The authors have declared that no competing interests exist.
Large-scale regeneration programmes to improve the personal conditions and living circumstances in deprived areas may affect health and the lifestyle of the residents. Previous evaluations concluded that a large-scale urban regeneration programme in the Netherlands had some positive effects within 3.5 years. The aim of the current study was to evaluate the effects at the longer run.
With a quasi-experimental research design we assessed changes in the prevalence of general health, mental health, physical activity, overweight, obesity, and smoking between the pre-intervention (2003–04 –mid 2008) and intervention period (mid 2008–2013–14) in 40 deprived target districts and comparably deprived control districts. We used the Difference-in-Difference (DiD) to assess programme impact. Additionally, we stratified analyses by sex and by the intensity of the regeneration programme.
Changes in health and health related behaviours from pre-intervention to the intervention period were about equally large in the target districts as in control districts. DiD impact estimates were inconsistent and not statistically significant. Sex differences in DiD estimates were not consistent or significant. Furthermore, DiD impact estimates were not consistently larger in target districts with more intensive intervention programmes.
We found no evidence that this Dutch urban regeneration programme had an impact in the longer run on self-reported health and related behaviour at the area level.
Residents of deprived areas generally have worse health than those living in non-deprived areas. These health differences cannot be explained completely by individual characteristics, such as individual socioeconomic status [
Urban regeneration programmes aim to ameliorate the physical and social environment of deprived areas and may in addition address socioeconomic problems that are common in these areas, such as unemployment and high levels of school dropout. As these urban regeneration programmes target the social determinants of health, they may contribute to health improvements of residents and reduce geographical health inequalities [
The Dutch District Approach is an urban regeneration programme that may impact the health of residents because it addressed the social determinants of health. The Dutch government launched this programme in 2007 with the aim to improve the living conditions of the 40 most deprived districts of the Netherlands within the next ten years [
A first evaluation of the health impact of the Dutch District Approach showed that for some health indicators trends in the ‘target districts’ were more favourable than in control districts [
The initial evaluation of the Dutch District Approach showed the importance of considering specific population groups and selected districts [
In the current study, we examine the health impact of the Dutch District Approach up to 6.5 years after the introduction of the programme. We compare the health and health-related behaviour in the 40 target districts in the years following the start of the regeneration programme with the health and health-related behaviour levels in the years prior to the implementation. These changes are compared with changes in similarly deprived control districts, using a quasi-experimental approach. Furthermore, we examine whether the health impact differs by sex and by the intensity of the urban regeneration programme. We investigate the effect on general health, mental health, physical activity, smoking, overweight and obesity.
In 2007, the Dutch Government launched the Dutch District Approach with the aim to improve the liveability in the 40 most deprived districts in the Netherlands, located in 18 large Dutch cities. These 40 districts were selected using registration data on physical and socio-economic deprivation, and survey information from residents concerning physical and social neighbourhood problems [
The previous evaluation of the Dutch District Approach included an inventory of the intensity of the regeneration programme in each district [
We used nationwide repeated cross-sectional health data from the Dutch Health Interview Survey (HIS), collected between January 2003 and December 2014 by Statistics Netherlands. Respondents are of all ages, living in private households in the Netherlands. Each month, Statistics Netherlands draws a person-based sample from the Dutch population register (approximately 15,000 persons per year). In the years 2003 to 2009, respondents were interviewed at home for the basic survey, which included questions about their general health, height, weight and smoking behaviour. A second, additional questionnaire was left behind after the interview for respondents older than 12 years to ask them about more sensitive topics, such as mental health, but also about their physical activity behaviour. Since 2010, the HIS employs a stepwise approach. First, the selected persons are asked to participate through internet. Second, non-respondents are approached and interviewed by telephone. Third, those who cannot be reached by internet or telephone are approached for a personal interview. The second, additional questionnaire could be filled out by internet or using a written questionnaire. In 2014, the basic survey and additional survey were combined into one survey that can be filled out either using the internet or through a personal interview. Between 2003 and 2014, the annual response rate for the main (in 2014 total) survey was 60–65%. Among those who responded, the response rate for the additional survey was around 80% (years 2003–2009) and around 55% (years 2010–2013). We selected data from adults who were aged 18 years and older and lived in the 40 target district or the control areas at January 1st 2008. We selected adults, because we expected that the impact of the intervention programme might differ between adults and adolescents or children. Unfortunately, we had not enough respondents under the age of 18 years to examine this group separately.
We used a quasi-experimental design to investigate the effect of the Dutch District Approach on individual-level outcomes. We selected control districts similar to the target districts in terms of neighbourhood and individual characteristics. These characteristics were measured at January 1st 2008, i.e. just prior to the start of the implementation, to ensure that the programme could not have influenced the characteristics, and ultimately the outcome. In matching target and control districts, we made the simplifying assumption that individuals did not migrate after 2008. The following is an outline of the matching steps; for details see
Two steps were taken to ensure that the
Second, we selected
For the estimation of the effect of high intensity and low intensity urban regeneration programmes, respectively, versus no programme, we could not use the previously described approach to adjust for neighbourhood level characteristics based on subclassification, because of the smaller sample size. Instead, we used 1:1 nearest neighbour matching on the propensity score to find a suitable control district for each target district. Stratification on individual characteristics was similar to the description above.
After the selection of respondents from the target and control districts, health data of 13,651 respondents were included in the analyses.
We included the same health outcomes as used in the previous evaluation of the programme (i.e. general health, mental health, smoking, leisure-time walking, leisure-time cycling, and sport participation) and added overweight and obesity.
For general health, smoking, overweight and obesity the pre-intervention period included the years 2003 to mid-2008 and the intervention period mid-2008 to 2014. For mental health and PA, we excluded data from 2014, because of methodological changes in the additional survey in 2014. To maintain balance in the pre-intervention and intervention period, we excluded data from 2003 as well, resulting in a pre-intervention period of 2004 to mid-2008 and intervention period of mid-2008 to 2013.
First, we obtained stratum-specific treatment effects by estimating the difference in outcome over time separately for the target and control districts per stratum and calculating the difference between these differences (Difference-in-Differences or DiD). The difference in outcome for the control group served as a way to control for unmeasured confounding (e.g. general trends in lifestyle or health) under the assumption that this affected the target and control districts similarly. We estimated stratum-specific treatment effects through a linear regression model, in which the outcome was estimated as a function of a period indicator (intervention period compared to its reference pre-intervention period), treatment indicator (treatment group, i.e. the target districts, compared to the reference group control) and the interaction between the period and treatment indicator. Next, the overall treatment effect was calculated by pooling the stratum-specific effects over all strata. The overall treatment effects were estimated in similar fashion for the high and low intensity urban regeneration programmes, with the main difference that the strata for the high/low intensity programme analyses were based on individual characteristics only. Sex-specific effects were estimated by pooling over strata concerning men and women, respectively.
Residents of the target areas were similar to the residents of the control districts in terms of age, sex and household composition, but were more likely to be lower educated, of non-western origin and with a lower income (
40 target districts | Control districts | |||
---|---|---|---|---|
Pre-intervention | Intervention | Pre-intervention | intervention | |
52.7 | 52.4 | 53.5 | 53.8 | |
18–35 | 38.9 | 34.9 | 36.0 | 31.4 |
35–55 years | 31.3 | 35.7 | 29.8 | 35.1 |
55 years and older | 29.8 | 29.4 | 34.2 | 33.6 |
Single with/without child(ren) | 39.2 | 40.5 | 36.7 | 37.8 |
Couples with/without child(ren), with others | 60.8 | 59.5 | 63.3 | 62.2 |
Ethnic Dutch, non-Dutch—Western | 64.7 | 69.4 | 82.2 | 84.2 |
Non-Dutch, non-Western | 35.3 | 30.6 | 17.8 | 15.8 |
Primary, secondary | 83.9 | 71.6 | 77.5 | 66.1 |
Tertiary | 16.1 | 28.4 | 22.5 | 33.9 |
First tertile (> €16,493) | 40.4 | 35.3 | 32.8 | 30.1 |
Second tertile (€16,493 – €23,993) | 34.1 | 34.8 | 33.3 | 32.5 |
Third tertile (> €23,993) | 25.5 | 29.9 | 33.9 | 37.4 |
a Characteristics represent average values over the period 2003 to 2014
During the whole study period, residents in the target districts more often reported poor health and unhealthy behaviour compared to residents in the control districts except for the prevalence of smoking in both periods and leisure time walking in the pre-intervention period, which were almost similar in the target and control districts (
Target districts | Control districts | ||||||||
---|---|---|---|---|---|---|---|---|---|
Pre-intervention | Intervention | Intervention versus pre-intervention period |
Pre-intervention | Intervention | Intervention versus pre-intervention period |
DiD (C.I.) |
p-value | ||
Good general health | 9,900 | 68.0 | 66.2 | -1.8 | 74.3 | 72.8 | -1.5 | -0.2 (-5.5;5.5) | 0.94 |
Fair or good mental health | 5,431 | 83.0 | 82.8 | -0.2 | 86.8 | 85.9 | -0.9 | 0.7 (-5.3;6.7) | 0.81 |
Leisure-time walking | 5,679 | 60.1 | 64.0 | 3.9 | 60.4 | 66.3 | 5.9 | -2.0 (-9.9;5.8) | 0.62 |
Leisure-time cycling | 5,431 | 41.5 | 45.0 | 3.5 | 47.7 | 52.7 | 5.0 | -1.5 (-10.2;7.2) | 0.73 |
Sports participation | 5,782 | 40.0 | 38.2 | -1.8 | 45.7 | 41.7 | -4.0 | 2.3 (-5.8;10.4) | 0.58 |
Overweight | 8,853 | 45.1 | 50.3 | 5.2 | 41.3 | 45.2 | 3.9 | 1.3 (-5.2;7.7) | 0.70 |
Obesity | 8,853 | 14.8 | 15.0 | 0.2 | 10.4 | 12.5 | 2.1 | -1.9 (-6.2;2.4) | 0.40 |
Smoking | 9,889 | 33.1 | 31.5 | -1.6 | 34.5 | 32.6 | -1.9 | 0.2 (-5.5;5.9) | 0.94 |
a Reference category
b The n is the sum of all four groups used in the analysis
c Difference in Difference (Confidence Intervals)
Target districts | Control districts | ||||||||
---|---|---|---|---|---|---|---|---|---|
Pre-intervention | Intervention | Intervention versus pre-intervention |
Pre-intervention | Intervention | Intervention versus pre-intervention |
DiD (C.I.) |
p-value | ||
sex | |||||||||
male | 4,641 | 70.0 | 67.7 | -2.3 | 78.5 | 75.2 | -3.3 | 1.0 (-6.3;8.4) | 0.78 |
female | 5,213 | 65.9 | 65.7 | -0.2 | 71.1 | 71.1 | 0.0 | -0.3 (-8.2;7.6) | 0.95 |
male | 2,526 | 88.4 | 84.1 | -4.3 | 91.6 | 89.6 | -2.0 | -2.3 (-9.8;5.2) | 0.55 |
female | 2,905 | 78.3 | 81.7 | 3.4 | 82.6 | 82.6 | 0.0 | 3.4 (-5.7;12.4) | 0.47 |
male | 2,605 | 59.5 | 63.1 | 3.6 | 58.5 | 62.9 | 4.4 | -0.8 (-12.2;10.7) | 0.90 |
female | 3,074 | 60.7 | 64.4 | 3.7 | 62.1 | 69.2 | 7.1 | -3.3 (-14.1;7.4) | 0.55 |
male | 2,337 | 43.0 | 46.6 | 3.6 | 48.9 | 55.5 | 6.6 | -3.1 (16.6;10.5) | 0.66 |
female | 2,696 | 42.3 | 46.1 | 3.8 | 47.5 | 52.7 | 5.2 | -1.3 (-13.5;10.8) | 0.83 |
male | 2,494 | 39.7 | 42.7 | 3.0 | 44.3 | 43.6 | -0.7 | 3.7 (-8.1;15.5) | 0.54 |
female | 3,020 | 41.1 | 34.6 | -6.5 | 46.4 | 39.7 | -6.7 | 0.1 (-10.7;11.0) | 0.98 |
male | 4,164 | 48.1 | 56.7 | 8.6 | 46.1 | 48.7 | 2.6 | 6.0 (-3.3;15.4) | 0.21 |
female | 4,675 | 43.2 | 43.3 | -0.1 | 36.6 | 42.1 | 5.5 | -5.4 (-14.4;3.7) | 0.24 |
male | 4,164 | 13.2 | 14.7 | 1.5 | 9.3 | 10.7 | 1.4 | 0.1 (-5.7;6.0) | 0.97 |
female | 4,675 | 16.9 | 15.0 | -1.9 | 11.7 | 14.3 | 2.6 | -4.4 (-10.8;2.0) | 0.18 |
male | 4,629 | 40.7 | 36.8 | -3.9 | 38.9 | 38.4 | -0.5 | -3.5 (-12.0;5.0) | 0.41 |
female | 5,260 | 26.4 | 26.7 | 0.3 | 30.7 | 27.5 | -3.2 | 3.5 (-4.3;11.3) | 0.38 |
a Reference category
b The n is the sum of all four groups used in the analysis
c Difference in Difference (Confidence Intervals)
Additionally, we examined the DiD impact estimates separately for high and low intensity target districts, but found no significant effects of the regeneration programme in either of these two groups (
Finally, in order to examine whether effects would occur when only the later intervention years were considered, we additionally divided the intervention period into an early (2008–2011) and late (2012-2013/2014) impact period, but found no significant impact of the regeneration programme either (
We examined the changes in health and health related behaviour during and following the implementation of a Dutch urban regeneration programme in 40 deprived districts. Our longer follow-up time (up to 6.5 years) compared to earlier evaluations allowed us to examine the longer-term health effects of this programme. The DiD impact estimates were inconsistent and not statistically significant, i.e. we did not observe significantly different changes in health or health-related behaviour in the 6.5 years after the start of the programme in the target districts compared to the control districts. There was no significant or consistently different pattern in the DiD impact estimates between men and women. Furthermore, DiD estimates were not consistently larger in target districts with a more intensive intervention programme.
Several limitations of our study should be mentioned before we discuss the findings. Although we used propensity score matching (PSM) to select our control areas, the target and control districts still differed in terms of their deprivation levels. Because the Dutch District Approach aimed to address the most deprived areas in the Netherlands, it was difficult to find equally deprived areas for our control group. On average, it is fair to state that the control districts were less deprived and the problems in these districts were less severe than in the target districts. Dissimilarity of our control areas could have biased our findings to some extent.
Because of the relatively small number of respondents in our study, we were limited in the number of individual characteristics we could use for stratification at the individual level. This did not seem to play a major role, however. Sensitivity analyses showed that controlling for these individual characteristics had only marginal effect on our estimates, suggesting that stratification on all potentially relevant individual characteristics would have had an equally limited impact on the estimates.
During our study period, social and physical interventions have most likely been implemented in the control districts as well. These interventions could have contributed to an underestimation of the intervention effect as it affected our case-control design. However, the investments in the 40 target districts were considerably larger and more comprehensive than the activities implemented concurrently in other deprived districts [
Our study comprised the adult population; not children or adolescents. This is unfortunate, since a substantial part of the interventions included the school environment and school system. We could not evaluate the possibility that the regeneration programme impacted the health of the younger population positively.
Finally, in our study design we excluded respondents who moved into the target and control districts after 2008 to prevent that potential positive health effects are caused by the inward migration of healthier residents. As a result, mobility patterns inwards could not have affected our findings. Outward migration, on the other hand, could have. For the individuals in the period 2006–2008, 18.2% of all individuals in control districts and 19.5% of all individuals in target districts have moved out at some point during 2009–2014. Outward migration to dissimilar districts (that is, individuals from target districts moving to non-target districts and individuals from control districts moving to non-control districts) could have affected our results, as it might decrease the actual difference between the two groups in exposure to the intervention. In that case, the health impact of the Dutch District Approach would be underestimated. However, such underestimation is probably mitigated, because we applied a matching method across target and control districts, as well as over time, and we thus controlled as much as possible for any differences in population characteristics across districts and over time.
In the current study, with a maximum follow-up time of 6.5 years, we did not observe a health impact of the Dutch District Approach. Three reasons might possibly explain these findings.
Firstly, the programme duration was relatively short, whilst these types of intervention programmes may need considerable time and funding to become effective [
Secondly, health was not one of the five goals defined at the start of the programme. We hypothesised that the health of the residents could improve over time, because this regeneration programme addressed various social determinants of health. However, maybe health goals should be made explicit in this kind of complex regeneration programmes to make sure that the interventions and activities selected to improve any of the themes of the District Approach will benefit (or at least not harm) health and related behaviour. A Health-in-All-Policies approach (HiAP), where health is an explicit aim within other policy fields, may have led to other activities and maybe other results.
Thirdly, the interventions that were implemented in the four years of the Dutch District Approach were perhaps not intensive enough to induce a noticeable health impact. A Dutch study that evaluated the programme’s primary outcomes reported that changes in liveability, social cohesion, safety, and social mobility in the 40 target districts were generally similar to comparably deprived control districts [
We did not find health effects of the Dutch District Approach at the area level, but this does not mean that there were no health benefits among some individuals. Our analyses of the treatment group (the target districts) included all respondents living in the districts, instead of only those residents who may have benefitted from activities specifically targeted to them. The evaluation of the New Deal for Communities (NDC) in the UK found no substantial effects in community outcomes at the area level. However, residents who participated in NDC activities reported greater changes in community outcomes than those who did not get involved [
We found no evidence that the Dutch District Approach had an impact in the longer run on self-reported health and related behaviour at the area level.
(DOCX)
(DOCX)
(DOCX)
(DOCX)
We are grateful to Statistics Netherlands for providing access to the data. We thank Danielle Kramer for her help with the preparation of the data on physical activity.