Do Food Deserts Cause Worse Health Outcomes in Scotland?

Introduction

Access to nutritious and affordable food is a fundamental determinant of health outcomes and mental well-being. However, food insecurity is ever increasing across the UK as supermarkets relocate further out of city centres, further stretching household budgets to accommodate these increased transport costs (Blake 2018).

Allcot et al. (2019) define a food desert as a region with low availability or high prices of healthy foods, arguing that it is important to define as food deserts may be directly responsible for unhealthy eating and poorer health outcomes.

This issue is of particular significance in Scotland, where diverse urban and socioeconomic landscapes exist (Walsh et al. 2010)

The central question is “Do food deserts cause worse health outcomes in Scotland?” which will be analysed using data from the SIMD and ONS Business Counts (used as a proxy for food deserts as a lower count of retail units will limit the availability of healthy foods) and applied using a multivariate regression model and a directed acyclic graph to understand the causal relationship of the variables involved.

Variable	Obs	Mean	Std. Dev.	Min	Max
CIF	7274	102.894	57.092	5	370
depress	7276	.191	.054	0	.472
retail	7277	3.479	3.581	0	30
deprived	7277	.241	.428	0	1
income rate	7274	.123	.096	0	.59
employment rate	7274	.096	.071	0	.47
no qualifications	7277	102.2	55.317	2.995	353.078
drive gp	7277	3.587	2.789	.56	89.377
drive retail	7277	5.387	6.633	.785	190

Table 1: Summary statistics of the variables. Grey rows feature the dependent variables, whilst all other rows feature the independent variables.

Fuchs (2008) asserts the concept that health is multifaceted, which is reflected through the DWP’s distinct definitions for these two different health variables. DWP (2005) provides definitions for the key variables in the analysis. CIF (Comparative Illness Factor) is a measure often used in impact assessments to quantify the societal or economic impacts (such as healthcare costs or loss of productivity) of ill-health which may be captured through the combined count of recipients on Disability Living Allowance, Attendance Allowance, Incapacity Benefit, and Severe Disablement Allowance (DWP 2005). Additionally, the DWP (2005) defines Depress as the proportion of the population being prescribed drugs for anxiety, depression, or psychosis.

Definitions for the independent variables are then given by SIMD (2020). The retail variable represents a count of non-specialised retail units within the data zone that sell food, beverages, and tobacco products. Deprived is a binary variable representing if the individual is within a 'most 25% deprived' area.

Income rate and employment rate are other measures of deprivation, which is captured by the Income Support and Job Seekers Allowance (or Employment Support Allowance if the individual is still on Legacy Benefits) elements of Universal Credit, respectively (SIMD 2020).

No Qualifications is a more straightforward variable, measuring if the individual is of working age, and left school with a lack of qualifications, this includes leaving without National awards (SIMD 2020).

Lastly, Drive GP and Drive Retail both represent drive times in minutes, to either a General Practitioner or a Retail Centre.

Model and Interpretation

Analysis of the question employs a multivariate regression model to assess the combined impact of several variables on health outcomes of interest.

The above equation reveals how these variables are modelled to jointly influence the health outcome (Y), offering a comprehensive understanding of the complex interplay. By controlling for multiple factors simultaneously, we can derive insights that go beyond the standard bivariate model, and provide a more a more nuanced perspective on the research question.

	(CIF)	(DEPRESS)
retail	-0.001	-0.000***
	(0.001)	(0.000)
income_rate	4.132***	-0.029
	(0.178)	(0.018)
employment_rate	4.593***	0.483***
	(0.219)	(0.023)
no_qualifications	0.005***	0.000***
	(0.000)	(0.000)
drive_gp	-0.010***	-0.001***
	(0.001)	(0.000)
drive_retail	-0.002***	-0.000
	(0.000)	(0.000)
deprived	0.014*	0.001
	(0.008)	(0.001)
_cons	-1.391***	0.130***
	(0.010)	(0.001)
R^2	0.92	0.66
N	7,274	7,274

Table 2: Regression results, * p<0.1; ** p<0.5; *** p<0.01

According to Table 2, the greatest predictor of being in receipt of the aforementioned benefits is being income or employment deprived, and this is captured by CIF, suggesting that 1 unit increase in income or employment deprivation increases CIF by 4.13% and 4.59%, respectively.

However, retail count is rejected due to a -value of -1.29, failing to pass -crit of ± 1.65, 1.96 or 2.33 at 10%, 5% or 1% significance levels, respectively.

A similar method is used to reject income rate, drive retail, and deprived for the Depress model.

As for Depress, employment deprivation appears to have the largest impact, with a 1 unit increase in employment deprivation leading to a 48% increase in Depress, suggesting a larger likelihood to be prescribed drugs relating to anxiety, depression, or psychosis.

  However, it is worth making note of the high correlation between income rate and employment rate.

Variables	(1)	(2)
(1) income_rate	1.000
(2) employment_rate	0.962	1.000

Table 3: Matrix of Correlations

Whilst there is notable correlation between these variables, they serve distinct purposes in our analysis. SIMD (2020) emphasises that the employment rate not only accounts for the unemployed, but also encompasses individuals who are inactive/discouraged, underemployed (those seeking additional hours), and those with qualifications and skills but struggle to secure suitable employment.

Income rate focuses on identifying individuals who face income deprivation relative to the wider population, captured through data for Universal Credit and Legacy Benefits (Employment Support Allowance and Job Seekers Allowance (SIMD 2020).

The R^2 values for both CIF and Depress are noteworthy, with CIF featuring 92% and Depress featuring 66%. These values indicate that CIF is well-explained by the chosen variables, while Depress, though well explained, falls short of reaching the same explanatory power.

The F-values follow a similar pattern with CIF registering a significant F-value of 8554.15 and Depress also demonstrating a significant F-value of 1557.40.

These values confirm the substantial explanatory power of both estimated models.

The rejection of retail in the CIF model and its exceptionally low coefficient in the Depress model serve as evidence that food deserts, on their own, do not lead to worse health outcomes. To achieve meaningful explanatory power, these variables must be considered in conjunction with other factors, such as income rate, employment rate, and no qualifications.

Directed Acyclic Graph

Figure 1: Directed Acyclic Graph describing the causal relationship between the variables

In the context of regression analysis, a directed acyclic graph (figure 1) was employed to explain causal relationships amongst the variables. However, Drive GP and Drive Retail were removed from this causal model as it became evident that the influence of these variables could be explained by other variables, such as lower income dictating where an individual may live.

This is especially the case with those who are in social and community housing, as this is typically in designated areas of local authorities (Scottish Government 2011).

The inclusion of retail as an exposure variable was driven by its relevance to the research question, as well as by Allcott’s (2019) arguing of food deserts being a directly responsible for poorer health outcomes.

No Qualifications also holds particular importance as it often serves as a foundational factor contributing to a lack of employment and income (Townsend 2009). It’s significance across multiple health variables further justifies its inclusion in the causal model as an exposure variable.

By incorporating retail and no qualifications as exposure variables, the model benefits from a comprehensive approach that considers both the specific aspects of the question, and the broader socioeconomic context.

Unobserved data is present across both models, this is evident in the R^2 variables as neither health variable is 100% explained.

This unobserved data can be explained by a multitude of factors, including an individual’s background. These backgrounds encompass elements such as parental income, education, and well-being which can influence the individual and what opportunities they may have (Walsh et al. 2010).  

Once more, it’s important to highlight that retail, serving as a proxy for food deserts, is not used in isolation to explain health outcomes, a method similar to that taken by Allcott et al. (2019), with their including of employment status and income in their model for analysing health outcomes. Regression results from Table 2 strongly indicate that this variable should be considered in conjunction with other variables to further understand the outcome.

Conclusion

The impact of food deserts and health outcomes is an incredibly complex and dynamic relationship, especially in Scotland with the varied socioeconomic landscape (Walsh et al. 2019). Whilst analysis showed that there may be a relationship between food deserts and those being prescribed drugs for anxiety, depression, and psychosis, other variables proved more effective at explaining a relationship, such as the impact of being income or employment deprived on the individual and their readiness to be captured by CIF data.

This underscores the importance of not solely relying on retail counts within this data, as it lacks the necessary explanatory power to support a direct causal relationship. Instead, a comprehensive model, similar to Alcott et al.’s (2019) model to analyse health outcomes, should be considered in conjunction with various relevant variables that can be identified as to having a causal relationship. 

References

Allcot, H., Diamond, R., Dubé, J.P., Handbury, J., Rahkovsky, I., Schnell, M. (2019) Food Deserts and the Cases of Nutriontial Inequality. The Quarterly Journal of Economics, 134 (4), pp. 1793-1844. Available: https://academic.oup.com/qje/article/134/4/1793/5492274?login=true [Accessed 6 November 2023].

Blake, M (2018) 1.2 million living in UK food deserts, study shows. University of Sheffield Faculty of Social Sciences, 16 October. Available: https://www.sheffield.ac.uk/social-sciences/news/12-million-living-uk-food-deserts-studys-shows#:~:text=1.2%20million%20people%20in%20the,neighbourhoods%20across%20the%%2020United%20Kingdom[Accessed 6 November 2023].

Department for Work and Pensions (2005) Scottish Index of Multiple Deprivation 2006 Technical Report. Scottish Executive. Available: http://doc.ukdataservice.ac.uk/doc/6870/mrdoc/pdf/6870technical_report_2006.pdf [Accessed 20 October 2023].

Fuchs, V. R. eds. (2008) Economic Aspects of Health. 2nd ed. Chicago: University of Chicago Press.

Scottish Government (2011) Independent Living, National Strategy for Older People. Health and Social Care. Available: https://www.gov.scot/policies/independent-living/national-strategy-for-older-people/ [Accessed 6 November 2023].

SIMD (2020) SIMD 2020 Technical Notes. National Statistics. Available: https://www.gov.scot/binaries/content/documents/govscot/publications/statistics/2020/09/simd-2020-technical-notes/documents/simd-2020-technical-notes/simd-2020-technical-notes/govscot%3Adocument/SIMD%2B2020%2Btechnical%2Bnotes.pdf [Accessed 20 October 2023].

Townsend, P. (2009) Deprivation. Journal of Social Policy, 16 (2). Available: https://www.cambridge.org/core/journals/journal-of-social-policy/article/deprivation/071B5D2C0917B508551AC72D941D6054 [Accessed 6 November 2023].

Walsh, D., Bendel, N., Jones, R., Hanlon, P. (2010) It’s not ‘just deprivation’: Why do equally deprived UK cities experience different health outcomes? Public Health, 124 (9), pp. 487 – 495. Available: https://www.sciencedirect.com/science/article/abs/pii/S0033350610000338 [Accessed 6 November 2023].