or over 61.7% 62.2% 63.3% 58.6% 62.7% 63.0% 61.9% χ2 test 10.0** 11.1** 1.0 3.5 4.8+2.9 17.7*** MEASURING THE DARK FIGURE OF CRIME IN GEOGRAPHIC AREAS 371 Downloaded from https://academic.oup.com/bjc/article/61/2/364/5924614 by UNIVERSIDAD DE SEVILLA user on 29 August 2022 2011–12 2012–13 2013–14 2014–15 2015–16 2016 –17 2011–17 Total F8,874 6,205 5,669 4,868 5,038 4,753 35,407 %61.0% 61.5% 62.3% 60.6% 60.8% 61.3% 61.2% Relationship to offender Knew well 50.9% 52.5% 50.8% 50.5% 49.0% 49.4% 50.6% Casual 49.7% 53.2% 52.2% 47.8% 51.7% 51.4% 51.0% Stranger 63.3% 63.5% 64.7% 63.2% 63.3% 63.9% 63.6% χ2 test 148.4*** 71.0*** 99.0*** 96.7*** 89.8*** 95.3*** 592.4*** Crime Violent/sexual offences 53.7% 53.5% 51.9% 48.8% 44.9% 53.7% 51.4% Robbery/theft 61.7% 62.1% 65.4% 63.1% 64.6% 63.6% 63.2% Burglary 27.7% 26.7% 26.2% 32.3% 32.5% 27.3% 28.6% Theft of motor vehicle 7.3% 8.1% 5.9% 12.6% 5.0% 7.1% 7.5% Theft of cycle 53.7% 56.0% 52.9% 50.2% 48.8% 48.4% 52.2% Criminal damage 66.8% 66.8% 68.4% 67.9% 68.4% 67.7% 67.5% Threat/ intimidation 59.3% 63.8% 63.7% 59.1% 61.0% 61.8% 61.5% χ2 test 665.4*** 460.9*** 504.2*** 339.3*** 449.6*** 385.3*** 2,748.9*** Source: CSEW. ***p<0.001, **p<0.01, *p<0.05, +p<0.10. Table 2 Continued BUIL-GIL ET AL. 372 Downloaded from https://academic.oup.com/bjc/article/61/2/364/5924614 by UNIVERSIDAD DE SEVILLA user on 29 August 2022 Therefore, we will use the STEBLUP approach to analyse the dark figure of crime at the LAD level, and the SEBLUP at the MSOA level. Moreover, the LAD-level STEBLUP model and the MSOA-level SEBLUP model show the best indicators of goodness-of-fit. At the LAD level, the log-likelihood indicator of the Rao–Yu model equals −8,166.6, whereas the STEBLUP model’s log likelihood is −8,163.1; and the log-likelihood parameters of the models fitted at the MSOA scale are −30,562.8 for the EBLUP model and −30,560.4 for the SEBLUPmodel. Additionally, in order to present further robustness checks, we also computed all estimates after selecting only those crimes that took place in the local area, and computed the Spearman’s rank correlation between those and our small area estimates of crimes unknown to the police produced from all crimes. These show a very large coefficient in both cases ( ρ=0.91, p<0.001 at the LAD level and ρ=0.88, p<0.001 at the MSOA level). Thus, the analytic decision of producing estimates from all crimes recorded in the CSEW, as opposed to only crimes that take place in the local area, does not have a large impact on final results. SAE of Crimes Unknown to thePolice First, we present SAE model results to analyse the predictors of the dark figure of crime for area victimization rates. Although the main objective of SAE is to increase estimates’ reliability, it provides valuable information to explain our outcome measure. Second, we map our estimates. Explaining the dark figure of crime for area victimizationrates Table6 shows the spatial–temporal model used to produce estimates at the LAD level across six time periods. The STEBLUP model is used to explain both the spatial and temporal variation of crimes unknown to the police, and model results do not vary across years. For instance, a large positive coefficient in one covariate indicates that areas with higher values of such variable will suffer from more crimes unknown to the police but also that years with increases in such value will tend to have a larger Figure 3. RRMSE% of estimates produced for MSOAs (ordered by sample size). MEASURING THE DARK FIGURE OF CRIME IN GEOGRAPHIC AREAS 379 Downloaded from https://academic.oup.com/bjc/article/61/2/364/5924614 by UNIVERSIDAD DE SEVILLA user on 29 August 2022 dark figure of crime. The STEBLUP accounts for spatial ( ρ1=0.21 ) and temporal ( ρ2=0.09 ) autocorrelation parameters. Although these show relatively small scores of spatial and temporal clustering, their incorporation increases the estimates’ reliability and model’s explanatory capacity. At the LAD level, three covariates show significant coefficients to explain the percentage of crimes unknown to the police. The strongest coefficient is observed for the measure of small urban areas as opposed to conurbations and rural areas. The dark figure of crime is significantly larger in small urban districts, while the measure of large conurbations shows a positive but not significant coefficient. The ASS of income has the second largest coefficient. LADs whose income is far from the average income (i.e. wealthy and deprived municipalities) have a larger percentage of crimes unknown to the police, while middle-classLADs have lower dark figures. The coefficient of mean house price shows that the percentage of crimes unknown to the police is slightly lower in expensiveLADs. Table7 shows the results of the SEBLUP model at the MSOA level for the combined survey editions. The spatial autocorrelation parameter is ρ1=0.13 , showing that the level of spatial clustering issmall. Seven of our covariates show significant coefficients. The largest coefficient is observed for the percentage of citizens with higher/intermediate occupations: the dark figure of crime is lower in areas where more neighbours have occupations of high social grade. The second largest effect is observed for the percentage of UK-born citizens: MSOAs with more UK-born citizens have a lower dark figure. The percentages of Asians and whites in the area, as opposed to blacks and other minorities, show negative coefficients, but only the coefficient of Asians is significant. Areas with more citizens without qualifications have a larger dark figure of crime. The mean house price also shows a significant negative coefficient at the MSOA level, and the percentage of crimes unknown to the police is larger in urban neighbourhoods, which are not part of large conurbations. The ASS of income also shows a significant positive coefficient. Table 6 STEBLUP models of crimes unknown to the police at the LAD level (standardized coefficients) Beta SE t-value p-value (Intercept) 60.96 2.1 28.37 0.000 ASS income 1.558 0.8 2.04 0.031 Conurbation 0.496 1.1 0.44 0.461 Small urban 2.246 1.0 2.24 0.025 Employed (%) −0.493 0.8 −0.58 0.459 Whites (%) −0.374 3.3 −0.11 0.511 Asians (%) −0.025 2.9 −0.01 0.593 Mean house price −1.071 1.1 −0.99 0.041 Crime rate −0.543 0.9 −0.63 0.128 Males (%) 0.514 0.8 0.64 0.517 Spatial autocorrelation 0.21 Temporal autocorrelation 0.09 Log likelihood −8,171.1 Number of areas 331 SE, standard error. BUIL-GIL ET AL. 380 Downloaded from https://academic.oup.com/bjc/article/61/2/364/5924614 by UNIVERSIDAD DE SEVILLA user on 29 August 2022 Mapping the dark figure of crime for area victimizationrates Figure4 shows the wide distribution of the dark figure of crime for area victimization rates. The median percentage of crimes unknown to the police remains stable around 60 per cent, but the variation between estimates is very large. The dark figure of crime is unequally distributed across areas. The variation of the dark figure of crime is particularly large across neighbourhoods (MSOAs), which shows the need to account for crimes unknown to the police in crime mapping. Model-based estimates of the dark figure of crime produced at the LAD level are shown in Figure5. STEBLUP estimates are produced for 331 out of 348 municipalities, while synthetic estimates are used for areas with zero and one sample sizes. Darker shades of grey represent a larger dark figure of crime for area victimization rates, and lighter tones show lower percentages of crimes unknown to the police, according to groups defined by equal intervals. The level of spatial clustering is medium and the temporal variability is large in many districts. Estimates show that the dark figure of crime has increased in 180 out of 348 LADs (51.7 per cent) between 2011 and 2017. Nine out of the ten most populated LADs show decreases in the dark figure of crime, with the only exception of Liverpool, where the observed percentage of crimes unknown to the police increased by 4.3 per cent between 2011 and 2017. Such reduction was very large in Sheffield (−15.0 per cent) and Bristol (−9.8 per cent). In Greater London, the dark figure of crime increased in 23 out of 33 LADs. The City of London is among the five Police Force Areas (PFAs) with the lowest dark figure of crime in four out of six years. Besides West Midlands Police, the other two largest police forces (Metropolitan Police Service and Greater Manchester Police) maintain the dark figure of crime stable around 58 percent. Figure6 shows model-based estimates produced for MSOAs (groups defined by equal intervals). SEBLUP estimates are used in 6,657 of the 7,201 MSOAs, while synthetic estimates are used in areas with zero or one sample sizes. On average, the PFAs with the Table 7 SEBLUP models of crimes unknown to the police at the MSOA level (standardized coefficients) Beta SE t-value p-value (Intercept) 59.965 0.3 198.05 0.000 ASS income 1.257 0.7 1.71 0.048 Conurbation −0.814 1.0 −0.81 0.419 Small urban 2.127 0.9 2.33 0.019 Employed (%) −0.639 0.8 −0.81 0.420 Whites (%) −2.747 4.4 −0.62 0.532 Asians (%) −5.614 3.0 −1.85 0.044 Mean house price −4.322 1.2 −3.56 0.000 Crime rate −1.091 0.9 −1.27 0.204 Males (%) 0.616 0.7 0.84 0.401 No qualification (%) 4.453 1.7 2.63 0.008 High/int. occupations (%) −5.989 1.9 −3.18 0.001 Born in the United Kingdom (%) −5.666 2.4 −2.41 0.016 Workday population density 1.237 0.9 1.39 0.094 Spatial autocorrelation 0.13 Log likelihood −30,556.88 Number of areas 6,657 SE, standard error. MEASURING THE DARK FIGURE OF CRIME IN GEOGRAPHIC AREAS 381 Downloaded from https://academic.oup.com/bjc/article/61/2/364/5924614 by UNIVERSIDAD DE SEVILLA user on 29 August 2022 largest dark figure of crime are Sussex, Staffordshire and Hertfordshire. The PFAs with the lowest dark figures of crime are the City of London (where the police know 50 per cent of crimes), Cumbria and West Yorkshire. In Greater London, 441 of 982 MSOAs show dark figures of crime larger than 65 percent. Discussion and Conclusions This research presents the first map of the dark figure of crime (for area victimization rates) in the United Kingdom and illustrates with estimates which areas may require further efforts to increase the police effectiveness to register offences. The main conclusion is that the dark figure of crime varies across neighbourhoods. Crime maps produced solely from police records may not show a valid visualization of the geography of crime, and these may be complemented by producing survey-based small area estimates of the dark figure of crime. Police records and survey-based estimates may be used as complementary sources of data to obtain more reliable representations of the geography ofcrime. Moreover, although the main objective of SAE is to produce reliable estimates, our models provide a significant set of information for advancing the understanding of the dark figure of crime. At the local level, the main predictor of the dark figure of crime is the measure of small urban areas as opposed to large conurbations and rural areas. This variable is also significant at the MSOA scale, showing that the dark figure of crime is Figure 4. Boxplots of model-based estimates of crimes unknown to the police. BUIL-GIL ET AL. 382 Downloaded from https://academic.oup.com/bjc/article/61/2/364/5924614 by UNIVERSIDAD DE SEVILLA user on 29 August 2022 Figure 5. Model-based estimates of crimes unknown to the police at the LAD level. MEASURING THE DARK FIGURE OF CRIME IN GEOGRAPHIC AREAS 383 Downloaded from https://academic.oup.com/bjc/article/61/2/364/5924614 by UNIVERSIDAD DE SEVILLA user on 29 August 2022 larger in suburban neighbourhoods outside the main conurbations. This adds evidence to previous research (Hart and Rennison 2003; Langton etal. 2012). Data from the UK Community Life Survey 2016–17 show that citizens from small urban areas tend to live in those places less time than residents from rural areas and conurbations, and feelings of belonging and social harmony are the lowest in small urban areas (Department of Culture, Media and Sport 2017). One might expect that residents from areas with a low sense of community do less to maintain the security in places where they do not feel Figure 6. Model-based estimates of crimes unknown to the police at the MSOA level (2011–17). BUIL-GIL ET AL. 384 Downloaded from https://academic.oup.com/bjc/article/61/2/364/5924614 by UNIVERSIDAD DE SEVILLA user on 29 August 2022 they belong (Goudriaan etal. 2006; Jackson etal. 2013). In other words, ‘sense of community is expected to thrive in a socially cohesive context whereby residents would be willing to engage in activities to improve their community and to prevent crime’ (Aiyer etal. 2015: 141). The decision to report crimes to the police (especially minor offences) tends to be driven by the residents’ will to do something to keep their area safe rather than a hope or need for restoration of harm (Hart and Rennison 2003; Tarling and Morris 2010). The dark figure of crime in small urban areas may be mediated by the low perception of social harmony and feeling of belonging in thoseareas. The ASS of income also explains the dark figure at the local level. Both wealthy and deprived districts suffer from more crimes unknown to the police, while crimes suffered by residents from medium-classLADs are more likely to be known to the police. Baumer (2002) showed that the relationship between the area’s wealth and crime reporting is curvilinear. He argued that this is explained by the area’s social cohesion, which would be larger in wealthy and deprived communities than middle-class areas and would allow residents to cope with victimization by alternative non-police ways (Black 2010). Nevertheless, research conducted in the United Kingdom and the Netherlands shows that cohesive communities tend to cooperate more with the police (Goudriaan etal. 2006; Jackson etal. 2013). Moreover, complementary analyses conducted from the CSEW 2016–17 show that the main reason for not reporting to the police among residents living in the 20 per cent most deprived neighbourhoods is the low confidence in police work (41.7 per cent believe that the police would do nothing, would not be bothered or would not be interested). Instead, the most common reason for not reporting in the 30 per cent least deprived areas is that crimes are trivial or not worth reporting (answered by 28.7 per cent who did not report). The proportion of crimes dealt by victims themselves or by other authorities is very small in both cases (smaller than 9 per cent). Therefore, while the large dark figure of crime in deprived areas is likely affected by low levels of confidence in policing (Jackson etal. 2013; Xie 2014), many minor crimes in wealthy areas are unreported due to their small effect on victims’ lives. The mean house price also shows a negative significant relationship with the dark figure, which shows that residents from more expensive areas cooperate more with police services. At the MSOA level, the measures of small urban areas, ASS of income and mean house price remain significant and show the same directionality, but the covariates with the major explanatory capacity are the percentage of citizens with higher/intermediate occupations, born in the United Kingdom, Asians, and without any qualification. The dark figure is smaller in areas where citizens have a higher social grade. Neighbourhoods with larger proportions of citizens with high occupations are typically expensive areas, where previous research had shown that the sense of community is larger (Stewart etal. 2009). Thus, residents living in high social grade areas are likely to develop proactive roles to maintain their areas safe. Xie and Baumer (2019) found that crime reporting rates are lower in neighbourhoods with a large proportion of immigrants. Our research shows that areas with larger percentages of whites and Asians, as opposed to blacks and other ethnic groups, have lower dark figures of crime, although only the proportion of Asians is significant. In the United Kingdom, Asian communities have the highest level of sense of belonging to their neighbourhoods, followed by white citizens (Department of Culture, Media and Sport 2017). Asian communities may develop more active contributions to their neighbourhoods’ safety. Contrarily, MEASURING THE DARK FIGURE OF CRIME IN GEOGRAPHIC AREAS 385 Downloaded from https://academic.oup.com/bjc/article/61/2/364/5924614 by UNIVERSIDAD DE SEVILLA user on 29 August 2022 other ethnic groups tend to show lower values of sense of community (Department of Culture, Media and Sport 2017) and have lower values of trust in the police (data from CSEW 2016–17). Further research should analyse the impact of perceptions about police services on different communities, as well as the composite effect of the ethnic concentration and income deprivation, since certain minorities are overrepresented in deprived areas (Xie and Lauritsen 2012; Jackson etal. 2013). Uneducated citizens are also overrepresented in deprivedareas. There are significant limitations to our results that need to be considered and accounted for in future research in order to offer more reliable estimates ofcrime: (1) It would be more appropriate to produce estimates of the dark figure of each crime type (as opposed to aggregating crime types), but we would encounter a zero-inflated data set that requires different types of SAE methods that are still to be investigated. (2) Our estimates produced at the MSOA level suffer from low reliability in some areas. (3) Crime surveys have their own methodological issues and measurement error may arise from victims’ non-recall, overestimation or underestimation of situations. Moreover, no information is recorded about so-called victimless crimes (drugrelated offences and corporate crimes) and homicides. (4) Our estimates show crimes unknown to the police for area victimization rates rather than area offence rates. This last limitation might complicate efforts to combine estimates of the dark figure and police records, but it can be mitigated in future research by selecting crimes that took place in the local area and applying methods to deal with zero-inflated data. Acknowledgements Survey data used for this research have been provided by the UK Data Service under project ID 108913; SN 7280. Data on modelled ethnicity proportions and population churn have been provided by the Consumer Data Research Centre, an ESRC Data Investment, under project ID CDRC 260, ES/L011840/1; ES/L011891/1. References Aiyer,S., Zimmerman,M., Morrel-Samuels,S. and Reischl, T. (2015), ‘From Broken Windows to Busy Streets: ACommunity Empowerment Perspective’, Health Education & Behavior, 42: 137–47. Baumer, E. (2002), ‘Neighborhood Disadvantage and Police Notification by Victims of Violence’, Criminology, 40: 579–616. Berg,M., Slocum,L. and Loeber,R. (2013), ‘Illegal Behaviour, Neighborhood Context, and Police Reporting by Victims of Violence’, Journal of Research in Crime and Delinquency, 50: 75–103. Black,D. (2010), The Behaviour of Law. Special edition. Emerald Group. Bottoms,A., Mawby,R. and Walker,M. (1987), ‘A Localised Crime Survey in Constrasting Areas of a City’, British Journal of Criminology, 23: 125–54. BUIL-GIL ET AL. 386 Downloaded from https://academic.oup.com/bjc/article/61/2/364/5924614 by UNIVERSIDAD DE SEVILLA user on 29 August 2022 Bottoms,A. and Wiles,P. (1997), ‘Environmental Criminology’, in M.Maguire, R.Morgan and R.Reiner, eds, The Oxford Handbook of Criminology. 2nd edn, 305–59. Oxford University Press. Bradford, B. (2014), ‘Policing and Social Identity: Procedural Justice, Inclusion and Cooperation Between Police and Public’, Policing & Society, 24: 22–43. Braga,A. (2007), ‘Effects of Hot Spots Policing on Crime’, Campbell Systematic Reviews, 3: 1–36. Brakelvander,J. and Buelens,B. (2014), ‘Covariate Selection for Small Area Estimation in Repeated Sample Surveys’, Statistics in Transition New Series and Survey Methodology, 16: 523–40. Brantingham,P. (2018), ‘The Logic of Data Bias and Its Impact on Place-Based Predictive Policing’, Ohio State Journal of Criminal Law, 15: 473–86. Department for Culture, Media and Sport (2017), Community Life Survey, 2016–2017 [data collection]. UK Data Service. Dreassi,E., Petrucci,A. and Rocco,E. (2014), ‘Small Area Estimation for Semicontinuous Skewed Spatial Data: An Application to the Grape Wine Production in Tuscany’, Biometrical Journal, 1: 141–56. Fay,R. and Diallo,M. (2015), ‘Package “sae 2.” R package version 0.1–1, available online at https://mran.microsoft.com/snapshot/2016-11-30/web/packages/sae2/sae2.pdf. Accessed 3 October 2020. Fay,R. and Herriot,R. (1979), ‘Estimates of Income for Small Places: An Application of James-Stein Procedures to Census Data’, Journal of the American Statistical Association, 74: 269–77. Gelman,A. and Hill,J. (2007), Data Analysis Using Regression and Multilevel/Hierarchical Models. Cambridge University Press. González-Manteiga,W., Lombardía,M., Molina,I., Morales,D. and Santamaría,L. (2008), ‘Bootstrap Mean Squared Error of a Small-Area EBLUP’, Journal of Statistical Computation and Simulation, 78: 443–62. Goudriaan, H., Lynch, J. and Nieuwbeerta, P. (2004), ‘Reporting to the Police in Western Nations: ATheoretical Analysis of the Effects of Social Context’, Justice Quarterly, 21: 933–69. Goudriaan, H., Wittebrood, K. and Nieuwbeerta, P. (2006), ‘Neighbourhood Characteristics and Reporting Crime: Effects of Social Cohesion, Confidence in Police Effectiveness and Socio-Economic Disadvantage’, British Journal of Criminology, 46: 719–42. Gove, W., Hughes, M. and Geerken, M. (1985), ‘Are Uniform Crime Reports a Valid Indicator of the Index Crimes? An Affirmative Answer With Minor Qualifications’, Criminology, 23: 451–501. Hart,T. and Rennison,C. (2003), Reporting Crime to the Police, 1992–2000. Special Report. BJS. HMIC (2014), ‘Crime-Recording: Making the Victim Count’, The final report of an inspection of crime data integrity in police forces in England and Wales. HMIC. Horvitz,D. and Thompson,D. (1952), ‘A Generalization of Sampling Without Replacement From a Finite Universe’, Journal of the American Statistical Association, 47: 663–85. Jackson,J., Bradford,B., Stanko,B. and Hohl,K. (2013), Just Authority? Trust in the Police in England and Wales. Routledge. Langton,L., Berzofsky,M., Krebs,C. and Smiley-McDonald,H. (2012), Victimizations Not Reported to the Police, 2006–2010. Special Report. BJS. MEASURING THE DARK FIGURE OF CRIME IN GEOGRAPHIC AREAS 387 Downloaded from https://academic.oup.com/bjc/article/61/2/364/5924614 by UNIVERSIDAD DE SEVILLA user on 29 August 2022 Marhuenda,Y., Molina,I. and Morales,D. (2013), ‘Small Area Estimation With Spatio- Temporal Fay–Herriot Models’, Computational Statistics & Data Analysis, 58: 208–325. Mawby,R. (1979), Policing the City. Saxon House. McCandless, R., Feist, A., Allan, J. and Morgan, N. (2016), Do Initiatives Involving Substantial Increases in Stop and Search Reduce Crime? Assessing the Impact of Operation BLUNT 2. Home Office Report. Molina,I. and Marhuenda,Y. (2015), ‘sae: An R Package for Small Area Estimation’, The R Journal, 70: 81–98. Office for National Statistics (2017), User Guide to Crime Statistics for England and Wales. ONS. —— (2018), Crime Survey for England and Wales, 1996–2017: Secure Access [data collection]. UK Data Service. Pratesi,M. and Salvati,N. (2008), ‘Small Area Estimation: The EBLUP Estimator Based on Spatially Correlated Random Area Effects’, Statistical Methods and Applications, 17: 113–41. Rao,J. and Molina,I. (2015), Small Area Estimation. 2nd edn. Wiley. Rao,J. and Yu,M. (1994), ‘Small‐Area Estimation by Combining Time‐Series and Cross‐ Sectional Data’, The Canadian Journal of Statistics, 22: 511–28. Sampson, R. and Groves, W. (1989), ‘Community Structure and Crime: Testing Social- Disorganization Theory’, American Journal of Sociology, 94: 774–802. Schnebly,S. (2008), ‘The Influence of Community‐Oriented Policing on Crime‐Reporting Behaviour’, Justice Quarterly, 25: 223–51. Skogan, W. (1977), ‘Dimensions of the Dark Figure of Unreported Crime’, Crime & Delinquency, 23: 41–50. Slocum, L., Taylor,T., Brick, B. and Esbensen, F. (2010), ‘Neighborhood Structural Characteristics, Individual-Level Attitudes, and Youths’ Crime Reporting Intentions’, Criminology, 48: 1063–100. Stewart, M., Makwarimba,E., Reutter,L., Veenstra,G., Raphael,D. and Love,R. (2009), ‘Poverty, Sense of Belonging and Experiences of Social Isolation’, Journal of Poverty, 13: 173–95. Tarling, R. and Morris, K. (2010), ‘Reporting Crime to the Police’, British Journal of Criminology, 50: 474–90. Weisburd,D., Groff,E. and Yang,S. (2012), The Criminology of Place. Street Segments and Our Understanding of the Crime Problem. Oxford University Place. Xie,M. (2014), ‘Area Differences and Time Trends in Crime Reporting: Comparing New York With Other Metropolitan Areas’, Justice Quarterly, 31: 43–73. Xie, M. and Baumer,E. (2019), ‘Neighborhood Immigrant Concentration and Violent Crime Reporting to the Police: AMultilevel Analysis of Data From the National Crime Victimization Survey’, Criminology, 57: 237–67. Xie,M. and Lauritsen,J. (2012), ‘Racial Context and Crime Reporting: ATest of Black’s Stratification Hypothesis’, Journal of Quantitative Criminology, 28: 265–93. BUIL-GIL ET AL. 388 Downloaded from https://academic.oup.com/bjc/article/61/2/364/5924614 by UNIVERSIDAD DE SEVILLA user on 29 August 2022