University of Groningen
Faculty of Economics and Business
The relationship between neighborhood
SES and health:
supermarkets as a potential mechanism
A thesis by Marjette Dreijer, S2906139 MSc student Business Administration: Health
Thesis supervisor: Jochen O. Mierau, PhD Co-assessor: Maarten J. Postma, PhD
Date: 17th of January, 2020 Word count: 6159
1
Abstract
Background. Many researchers have found a neighborhood socioeconomic gradient in
health. However, less is known about the mechanisms explaining this relationship. A potential mechanism in the socioeconomic gradient might be the availability of supermarkets. Previous literature shows that there exists a potential relationship between proximity to supermarkets and neighborhood health.
Objectives. To estimate the relationship between neighborhood socioeconomic status (SES)
and overweight and long-term conditions or illnesses in the Netherlands. Can the availability of supermarkets be a potential mechanism in this relationship?
Methods. Using 2016 data on Dutch neighborhoods, the relationship between neighborhood
SES and both overweight and long-term conditions or illnesses are evaluated. Classifying neighborhoods by the quintile of their SES, the differences in the prevalence of the health outcomes have been determined. In an additional model, the availability of supermarkets has been included. There has been controlled for the share of elderly and population density.
Results. A neighborhood socioeconomic gradient in overweight and long-term conditions or
illnesses was found. The data showed that low SES neighborhoods have higher availability of supermarkets. Nevertheless, the analyses indicated that the availability of supermarkets has a positive effect on health.
Conclusions. This study provides additional evidence that there exists a neighborhood
socioeconomic gradient in health. There is still unclarity on how supermarket availability mediates in this relationship. Overall, supermarket availability seems to have a positive effect on health. However, low SES neighborhoods, that have higher availability of supermarkets, do not seem to benefit from these health effects.
Implications. This study raises the question as to why low SES neighborhoods do no benefit
2
Introduction
In the last decades, researchers trying to explain health differences have realized that individual characteristics only cannot provide a full explanation. Spatial contexts, in particular neighborhoods, have been concepts of interest. The characteristics of a neighborhood possibly affect the health in that neighborhood.10 In a recent book of Diex Roux (2018), this relationship
between neighborhoods and health is captured. The book summarizes relevant research findings and methodological insights in this field.11 This study aims to further research the relationship between neighborhood characteristics and neighborhood health.
One neighborhood characteristic that has received particular attention is socioeconomic status (SES). Many researchers have found an inverse relationship between SES and health. This relationship, also referred to as the socioeconomic gradient in health, occurs at every level of SES.18,19 On the neighborhood level, such a relationship is found as well. Residents living in neighborhoods with low SES appear to have worse health than residents living in neighborhoods with higher SES. Associations between neighborhood SES and several adverse health outcomes have been found, such as mortality9, quality of life12, self-rated health29, and obesity2. Even when controlling for individual SES, health differences remain.31 This indicates that neighborhood SES indeed influences health. This study aims to further research the relationship between neighborhood SES and neighborhood health.
This study will contribute to the existing literature in several ways. Firstly, while most research in this area has been done in the United States, this study attempts to find whether the neighborhood socioeconomic gradient in health exists in the Netherlands as well. Secondly, the way of operationalizing in this study is unique. Overweight and the prevalence of long-term conditions or illnesses have been operationalized as health outcomes, which to the best of my knowledge have not been used in that combination before. Thirdly, this study addresses the potential mechanisms in the relationship between neighborhood SES and health. The availability of supermarkets is one of these potential mechanisms.
3 Literature review
Much earlier research has been done in an attempt to get a grasp on the relationship between neighborhoods and health. For example, by looking at previous relevant research, Diex Roux et al. (2010) summarized the processes through which neighborhoods affect health and health inequalities. They made a distinction between the physical and social environment, included behavior as mediators, and came to several mechanisms. For example, stress can lead to unhealthy behaviors (e.g. unhealthy diet), while in turn, unhealthy behaviors (physical inactivity) miss out on adverse effects on stress. Characteristics of the social and physical environment influence each other as well.10 More recently, Travert et al. (2019) attempted to unravel the interactions between the built environment and health behaviors. They developed an extensive model including personal factors, capability, opportunity, motivation, internal and external environment, and finally physical activity and diet behavior.27 These reviews made clear that the relationship between neighborhoods and health is complex.
While it seems to be clear that several mechanisms are working in the neighborhood socioeconomic gradient, there is still unclarity on which mechanisms this exactly are and what influence they have on the relationship. A better understanding of the mechanisms working in the gradient can contribute to determining the best health policy or intervention.
A potential mechanism in the relationship of neighborhood SES and health is the availability of supermarkets. Low SES neighborhoods seem to have lower access to supermarkets.17,21 Moreover, literature shows that there is a relationship between proximity to supermarkets and neighborhood health. Better availability of supermarkets can be related to healthier diets because supermarkets offer a high variety of healthful foods. Accordingly, better access to supermarkets is associated with lower levels of obesity17, Body Mass Index24, and overweight21. It should be noted that supermarkets in this case, and in the remainder of this thesis, mean big (chain) supermarkets. Convenience stores namely show a contradicting relationship. High access to convenience stores increases levels of obesity17, Body Mass Index24, and overweight21.
Leading in the literature on the relationship between supermarket availability and health is the United States. In the United States, there is spoken of so-called food deserts. Although there is a lack of consensus on the actual definition of food deserts, they are, in essence, poor urban areas with low availability of healthful food stores.16 For several reasons, including financial
4
Moreover, people tend to make food choices based on the availability of food outlets in their immediate neighborhood. In poor neighborhoods, food prices are higher, food quality is poorer, and are offered in a smaller quantity and variety. There is increased exposure to convenience stores and fast-food restaurants. People living in those poor urban neighborhoods thus buy their foods more frequently from unhealthy food outlets and have poorer diets.30
American policy has made this a health priority and attributed a lot of attention and money aimed at eliminating such food deserts. Through the Agricultural Act of 2014 and its precursors, almost $500 million is spent to improve access to healthful foods in food deserts since 2011.14 Moreover, food deserts have become central to health policy research.
The literature on the relationship between food deserts and health, however, is inconclusive. Early research showed that supermarket access is positively associated with fruit and vegetable consumption.17,25 Moreover, supermarket presence is supposed to be associated with a lower
prevalence of obesity and overweight.22 However, later studies, often using better designs, did
not find such associations. For example, the longitudinal study of Boone-Heinonen et al. (2011) found that supermarket availability was generally unrelated to diet quality and fruit and vegetable intake.4 Moreover, several other studies with a longitudinal design estimating the relationship between food desserts and diet and obesity found small or no effects.3, 8, 13,15 Among the elderly, food deserts do not substantially affect diet-related diseases such as diabetes, heart disease, and high blood pressure either.14
5
Based on the literature, this study will focus on the relationship between neighborhood SES and health, including supermarket availability as a potential mechanism. While findings on the effects of supermarket availability on health are inconclusive, there is enough evidence to believe it is a relevant factor in the relationship between neighborhood SES and health and its effect is worthwhile to research in the Netherlands as well. Moreover, interestingly, the Netherlands is a far more compact country than the United States. Actual food deserts are less likely to occur in the Netherlands. It might, therefore, not be surprising when the use of Dutch data results in different findings.
While previous researchers focused on the positive association between supermarket availability and health, there can also be reasoned in the other direction. It can be argued that too high availability of supermarkets can result in worse health. High exposure to supermarkets may result in people buying more, excessive, foods.
Methods and data
In this study, we focus on the relationship between neighborhood SES and health, operationalizing overweight and long-term conditions or illnesses as health outcomes. We will examine the effect of the availability of supermarkets as a potential mechanism in this relationship. For this, the following data and analyses are needed.
Data
In my analysis, data aggregated on the neighborhood level is used. Data on the dependent variables, long-term illnesses or conditions and overweight, are derived from the Adult Health Monitor 2016 from GGDs, CBS, and RIVM. For long-term conditions, the over 457,000 respondents were asked the question of whether they have one or more long-term illnesses or conditions. The data is given as the percentage of these respondents in the neighborhood that has answered ‘yes’ to this question. Long-term is considered (expected) 6 months or more. Overweight is defined as the percentage of persons with a BMI of 25.0 kg/m2 or higher. This is based on self-reported height and weight.
6
Statistics Netherlands also provides data on neighborhood proximity of facilities. The 2016 version of these numbers is used for the proximity to supermarkets. Proximity is given both as the average distance of every resident in the neighborhood to the closest facility, as the average amount of the facility within a fixed distance (1 km, 3 km, and 5 km) for all residents of the neighborhood. Distances are calculated by road. The different measures have been tested. The methods gave different results, so they all have been included and compared in the analysis. Population density, a potential mediator of the association between supermarkets and health, is measured as the average amount of addresses per km2 within a circle of a 1 km radius. Definitions of all the variables are summarized in Table 1 of the Appendix.
Table 1 below shows some descriptive statistics of the data. It shows the averages and standard deviations of all the variables for each SES quintile separately and in total. It can be seen that for both health outcomes, prevalence decreases for every higher level of SES. Moreover, noticeable is that the availability of supermarkets is highest in the lowest SES neighborhoods. The availability seems to gradually decrease for every higher level of SES, although the number of supermarkets within the bigger distances is again higher at the highest level of SES. This might be because most supermarkets are located in urban areas. From the population density can be observed that urbanity is higher in low SES neighborhoods. Moreover, high SES neighborhoods may be more located in suburban areas.
In total, 12.922 Dutch neighborhoods exist. After removing 3110 observations with missing values, data on 9812 neighborhoods were left.
Table 1. Descriptive statistics (N = 9812)
Mean (SD)
Variable SES 1 SES 2 SES 3 SES 4 SES 5 Total
7 Analysis
For the analyses, the data was first ordered according to SES and divided into roughly equal-sized quintiles (with 1 being the lowest SES and 5 being the highest SES). For a direct overview of the differences between different levels of SES, means and standard deviations for all variables are calculated for the different SES quintiles.
Then, there has been taken a deeper look at how the health outcomes differ between groups of SES. Firstly, this has been visualized with box plots. The box plots represent the distribution (minimum and maximum value, median, and 25th and 75th centiles) for every group of SES separately. Thereafter, linear regression analyses are run. The dependent variables of interest are overweight and long-term conditions or illnesses. For both dependent variables, separate regression analyses are done. Since both overweight and long-term conditions or illnesses are considerably more frequent among older people, there has been controlled for the share of elderly. SES is included as an independent variable, whereby the different SES quintiles are used. The middle quintile is used as a reference category. This way, the regressions show clearly how the low and high SES neighborhoods relate to the average neighborhoods.
After this, additional analyses have been done to measure the effect of the availability of supermarkets. For this, the proximity variables of supermarkets have been added to the same model. Different regressions have been done for every measure of proximity. Thus, distance to the closest store and the number of stores within 1 km, 3 km, and 5 km have all been tested. Availability of stores is not only dependent on the proximity, but also on the number of residents surrounding the stores. Therefore, there has been controlled for population density. Using these independent and control variables, the regression has been run with the third quintile of SES as a reference again.
The analyses were performed using Stata/SE version 15.0 (StataCorp. 2017. Stata Statistical Software: Release 15. College Station, TX: StataCorp LLC.).
Results
Using the methods and analyses described in the previous section, the results discussed in the following section were found.
8
quite some variation. Some low SES neighborhoods have extraordinary good health, while the highest SES neighborhoods can still have poor health. Noticeable is that for overweight, variation clearly stems from the lower centiles, while for long-term conditions or illnesses no such clear distinction can be made. Moreover, this within-group variability increases for every lower level of SES.
A B
Figure 1. Boxplots by SES for (A) overweight and (B) long-term conditions or illnesses
Now that we know how our data is distributed, we will have a look at the regressions. Table 2 shows the results of the regression analyses. As mentioned earlier, the third quintile has been used as a reference level. The interpretation of the coefficients is therefore straightforward. Positive estimates indicate worse health and negative estimates indicate better health. As expected, the lower levels of SES show worse health and the higher levels of SES show better health than the middle category. The first column shows the results for overweight. Here, in particular, the highest quintile highly relates to less overweight. For long-term conditions or illnesses, given in the second column, it seems that the lowest quintile especially relates to increased prevalence. Overall, the effect of SES on health seems to be higher for long-term conditions or illnesses than for overweight.
These results show that there is indeed a relationship between neighborhood SES and health. As identified earlier, a potential mechanism in this relationship is the availability of supermarkets. Therefore, additionally, the effect of the availability of supermarkets has been estimated by adding this variable to the model. The results of these analyses will be evaluated next.
Table 1 showed, contradicting previous literature, that supermarkets are closer in low SES neighborhoods than in high SES neighborhoods. A possible explanation for this could be that,
9
as opposed to the United States, in the Netherlands urban areas have more supermarkets. We already observed that population density is indeed substantially higher in low SES neighborhoods, and are thus most likely to be located in urban areas. In the regressions, there is controlled for population density. The results of the regression analyses will be evaluated next.
The results of the analyses including supermarkets in the regression are given in Table 2. Results of the different proximity measures have been included. The socioeconomic gradient can still be clearly observed for every regression. For overweight, the gradient is even greater than at the main analysis. For long-term conditions or illnesses, the gradient seems to be slightly smaller.
Particularly interesting in this analysis, however, is the supermarket parameter. Comparing the parameters of the different regressions results in some interesting findings. Firstly, the distance to the closest store has a significant positive association with overweight. Thus, neighborhoods with a supermarket closer to the residences have less overweight. Contradicting, long-term conditions or illnesses show a significant negative association with the distance to the closest store. Surprisingly, this would mean that having a supermarket closer in a neighborhood results in more long-term conditions or illnesses.
For the number of stores in a neighborhood, negative associations are found for both overweight and long-term conditions or illnesses. This indicates that having more supermarkets in a neighborhood results in better health. These associations decrease for every higher distance. Thus, the associations are strongest for the number of stores within 1 km and weakest for the number of stores within 5 km. For long-term conditions or illnesses, the number of stores within 5 km results in a non-significant association. Overall, the associations are stronger for overweight than for long-term conditions or illnesses.
Table 2. Association between SES and overweight and long-term conditions or illnesses
Main analysis Distance to the closest store Number of stores within 1 km Overweight
Long-term
condition or illness Overweight
Long-term
condition or illness Overweight
10
(0.1599) (0.0990) (0.1342) (0.0980) (0.1328) (0.0972) Share of elderly 0.191*** 0.318*** 0.148*** 0.325*** 0.149*** 0.332***
(0.0056) (0.0035) (0.0047) (0.0034) (0.0047) (0.0034) Population density N/A N/A -0.002*** 0.000*** -0.0016*** 0.001***
(0.0000) (0.0000) (0.0000) (0.0000) Supermarkets N/A N/A 0.132*** -0.245*** -0.217*** -0.187***
(0.0394) (0.0287) (0.0405) (0.0296) Number of stores within 3 km Number of stores within 5 km
Overweight
Long-term
condition or illness Overweight
Long-term condition or illness SES 1 4.126*** 5.023*** 4.056*** 5.019*** (0.1321) (0.0976) (0.1326) (0.0980) SES 2 1.609*** 1.840*** 1.562*** 1.842*** (0.1309) (0.0967) (0.1313) (0.0971) SES 3 Reference Reference Reference Reference SES 4 -1.745*** -1.591*** -1.758*** -1.640*** (0.1308) (0.0967) (0.1312) (0.0970) SES 5 -4.273*** -3.223*** -4.308*** -3.372*** (0.1339) (0.0989) (0.1346) (0.0995) Share of elderly 0.140*** 0.326*** 0.139*** 0.328*** (0.0046) (0.0034) (0.0046) (0.0034) Population density 0.001*** 0.001*** -0.001*** 0.000*** (0.0001) (0.0001) (0.0001) (0.0000) Supermarkets -0.155*** -0.061*** -0.057*** -0.004 (0.0098) (0.0072) (0.0041) (0.0030)
Discussion and limitations
In this thesis, a clear neighborhood socioeconomic gradient in health was found. With Dutch neighborhood data, we found that SES is negatively associated with both overweight and long-term conditions or illnesses, controlling for the share of elderly. This association is consistent for every quintile of SES. These findings confirm the negative relationship between neighborhood SES and health that others have found in other countries as well.
11
Furthermore, interestingly, the Netherlands differ substantially in structure and characteristics from the United States.
Indeed, the data in this study showed to differ from the findings of other researchers. In this study, it appeared that in the Netherlands, low SES neighborhoods have higher availability of supermarkets. Since the low SES neighborhoods do have worse health, this raises the question of whether too high availability of supermarkets may result in worse health. Yet, the analyses in this study indicated that the availability of more supermarkets in a neighborhood is positively associated with health. While this is in line with the findings of many previous researchers, given that in this study low SES neighborhoods have higher availability of supermarkets, this is a surprising result.
The availability of supermarkets thus does not seem to explain the socioeconomic gradient. Supermarkets might not be seen as a mechanism but rather as a health factor. Taking into account the environmental factors of population density and the share of elderly, the availability of more supermarkets in a neighborhood has a positive effect on health. However, the effects of the closest supermarket in a neighborhood are ambiguous.
Limitations
12
Secondly, this study lacks causal inference. I have only researched the associations between neighborhood SES and health. Because cross-sectional data is used, no conclusions can be drawn on the causation in this relationship. However, Hill’s criteria can indicate the likelihood of causal inference.26 Applying these criteria to my findings can thus provide some insights concerning causality. One of these criteria is the strength of the associations, which can be indicated by the R-value. Thus, I determined the R-values of all associations, which are given in Table 3 of the Appendix. It can be concluded that the associations are quite strong. Because the findings can be considered consistent and coherent as well, a causal relationship is plausible. However, to be able to draw more concrete conclusions on causal inference, a longitudinal study is needed.
Lastly, long-term conditions or illnesses were self-reported and therefore prone to reporting bias. Overall, self-reported health results in reasonably accurate estimates, but the validity varies for different conditions.20,23 Data which was verified by medical professionals might
have resulted in more accurate estimates.
Implications
This study implies that the availability of supermarkets does not seem to explain the neighborhood socioeconomic gradient in health. Future research should continue to focus on the mechanisms behind the socioeconomic gradient. Other relevant characteristics of neighborhoods should be identified.
Moreover, more clarity is needed on the effect of the availability of supermarkets in a neighborhood on health. For this, more information on supermarkets is needed. This study found a significant positive effect of supermarket availability on health. While low SES neighborhoods have more availability of supermarkets, they have worse health than higher SES neighborhoods. Perhaps this could be explained by differences in the supermarkets. Neighborhoods with different SES might have different kinds of supermarkets, such that low SES neighborhoods do not experience the positive health effects high SES neighborhoods experience.
13
drinks and fewer fruits and vegetables than in the wealthiest neighborhoods.5 This could be true for the Netherlands as well, which would influence the way supermarkets behave in relationship to SES and health. If the supermarkets in low SES neighborhoods offer less healthy foods, it could explain why high supermarket availability in low SES neighborhoods does not improve health in those neighborhoods. Thus, the groceries available in supermarkets can be a potential expansion to the research of supermarket availability as a mechanism within the neighborhood socioeconomic gradient.
Furthermore, it might be interesting to explore more aspects of the food environment beyond supermarket availability. Recent studies in the United States have focused on food swamps rather than food deserts. Food swamps have been described as areas with a high-density of establishments selling high-calorie fast food and junk food, relative to healthier food options. Cooksey-Stowers et al. (2017) found that food swamps are a better predictor of obesity than food deserts.6 Thus, future research could focus on such high-calorie selling establishments,
such as fast-food restaurants and convenience stores. The availability of such stores in the neighborhoods might behave as a mechanism behind the neighborhood socioeconomic gradient, perhaps in relation to supermarket availability. A follow-up study could include these factors in the analyses to better estimate the impact of the food environment on the relationship between neighborhood SES and health in the Netherlands.
14
References
1. Allcott, H., Diamond, R., Dubé, J.-P., Handbury, J., Rahkovsky, I., & Schnell, M. (2019). Food deserts and the causes of nutritional inequality. The Quarterly Journal of
Economics, 134 (4), 1793-1844.
2. Black, J. L. & Macinko, J. (2008). Neighborhoods and obesity. Nutrition Reviews, 66 (1), 2–20.
3. Block, J. P., Christakis, N. A., O’Malley, A. J., & Subramanian, S. V. (2011). Proximity to food establishments and body mass index in the Framingham heart study offspring cohort over 30 years. American Journal of Epidemiology, 174 (10), 1108-1114. 4. Boone-Heinonen, J., Gordon-Larsen, P., Kiefe, C. I., Shikany, J. M., Lewis, C. E., &
Popkin, B. M. (2011). Fast food restaurants and food stores: longitudinal associations with diet in young to middle-aged adults: the CARDIA study. Archives of Internal Medicine, 171 (13), 1162–1170.
5. Cameron, A. J., Thornton, L. E., McNaughton, S. A., & Crawford, D. (2013). Variation in supermarket exposure to energy‐dense snack foods by socio‐economic position. Public
health nutrition, 16 (7), 1178‐1185.
6. Cooksey-Stowers, K., Schwartz, M. B., & Brownell, K. D. (2017). Food swamps predict obesity rates better than food deserts in the United States. International Journal of
Environmental Research and Public Health, 14 (11), 1366.
7. Cummins, S., Findlay, A., Petticrew, M., & Sparks, L. (2005). Healthy cities: the impact of food retail-led regeneration on food access, choice and retail structure. Built
Environment, 31 (4), 288-301.
8. Cummins, S., Flint, E., & Matthews, S.A. (2014). New neighborhood grocery store increased awareness of food access but did not alter dietary habits or obesity. Health
Affairs, 33 (2), 283–291.
9. Daly, M. C., Duncan, G. J., McDonough, P., & Williams, D. R. (2002). Optimal indicators of socioeconomic status for health research. American Journal of Public
Health, 92 (7), 1151–1157.
10. Diez Roux, A. V., & Mair, C. (2010). Neighborhoods and health. Annals of the New York
15
11. Diez Roux, A. V. (2018). Neighborhoods and health (2nd ed.). New York, United States:
Oxford University Press.
12. Drukker, M., & van Os, J. (2003). Mediators of neighbourhood socioeconomic
deprivation and quality of life. Social Psychiatry and Psychiatric Epidemiology, 38 (12), 698–706.
13. Dubowitz, T., Ghosh-Dastidar, M., Cohen, D. A., Beckman, R., Steiner, E. D., Hunter, G. P., Florez, K. R., Huang, C., Vaughan, C. A., & Sloan, J. C. (2015). Diet and perceptions change with supermarket introduction in a food desert, but not because of supermarket use. Health Affairs, 34 (11), 1858–1868.
14. Fitzpatrick, K., Greenhalgh-Stanley, N., & Ver Ploeg, M. (2016). The impact of food deserts on food insufficiency and snap participation among the elderly. American Journal
of Agricultural Economics, 98 (1), 19–40.
15. Handbury, J., Rahkovsky, I., & Schnell, M. (2015). Is the focus on food deserts fruitless? Retail access and food purchases across the socioeconomic spectrum. National Bureau of
Economic Research. NBER Working Paper No. 21126.
16. Hendrickson, D., Smith, C., & Eikenberry, N. (2004). Fruit and vegetable access in four low-income food deserts communities in Minnesota. Agriculture and Human Values, 23 (3), 371–383.
17. Larson, N., Story, M., & Nelson, M. (2009). Neighborhood environments: disparities in access to healthy foods in the U.S. American Journal of Preventive Medicine, 36 (1), 74– 81.
18. Marmot, M. G., Shipley, M., & Rose, G. (1984). Inequalities in death: specific explanations of a general pattern? Lancet, 323 (8384), 1003–1006.
19. Marmot, M. G., Smith, G. D., Stansfeld, S., Patel, C., North, F., Head, J., White, I., Brunner, E., & Feeney, A. (1991). Health inequalities among British civil servants: the Whitehall II study. Lancet, 337 (8754), 1387–1393.
20. Martin, L. M., Leff, M., Calonge, N., Garrett, C., & Nelson, D. E. (2000). Validation of self-reported chronic conditions and health services in a managed care population.
16
21. Moore, L., Roux, A. D., Nettleton, J., & Jacobs, D. (2008). Associations of the local food environment with diet quality—a comparison of assessments based on surveys and geographic information systems: the multi-ethnic study of atherosclerosis. American
Journal of Epidemiology, 167 (8), 917–924.
22. Morland, K., Diez Roux, A. V., & Wing, S. (2006). Supermarkets, other food stores, and obesity: the atherosclerosis risk in communities study. American Journal of Preventive
Medicine, 30 (4), 333–339.
23. Okura, Y., Urban, L. H., Mahoney, D. W., Jacobsen, S. J., & Rodeheffer, R. J. (2004). Agreement between self-report questionnaires and medical record data was substantial for diabetes, hypertension, myocardial infarction and stroke but not for heart failure. Journal
of Clinical Epidemiology, 57 (10), 1096–1103.
24. Powell, L. M., Auld, M. C., Chaloupka, F. J., O'Malley, P. M., & Johnston, L. D. (2007). Associations between access to food stores and adolescent body mass index. American
Journal of Preventive Medicine, 33 (4), 301–307.
25. Rose, D., & Richards, R. (2004). Food store access and household fruit and vegetable use among participants in the US Food Stamp Program. Public Health Nutrition, 7 (8), 1081-1088.
26. Rothman, K. J., & Greenland, S. (2005). Hill’s criteria for causality. In P. Armitage & T. Colton (Eds.), Encyclopedia of biostatistics, 4 (2).
27. Travert, A.-S., Sidney Annerstedt, K., & Daivadanam, M. (2019). Built environment and health behaviors: deconstructing the black box of interactions—a review of reviews.
International Journal of Environmental Research and Public Health, 16 (8), 1454.
28. Ulrich, V., Hillier, A., & Isselmann DiSantis, K. (2015). The impact of a new nonprofit supermarket within an urban food desert on household food shopping. Medical Research
Archives, (3).
17
30. Walker, R. E., Keane, C. R., & Burke, J. G. (2010). Disparities and access to healthy food in the United States: A review of food deserts literature. Health and Place, 16 (5), 876– 884.
18 Appendix Table 1. Definitions Variable Definition Long-term condition or illness
Percentage saying to have one or more disease or condition expecting to take longer than six months
Overweight Percentage with a BMI of 25.0 kg/m2 or higher Social Economic Status Average real estate valuation (x1000 euros) Share of elderly Percentage of persons that is 65 years or older
Population density Average amount of addresses per km2 within a circle of a 1 km radius Supermarkets Store with multiple kinds of daily articles, and a surface of at least 150 m2
Table 2. Averages of variables of missing values and non-missing values
Variable Missing values Non-missing values
Number of business locations 43.23 137.03
Population density 485.71 1360.07
Number of households 27.33 773.96
Overweight 49.13 50.03
Long-term condition or illness 30.45 33.54
Table 3. R-values
Analysis Health outcome R-value
Main analysis Overweight 0.50
Long-term condition or illness
0.80
Distance to closest store Overweight 0.70
Long-term condition or illness
0.82
Number of stores within 1 km Overweight 0.70
Long-term condition or illness
0,82
Number of stores within 3 km Overweight 0.71
Long-term condition or illness
0.82
Number of stores within 5 km Overweight 0.70
Long-term condition or illness
