This study was based on secondary analyses of anonymized survey data. The Medical Ethics Committee of the Academic Medical Centre in Amsterdam, the Netherlands, has confirmed that ethics approval is not necessary, because the Medical Research Involving Human Subjects Act (WMO) does not apply to our study.
Implementation of the District Approach
For 36 out of the 40 target districts, data on the content, duration, and scale of interventions implemented as part of the District Approach since 2008 were retrospectively collected using standardized questionnaires and face-to-face interviews with local district managers [4]. Most target districts addressed all six main themes that were central to the approach. The type and scale of interventions that were implemented to address each main theme varied greatly across the target districts (Fig. 1). Two types of interventions were identified that could potentially improve perceptions of area safety and remove underlying problems. A first group of potentially effective interventions aimed to tackle underlying safety problems like general social disorder, youth social disorder, physical disorder, and burglary. Examples of interventions include extra police surveillance, youth leisure activities, youth counselling, bins, and cleaning services. A second group of potentially effective interventions aimed to improve neighbourhood conditions such as housing quality, housing stock, green space, playgrounds, sports facilities/activities, trails, and social capital. Examples of interventions include demolition of rundown homes, housing renewal, (re)construction of green space and playgrounds, extra sports facilities and activities.
Data and study population
Repeated cross-sectional data were derived from the National Safety Monitor (NSM) years 2005–2008 and its successor the Integrated Safety Monitor (ISM) years 2008–2011. Both surveys were targeted at non-institutionalized persons of 15 years and older nationwide. The sampling frame was derived from the national population registry. The sampling frame was renewed each year to assure independence of observations, and it was stratified by police region to assure coverage of each Dutch police region. Monthly samples were drawn from the sampling frame using a two-step design, with sub-municipalities in step one and individuals in step two. For NSM, individuals were approached by telephone or interviewer between January and March. For ISM, individuals were sent a letter between mid-September and December in which they were asked to participate by internet or paper-and-pencil survey. Non-respondents were approached by telephone or interviewer. A total of 226,165 individuals were approached between 2005 and 2011. Overall response rate was 62 %. Respondents were excluded when they had no personal identification number (N = 269), no zip code information (N = 362), or were under 18 years old (N = 6609). A total remained of 133,522 adult respondents. Of these respondents, 3,595 resided in the target districts and 12,9927 resided elsewhere in the Netherlands.
Measures
Perceived area safety and underlying problems
Four outcome variables were included:
-
Perceived general safety: in NSM as well as ISM, respondents were asked whether they sometimes felt unsafe in their own neighbourhood. They could answer yes or no.
-
Perceived physical and social order: in NSM as well as ISM, respondents were asked whether they judged nine problems to occur often (1), sometimes (2), or (almost) never (3) in their neighbourhood. A physical order variable was computed by averaging the scores on graffiti, litter, dog waste, and demolition of phone booths/bus-cubicles/tram-cubicles. Cronbach’s alpha of the four items was 0.61, indicating fair reliability. A social order variable was computed by averaging the scores on nuisance from youth, nuisance from drugs, nuisance from neighbours, drunken people on the street, and people who get harassed on the street. Cronbach’s alpha of the five items was 0.68, indicating fair reliability. As the distribution of mean scores on both disorder variables was highly skewed, mean scores were dichotomized into ‘disorder generally occurs sometimes or often’ (mean score ≤2) and ‘disorder generally occurs (almost) never’ (mean score >2).
-
Self-reported victimization: in both NSM and ISM, respondents were asked whether they had been a victim of any of the following fourteen crimes in the past five years: attempted burglary, burglary, bicycle theft, car theft, theft from their car, car damaging, pick pocketing, violent robbery, other thefts, other damaging, sexual abuse, threat of physical abuse, physical abuse, and other crimes. Respondents could answer yes or no. If they answered yes to any of the crimes, they were asked if they were victimized before or after January 1st of last year (NSM), or this year, last year, or earlier (ISM). If respondents were victimized after January 1st of last year (NSM) or this year (ISM) they were asked whether they were last victimized in the own neighbourhood, somewhere else in the municipality, somewhere else in the Netherlands, or in a foreign country. This information was used to compose a dichotomous variable that measured whether or not the respondent had been a victim of one or more crimes after January 1st of last year (NSM) or this year (ISM) in their own neighbourhood.
Time variables
The main predictor variable was survey year. We also included the variable survey period, which was dichotomized into ‘pre-intervention period’ (years 2005 to 2008 from the NSM) and ‘intervention period’ (years 2008 to 2011 from the ISM).
Districts
The respondents’ district of residence was measured using data on the 4-digit zip codes obtained from the national population registry. The intervention group consisted of the 3,595 respondents living in the 40 target districts. This group comprised 83 zip codes distributed across 18 cities throughout the Netherlands. Nearly three quarter of the zip codes were located in the four largest cities of the Netherlands. Three control groups were included:
-
1.
Rest of the Netherlands: consisting of 129,927 respondents living anywhere in the Netherlands but the target districts. This group comprised 3,697 zip codes.
-
2.
Other deprived districts: consisting of 11,248 respondents living in districts number 41 to 140 on the official list of most deprived districts of the Netherlands (the target districts are number 1 to 40 on the list). These districts were slightly less deprived than the target districts. This group comprised 257 zip codes distributed across 114 cities and villages throughout the Netherlands. Nearly one quarter of the zip codes were located in the four largest cities of the Netherlands.
-
3.
Other deprived districts same city: consisting of 6,022 respondents living in those districts listed under number 2 that were located in the same cities as the target districts. This group comprised 119 zip codes distributed across 18 cities throughout the Netherlands. Over half of the zip codes were located in the four largest cities in the Netherlands.
Because of lower statistical power and the possibility of spillover effects when using the latter two control groups, we used the first group as the main control group.
Intensity of safety interventions
In stratified analyses, the intervention group was split based on the intensity of their safety interventions. For 36 out of the 40 target districts, information on intervention content and scale was available to determine programme intensity [4]. First, for each district a list was composed of all interventions that primarily aimed to improve neighbourhood safety by addressing one of four safety related problems: general social disorder, youth social disorder, physical disorder, burglary [4]. Minimum duration was set at one year. Second, for each safety problem, the scale of combined interventions was graded as small (no change expected), intermediate (small changes expected), or large (substantial changes expected). Third, per district, an overall intensity score was calculated by summing the grades for all four safety problems (small = 0, intermediate = 1, large = 2). Target districts with less intensive safety interventions (score <5, n = 13) were distinguished from those with more intensive safety interventions (score ≥ 5, n = 23). Figure 1 provides an overview of intensity score and intensity classification of the 36 target districts.
Covariates
Control variables included age (seven categories: 18–24, 25–34, 35–44, 45–54, 55–64, 65–74, 75 years and older), gender (men, women), ethnicity (ethnic Dutch, non-ethnic Dutch) and educational level (primary-, lower secondary-, higher secondary-, and tertiary level, based on the International Standard Classification of Education).
Statistical analyses
Interrupted time series analyses were used to assess whether trends in perceived area safety and underlying problems have changed with the implementation of the District Approach in 2008. Multilevel logistic regression models were applied to assess the association between year and any of the outcome variables, i.e., the annual rate of change in the outcome variable. Hereafter, this is called the trend. The variable district was included to measure differences in outcome between the target districts and various control groups at the start of the District Approach. The variable period was included to account for any difference in outcome related to the change in survey design in 2008. An interaction term for the variables year and district was included to assess differences in trend between the target districts and various control groups. An interaction term for year and period was included to assess differences in trend between the pre-intervention period and the intervention period. Hereafter, this is called the trend change. An interaction term for the variables year, district, and period was included to assess whether trend change varied between the target districts and various control groups.
All analyses were controlled for age, gender, ethnicity and education. Additional analyses were stratified by gender (men versus women), age (under 55 years old versus 55 years and older), education (primary- and lower secondary level versus higher secondary- and tertiary level), and intensity of the safety interventions (less intensive interventions versus more intensive interventions). Multilevel regression analyses were applied to take into account clustering of respondents in districts. Level 1 represented individuals and level 2 represented zip codes. All analyses were carried out using STATA 11.0 software. Statistical significance was set at 0.05.