Help with writing a short analytical summary of 150-200 words on each of the 2 articles below.

Article 1: Exploring community-based options for reducing youth crime.

The BackTrack program was established in Armidale in northern New South Wales (NSW), Australia, in 2006, for 14-17-year-old high risk young people (http://www.backtrack.org.au, accessed on 15 April 2021). It is underpinned by six key principles derived from previous reviews of the literature [1,2] together with feedback from staff [3]: (i) in recognition that its participants are more likely to engage in multiple risk behaviour, the program is comprised of multiple components that target different areas of need simultaneously (e.g., personal development, skills training and legal issues); (ii) flexibility in the delivery of the program components, which reflects that the focus of young people's needs shifts over time; (iii) flexibility in program attendance, so that participants are able to start, leave, and re-enter the program as they wish, or as their life circumstances permit; (iv) a requirement that young people in the program eventually actively participate in all components of the program; (v) active engagement of local businesses, local media, key stakeholders (e.g., police, and magistrates), and community members in delivering program elements, resolving bureaucratic problems, providing infrastructure and funds, and facilitating communication about the benefits of the program; and (vi) recognition that achieving sustained change among high-risk young people will take a number of years.

The BackTrack program comprises a range of activities organised into five standardised core program components: effective engagement, to optimise participation in the program; individualised life-skills management, to address participants' immediate and practical needs, such as attending court or finding secure housing; diversionary activities, to reduce participants' exposure to high-risk situations, such as night-time encounters with police in public places or volatile situations at home; personal development, identity and team identity, to improve participants' social and emotional (or psychological) wellbeing, increase their range of personal coping strategies (especially for using in high-risk situations) and to enhance their sense of connection to their peers and community; and learning and vocational skills, to increase their opportunities for active participation in education or training likely to lead to employment. This model of standardisation (the five core program components) with in-built flexibility (the specific activities that operationalise each component are selected and designed by staff), provides a mechanism to both standardise the intervention across multiple communities and tailor it to the resources available in different communities. Further details of the BackTrack program can be found on the BackTrack website (http://www.backtrack.org.au, accessed on 15 April 2021) and other sources [4].

Participants in the BackTrack program are typically at a high risk of drug and alcohol harm, psychological distress, and suicide. The eligibility criteria and the procedure for referral and acceptance into the program are detailed elsewhere [3]. Briefly, young people are eligible to participate in the service if they: (i) resided in a community where the service was available; (ii) were aged 14-21 years; and (iii) were currently experiencing more than one of the following behavioural risk factors: involvement in criminal activity; substance use; violent behaviour; homelessness; poor mental health and wellbeing; poor engagement with school (including suspensions and unexplained absences); and un- or under-employment. All program participants report experiencing risk factors in at least two domains of risk, and more than half experience risk factors in all four domains [3]. The most common risk factors were involvement in crime or with the juvenile justice system, school absence, unemployment, suicide ideation, psychological distress, substance use, low levels of physical activity, and low health service utilisation [3]. The presence of these risk factors places these young people at both short-term risk of harm and long-term risk of entrenched unemployment, criminal involvement, and incarceration [5,6,7,8,9]. Recent research has suggested that one in three serious young offenders strongly endorsed the view that crime had become their way of life with age of onset and frequency of offending reinforcing this view [10].

In addition to personal hardship, the harms experienced by these young people have a negative impact on their communities through increased social disruption, potential loss or damage to property, fear for personal safety, and increased health costs, as well as police, court and incarceration costs [11,12]. Given this impact on both individuals and the community, it is important to consider community value when evaluating the effectiveness or conducting an economic evaluation of a program like BackTrack. Such views have the potential to impact on program uptake, funding, sustainability and estimates of its benefits. Despite the importance of economic evaluation, a systematic review of community-based programs for high-risk young people found that no published evaluations to date have conducted an economic analysis or systematically quantified the costs or economic benefit derived from such programs [2]. The aim of the current study is to address this lack of data and seeks to explore community value and preferences for reducing youth crime and improving community safety using BackTrack in a rural setting in Armidale, New South Wales, Australia.

2. Materials and Methods

Discrete choice experiments (DCEs) provide a useful method for quantifying preferences for nonmarket goods and services [13,14]. DCEs are based on the premise that any good or service can be described according to its attributes, and the levels of these attributes determines the relative value they place on them. Respondents in a DCE must choose their preferred option from a group of hypothetical scenarios, called a choice set. It is proposed that aligning health care policy with patient or community preferences could improve the effectiveness of health care interventions by improving adoption of, satisfaction with, and adherence to clinical treatments or public health programs [15].

The application of DCEs has increased rapidly over the past decade and has been successfully applied to a diverse range of health applications including cancer treatment [16]; depression [17]; dermatology services [18]; diabetes [19] and treatments for Alzheimer's disease [20]; and weight-loss programs [21]. To date, there has been limited application of DCE methods to community-based interventions for vulnerable youth such as the BackTrack program.

In order to develop consensus-based methodological standards, the International Society for Pharmacoeconomics and Outcomes Research (ISPOR) developed a 10-item checklist covering: research question; attributes and levels; construction of tasks; experimental design; preference elicitation; instrument design; data collection; statistical analyses; results and conclusions; and study presentation [22]. This study follows these guidelines. Although the checklist does not endorse any specific methodological approach to conjoint analysis such as choice of attributes or analytical approach, they do promote good research practices for the application of conjoint-analysis methods. For further discussion on methodological challenges inherent in conducting DCEs, the reader is referred to additional sources [23,24,25,26,27].

2.1. Research Question

The research question for the current study is: Does the community prefer community programs like BackTrack compared to greater police presence and what factors affect these preferences?

2.2. Attributes and Levels

Attributes and levels were identified using mixed methods. Researchers engaged in discussions with the research and BackTrack program teams in order to characterise the choice decision, identify the alternative programs, and determine which factors (attributes) might drive the decision process, and which were most important.

Given the tension between drawing a large sample from a regional area with a limited budget, the choice was restricted to three intervention options (BackTrack, greater police presence and current practice), each of which had two attributes (one for effectiveness and one for cost) and three levels of difference. This simple design helps to ensure adequate power to detect differences between the key variables of interest, program efficacy and cost. An additional literature search informed the identification of realistic values for these variables [1,2].

These measures led to a best-worst DCE design with three labelled options; current practice, BackTrack, and greater police presence, with two attributes each; less crime (as a percentage) and cost. The current practice option had fixed levels of 0% less crime and AUD 0 cost, while the BackTrack and greater police presence had three levels of 10%, 20%, and 30% and AUD 10, AUD 20, and AUD 40 for less crime and cost, respectively. These were initially arranged in a factorial design with each possible combination allowed, which was then reduced to a more efficient design. This design prohibited equal levels of one attribute across options, e.g., 20% less crime for both BackTrack and greater police presence, and removed dominating scenarios, e.g., where BackTrack was AUD 10 and 30% less crime while greater police presence was AUD 40 and 10% less crime. These dominating scenarios were defined by the research team.

This design was then tested in a pilot study of 43 participants to establish if the survey were easy to understand and complete, and to determine a magnitude of difference between the three levels of effectiveness that would allow respondents to easily discriminate and trade-off between percentage reductions in crime and cost. It was found that current practice was only chosen as the best option 2.7% of the time and it was chosen as the worst option 90.6% of the time. It was therefore removed for the final DCE as it was clearly a redundant option. Furthermore, it was found that people's responses generally made sense in that higher costs and lower reductions in crime lead to a greater chance of selecting an option as best; however, the costs were too low to provide insight into the maximum value that people would be willing to pay. For the final DCE, the levels of cost were therefore adjusted to AUD 30, AUD 60, and AUD 120 based on model predictions from the pilot data. The pilot data did show differences between the efficacy levels however the difference between 20% and 30% tended to reduce as the cost increased; therefore, the third level of efficacy was increased slightly for the final DCE. Finally, there were some comments regarding the clarity and wording of the attributes which led to the descriptions seen in Table 1. Table 1 provides the components used to construct the choice sets for the final DCE.

2.3. Construction of Tasks

Each program type (BackTrack or greater police presence) generated nine possible scenarios (combinations of program characteristics and cost), so the possible pairs (i.e., choice sets) numbered 81. Choice sets with duplicate combinations of levels (n = 45) and with dominated scenarios (n = 18) were removed to minimise redundancy and maximise efficiency. Eighteen choice sets remained, the minimum number of choice sets required to incorporate every possible trade-off of the attributes and levels. These were organised into two blocks with identical structure, each containing half the choice sets (nine per block). The nine choice sets were then randomised within each block. To minimise respondent burden, participants were randomly allocated one of the two blocks.

2.4. Experimental Design

The removal of duplicate levels and dominating scenarios to increase the amount of useful preference information resulted in a non-factorial design. Across the two options of BackTrack and greater police presence there were six possible pairs of efficacy and six of cost, excluding duplicates, e.g., 10% for both options. A fully factorial design would then combine these six pairs with each other to produce 36 possible choice sets. Our design restricted the combinations of pairs such that an option could not have both a lower (or greater) cost and greater (or lower) percentage than the other option, and thus a trade-off needed to occur, as seen in Table 2. This resulted in each efficacy or cost pair occurring three times each across the 18 choice sets. Each combination of efficacy with cost, did not occur equally however with lower/higher costs more likely to occur with lower/higher efficacy. This can cause difficulties in interpreting marginal probabilities as it can appear that higher costs are preferred to lower costs (because they more often occur with higher efficacy); however, the cumulative link model used for analysis appropriately manages such a design.

2.5. Preference Elicitation

Preference elicitation was by trade-off. The respondent was asked to choose one option from the choice set, no allowance for indifference was provided. Explanations about how to complete the task and cheap talk were included. Background information also provided context of crime within the local catchment area. An example of the survey is provided in the supplementary material.

2.6. Instrument Design

The DCE utilised a survey for face-to-face delivery. Demographic and clinical background information was collected so that characteristics of respondents could be examined, and subgroup analyses could be performed, and responder bias examined. The survey included detailed descriptions on attributes and levels and an explanation of the decision scenario to provide good explanations in an attempt to inform participants, the choice they were asked to make, and how to complete the choice tasks.

2.7. Data Collection and Setting

Data was collected over a 12-day period between 30 May 2016 and 10 June 2016 by trained and experienced researchers. Data collection was scheduled at different times and days with researchers allocated to various locations to maximise reach. In order to minimise the potential for selection and volunteer bias, respondents were selected randomly by systematically counting passersby who were within the target age, to a pre-specified number and approaching the nth person (e.g., the 8th in a busy traffic area or 3rd in a low traffic area). The number of non-responders or individuals approached who did not consent to participate in the DCE was recorded to calculate the response rate. This record included gender and approximate age within a range, determined by the interviewer.

All researchers followed the same implementation protocol. They began with an introduction to the research project to elicit informed consent from potential respondents. In order to promote incentive compatibility, respondents were informed that the results of the study were part of an economic analysis that would be used to inform policy. Each respondent was then shown a sample choice set to ensure they understood the DCE survey format, the respective program attributes and efficacy and cost levels, as well as the task they were required to complete. Respondents were then asked to repeatedly choose between the two scenarios based on their efficacy and cost. After completing the nine choice sets, the respondents were given an additional, separate question which asked if there was an intervention or option other than BackTrack or greater police presence they would prefer. If the answer was affirmative, they were asked to explain their alternative. This question was followed by six demographic questions designed to elicit factors that may be influencing choice such as gender, age, employment status, education, and income. Postcode was included to confirm that respondents were residents of the specified community catchment.

2.8. Statistical Analyses

We used a cumulative/probit link model because it provides a psychological interpretation of the choices in terms of a Thurstonian, or random utility model, with normally distributed utility [28]. The DCE model excluded intervention-specific co-efficients for cost and benefit and excluded an alternative specific constant. The former specification was selected to facilitate ease of data collection and model interpretation, which was judged to be a higher priority than optimising the precision of the estimate, given this is the first economic modelling that has ever been applied to community-based programs for high-risk young people internationally [1,2]. The latter specification is appropriate given the aim of the study is to estimate preferences for existing approaches to reducing youth crime, rather than predicting the likely adoption of new strategies [23].

Statistical analysis of the data used a cumulative link model, an approach like logistic regression, to appropriately treat the non-factorial experimental design. We assumed a probit link, which also provides a psychological interpretation of the choices in terms of a Thurstonian, or random utility model, with normally distributed utility. We estimated a linear model on the mean of the utility distribution within the four factors of the design as predictors, variance fixed arbitrarily at 1, and an estimated decision threshold parameter (which measures overall bias towards or against BackTrack or Greater Police Presence). All analyses were conducted using R [29].

3. Results

A total of 282 respondents who satisfied the inclusion criteria completed useable DCE surveys. A total of 805 people were approached, a survey response rate of 35%. Women were more prevalent responders than men (60% vs. 40%), with slightly elevated numbers of female respondents in the 30-49-year age group. Ninety per cent of respondents were residents of Armidale (postcode 2350); of the remaining 10% in the survey catchment, 4% came from Uralla (postcode 2358) and 2.5% from Guyra (postcode 2365). The sample also comprised a variety of educational levels, with almost half (48%) having completed a university education, 42% of these with postgraduate qualifications. Only 14% of the sample had less than a Year 12 level of education. There was however, a more even spread of employment types and incomes. A total of 39% of respondents were in fulltime employment, 26% had part time or casual employment and 29% were not in the workforce. Of the 79% of the sample who answered the income question, 30% earned over AUD 80,000 per year and 27% earned AUD 39,000 or less.

Forty per cent of respondents responded affirmatively to the opt-out alternative question. Half of the suggestions provided by these respondents related to youth programs like BackTrack that addressed social needs, employment needs, educations needs or a combination of these. Eighteen per cent of respondents suggested family programs rather than youth programs; 10% opted for a combination of greater police presence and BackTrack; and 8% for mental health support or rehabilitation. The remainder suggested systemic, policy level changes; Indigenous specific; religion based; or justice-based approaches.

The DCE generated results that consistently demonstrated a preference for social programs to address youth crime and community safety in the Armidale area. Overall, the proportion of times in which respondents chose BackTrack over greater police presence was very high: 74.8%. The estimate of the decision threshold was significantly below zero, (0.6385, z = 4.02), which reflects the strong tendency to favour BackTrack over greater police presence in these data. Each of the four attributes had a statistically significant effect on the utility, reflecting their importance to choose. The standardised regression coefficients (column "Z" in Table 3) indicates the relative importance of the different attributes. Two coefficients were positive (BackTrack benefit and police cost) indicating that greater values of those attributes led to more people choosing BackTrack over greater police presence, and two coefficients were negative (BackTrack cost and police benefit). These directions are exactly as expected. The effect on utility was greatest for BackTrack benefit, moderate for each of BackTrack cost and police benefit, and small for police cost. This pattern indicates that respondents were mostly influenced by the benefit of BackTrack. For example, adding one extra dollar to the cost of BackTrack has the same influence as subtracting approximately three dollars from the cost of police.

Estimated choice proportions from the probit regression are used to show how the different factors influence choice (Figure 1). Each of the four panels in the figure takes one of the factors and shows how the proportion of people choosing BackTrack changes as this factor is increased (while holding all the other factors constant at their median values). The dashed red lines show the point of indifference, 50% choice. The top left panel shows that people more often choose BackTrack than greater police presence (i.e., choice proportions above 50%) until the cost of the BackTrack option reaches approximately AUD 150. The top right panel shows that people always choose BackTrack more often than policing, no matter what cost we gave to the police option (at least for the median values of the other factors). The bottom left panel shows that BackTrack is mostly preferred over policing, even for very small benefits of the BackTrack option. The bottom right panel shows that respondents only choose policing more often than BackTrack once the benefit of the police option exceeds about 53%.

Analyses were re-run after splitting the data by gender (168 female, 112 male, and 1 unspecified), and by education level (96 with secondary education only, 185 with tertiary education, 2 unspecified), and by household income (110 with >AUD 60,000, 100 with

4. Discussion

In a rural/regional community where there is concern about youth unemployment and associated youth crime and antisocial behaviour [30], public perception of a community program designed to address the needs of high risk young people has the potential to be a powerful determinant of program acceptability, uptake, success, and sustainability. Our reviews of the literature identified a lack of outcome evaluation studies of interventions that targeted multiple risk factors, relative to single risk factors, among high-risk young people [1] or economic evaluations of interventions for high risk young people [2]. Further, very few studies have considered the viewpoint of the community itself or the value that they may attach to community-based programs that address youth crime.

This research leverages off the implementation of the BackTrack program implemented in Armidale, NSW, since 2006. Discrete choice experimental methodology was implemented to explore preferences for, and value of, implementing the BackTrack program to reduce youth crime and improve community safety.

4.1. Overall Findings

The results from this study showed that in a representative sample of the population of Armidale, there was strong preference for BackTrack. Overall, respondents chose BackTrack over greater police presence 75% of the time and continued to choose BackTrack to a cost of AUD 150 per household per year equivalent to a total benefit of AUD 2.04 million per annum.

The effectiveness of BackTrack was the strongest predictor of choice; however, all four attributes or predictors (the cost and benefit for each of the two choice optionsBackTrack and greater police presence) had a statistically significant effect on utility. The direction of the effect of the four attributes on utility was as expected. For example, greater BackTrack benefit and greater police cost resulted in respondents choosing BackTrack over greater police presence, whereas greater BackTrack cost and greater police benefit had a negative effect on the choice of BackTrack over greater police presence. Split sample analyses, conducted to account for differences in preferences that arise from differences in individual characteristics such income, education, and gender, revealed no significant differences between the subsets analysed. These results provide further evidence of the strong community support for BackTrack. Interestingly, when respondents were given the option of suggesting an alternative program to the two offered in the DCE, most suggestions (approx. 80%) were similar social/education/health programs or a combination of greater police presence and BackTrack.

This research fills a void in the literature in terms of understanding community values and preferences for programs like BackTrack. Such information is important in the context of program uptake, funding, and sustainability. The research methods also extend the application of DCE methods and provide important inputs into economic evaluations by valuing community benefit. The Armidale community has embraced the BackTrack program and graduates of the BackTrack program are seen as important community members [4]. This is in stark contrast to community attitudes of participants first enrolling in the program that have a legacy of crime and community disruption.

4.2. Limitations

Data collection was somewhat limited by time and funding constraints; however, a sample size of 282 was considered adequate for the purposes of the analysis. This study removed the opt-out alternative because it was judged to be the worst alternative by 90% of respondents. Excluding choices/pairs in a labelled DCE needs great care because of relatively subjective judgements about the threshold for what proportion of responses ought to be properly regarded as a marginal result, and because retaining more alternatives can be used to help more precisely understand the outcomes. In this study, for example, we have estimated respondents' willingness to pay for different interventions but retaining more alternatives could have also helped determine whether respondents made their choices because of their preferences related to the effectiveness of each alternative or the type of intervention (e.g., some respondents may be willing to pay more for greater police presence even though it is less effective because they simply prefer more police on the streets). Future research could start to examine the decision-making process of respondents' in determining their preferences in addition to identifying their preferences per se.

We used a cumulative/probit link model because it provides a psychological interpretation of the choices in terms of a Thurstonian, or random utility model, with normally distributed utility [28]. Although the cumulative/probit model was selected because of its choice and technical features were appropriate, a mixed model could have been used to demonstrate the distribution of preferences across the population. Future research could utilise both approaches to quantitatively examine the robustness of the results to the choice of model and analysis.

The key design features of this study were the exclusion of intervention-specific co-efficients for cost and benefit, and the exclusion of an alternative specific constant. Having a cost-specific coefficient (separately to a co-efficient for effectiveness) would allow a more precise estimation of WTP and should be integrated into future DCE evaluations of programs for high-risk young people. Although the inclusion of an alternative specific constant is generally recommended in DCEs to avoid forced choices, the decision to exclude it in this study was appropriate for two reasons. First, neither of the alternatives under consideration were new services in the community, meaning that respondents were asked to choose between realistic, existing alternatives rather than more abstract experimental options for which an opt-out option would be appropriate. Second, the way in which an opt-out alternative could have been presented to respondents would have been arbitrary and of unknown impact on respondent's choices [23].

A further methodological issue is that the extent to which the results can be reasonably extrapolated to the whole community might be limited by two factors. First, this DCE only considered one alternative (BackTrack) to more policing, rather than multiple alternative options, such as harsher penalties for offending or increasing youth detention. The extent to which the community would preference BackTrack over alternatives other than more policing across the entire population of the community remains unclear. Second, is the extent to which the sample was representative of the population. A response rate of 35% was lower than expected but, given the unfamiliar nature of a DCE survey and the potential for respondent burden, was an acceptable outcome. The survey results showed that the DCE captured a representative sample of the population of the Armidale region [31]. The largest proportion of non-responders were women in the 30-49-year age group; a group who tended to be apologetic, citing lack of time, being at work, busy with children or going to collect children, as their reasons for not participating.

5. Conclusions

This study estimates a strong community preference for BackTrack relative to more policing (a community WTP of AUD 2.04 million). Although it is a compelling result, the exact strength of the estimated preference may lack some precision as a consequence of the methods. Nevertheless, the apparent strong preference for community-based programs to reduce youth crime relative to more policing, coupled with the new availability of refined DCE methods [23,27], highlights the clear value of replicating this DCE with more community-based programs for high-risk young people.

Although the BackTrack program commenced in Armidale, the program has been implemented in several other rural communities. The BackTrack strategy is to build capacity and capability to positively impact the lives of many more young people across a range of disparate communities. Building an evidence base for programs like BackTrack are essential for ongoing investment and sustainability. This research adds to this evidence base by highlighting strong community preferences for youth based programs that are community based rather than traditional means of dealing with youth crime through punitive measures.

Article 2: Classifying neighbourhoods for reassurance policy.

Introduction High rates of recorded crime and high levels of fear of crime have emerged as characteristics of late modern societies (Garland, 2001: xi). According to Beck (1992), the technological advances as a result of globalization are creating sociological changes that are contributing to what he terms a ''global risk society''. Giddens (2002: 65) argues that society is increasingly confronted with various types of manufactured risk and one of the central tasks of governments has been to control the risks that cause public concern, including that of crime and anti-social behaviour. This has fundamental implications for policing as we have traditionally known it in liberal democracies. Ericson and Haggerty (1999) argue that in late modern societies, policing is moving from a traditional focus on maintaining law and order and controlling crime to a role that is more about detecting and managing risk and communicating knowledge of that risk to other institutions in society. Similarly, it is argued that community policing is giving way to ''policing communities of risk'' (Johnston, 2000: 52/56), and the notion of the police controlling a hub measuring people's experience, thoughts and feelings about crime in their neighbourhood is an important part of a new Reassurance Policing Strategy Model (Innes et al., 2004: vxxvi). Innovative new models of policing may be a response to the risk society that has led to increasing demand for security and a proliferation of providers, including those from the private sector. Button (2002: 26/27) points out, with some irony, the more that people become aware of security strategies and products, the more aware they become of the risk of victimization and the more conscious of the risks they face. Risk in this context is used to describe the fears and uncertainties that reflect an ambient sense of insecurity, rather than actuarial risk. This ambient sense of risk is not helped by the debate on law and order being conducted without reference to the local context, or by the publication of recorded crime statistics and national crime surveys that predominantly lack sufficient contextual information. The British Crime Survey (BCS) is an example where the intention was to provide a less alarmist and more balanced picture of crime. Indeed, the surveys have established that, for the statistically average person aged over 16, crime is a rare event. However, a frequent failure to fully acknowledge and account for variations in local context prevents a detailed and widespread appreciation of the very different experiences (and expectations/perceptions) of crime that one is likely to encounter dependent on neighbourhood. Left-realist criminologists have quite properly been critical of British Crime Survey data. Frequent concerns are expressed regarding what has been termed the ''moral symmetry'' conception of victimization*/namely that victims and offenders are very similar in social characteristics, in sharp contrast to the notion of a predatory offender and an innocent victim. In particular, these ''global'' surveys could also mask the kind of victimization experienced by women and minority ethnic communities 190 T. Williamson, D.I. Ashby & R. Webber (Lea & Young, 1984; Young, 1994). Furthermore, they often fail to identify or differentiate those who are at particular risk of victimization. The notion that some neighbourhoods are ''riskier'' than others is something that we are all familiar with and learn early in life. In 1998, the BCS produced a ''risk index'' for the first time indicating which segments of the population are at the greatest risk from certain crimes. Subsequently, it has been argued, for example, that there are strong links between crime and social exclusion (Young, 1999). This may be experienced in local communities that suffer from a loss of stable livelihoods, transient populations and weakening social cohesion. Community fragmentation places strains on family life and the socialization of children is weakened resulting in communities that are socially impoverished (Currie, 1998). For people trapped in such communities, legitimate channels for change such as the political system or a community organization may be rejected in favour of crime. Some communities are also better able to organize themselves in order to address problems and press for their amelioration. It is argued that different communities have differential levels of social capital,1 which can be inferred from the social and other networks that the community has to draw upon to address its needs (Putnam, 1995). The explanation for such differences are likely to be multi-factorial rather than poverty per se, and indeed it is insufficient to characterize the offender/crime profiles of neighbourhoods by deprivation alone as the correlation between these and crime is not perfect (Bottoms & Wiles, 2002: 642). One example of this can be found in the detailed analysis of two estates in Sheffield featured in a range of academic research (see Bottoms & Wiles, 2002). Two small adjacent areas, separated only by a main road, had a 300 per cent difference in recorded offender rates and a 350 per cent difference in recorded offence rates against individual residents and households. Of particular relevance to our studies is that no statistically significant differences were found on a set of key demographic variables, including: sex; age; social class; ethnic origin; average household size; percentage single; percentage male unemployment; age of termination of full-time education; and length of stay in current dwelling. Both areas had been built at the same time and had begun as crime-free, but one estate had tipped with the resulting differences in offending and victimization. Ethnographic studies showed that housing allocation strategies appear to have been partially responsible for the difference, drawing to the two estates new residents with a differential propensity to offend. Bottoms and Wiles (2002: 636) conclude that this research indicates that differential area offender rates are not simply the product of macro-level aspects of social stratification worked through to a local level. Such findings and the consistent evidence of high crime rates in certain deprived areas is reminiscent of some of the social disorganization theorising provided by the Chicago School. In recent years, this has contributed to the renaissance of much research and theory in this vein, particularly regarding the concepts of ''social capital'' and ''collective efficacy'' (see, e.g., Sampson et al., 1997; Sampson & Raudenbush, 1999). Policing & Society 191 Bottoms and Wiles (2002) continue to outline the return to social disorganization theorizing in environmental criminology. Another challenge for social disorganization theories is that some criminal gangs are capable of influencing neighbourhoods in a way that demonstrates collective efficacy, but whose effects are undesirable. Walklate and Evans (1999), for example, report one high-crime area in Salford where there were strong family ties and a well-known criminal group that policed local criminal incidents by giving the suspected local youths a ''smacking'' and discouraging police involvement by writing on a wall in a central location the names of people suspected of being police informants. The study showed that residents were ambivalent about the kind of social order in operation, but appreciated the personal advantage it afforded them, while at the same time being concerned about their children growing up in such an environment (Walklate & Evans, 1999). The tipping back process may rely heavily on the activities of a small number of key individuals who assume community leadership for dealing with the indigenous criminal element, or acquiescing in it. It is evident that in recent years, central government in the United Kingdom has been keen to develop social capital research and practice across a range of departments including the Home Office, Performance and Innovation Unit and National Statistics (cf. Blunkett, 2002). Similarly, recent debates concerning ''collective efficacy''2 in criminological contexts (see Sampson & Raudenbush, 1999; Sampson et al., 1997; Nolan et al., 2004) highlight the importance of community cohesion and shared values in maintaining social control within neighbourhoods. Furthermore, the recent divergence in both central government and the police service from the hard performance culture of recent years to an environment in which key players continually declare the importance of community engagement and citizenfocused service delivery may have been invoked by the loss of public confidence. Indeed, despite the level of victimization, recorded by the BCS, having fallen by 39 per cent since 1995, two-thirds of BCS respondents believe that crime has risen in the past two years (Dodd et al., 2004). This divergence between public perception and observed achievements in crime reduction statistics has been coined the ''reassurance gap'' and given rise to new policing initiatives focused upon both reconciling this disparity and ultimately providing more secure environments. However, the global reduction in crime observed over recent years does not necessarily indicate the cooling of crime ''hot spots'' where people continue to be at greater risk of victimization. Indeed, a blinkered focus on global crime statistics may disguise increased polarization between high- and low-crime neighbourhoods at a local level. It has been frequently illustrated that high-crime neighbourhoods often correlate with relative levels of poverty, and some further evidence for this is contained in those analyses that we will present in this article. However, the exploration of important additional dimensions and perspectives beyond poverty is key to our methodology and analytical approach. The illustrative analyses presented here are based upon the analysis of BCS data and recorded crime data segmented by neighbourhood type. The precise definition of different neighbourhood types in this sense is determined by 192 T. Williamson, D.I. Ashby & R. Webber statistical cluster analysis of a range of socio-economic, demographic, commercial and lifestyle data (see Webber (2004) for discussion). The classification of small areas and the creation of neighbourhood typologies have etched out significant research domains in both the disciplines of geography and sociology. Specifically, we refer here to geodemographics: ''the analysis of people by where they live'' or ''locality marketing'' (Sleight, 2004). Detailed definitions and histories are offered in a number of works (Brown, 1991; Batey & Brown, 1995; Birkin, 1995) and it is generally accepted that the term refers to small area typologies that discriminate neighbourhood type and often ''consumer'' behaviour. In essence, the maxim of ''birds of a feather flock together'' is used to characterize neighbourhoods and analyze likely behavioural patterns. While the dominant applications of geodemographics throughout the late 1970s, 1980s and 1990s were in the financial and commercial sectors, the origins of the technique lie in deprivation analysis and thus the public sector (see Webber, 1975, 2004). From the late 1990s to date, geodemographics in public-sector applications has experienced something of a renaissance, not least due to the consistencies between the local intelligence benefits and ''business best-practice'' angles offered by the geodemographic approach with the political rhetoric surrounding public service reform and the new localism agenda in the United Kingdom. It is proposed here that without detailed contextual information (e.g., geodemographic analysis or similar) many of the statistical findings presented in government statistics and in much criminological research could be misleading and/or misinterpreted. However, although the linkage of geodemographic data with other data sets can be illuminating in providing a more detailed local context, it does not claim to profile a particular person or specific neighbourhood. Of fundamental importance here is that the neighbourhood typology used in this article provides a generalization of unit postcodes (c. 15 households) into similar, comparable neighbourhood types (''ideal types''). A key distinguishing feature of this approach is that it provides a framework for the implementation of targeted communications campaigns. To this extent, it is argued that a most significant facet of geodemographics is their use in the delivery of targeted policing strategies, notwithstanding the fact that the underlying analytical research and findings are significant in their own right. It is the full package of diagnosis, prognosis and the delivery of appropriate intervention strategies facilitated through a geodemographic approach that is argued here to be of utmost importance. In demonstrating the observed link between the risk of victimization and deprivation we do not seek to suggest here any causal relationship and so some caveats are appropriate. We seek to identify the geodemographic data that might help us identify areas of highest risk in order for further research to be undertaken into the reality of people's experience of crime in a particular neighbourhood. Furthermore, if ''high-risk neighbourhoods'' (however defined) can be identified robustly and successfully, this has very significant implications for both policing strategy and public policy. To this end, we illustrate that geodemographic typologies can help to Policing & Society 193 develop heuristic devices for determining customized local service delivery strategies. Our approach is nomothetic in that we are endeavouring to describe, explain, understand and predict general principles, relationships and patterns of behaviour and experience, and we are engaged in seeking increasingly better understandings of representations of reality. It is our expectation that the analysis of data by geodemographic classification can be helpful in identifying where more detailed analyses, perhaps of an ethnographic or idiographic nature, would assist in painting a more veridical picture. We do not envisage geodemographics as a standalone technique, but as one that is complementary of other forms of analysis such as that developed for the Reassurance Policing Strategy Model (Innes et al., 2004). It is important to acknowledge that the descriptions and pen-portraits of geodemographic neighbourhood types are consistent with the notion of ''ideal types'' and thus portray the archetypal essence of neighbourhoods. To varying degrees, those individuals residing within an area aggregate to the typical characteristics of the prescribed neighbourhood type. It is fair to say that inevitably when one aggregates to any unit (be it spatial aggregation, or that based on social similarity) relationships apparent at the area level may not hold true at that of the individual. This wisdom (and associated matters of the Modifiable Areal Unit Problem, MAUP) and ecological fallacy; see Openshaw (1984a,b) is widely acknowledged and appreciated in areal studies, but frequently misunderstood in that of geodemographics. While such concerns do remain in geodemographic research, it can be illustrated that the risks of falling foul of the MAUP and ecological fallacy are indeed less pronounced than one would experience in more traditional analyses by spatial units (Webber & Longley, 2003: 252/254). Geodemographic classifications retain a greater portion of the variance observed at the individual level than do spatial aggregations based on proximity, thus ameliorating the risk of such misinterpretation. In particular, we do not make the assumption that any specific individual necessarily shares the general characteristics of her or his neighbourhood and its population. Many branches of the social sciences support an idiographic approach and operate on a case-by-case basis with specific facts. Similarly, not all neighbourhoods with equal levels of deprivation will be similarly criminogenic, there will be other factors relating, inter alia, to levels of social cohesion, the influence of families, schools and, of course, the individual (Farrington, 2002). Geodemographic analyses go some way to identifying these factors; not only are relative levels of socioeconomic status and deprivation considered, but lifestyles data and psychographics3 can also be incorporated into such analyses to provide a richer perspective of neighbourhoods than can be provided by deprivation indices alone (see Sleight, 2004). Williamson et al. (2005) illustrate one such example in their analysis of youth crime in Nottinghamshire. Here a peculiar feature of one of the highest risk neighbourhood types for youth crime was the almost complete absence of economically successful people. Polemically, it was hypothesized that it may be the absence of successful role models within the immediate community that contributes to anti-social behaviour as much as poverty itself. Inevitably, if such contentions were 194 T. Williamson, D.I. Ashby & R. Webber established, this would have far-reaching policy implications on those methodologies typically used to identify areas of deprivation and allocate public-sector resources. We begin our illustrative analysis here with a brief review of approaches to poverty and crime. Profiling Neighbourhoods and Their Experience of Crime For a more detailed description of the geodemographic approach and methodology used here to examine the relative levels of risk across neighbourhoods see Ashby (2005). This article focuses on the substantive applications of geodemographics in policing and public-service delivery particularly focused towards social capital and neighbourhood development, rather than the methodology per se. The commercial success of neighbourhood typologies for targeting customers has developed over recent decades to support a thriving geodemographic industry with a growing number of proprietary geodemographic typologies now available both in the United Kingdom (see Sleight, 2004; Harris et al., 2005) and overseas. These classifications are typically used in a number of ways: to enable administrative databases to be used as sources of research information; to prioritize areas for investment in new facilities; in the targeting of communications to consumers most ''at risk'' or most likely to respond to a marketing promotion; and to vary the manner in which communications are undertaken with existing customers or clients. The linkage of census and survey data provides an opportunity to examine the level of social capital within each of the neighbourhood types. Therefore, we are able to quantitatively illustrate that those neighbourhoods that are likely to exhibit higher levels of social capital are less frequently victimized and hence potentially better able to resist victimization. It has also been illustrated that neighbourhood analyses of this sort can be of significant benefit in resource determination and the assessment of comparative performance (see Ashby & Longley, 2005; Ashby, 2005). Much contemporary crime reduction research can be criticized for seeking an answer to the question ''what works'', but not taking into consideration the local, geographical context (see Pawson & Tilley (1997) for a review). Using a neighbourhood typology, we have developed a sustainable model of change to create a menu of Policing & Society 199 options to test in each neighbourhood context and so provide an evidence basis for policy and improved practice. Early examples of this are the opportunities to target communications to counter potential intruders posing as meter readers, where we may reasonably conclude that Bungalow Retirement neighbourhoods are particularly at risk, and the targeting of postcode marking and improved security on entrances to flats in those Town-Gown Transition neighbourhoods where evidence both from the BCS and from North and East Devon Basic Command Unit (BCU) shows breaking and entering to be particularly rife. Furthermore, unsurprisingly our research shows that among the areas at high risk of crime are often those neighbourhoods associated with high levels of deprivation. However, while Indices of Deprivation are of value to strategic policing, we argue that multivariate typologies can offer significant and complementary insight; two specific advantages are outlined below. Primarily, traditional analytical methodologies regarding deprivation studies utilize aggregated data that are often further reduced to some classification scale from least to most deprived. This may even be a binary code: for example, whether or not an area is considered within the most deprived 20 per cent. One fundamental concern here is the nature of the data aggregation. To elaborate, if income is aggregated for a Local Authority District, the mean or median statistic used will invariably conceal a variety of conditions. Using these summary measures for such an area with an average household income of 30,000 may be appropriate for a homogeneous population where this is typical of the constituents. However, if such an area contained two or more very different estates (e.g., one with an average income approaching 50,000 and the other below 20,000), the social, environmental and criminological conditions are likely to be very different. Such a tendency towards a potentially misleading crude aggregate average statistic is ameliorated by a multivariate geodemographic approach that explicitly recognizes such non-uniformities and operates at a finer spatial scale. Furthermore, while a deprivation score alone may serve as an effective proxy for crime rate, it provides little assistance in identifying the most efficient form of service delivery or of targeting communication. Geodemographics is particularly valuable in the targeting of specific interventions such as the distribution of literature promoting the need for heightened awareness as in the case of rogue meter readers given above. In this article we use the Mosaic UK7 proprietary geodemographic typology. Mosaic classifies each of the 1.6 million postcodes in the United Kingdom into one of 61 neighbourhood types and 11 aggregate groups. Each of the 61 types and 11 groups are identified by a 20-character label and code (1/61 and A/K). These labels have met some resistance in public-sector applications with some police forces adopting the convention to only refer to the codes (e.g., G43) rather than the full labels (e.g., G: Municipal Dependency, G43: Ex-Industrial Legacy). The inherent difficulties in reducing 400 data variables to snappy ''characteristic'' labels are acknowledged and while issues regarding nomenclature remain, we use the labels in this article for identification purposes.8 200 T. Williamson, D.I. Ashby & R. Webber Figure 1a and 1b illustrate the diversity of neighbourhoods in an example of a small area geography that is increasingly used for reassurance policing and by crime analysts to examine local trends. Ashby and Longley (2005) discuss the various administrative policing geographies from the National, to the Force Level, to Basic Command Units, to Crime and Disorder Reduction Partnership areas, and to more ''local'' geographies. Wards (as in the Talbot Green example in Figure 1a and 1b) are administrative divisions of Local Authority Districts that recently have been adopted by many police forces in the United Kingdom for local reassurance/community/ neighbourhood policing purposes (e.g., the Safer Neighbourhoods teams in London). While wards could be considered among the smallest, most disaggregate spatial units used in strategic policing strategy, the demographics, attitudes, perceptions and lifestyles of the underlying population should not be assumed to be homogeneous. Furthermore, the variation in neighbourhood types will inevitably increase as one aggregates up to larger standard policing spatial units (such as BCUs). Such variation is often lost in crude summary statistics and broad-brush strategies applied to large areal units. People's perception of crime varies depending on the type of neighbourhood in which they reside. Our analysis of the BCS for 2000 illustrates that perception of the local crime rate varies according to neighbourhood composition. Similarly, crime rates, crime mix, fears, anxieties and local concerns all vary by neighbourhood type. One example index profile is given in Figure 2. Here we use standardized ratios indexed to the national average; under this convention, a rate equivalent to the national average would be indicated by a value of 100, a rate twice the national average by a value of 200 and a rate half the national average by a value of 50 (i.e., percentages of the national mean). In this example it can be observed that the people in neighbourhoods that are less affluent tend to keep themselves to themselves and do not have neighbours that help each other. We found the reverse to be the case in rural communities where, despite their lower than average incomes, levels of social cohesion in neighbourhoods are much higher. The use of geodemographics facilitates the leverage of extra value from national surveys such as the BCS. While the analyses of trends by neighbourhood type can be most insightful when taking a national perspective, this methodology can also help to illustrate local propensities for crime-and-disorder-related issues for which there may otherwise be no reliable data source at a fine spatial granularity. Figure 3 illustrates the likelihood of rubbish and litter being considered a problem by local residents in the Birmingham ward of Aston. Here the relative index values have been calculated from the BCS national data set and, under the assumption that attitudes, perceptions and experiences of residents within a given neighbourhood type are consistent across the country, we are able to map the expected local variation in the level of this concern on the basis of the results of a national sample survey. A geodemographic approach can therefore unlock a valuable impression of social and physical disorders, alongside likely perceptions and fears, which may aid local police forces in addressing issues of reassurance. Significantly, such an approach can be relatively inexpensive. Conclusion The trends resulting in the transition of traditional local social structures giving rise to the ''risk society'' means that public concerns about the risk of high levels of crime and victimization are likely to be a permanent feature of life in the twenty-first century. A perception that the state is failing to minimize and manage effectively the risk inevitably will result in erosion of trust and confidence, and to the attribution of blame to those organizations and professions charged with managing the risk, and to the state as being ultimately responsible for them. Top-down command and control structures of managing the risk of crime tend to assume a homogeneous population and published performance data may simply reinforce an ambient sense of risk of victimization. Actuarial analysis of national and operational databases using geodemographic classifications of neighbourhoods reveals that the risk of crime varies considerably depending upon the type of neighbourhood. Even within a ward there is usually considerable heterogeneity of different types of neighbourhoods, and hence related attitudes, perceptions, lifestyles, priorities and perspectives. Some neighbourhoods would appear to be richer in social capital than others and therefore better able to help themselves. Some, although disadvantaged, appear to have a degree of collective efficacy that is missing in other similarly disadvantaged areas. Extending our library of geodemographic profiles may assist in successfully diagnosing such neighbourhood traits, and further empirical review of policing options help develop the prescription of appropriate neighbourhood policing strategies. Our research demonstrates once more the link between levels of victimization and poverty without defining any cause-effect relationship as there are inevitably many other contributing and relating factors. It is the urban poor who are more likely to be victimized and to be insufficiently supported by social capital in their neighbourhood to be able to address such issues. With some recent exceptions, since the introduction of Crime and Disorder Reduction partnerships under the Crime and Disorder Act 1998 there has been a general failure of crime reduction strategies and structures to engage at the neighbourhood level and substantiate the considerable political rhetoric and sound intentions for the Act. Furthermore, community engagement strategies to promote social capital, or drive towards collective efficacy, require a sound and Policing & Society 213 detailed understanding of local community geodemographics. Rebuilding reassurance in local communities is unlikely to occur through existing structures that have tended to ignore the local geodemographic context. Public reassurance is more likely to transpire (and interventions are more likely to be effective) when the needs of particular neighbourhood types are taken into consideration and strategies developed that address them and engage with the people in building social capital. This is no easy task and it is unlikely to prove inexpensive, but without further investment in financially, practically and intellectually customizing the delivery of services to the needs of particular communities, it is difficult to envisage how high-risk communities may ever develop the collective efficacy to enable a sustainable reduction in locally observed crime, anti-social behaviour and related social disorders. Thus far, little has been discussed regarding how such an approach nests with contemporary reassurance policing trials and policy practice. To conclude, we outline our perspective and outlook in this context, with particular reference to the recent reassurance policing pilot programme. The Signal Crimes Perspective (SCP; see the article by Innes and Fielding in this special issue) as an approach to reassurance policing is very much the lynchpin of the National Reassurance Policing Programme trial in the United Kingdom (Innes, 2004). While the SCP aims to identify and address those issues of most importance to the local community, as does the geodemographic approach outlined above, important distinctions should be made. The SCP necessarily, and by definition, moves away from the centralized targets and performance regimes that dominated policing in the 1990s and focuses upon issues of local concern. Whereas SCP requires the identification of risks, signal crimes and signal disorders within local communities through the direct interaction of the police and local community, the geodemographic approach seeks to highlight neighbourhood trends through leveraging extra value from current national data sets (e.g., the BCS) and local administrative and operational data sets. While very localized problems and concerns would only be addressed through the SCP, the geodemographic approach enables the benchmarking of expected conditions given the neighbourhood composition (Ashby, 2005). This further enables the dissemination of best-practice, cross-area comparisons and the setting of targets, if deemed appropriate. The geodemographic approach is not dissimilar or mutually exclusive to the SCP. Indeed, these methodologies could prove very complementary in the reassurance policing agenda. Geodemographic analyses may prove most insightful for reassurance policing at a strategic level, and for the development of communications strategies across priority neighbourhoods. Both the SCP and geodemographic approaches are intelligence-led, citizen-focused methodologies that are beginning to help us better understand the challenges of reassurance policing and rectify the perception gap.