/I]\*SSH7SYVGIW/LE]IPMXWLE

**Figure 6.** Household sources of key foods in Khayelitsha—numbers on the right reflect total counts of households consuming each food.

**Figure 7.** Household sources of key foods in Ahodwo—numbers on the right reflect total counts of households consuming each food.

#### *3.2. Neighbourhood Food Environment*

The Ahodwo sample frame contained a far larger number of both food retail and food service outlets than the Khayelitsha sample frame due to the larger geographical area (Table 5). However, food retail outlets are represented at similar densities of approximately 39/km<sup>2</sup> in Khayelitsha and 41/km2 in Ahodwo. Food service outlets are represented slightly more densely in Ahodwo, with 32/km2 as opposed to 23/km2 in Khayelitsha. The proportions of the different outlet classes are similar in both sites. Ahodwo has more outlets per square kilometre, but in Khayelitsha the proportions of outlets providing access to obesogenic as well as those providing protective food are larger (Figure 8). The difference in distribution between the two sites is statistically significant at *p* = 0.0001. This indicates that Khayelitsha residents have fewer outlets to choose from, and that, of those, more than half (57%) are risky, as opposed to one in three in Ahodwo (39%). However, Ahodwo residents appear to have poorer aggregate access to protective foods (16% vs. 43%).

**Table 5.** Number and composition of food outlets.


#### 6IWIEVGL7MXI %LSH[S /LE]IPMXWLE 0S[VMWOERHTVSXIGXMZI 0S[VMWO ,MKLVMWOERHTVSXIGXMZI ,MKLVMWO

#### \*SSHSYXPIXWTIVWUYEVIOMPSQIXVIF]SFIWMX]VMWO

63\*)%LSH[SERH/LE]IPMXWLET!

**Figure 8.** Food outlets per square kilometre by risk and protection class—numbers reflect class and aggregate totals.

The distribution of household and neighbourhood food environment classes show a strong (r = 0.862) positive correlation in Ahodwo, while the correlation in Khayelitsha is less strong (r = 0.624), suggesting confounding variables. Poverty was identified as the most likely confounding variable. Income levels for both populations were low, with 49% in Khayelitsha earning below R3000/month (187 Euro), and 37% earning less than 964GHc/month (166 Euro) in Ahodwo. However, the 2016 upper-bound poverty line (UBPL) in Ghana was 1314 GHS [114], thus 47% of respondent households can be considered income-poor. In Khayelitsha, 70% of households reported aggregate incomes below the South African 2018 UBPL of R1183/person [115]. Moreover, the PACSA annual food price report of 2017 indicates that the cost of a basic basket of food to feed a household of seven persons comes to R1912.98, thus roughly half that to feed the average Khayelitsha household. Consequently, about half of the population in the Khayelitsha sample would have to spend more than a third of household income just to satisfy basic food needs [116].

Noting the prevalence of poverty in the research sites, additional analysis was done cross-tabulating risk classes with the experience of income poverty documented by the Lived Poverty Index component of the survey [24]. Lack of access to cash incomes was a key issue (46% in Khayelitsha and 20% in Ahodwo). Chi-test analysis yielded statistically significant (*p* = 0.005) differences in dietary risk distribution between households in Khayelitsha experiencing income deprivation and those who did not. Eighteen per cent fewer poor households exceeded obesogenic food consumption cutoffs than those who had not reported income deprivation. Cutoffs for protective food consumption were met by slightly fewer (6%) income-deprived households than by non-deprived, likely because consumption levels were already low. This indicates that households experiencing income deprivation reduce consumption of obesogenic and, to a lesser extent, protective foods in favour of dietary staples. The Ahodwo sample showed a slight increase in consumption of protective foods among poor households (7%) and a slight decrease in consumption of

obesogenic. However, the association was not statistically significant (*p* = 0.54), suggesting that income poverty may have a more ambiguous impact on household food provisioning in Ahodwo.

#### *3.3. Mapping Obesogenic Food Provision and Consumption*

The following section presents maps reflecting food provision and consumption patterns. The food provision maps reflect the distribution of outlets belonging to the different risk classes. These are each juxtaposed with maps reflecting the distribution of household food environments in terms of the aggregate risk index.

#### 3.3.1. Food Provision and Consumption—Ahodwo

The overview of food provision outlets in Ahodwo reveals a large number of outlets, concentrated along major roads and densely clustered around key intersections (Figure 9).

The Aggregate Consumption Index Map for Ahodwo (Figure 10) reveals no definite spatial clustering or pattern except for an apparent concentration of low risk, high protection diets around Kufuor I and Kufuor II streets and near the outlet clusters in the Northeast and Northwest corners of the sample frame (circled in red). The map reflects widespread prevalence of low-protection diets, with a fairly even distribution of low risk and high risk diets.

**Figure 9.** Ahodwo—food provision.

**Figure 10.** Ahodwo—aggregate food consumption.

3.3.2. Food Provision and Consumption—Khayelitsha

The food provision outlet map of Khayelitsha Site B (Figure 11) reveals a fairly low density of outlets, most clustered around the Nonkqubela shopping mall.

**Figure 11.** Khayelitsha Site B—Food provision.

The Makhaza area (Figure 12) shows dense clustering around the Makhaza shopping mall area, and a concentration of informal outlets along Ntlazane road in the informal settlement of Enkanini.

**Figure 12.** Makhaza-Enkanini—food provision.

In neither of these transects is there a clear spatial clustering of different food consumption risk categories (Figures 13 and 14).

**Figure 13.** Map 7: Khayelitsha Site B—aggregate food consumption.

**Figure 14.** Map 8: Makhaza-Enkanini—aggregate food consumption.

#### *3.4. Summary*

In summary, household consumption patterns revealed that a large proportion of the Khayelitsha household food environments were high risk (i.e., frequent consumption of ultra-processed and obesogenic foods—71%), but with low consumption of protective foods (only 16% meeting the threshold). The most commonly-consumed obesogenic foods in Khayelitsha were industrially produced bread, processed meat, sugar-sweetened beverages, and sugar (typically in hot beverages and added to porridge). These foods can be interpreted as a response to the poverty experienced, as they are typically cheap [9,117,118], and do not require much preparation. The most commonly consumed protective foods were cooked vegetables.

Ahodwo household food environments differed slightly: Only 26% of the respondents reported risky diets, and only 23% met cutoffs for protective food consumption. While commercial bread and sugar-sweetened beverages were also frequently consumed, consumption of processed meat was less prevalent and confectionery was instead eaten more widely. Sugar was also among the obesogenic foods consumed regularly in Ahodwo. Protective foods commonly eaten included fish, cooked vegetables, as well as a slightly higher intake of fresh fruit and vegetables than in Khayelitsha (Figure 5). Consumption of low-risk, vulnerable diets is clearly higher, indicating a larger segment of the population relies on staple foods sourced from convenience stores and open-air vendors, while lacking other foods rich in micronutrients including fruits, vegetables and legumes.

Respondents' reports of their usual food sources show that in Khayelitsha, supermarkets play a key role in providing access to obesogenic foods. Roadside stalls, although often selling sweets on the side, are important sources of fresh produce in Khayelitsha. This confirms earlier AFSUN findings [37,78]. However, supermarkets are less important as a source of sugar-sweetened beverages. Instead, large proportions of the population access obesogenic foods such as commercial bread and sugar-sweetened beverages from small shops. People frequently access protective foods through roadside stalls and through local markets.

The spatial patterning of the two research sites appears slightly different. In the more spatially extensive Ahodwo site, there was a higher number of food outlets which seem evenly dispersed except for dense clusters at busy intersections and markets. This may reflect the market culture of Kumasi, where outlets agglomerate in particular areas. By contrast, the transects surveyed in Khayelitsha suggest that in Site B food outlets are spatially clustered near the new malls and supermarkets, often close to public transport nodes and interchanges. In the Enkanini-Makhaza transect, there also appears to be a clustering of outlets along main access roads in the informal Enkanini settlement, providing convenient food access to residents far from the Makhaza mall. In Khayelitsha, the Site B and Makhaza malls are hotspots of obesogenic food provision, but also attract fruit and vegetable traders who provide access to healthier options. Mapping of aggregate food consumption risk indicated no obvious spatial clustering of risky household food intake in any of the research sites, nor any clear spatial relationship to food outlet location.

Outlets offering a variety of healthier food in Ahodwo were few and far between, and are clustered along Afua Ampomah street and Asante Frempong Avenue on the far eastern and western edge of the study area. However, many of these outlets also stock obesogenic food, and there is no obviously discernible pattern to the distribution of households consuming high levels of protective foods. In Khayelitsha, outlets providing access to healthier foods clustered around Site B mall, Ntlazane road and Makhaza mall. Supermarkets provide access to a range of healthier options, although of course they stock many obesogenic and ultra-processed foods, too.

#### **4. Discussion**

#### *4.1. Limitations*

Although the findings reveal information that speaks to previous research and has various implications for research and policy, the study has several limitations which constrain permissible inferences. The first is that approximately half of the food outlets mapped in Khayelitsha refused to participate in the survey (see below). In these cases, georeferences were recorded along with the store typology and a basic description (e.g., vegetable stall) to infer their level of nutritional risk. Secondly, the household survey did not document outshopping, i.e., that consumers travel to retail outlets outside of their local neighbourhoods to access food due to better prices, particular quality, or convenience along commuting routes. Thirdly, some informal outlets operate at times during which enumerators could not be in the field. Finally, the findings cannot be extrapolated as representative of broader consumption patterns in the immediate environments of the survey sample areas. Household scale of analysis means that internal dietary differences related to age, gender and power remain uncaptured in this particular analysis. Indicators of vulnerable household food environments are based on reported household food consumption, not individual food consumption, which will be published elsewhere.

#### *4.2. Key Insights*

The findings confirm the conclusions reached by several previous studies, namely that: (i) there is a co-existence of a diverse range of formal and informal food outlets [66,68,82,85,86]; (ii) obesogenic foods are widely prevalent and available [65,87]; (iii) supermarket expansion in particular is making ultra-processed and other obesogenic foods more accessible, although also offering access to a range of healthy options [25,38,46,85,87,117]; and (iv) there are high levels of consumption of obesogenic foods. This appears correlated with a local food geography which presents a large proportion of outlets with high prevalence of obesogenic foods.

Comparison also reveals important differences between the two sites. Both sites are urban and of comparable density, yet degrees of obesogenic risk in household and local food environments are

clearly different. **Firstly**, the overall number of outlets per square kilometre is slightly higher in Ahodwo than in Khayelitsha, offering consumers a greater range of conveniently-located options, particularly of prepared foods. **Secondly**, the proportion of outlets stocking a range of obesogenic foods appears higher in Khayelitsha. **Thirdly**, from a spatial perspective, food outlets in Khayelitsha are far more densely clustered around the supermarkets and transport nodes, which likely influenced the location of the supermarkets. By comparison, in Ahodwo, although there is also some clustering around key intersections and open-air markets, there is a far broader distribution of food outlets spatially, making foods more easily accessible. However, the correlation between distributions of consumption and provision is far stronger in Ahodwo. Despite far greater relative availability of protective foods in Khayelitsha than Ahodwo, the household food environments suggested lower levels of protective food consumption in Khayelitsha (23% Ahodwo; 16% Khayelitsha). This finding suggests that the local availability of protective foods alone plays only a partial role in promoting their consumption, and that other factors, such as cost, availability of refrigeration, and cost of cooking fuel [25,80,82,119–122], may constrain consumption of protective foods. In particular, poverty appears to play a strong role in influencing household food consumption in Khayelitsha. These findings have implications for the study of food environments, for our understanding of the nutrition transition and its drivers, and ultimately, for planning, policy and governance.

#### *4.3. Implications for Food Environment Research Methods*

The study results demonstrate that the widely accessible capabilities of smartphones, geo-location technologies, and online enumeration and data management technologies offer new opportunities to gather and evaluate data on informal food environments. However, the study also revealed limitations in the usefulness of such technology in areas of high poverty and inequality, where they present a safety risk to enumerators. By surveying two scales of food environment analysis (household and neighbourhood) and comparing emerging patterns, we were nevertheless able to identify key obesogenic and protective foods constituting household food environments. Moreover, statistical analysis based on the geographical density of outlet classes and the distribution of household risk classes revealed suggestive correlations and disjunctures. However, these quantitative and spatial perspectives should be complemented with qualitative and participatory approaches to document the experience of food environments by consumers and to interpret the findings of this survey.

#### *4.4. Implications for the Dietary Transition and Non-Communicable Disease*

The findings confirm that the dietary transition in LMICs [9] and Africa [13] is progressing apace, although further along in South Africa [14,15,26,91] than in West Africa [16]. Household food environments in Khayelitsha appear more severely obesogenic than in Ahodwo, mainly due to the higher consumption of risky foods and slightly lower consumption of protective foods. This appears to accord with the higher levels of obesity in the Western Cape than in Ashanti Region. The transition is often explained with reference to urbanisation and greater disposable incomes. Indeed, global sales of ultra-processed food correlate with higher levels of urbanisation and higher income countries. Nevertheless, sales growth of ultra-processed food over the last decades have been higher in low- and middle-income countries, compared with high-income ones [59]. The present study confirmed high levels of purchase and consumption of ultra-processed foods in urban settings in South Africa and Ghana. However, the higher prevalence of obesogenic food consumption in Khayelitsha in comparison with Ahodwo, where poverty is less extreme, calls the generic correlation of obesogenic food consumption with increased incomes into question. Nevertheless, within the Khayelitsha sample, households experiencing income deprivation consumed far less obesogenic foods. Therefore, while the availability of obesogenic foods in the Khayelitsha neighbourhood food environment appears to promote greater obesogenic household food

consumption despite deeper levels of poverty than in Ahodwo, within the Khayelitsha population, higher incomes do appear to be correlated with greater obesogenic food consumption. The contradictions noted above suggest that other factors play a key role in explaining the disjunctures noted above, including levels of poverty, the structure of the food economy, as well as the spatial patterning of urban settlements.

These data suggest that a cautious and nuanced discussion of the role of supermarkets and their interaction with the informal economy in Khayelitsha and Ahodwo is necessary. While they provide access to obesogenic foods and may out-compete some small shops, supermarkets also provide access to legumes, fruit and vegetables which are healthier foods and attract footfall, which in turn draws fruit and vegetable traders. Transnational supermarket and mall expansion into the Ahodwo area has begun, evidenced by the nearby construction of a modern shopping mall including a Shoprite supermarket. This confirms that the South African supermarket retail model is being exported to other countries in the continent [57,123]. This may cause a contraction in the informal economy which could affect livelihoods and increase the availability of ultra-processed and obesogenic foods while making healthier options less attractive and less conveniently available. However, the informal food economy has proved resilient and adaptive, developing a complementary relationship with formal outlets [66,68,124]. Small shops in Khayelitsha may have adapted by specialising in the provision of regularly-consumed obesogenic foods such as commercial bread, SSBs, and confectionery, as well as grocery hampers composed of staples (not considered high-risk in this study as they are dietary staples). In Ahodwo, however, small shops also are a key source of healthier options.

#### *4.5. Implications for Food Environments Theory*

These findings are relevant to a more fundamental inquiry concerning the balance of forces between local food environments and endogenous household drivers (poverty and other forms of disadvantage), themselves conditioned by larger systemic and structural dynamics. The lack of clear geographical clustering of different household risk classes in relation to the location of food outlets suggests that, at the neighbourhood level, the aggregate density and composition of outlet types is more relevant than their location. The above-mentioned contradictions raise the question whether the greater consumption of protective food in Ahodwo is a result of the lack of big corporate penetration—or simply a reflection that this urban landscape, with less poverty than Khayelitsha, is one in which local demands can be met despite lower aggregate availability?

It thus appears that in Khayelitsha, household incomes are a stronger determinant of obesogenic food consumption patterns than local food availability—while most households are poor, the poorest tend to reduce consumption of ultra-processed foods and protective foods—their obesity risk derives from a reduction of protective foods and increased reliance on energy-dense staples. Poverty in Khayelitsha is reinforced spatially by the dislocation of Khayelitsha as a remote peri-urban dormitory settlement, a legacy of apartheid-era spatial planning, which traps the urban poor in areas far from job opportunities [82,83,114]. By comparison, Ahodwo is located fairly centrally in the Southeast of Kumasi metropolis, with a ubiquitous and vibrant street economy, presumably presenting more opportunities for equitable participation. This suggests that it is not urbanisation per se which is the issue, but that the spatial forms of urban spaces and the economic opportunities they offer which are perhaps more important.

#### *4.6. Implications for Planning, Governance and Policy*

Spatial analysis has enabled the identification of hotspots of obesogenic food provision, particularly around malls in Khayelitsha and busy intersections in Ahodwo, providing urban governance actors with potential points of geographic focus and leverage. This means that urban planners should consider the impacts of shopping malls and their immediate food environments on local livelihoods that property

developers and commercial landlords be required to take into account the needs and opportunities presented by informal traders—especially fruit and vegetable stalls. Local government officials should not only adopt less obstructionist attitudes to street trade, but also consider how its role in providing access to fresh, whole foods can be supported in terms of land-use regulations, infrastructure and services. Interventions which make fresh, protective foods more cheaply and abundantly available to street traders may indirectly counter the tendency of the poorest households to economise by reducing their consumption of these foods. Infrastructure, regulation and social capital development supporting local aggregation and distribution and sale of fresh produce through cold chains, as well as the development of distributed micro-processing facilities may enhance the availability and lower costs of fresh and minimally-produced food.

The findings concerning small and informal shops suggest that particular attention should be paid to the regulation of upstream suppliers of obesogenic and ultra-processed foods as the small size, widespread distribution and large numbers of the retail outlets would make any form of direct regulation costly and logistically challenging. Although engagement with local trading associations may present opportunities to create awareness and develop adaptive multi-stakeholder governance approaches [97–99], the fractious nature of informal trade makes it challenging to find effective points of governance engagement and co-ordination [125]. Unless fresh, whole foods can be supplied more cheaply, it is therefore likely that traders will continue to respond to the demands of the urban poor for by providing cheap and convenient, but obesogenic foods. In Ahodwo, the development of large fresh-produce wholesale markets may support the provision of more affordable fresh produce through small shops and stalls.

The strong role which household poverty appears to play in constraining the consumption of protective foods suggests that regulatory intervention in local food environments alone may be of limited benefit unless accompanied by economic policy ensuring greater and more equitable economic participation, comprehensive social safety nets, and lower prices for fresh, healthy foods. At the local scale, the improvement and subsidisation of public transport could facilitate greater mobility essential to accessing economic opportunities.

#### **5. Conclusions**

The methodologies developed offer planners, activists and officials ways to visualise, engage with and interpret local food environments more concretely. The study reveals important insights into these particular food environments. The findings show that household food environments promoting obesity, more prevalent in Khayelitsha than in Ahodwo, appear correlated with neighbourhood food environments, which make obesogenic foods accessible and available, despite greater poverty in Khayelitsha. They also suggest that poverty is a powerful determinant not only of household consumption and purchasing but also of local food environments, thus suggesting a systemic feedback loop contrary to the direction of causality commonly implied in food environments theory. Making these "foodscapes" visible and legible may enable state and civil society agents to frame them as more concrete objects of local governance discourse. This is essential to galvanise the "will to transform" them [126,127]. In light of the above interpretation of the findings, however, governance of food environments may offer only limited leverage to address obesity in the face of systemic poverty and inequality. It cannot substitute for more fundamental engagement with socio-economic and spatial drivers of obesity which transcends a narrow focus on food.

**Author Contributions:** This paper is based on research conducted by the ROFE team. The authors listed made the following contributions: Conceptualisation, F.K., E.C.S., R.A.A., A.M.T., C.A., A.D.T., R.A. and D.S.; Data curation, F.K.; Formal analysis, F.K.; Funding acquisition, E.C.S., R.A.A., A.D.T. and D.S.; Investigation, F.K., E.C.S., R.A.A., C.A., L.N.E.A. and N.A.F.A.; Methodology, F.K., E.C.S., R.A.A., A.M.T., D.N., C.A., L.N.E.A., J.-C.M., R.A. and D.S.; Project administration, F.K., E.C.S. and L.N.E.A.; Resources, F.K.; Software, F.K. and A.M.T.; Supervision, E.C.S., R.A.A., A.M.T., J.-C.M. and D.S.; Validation, E.C.S., R.A.A., A.M.T., D.N., C.A., L.N.E.A., N.A.F.A., J.-C.M., R.A. and D.S.; Visualisation, F.K.; Writing—original draft, F.K.; and Writing—review and editing, E.C.S., R.A.A., A.M.T., D.N., L.N.E.A., N.A.F.A., J.-C.M., A.D.T. and D.S.

**Funding:** This research was supported by grant No. 108425-001 from the International Development Research Center, Canada. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. URL: https://www.idrc.ca/en/project/researching-obesogenic-urban-food-environment-its-driversand-potential-policy-levers-south.

**Acknowledgments:** The researchers would like to acknowledge the three anonymous reviewers, whose comments have helped refine this paper. We would also like to acknowledge the generous support of the DSI-NRF (Centre of Excellence in Food Security) Grant UID 91490.

**Conflicts of Interest:** The authors declare no conflict of interest. The funders had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript, or in the decision to publish the results.

#### **References**




mortality, cardiovascular disease, and type 2 diabetes: Systematic review and meta-analysis of observational studies. *BMJ* **2015**, *351*, h3978. [CrossRef]





© 2019 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).
