Planning for food secure cities: Measuring the influence of infrastructure and income on household food security in Southern African cities

Planning for food secure cities: Measuring the influence of infrastructure and income on household food security in Southern African cities

Geoforum 65 (2015) 1–11 Contents lists available at ScienceDirect Geoforum journal homepage: www.elsevier.com/locate/geoforum Planning for food sec...

396KB Sizes 0 Downloads 21 Views

Geoforum 65 (2015) 1–11

Contents lists available at ScienceDirect

Geoforum journal homepage: www.elsevier.com/locate/geoforum

Planning for food secure cities: Measuring the influence of infrastructure and income on household food security in Southern African cities Bruce Frayne a, Cameron McCordic b,⇑ a b

School of Environment, Enterprise, and Development, University of Waterloo, 200 University Avenue West, Waterloo, ON N2L 3G1, Canada Department of Environment and Resource Studies, University of Waterloo, 200 University Avenue West, Waterloo, ON N2L 3G1, Canada

a r t i c l e

i n f o

Article history: Received 2 September 2014 Received in revised form 19 April 2015 Accepted 25 June 2015

Keywords: Urban Food security Poverty HFIAS MAHFP LPI

a b s t r a c t This paper presents correlation and regression analyses designed to assess the respective relationships between the Household Food Insecure Access Scale/Prevalence (HFIAS/HFIAP) (as a measure of food access), the Months of Adequate Household Food Provisioning (MAHFP) (as a measure of food access stability) and (1) the Lived Poverty Index (LPI) (as an infrastructure access measure) and (2) household income. The data is drawn from a survey of 6453 households from 11 Southern African cities. The findings indicate that infrastructure access significantly predicted HFIAP and MAHFP scores. The regression analyses demonstrated that households with inconsistent or no access to a cash income, cooking fuel, medical care, electricity, or water had 11 times greater odds of being categorized as food insecure in the HFIAP and 8.5 times greater odds of having less than 12 months of adequate food provisioning in the last year. Household income alone does not sufficiently account for these relationships. The correlation analyses demonstrate a strong association between all the LPI subscales and household food access. These results clarify the differential impact of social and physical infrastructure on household food security and demonstrate that the prevailing conditions of an urban environment may better explain (and predict) urban household food security than household income alone. This investigation emphasizes the central role that urban planning and development can play in reducing food insecurity in poor urban neighborhoods. Ó 2015 Published by Elsevier Ltd.

1. Introduction Urban food security is now recognized as a major development dynamic in rapidly growing cities of the global south (Tacoli et al., 2013; FAO, 2012; Crush et al., 2012; Sonnino, 2009; Krausmann et al., 2009; Steel, 2008). While urban environments have been associated with economies of scale and improved wellbeing, rapid urbanization is also responsible for rising poverty, increases in population density, escalating land costs and informal living and working conditions (Parnell and Pieterse, 2014; Sassen, 2012; Glaeser, 2011; Saunders, 2011; Yu et al., 2010; Pugh, 2000; Beall and Fox, 2009; Ruel et al., 1999). In many cities and towns in the global south, at least half the population lives below the poverty line (Ravallion, 2007; UN-Habitat, 2008). These communities typically live under conditions of extreme economic hardship and are often the most vulnerable to food insecurity (Frayne et al., 2010). However, while inadequate and unreliable incomes may well be ⇑ Corresponding author. E-mail addresses: [email protected] (B. Frayne), [email protected] (C. McCordic). http://dx.doi.org/10.1016/j.geoforum.2015.06.025 0016-7185/Ó 2015 Published by Elsevier Ltd.

a cause of food insecurity, Tacoli argues that ‘‘inadequate housing and basic infrastructure and limited access to services contribute to levels of malnutrition and food insecurity that are often as high if not higher than in rural areas’’ (Tacoli et al., 2013: iv). It is in this context of urban informality and economic marginality that this paper examines the extent to which infrastructure and income determine food insecurity at the household level. This paper is based on the baseline survey conducted by the African Urban Food Security Network (AFSUN) in 11 cities in nine countries in Southern Africa. The survey sampled 6453 households from poorer neighborhoods in these cities. The analysis demonstrates that the limited availability of infrastructure does impact household food security negatively; conversely, planning for and investing in physical and social infrastructure in poor urban communities may be an important strategy for improving household food security. This conclusion speaks clearly to the view that food supply and availability are not the primary food security challenges in urban areas. Rather, the challenge is about access to food and the ability of households to store and use food effectively – these conditions are ultimately about access to infrastructure and not about food production (Crush et al., 2012). The research

2

B. Frayne, C. McCordic / Geoforum 65 (2015) 1–11

objective of this paper is to empirically demonstrate the relationship (in terms of association and likelihood) between household infrastructure access and household food security among poor urban households in Southern Africa.

2. Literature review In her former role as the Executive Director of UN-Habitat, Anna Tibaijuka writes in the UNEP publication Our Planet that ‘while it may be difficult to overcome relative poverty, it is perfectly possible to ensure that the poor are provided with adequate shelter and basic services. The history of cities in the developed world proves the point’ (2005: 12). This apparent causative association between infrastructure and development is echoed elsewhere in the literature (Turok, 2014; Mitlin and Satterthwaite, 2013; Pieterse and Simone, 2013; UN-Habitat, 2011; Glaeser, 2011; Beall and Fox, 2009; World Bank, 2009). Infrastructure is viewed as a hallmark of urban development, yet the towns and cities of the global south have not been able to meet the demand for infrastructure arising from rapidly urbanizing populations. Since the implementation of the Millennium Development Goals, the target set under Goal 7(d) to improve the lives of people living in informal housing by 100 million has been surpassed (UN, 2015). However, there are now more people living in slums than ever before, with another two billion expected by mid-century (UN, 2015; Burdett and Sudjic, 2010; Neuwirth, 2005). Informality is therefore now recognized as a key dimension of urban systems in the global south and is considered a development challenge that must be addressed in order for societies to move out of poverty (Parnell and Pieterse, 2014; Parnell and Oldfield, 2014; UN-Habitat, 2014; Watson, 2009; Simone, 2004). Not only is informality now a major characteristic of rapidly growing cities, the nature of that informality is an important dimension of how populations experience urban environments. The extent of deprivation within the informal urban fabric is the most extreme in Sub-Saharan Africa (SSA) (Fig. 1). Not only does almost two thirds of the urban population in SSA live in slums, another third live under conditions described by the United Nations as ‘severely deficient’ (UN-Habitat, 2008: 93–95). Within this context, soft infrastructure (social services) and hard infrastructure (physical utilities) are recognized as important determinants of urban household livelihoods (Ogun, 2010). In the context of SSA, Pieterse and Parnell argue that ‘poverty, informality and the absence of a strong local state with a clear and unchallenged mandate to manage the city are arguably the leitmotifs of African urbanism today’ (2014, 10). Moreover, the failure of governments to adequately address the infrastructure needs in rapidly urbanizing contexts has roots in the persistence of

outdated planning systems and tools and a planning education inherited from Europe (principally Britain) that are not designed to cope with the nature and scale of urban growth in Africa (Rakodi, 1997; Robinson, 2011). These challenges combine to create a set of institutional, fiscal and political bottlenecks in the ability of urban managers to meet the infrastructure needs of their cities (Duminy et al., 2014). The impact of infrastructure access on urban household poverty appears to be mediated by the inequitable costs (or complete lack) of physical infrastructure access, where poor urban households tend to pay a higher cost (in absolute terms) either for access to physical infrastructure services than wealthy urban households or are required to cover the cost of installing physical infrastructure post hoc (Pieterse, 2014; Amis, 1995). This differential cost of infrastructural services among the urban poor is in part explained by informal living conditions. Poor urban households are often forced to set up semi-permanent shelter informally on land that has not been designated by municipalities as residential (Turok, 2014; Satterthwaite, 2014). As a result, urban planners provide post hoc household access to utilities and social services in these informal settlements, although in some cases these services are never provided by urban planners but by international NGOs. In a review of South African urban development, Joseph (2009) found that there is insufficient capacity, and political will, among local urban governments to implement holistic infrastructural development, leading to piecemeal infrastructure projects in urban areas. These piecemeal projects also result from municipal politics, which may put poor urban areas at a disadvantage. The challenge of planning effective infrastructure access in informal settlements is exacerbated by the continued immigration of poor rural migrants (Crush and Frayne, 2010). These rural–urban migrants at times aggregate in these informal settlements, resulting in the continued growth of settlements with limited infrastructural access. Infrastructural development may, therefore, offer one means of addressing the growing challenge of urban poverty (Parnell and Oldfield, 2014; Pendleton et al., 2006). In their seminal report, Canning and Bennathan (2000) argue that the social returns on soft or hard infrastructural development diminished quicker over time when the infrastructure was developed in isolation, suggesting that development dividends from investment in soft and hard infrastructure are complementary. In a recent series of computational simulations using national level data, Ogun (2010) concluded that investment in social infrastructure was more effective in reducing levels of household poverty than investment in physical infrastructure. While infrastructure may play a role in reducing urban poverty, is it possible that infrastructure could play a role in the associated phenomenon of urban food insecurity? There is a well-established literature that argues that urban poverty is associated with household food insecurity (for example,

100 90

Percentage (%)

80 70 60

In slums

50 Moderate (1-2) deficiencies

40 30

Severe (3-4) deficiencies

20 10 0 Sub-Saharan Africa

Lan America & Caribbean

Southern Asia

Fig. 1. Percentage of households living in slum conditions by region (2005). Adapted from Parnell and Pieterse (2014, p. 10) and UN-HABITAT (2008, p. 90).

B. Frayne, C. McCordic / Geoforum 65 (2015) 1–11

UNDP, 2012; Owuor, 2006; Crush and Frayne, 2009; Faber et al., 2009; von Braun et al., 1993). This is supported by the findings of the AFSUN (African Food Security Urban Network) food security baseline survey undertaken in 2008–9 in 11 cities where the data show that there is a statistically significant relationship between income poverty and food poverty (Frayne et al., 2010: 305–6). Moreover, UN-Habitat (2008) reports that almost half of SSA’s urban population live below the poverty line; it is therefore no surprise that the same AFSUN survey also found that four out of five poor urban households were chronically food insecure (Frayne et al., 2010). These findings speak again to the argument made in the introduction to this paper: that food access is much more of a problem for poor urban households than in the supply of food. In order to demonstrate these points, this paper relies on the Food and Agriculture Organization’s (FAO) definition of food security, which is defined as the access to sufficient and nutritious food by all people at all times to lead a healthy and active life (FAO, 2006: 1). This definition rests upon four pillars: food utilization (the consumption of food), food supply (the availability of food), food access, and stability (Barrett, 2010). In order to achieve food security, the three variables of availability, utilization, and access must be held stable over the long term. The dimension of stability is defined by the lack of shocks (sudden and unexpected experiences of deficiency) in household access to food (FAO, 2006, p. 1). The multifaceted nature of these definitions provided by the FAO indicates how a potentially complex array of shocks could destabilize a household’s access to food. In a survey of urban households in Lesotho vulnerable to food insecurity, the World Food Programme found that, in addition to the experience of food price shocks, these households also tended to experience other shocks. These additional shocks included growth in family size (due in part to hosting orphans), a recent death in the family, and the additional supports required for household members with disabilities or chronic illnesses (World Food Programme, 2008). This survey also indicated that the most common food security shocks experienced by the households included food price shocks, fuel price shocks, serious illness or accidents, and irregular rainfall (World Food Programme, 2008). A common theme in these findings is the role of medical emergencies in food security shocks among poor urban households in Sub-Saharan Africa. Crush et al. (2011) offer a potential mechanism to explain this relationship by describing the connection between HIV and urban household food insecurity. Among Sub-Saharan African cities, the rates of HIV prevalence are substantially higher in urban areas than in rural areas. Among affected urban households, the contraction of HIV has a dual effect on household food security. The progression of the disease requires higher household expenditures on food and medical care while the affected household member has an increasingly limited physical ability to earn a meaningful income. Among poor urban households in Sub-Saharan, the impact of HIV can be magnified by the lack of access to physical infrastructure such as clean water and electricity (Crush et al., 2011). The impact of physical infrastructure access as an urban household food shock in Sub-Saharan Africa also has some support in the literature. In a survey of urban households in Windhoek, Frayne (2004) found that transportation provided a social coping mechanism for urban households suffering food insecurity via food remittances. Access to personal transportation (and the affordability of fuel) supported household remittances of cash from urban to rural households and remittances of food from rural to urban households. Transportation infrastructure also allowed an additional livelihood opportunity for urban households via the provision of transportation services (i.e. taxi services). Similarly, Owuor’s case study of Nakuru, Kenya suggested that infrastructure in-access among the urban poor, and the deterioration of infrastructure in

3

Nakuru, can reduce the chances of a household maintaining consumption if household income is reduced (2006). Owuor suggests that urban infrastructure can provide a link to livelihood opportunities and the inability of households to access urban infrastructure can make a household vulnerable to food insecurity over the long-term. These infrastructure services appear to support urban household food security and may provide an avenue for intervening in urban food insecurity. While perhaps not as visible as other areas of urban development, Pothukuchi and Kaufman (1999) argue that food systems are a necessary element of urban planning. They argue, together with other scholars (for example, Morgan and Sonnino, 2010; Rocha and Lessa, 2009), that food security in urban areas relies upon key urban infrastructure (such as transportation, food sector employment, waste disposal, and housing affordability) and impacts both urban infrastructure (for example, the potential for agricultural pesticide pollutants in city water supply) and household livelihoods (as demonstrated by the fact that a substantial amount of household income is spent on food in poor communities). In addition, food desserts are receiving attention in the literature as zones that are by definition lacking in readily accessible, good quality food. For example, Battersby and Crush (2014) demonstrate that in Cape Town many poor urban communities are a significant distance from supermarkets and effectively live in ‘food deserts’. Furthermore, the lack of high quality and affordable food outlets leads to a reliance on cheaper, nutrient-poor foods, which are in turn an important factor in the rise of obesity and malnutrition amongst poor urban communities (Battersby and Peyton, 2014; Frayne et al., 2014). This discussion demonstrates that the development of urban infrastructure to support urban household food security is a multi-dimensional challenge which needs to be addressed (Steel, 2008; McLachlan and Garrett, 2008; Crush and Caesar, 2014). SSA appears to be in an unfolding situation where (1) rapid urbanization and rising levels of informality are a major challenge for local government; (2) governments appear to lack the desire or will to intervene and are poorly equipped and ineffective at meeting the current demands for urban infrastructure among many cities in SSA; (3) a lack of access to infrastructure disproportionately disadvantages the urban poor; (4) urban poverty rates are high and may be rising; (5) income poverty and food insecurity are directly linked; (6) limited infrastructure results in poor levels of access to quality and affordable food retail services; and (7) that investment in infrastructure has positive development dividends for the urban poor. What, then, does this state of knowledge imply for the role of infrastructure in determining levels household food security? Within this context, our paper focuses on the relationship between levels of access that poor urban households have to infrastructure and the relationship between that access and their levels of food (in)security. Currently, this relationship remains untested in the empirical literature. However, if infrastructure is an important dimension of food security, the direction and magnitude of this relationship needs to be understood in order to offer policy-oriented evidence regarding the question of how to better plan for food security cities.

3. Research objective The broad objective is to empirically demonstrate the relationship (in terms of association and likelihood) between household infrastructure access and household food security among poor urban households in Southern Africa. Specifically, this investigation defines the relationship between the Lived Poverty Index (LPI), as a measure of social and physical infrastructure access, the

4

B. Frayne, C. McCordic / Geoforum 65 (2015) 1–11

Household Food Insecurity Access Scale (HFIAS) and the Months of Adequate Household Food Provisioning (MAHFP) among poor urban households in Southern African cities. This investigation also determines whether this relationship is due to the influence of household income on infrastructure access, household food access, and the stability of household food access. 4. Methods 4.1. Sample The sample was drawn from a survey completed in 2008 by the African Food Security Urban Network (AFSUN). The survey was administered to 6453 households in 11 cities distributed between 9 countries in Southern Africa (Table 1). The household surveys were administered using systematic sampling of households in informal and low income areas of these cities. Once a household was selected for a survey, the survey was administered to either the household head or a responsible adult in the household who could respond on behalf of the entire household. The surveys were administered by undergraduate students who attended training workshops on the administration of these surveys. The quality of survey administration and data entry was checked by field supervisors. The surveys covered information regarding income, poverty, food security, remittances, and household demographic information. In addition to these variables, the survey also administered the HFIAP, the MAHFP and the LPI. 4.2. Measures This investigation relies upon three measures: the Household Food Insecurity Access Scale (HFIAS) to measure household food access, the Months of Adequate Household Food Provisioning (MAHFP) to measure the long term stability of food access and the Lived Poverty Index (LPI) to measure household access to social and physical infrastructure. Before continuing further, it is important to both define the concept of ‘‘access’’ and to demonstrate how this concept is operationalized in the HFIAS, MAHFP and LPI. In the context of this investigation, access is defined by the ability of a household to command resources via labor, legal right, production, or social capital (the resources open to a household via social networks). Household access to a resource should not be narrowly defined by physical proximity to a resource (which is better defined as the availability of that resource) or exclusively defined by the purchasing power of a household. This concept of ‘‘access’’ should be used when interpreting the results from the HFIAS measure. Access is operationalized in the HFIAS by the physical, economic, and social experience of food in-access, relying on reported household perceptions of, physical consequences of, economic impacts of, and social shame resulting from, a lack of

Table 1 Total household sample distribution across Southern African cities. City

Country

n (%)

Blantyre Cape Town Gaborone Harare Johannesburg Lusaka Manzini Maputo Maseru Pietermaritzburg Windhoek Total

Malawi South Africa Botswana Zimbabwe South Africa Zambia Swaziland Mozambique Lesotho South Africa Namibia

432 (6.7) 1060 (16.4) 400 (6.2) 462 (7.2) 996 (15.4) 400 (6.2) 500 (7.7) 397 (6.2) 802 (12.4) 556 (8.6) 448 (6.9) 6453 (100.0)

household access to food. The stability of household food access is operationalized in the MAHFP as the adequacy of food provisions over a 12 month period. Household ‘‘in-access’’ of a resource is defined in the context of this investigation by the inability of a household to command a resource when that resource is required. This definition should be differentiated from a lack of household access due to a lack of intention to access that resource rather than due to a lack of household capacity to access that resource. In-access is operationalized in the calculation of LPI scores used in this investigation as a household’s consistent, inconsistent, or complete lack of access to resources. The Household Food Insecure Access Scale (HFIAS) is a household food security survey instrument developed by the Food and Nutrition Technical Assistance program at the United States Agency for International Development (USAID, 2007). The instrument is composed of nine likert scale questions regarding various characteristics associated with food access within the previous four weeks. The final scores in the HFIAS scales are assessed using a scoring algorithm to determine whether the household fits into one of four different categories of household food access. These categorical scores are represented as the Household Food Insecure Access Prevalence (HFIAP). The HFIAP categories are represented as food secure (HFIAP = 1), mildly food insecure access (HFIAP = 2), moderately food insecure access (HFIAP = 3), or severely food insecure access (HFIAP = 4). When used as a dependent variable in the logistic regression analyses in this investigation, these HFIAP categories are collapsed into two categories: food secure (HFIAP = 1) and food insecure (HFIAP = 2–4). Alternatively, the ranked responses to the 9 HFIAS scales can be summed and presented as a scaled score to represent household food access insecurity (called the Household Food Insecure Access Scaled Score or the HFIASS). In the HFIASS, higher values represent worse household food access (see Table 2 for the distribution of HFIAS scores in the sample). In this investigation, the HFIASS is used in the Spearman’s rho correlation analyses. The distribution of sample scores from the HFIAP demonstrates that most of the households in this population were categorized as having severely insecure food access (as demonstrated in Table 3). The Months of Adequate Household Food Provisioning (MAHFP) is another household food security survey instrument developed by the Food and Nutrition Technical Assistance program at United States Agency for International Development (USAID, 2010). The survey instrument measures the number of months during which a given household had sufficient food provisioning. This index is administered to one household member who answers on behalf of the rest of the household. In the context of this survey, any adult member of the household (preferably the household head) could provide responses for these survey questions. While the survey enumerators were trained to only administer the survey to an individual over the age of majority, the identifying characteristics of the household respondent were not recorded in this survey. The respondent is first asked to confirm whether or not

Table 2 Sample distribution across Household Food Insecurity Access Scale scores. Statistic Min Max Mean Median Range Standard Deviation Skewness Kurtosis Total N

0 27 10.28 10.00 27 7.578 .300 .860 6327

Standard Error

.031 .062

5

B. Frayne, C. McCordic / Geoforum 65 (2015) 1–11 Table 3 Sample distribution across Household Food Insecurity Access Prevalence categories. Household Food Insecurity Access Prevalence

n (%)

Food secure Mildly food insecure access Moderately food insecure access Severely food insecure access Total

1008 (15.9) 442 (7.0) 1273 (20.1) 3604 (57.0) 6453 (100.0)

there have been months over the last year in which the household did not have sufficient food provisioning. If the respondent indicates that there have been months over the previous year in which there was inadequate monthly household food provisions, the respondent is asked to identify the months during which this lack of food provisioning occurred (beginning with the most recent month and working backward across the months in the previous year). The MAHFP score is then calculated by subtracting 12 from the number of months indicated by the respondent. The final score thereby indicates the number of months during which the household had adequate food provisions. In this investigation, this scaled score for the MAHFP is used in the Spearman’s rho correlations. In the logistic regression analysis, these scores are collapsed to two categories: households with consistently adequate monthly food provisions (MAHFP = 12) and household with inconsistently adequate monthly food provisions (MAHFP < 12). This binary categorical variable is used as the dependent variable in all logistic regression analyses. The table below demonstrates the frequency distribution of MAHFP scores in the sampled population (see Table 4). The Lived Poverty Index (LPI) is a household survey instrument used to define the household experience of poverty by measuring household in-access to social and physical infrastructure resources and services. The index is composed of five Likert scale questions on the consistency of household access to a cash income, electricity, clean water, medical treatment, and cooking fuel. These questions are designed to determine the frequency with which the household has gone without access (experienced in-access) to any of these infrastructure resources and services over the past 12 months. This index represents physical resources (electricity, water, cooking fuel) and social services (cash income and medical care) which rely upon the functioning of both physical networks (physical infrastructure) and systems of social institutions (social infrastructure) in order to support household livelihoods in urban environments. These physical networks and systems of social institutions require coordination and planning in order to function efficiently. Household access to these resources and services (the physical and social infrastructure) in a city relies upon both availability and household capacity (see Table 5). In the context of this investigation, the household frequency of going without infrastructure services and resources is aggregated Table 4 Sample distribution across Months of Adequate Household Food Provisioning. Months of Adequate Household Food Provisioning

n (%)

0 1 2 3 4 5 6 7 8 9 10 11 12 Total

576 (9.1) 36 (0.6) 105 (1.7) 74 (1.2) 97 (1.5) 136 (2.2) 206 (3.3) 233 (3.7) 389 (6.2) 671 (10.6) 879 (13.9) 835 (13.2) 2068 (32.8) 6453 (100.0)

into three categories: consistent access, inconsistent access, and no access over the past 12 months. These are the ordinal variables used in the spearman’s rho correlation of the LPI subscales with the HFIAP and MAHFP. These categories are collapsed into two categories for the logistic regression analysis: consistent access and inconsistent or no access. The total LPI scores from these sub-scale questions are also averaged across the five infrastructure resources and services and presented as a scaled score. These scaled scores are used in the Spearman’s rho correlation of LPI scores and the HFIASS and MAHFP. Alternatively the scaled scores from these five subscales can also be represented as a binary categorical variable, where, households are categorized according to whether the household has had (1) consistent access to all infrastructure resources and services over the past 12 months, or (2) inconsistent or no infrastructure access in the past 12 months. This LPI categorical variable is used in the logistic regression analysis with the HFIAP and MAHFP. Table 6 demonstrates the sample distribution across these LPI scale score categories. Household income was measured as the sum of income received from all household income sources over the previous month by a given household. Given that nine national currencies are represented in the sample, conversion of household income levels was required to allow for comparability across households. In order to attain a comparable interval level indicator of household income, all household income levels have been converted into South African ZAR at the 2008 ZAR value. These conversions were completed using the average monthly currency exchange rate from each respective national currency into South African ZAR. Thus each household income value was converted based on the conversion rate of its national currency into ZAR for the given month in which the household income amount was collected by the AFSUN survey. These surveys were completed between August and December of 2008. The monthly currency exchange rates used in this investigation were taken from OANDA records of historical monthly exchange rates (OANDA, 2013) (see Table 7). Some of the analyses used in this investigation (such as the logistic regression analysis), required the conversion of household income into a categorical variable. In order to create a categorical variable representing household income, the entire sample of household income was categorized into three income terciles (representing low, middle, and high income) within each national currency represented in the sample. These categories therefore represent the distribution of household income within each national currency across the sampled cities and allows for comparisons between cities irrespective of currency. The sample distribution of household income terciles is provided in Table 8. 4.3. Analysis In order to test the predictive relationship between the HFIAP, MAHFP and the LPI, these relationships are analyzed using binary logistic regression analysis. Binary logistic regression determines the odds of a sample of cases (households) being categorized in one of the values of a binary dependent variable, given a set of independent variables. As such, the analysis requires that the dependent variable in the analysis is a dichotomous nominal variable while all independent variables can be at the interval/ratio, ordinal, or nominal level of measurement. In the logistic regression Table 5 Sample distribution across Lived Poverty Index categories. Lived Poverty Index category

n (%)

Consistent access to all infrastructure Inconsistent or no access to any Total

1026 (18.0) 4676 (82.0) 5702 (100.0)

6

B. Frayne, C. McCordic / Geoforum 65 (2015) 1–11

Table 6 Sample distribution across Lived Poverty Index sub-scale ordinal variables. Lived Poverty Index sub-scale

Consistent access n (%)

Inconsistent access n (%)

No access n (%)

Total n (%)

Clean water for home use Medicine or medical treatment Electricity in home Enough fuel to cook food A cash income

4151 (65.0)

1883 (29.5)

349 (5.5)

6383 (100)

3748 (60.3)

2226 (35.8)

244 (3.9)

6218 (100)

2324 (38.0)

1887 (30.9)

1899 (31.1)

6110 (100)

2560 (41.2)

3336 (53.7)

312 (5.0)

6208 (100)

2000 (31.7)

3693 (58.6)

611 (9.7)

6304 (100)

analysis, the HFIAP is categorized into two categories (where HFIAP level 1 is categorized as food secure and HFIAP levels 2–4 is categorized as food insecure) and the MAHFP is also categorized into two categories as well (where MAHFP category 1 represents 12 months of adequate household food provisioning and category 2 represents less than 12 months of adequate household food provisioning). The LPI scaled score is categorized as a binary independent variable (where LPI category 1 represents consistent household access to all infrastructure resources and services while category 2 represents inconsistent or no access to any of the infrastructure resources and services) and household income was divided into terciles (three proportional categories) within each national currency and then reversed in the logistic regression analysis to represent the impact of a household’s downward progression from high to middle to low income on household food security. A logistic regression analysis is also applied to the relationship between the ordinal subscales of the LPI and the constructed dichotomous HFIAP categories. This analysis demonstrates the contribution of each independent variable in the overall predictive relationship between the LPI and both the HFIAP and the MAHFP. This analysis relied upon the same binary dependent variables representing the HFIAP and the MAHFP. The LPI sub-scales are represented as ordinal independent variables (ranked according to consistent access, inconsistent access, or no access to the respective infrastructure resources or services) (see Table 9). The correlational relationships between household income, the LPI, the HFIAS, and the MAHFP are assessed using a Spearman Rho analysis (all four of these variables are presented at the interval level of measurement). Spearman Rho correlations provide a non-parametric analysis of these relationships which does not assume the normal distribution of sample scores, accepts data at both ordinal and continuous levels of measurement, and is sensitive to non-linear relationships (Corder and Foreman, 2009). This correlational analysis will also be applied the underlying categories in the LPI to determine whether these subscales are most correlated with the HFIAS, MAHFP, or household income. These LPI sub-scales are ordinal level and indicate whether a household

Table 7 Sample distribution of household income received in the last month (2008 ZAR). Statistic Min Max Mean Median Range Standard Deviation Skewness Kurtosis Total N

0 110,000 2571.95 1320.00 110,000 4221.39 7.44 115.96 5457

had consistent, inconsistent, or no access to the respective infrastructure resources or services. In doing so, this analysis helps to determine any potential underlying relationships, which in turn helps to explain why a relationship between the LPI and the HFIAS exists in urban environments in Southern African cities. In the correlational analysis, any correlation values less than 0.3 will be categorized as a weak association, any correlation values between 0.3 and 0.6 will be categorized as a moderate association, and any correlation values greater than 0.6 will be categorized as a strong association (see Table 10).

5. Results 5.1. Relating infrastructure access to the Household Food Insecure Access Scale The binary logistic regression model relating total LPI scores and household income terciles to the HFIAP dichotomous variable accurately categorized 85.3% (with a cut-off value of 0.5) of the cases involved in the analysis according to these HFIAP categories (an increase from 83.3% accuracy observed in the null logistic regression model). The model also demonstrated a Nagelkerke R2 value of .307. A Hosmer and Lemeshow test demonstrated a Chi-square value of 22.306 (p < .001, df = 3) and a correlation value of .030 was the highest value observed between the LPI categories and household income terciles (while the middle and low income tercile dummy variables demonstrated a .316 correlation). The model demonstrated that both household income and the LPI categories were significant predictors of the dichotomous HFIAP categories. In this model, households in middle and low income terciles are both compared to households in high income terciles (high income terciles will serve as the base for all household income tercile comparisons of odds ratios). The Wald criterion demonstrated that the LPI was a much stronger predictor of household food access than household income. While holding other variables in the model constant, the odds of households with inconsistent or no infrastructure being categorized as food insecure were approximately 11 times greater (95% C.I. of 9.305– 13.315) than the odds of households with consistent infrastructure access being categorized as food insecure. The odds of middle income households being categorized as food insecure were approximately 2 times greater (95% C.I. of 1.882–2.827) than the odds of high income households (holding other model variables constant). The odds of low income households being categorized as food insecure were about three times greater (95% C.I. of 2.541–4.030) compared to the odds of high income households (holding other model variables constant) (see Table 11). The subscales of the LPI demonstrate a significant relationship with the HFIAP dichotomous variable and the likert subscales in the LPI. In this model, households with inconsistent or no access are both compared to households with consistent access (the consistent access value will serve as the base for all infrastructure access comparisons of odds ratios within each LPI sub-scale). The model accurately categorized 85.2% (with a cut-off value of 0.5) of cases according to the HFIAP dichotomous variable (an increase from 82.9% accuracy observed in the null logistic regression

Standard Error Table 8 Sample household income terciles.

.033 .066

Household income terciles

n (%)

Low income Middle income High income Total

1727 (31.5) 1834 (33.5) 1914 (35.0) 5702 (100.0)

7

B. Frayne, C. McCordic / Geoforum 65 (2015) 1–11

Among the subscales in the LPI, HFIAS scores were moderately correlated with going without consistent access to medicine or medical treatment (rho(6107) = .421, p < .001), going without consistent access to a cash income (rho(6187) = .470, p < .001), going without electricity (rho(5996) = .405, p < .001) and going without enough fuel to cook food (rho(6098) = .474, p < .001). This analysis also found that household frequency of going without clean water (rho(6265) = .262, p < .001) was weakly correlated with HFIAS. Total HFIAS and LPI scores were close to strongly correlated (rho(5606) = .592, p < .001) while household income was moderately correlated with LPI scores (rho(4870) = .488, p < .001) and moderately correlated with HFIAS scores (rho(5357) = .501, p < .001). These results indicate that there is a stronger association between the LPI and the HFIAS than between household income and either of these measures (Table 13).

Table 9 Binary logistic regression variables. Variables

Level

Categories

Dependent variables HFIAP Binary MAHFP Binary

Food secure 12 months

Food insecure <12 months

Independent variables Income terciles Ordinal

High income

Middle income

Consistent access Consistent access Consistent access Consistent access Consistent access Consistent access

Inconsistent/No access Inconsistent/No access Inconsistent/No access Inconsistent/No access Inconsistent/No access Inconsistent/No access

LPI categories

Binary

Clean water

Binary

Medical care

Binary

Electricity

Binary

Cooking fuel

Binary

A cash income

Binary

Low income

5.2. Relating infrastructure access to Months of Adequate Household Food Provisioning

model). The model demonstrated a Nagelkerke R2 value of .390 and a Hosmer and Lemeshow Chi-Square value of 56.121 (p < .001, df = 6). The subscales were also correlated against each other and the highest value attained was a .226 correlation between medical treatment and access to clean water inconsistent access dummy variables. Of the subscales, only access to medical treatment, electricity, fuel to cook food and a cash income were significant predictors of the HFIAP dichotomous variable. According to the Wald statistic, the most significant predictor was access to cooking fuel (followed closely by access to a cash income) and the least significant predictor was access to water. That said, the Exp(B) values reveal a slightly different story. While inconsistent household access to a cash income resulted in a household’s odds of being food insecure growing 3.235 times (95% C.I. of 2.696– 3.882) (holding other model variables constant), inconsistent or no household access to medical treatment increased the odds that a household would be categorized as food insecure by 3.540 times (95% C.I. of 2.688–4.664) (holding other model variables constant) (see Table 12).

The binary logistic regression model relating total LPI scores and household income terciles to the MAHFP dichotomous variable accurately categorized 76.0% (with a cut-off value of 0.5) of the cases involved in the analysis according to these MAHFP categories (an increase from 67.3% accuracy observed in the null logistic regression model). The model also demonstrated a Nagelkerke R2 value of .239. A Hosmer and Lemeshow test demonstrated an insignificant Chi-square value of 1.594 (p = .661, df = 3) and a correlation value of .041 was the highest value observed between the LPI categories and household income terciles (while the middle and low income tercile dummy variables demonstrated a .417 correlation). The model demonstrated that both household income and the LPI categories were significant predictors of the dichotomous MAHFP categories. In this model, households in middle and low income terciles are both compared to households in high income terciles (high income terciles will serve as the base for all household income tercile comparisons of odds ratios). The Wald criterion demonstrated that the LPI was a much stronger predictor of household food access than household income. While holding other variables in the model constant, the odds of households with inconsistent or no infrastructure having less than 12 months of

Table 10 Spearman’s rho variables. Variables

Level

Categories

Household income LPI HFIASS MAHFP

Ratio Ratio Ratio Ratio

N/A N/A N/A N/A

Clean water for home use Medicine or medical treatment Electricity in home Enough fuel to cook food A cash income

Ordinal Ordinal Ordinal Ordinal Ordinal

Consistent Consistent Consistent Consistent Consistent

(Higher (Higher (Higher (Higher

values values values values

represent represent represent represent

higher income) worse infrastructure access) worse food access) better food provisioning)

access access access access access

Inconsistent Inconsistent Inconsistent Inconsistent Inconsistent

access access access access access

No No No No No

access access access access access

Table 11 Binary logistic regression model statistics predicting HFIAP scores. Variables

B

**

LPI categories Income terciles** Income tercile (middle income)** Income tercile (low income)** Constant** ⁄ **

p < .05. p < .01.

S.E.

2.410

.091

.836 1.163 .638

.104 .118 .083

Wald

695.118 124.661 64.746 97.695 59.223

df

1 2 1 1 1

P value

<.001 <.001 <.001 <.001 <.001

Exp(B)

95% C.I. for EXP(B) Lower

Upper

11.131

9.305

13.315

2.307 3.200 .528

1.882 2.541

2.827 4.030

8

B. Frayne, C. McCordic / Geoforum 65 (2015) 1–11

Table 12 Binary logistic regression model statistics predicting HFIAP scores using LPI subscales. LPI sub-scales

B

S.E.

Household frequency of going without: Clean water .085 Medical care** 1.264 ** Electricity .751 Cooking fuel** 1.451 A cash income** 1.174 Constant** .150 ⁄ **

Wald

.129 .141 .102 .120 .093 .056

df

.431 80.826 54.314 146.569 159.374 7.102

1 1 1 1 1 1

LPI scales

HFIASS Rho (n, p-value)

Household frequency of going without: Clean water for home use .262** (6265, <.001) Medicine or medical treatment .421** (6107, <.001) Electricity in home .405** (5996, <.001) Enough fuel to cook food .474** (6098, <.001) A cash income .470** (6187, <.001) LPI .592** (5606, <.001) HFIASS 1 (6327, <.001) ⁄

.511 <.001 <.001 <.001 <.001 .008

Exp(B)

1.088 3.540 2.119 4.267 3.235 .860

95% C.I. for EXP(B) Lower

Upper

.845 2.688 1.735 3.374 2.696

1.401 4.664 2.587 5.396 3.882

p < .05. p < .01.

Table 13 Spearman’s rho correlations of the LPI, HFIASS, and household income.

**

P value

Household income Rho (n, p-value) .288** (5404, <.001) .272** (5281, <.001) .366** (5173, <.001) .334** (5284, <.001) .369** (5365, <.001) .488** (4870, <.001) .501** (5357, <.001)

p < .05. p < .01.

food provisioning were approximately 8.5 times greater (95% C.I. of 7.139–10.144) compared to the odds of households with consistent infrastructure access. The odds of middle income households were about 1.7 times greater (95% C.I. of 1.511–2.069) compared to the odds of high income households (holding other model variables constant). The odds of low income households having less than 12 months of food provisioning were about 2.5 times greater (95% C.I. of 2.158–3.029) compared to the odds of high income

households (holding other model variables constant) (see Table 14). The subscales of the LPI demonstrate a significant relationship with the MAHFP dichotomous variable and the likert subscales in the LPI. In this model, households with inconsistent or no access are both compared to households with consistent access (the consistent access value will serve as the base for all infrastructure access comparisons of odds ratios within each LPI sub-scale). The model accurately categorized 76.7% (with a cut-off value of 0.5) of cases according to the MAHFP dichotomous variable (an increase from 66.3% accuracy observed in the null logistic regression model). The model demonstrated a Nagelkerke R2 value of .312 and a Hosmer and Lemeshow Chi-Square value of 43.469 (p < .001, df = 7). The subscales were also correlated against each other and the highest value attained was a .306 correlation between the access to electricity and fuel variables. Of the subscales, only access to medical treatment, electricity, fuel to cook food and a cash income were significant predictors of the MAHFP dichotomous variable. According to the Wald statistic, the most significant predictor was access to cooking fuel (followed closely by access to a cash income) and the least significant predictor was access to water. The Exp(B) values reveal a slightly different story. While inconsistent household access to a cash income

Table 14 Binary logistic regression model statistics predicting MAHFP scores. Variables

B

LPI categories** Income terciles** Income tercile (middle income)** Income tercile (low income)** Constant ⁄ **

S.E.

Wald

df

P value

Exp(B)

95% C.I. for EXP(B) Lower

Upper

2.141

.090

<.001 <.001 <.001 <.001 <.001

7.139

10.144

.080 .086 .089

1 2 1 1 1

8.510

.570 .939 1.453

570.885 125.983 50.507 117.961 263.675

1.768 2.557 .234

1.511 2.158

2.069 3.029

p < .05. p < .01.

Table 15 Binary logistic regression model statistics predicting MAHFP scores using LPI subscales. LPI sub-scales

B

Household frequency of going without: Clean water .152 Medical care** .667 Electricity** .607 ** Cooking fuel .981 ** A cash income .871 ** Constant .956 ⁄ **

p < .05. p < .01.

S.E.

.082 .079 .075 .077 .072 .057

Wald

3.442 70.738 65.145 162.544 144.932 282.581

df

1 1 1 1 1 1

P value

.064 <.001 <.001 <.001 <.001 <.001

Exp(B)

1.165 1.949 1.835 2.666 2.389 .385

95% C.I. for EXP(B) Lower

Upper

.991 1.669 1.583 2.293 2.073

1.368 2.277 2.126 3.100 2.753

9

B. Frayne, C. McCordic / Geoforum 65 (2015) 1–11 Table 16 Spearman’s rho correlations of the LPI, MAHFP, and household income. LPI scales Household frequency of going without: Clean water for home use Medicine or medical treatment Electricity in home Enough fuel to cook food A cash income? LPI MAHFP ⁄ **

MAHFP Rho (n, p-value) .223** (6242, <.001) .303** (6082, <.001) .313** (5971, <.001) .382** (6073, <.001) .382** (6166, <.001) .465** (5575, <.001) 1 (6305, <.001)

Household income Rho (n, p-value) .288** (5404, <.001) .272** (5281, <.001) .366** (5173, <.001) .334** (5284, <.001) .369** (5365, <.001) .488** (4870, <.001) .392** (5338, <.001)

p < .05. p < .01.

resulted in a household’s odds of going without monthly adequate food provisions which were 2.389 times greater (95% C.I. of 2.073– 2.753) than the odds of households with consistent access to a cash income (holding other model variables constant), inconsistent household access to medical treatment increased the odds that a household would be categorized as food insecure by 1.949 times compared with households which had consistent access to a cash income (95% C.I. of 1.669–2.277) (holding other model variables constant) (see Table 15). Among the subscales in the LPI, MAHFP scores were moderately correlated with household frequency of going without consistent access to medicine or medical treatment (rho(6082) = .303, p < .001), going without electricity (rho(5971) = .313, p < .001), going without consistent access to a cash income (rho(6166) = .382, p < .001), and going without enough fuel to cook food (rho(6073) = . .382, p < .001). This analysis also found that household frequency of going without clean water (rho(6242) = .223, p < .001) was weakly correlated with MAHFP. This investigation found that household income was only moderately or weakly correlated with these LPI subscales. Total MAHFP and LPI scores were moderately correlated (rho(5575) = .465, p < .001). Total household income was moderately correlated with LPI scores (rho(4870) = .488, p < .001) and moderately correlated with MAHFP scores (rho(5338) = .392, p < .001). These results indicate that there is a stronger correlational association between the LPI and household income than between household income and the MAHFP (in contrast to the stronger predictive relationship between the LPI and MAHFP) (Table 16). 6. Conclusion 6.1. Household food access and infrastructure These results demonstrate that there is indeed a predictive relationship between infrastructure access and the HFIAP among poor urban households in this eleven-city survey. In addition, the relationship between the HFIAP and the infrastructure access index score (demonstrating the consistency of household access to any of the infrastructure resources included in this investigation) is not better explained by household income. These results indicate that, in addition to household income, household food (in)access is also determined by the level of household access to a range of social and physical infrastructure services. One explanation for this finding is that there is an increased cost associated with accessing utilities and social services by households that are located in informal areas, which are predominantly inhabited by the urban poor. Since these areas are not well integrated into electrical and water grids, and often have limited access to medical facilities, the cost of attaining these resources is higher than for residents in other better served areas. The different relationships explored in this paper

between the HFIAS and the LPI versus the HFIAS and household income was previously suggested by Crush et al. (2012). While that study assessed these different relationships within each surveyed city, this current study establishes this relationship across cities by converting the 9 currencies represented into South African Rand (or by categorizing household income within each currency into three terciles representing low, middle, or high income) and using regression analysis to demonstrate the different relationships between the LPI and HFIAS versus household income and the HFIAS. This investigation did not, however, determine whether the individual infrastructure resources included in this investigation (water, electricity, medical care, a cash income) are a better determinant of HFIAP scores than household income. The correlation analysis appears to suggest this may be the case, but a future regression analysis will need to compare the impact of these variables directly (by including both in a regression analysis in order to control other variables in the model during the calculation of the log-odds values). 6.2. Months of Adequate Household Food Provisions and infrastructure Social and physical infrastructure access was a significant predictor and correlate of long term household food access (as measured by the MAHFP). That said, the relationship between infrastructure access and household income with months of adequate household food provisions was different from the relationship between these variables and household food access in two qualities. First, the correlational strength of the relationship between infrastructure access and the MAHFP was weaker than the relationship between infrastructure access and the HFIASS and weaker than the relationship between household income and the MAHFP. The weaker correlational strength in the relationship between the independent variables and the MAHFP indicates that the independent variables are not as comprehensive in explaining household MAHFP scores as they were in explaining household HFIASS scores (in spite of the strong predictive relationship observed in the logistic regression analysis). As the MAHFP may be viewed as a proxy for food stability over a period of 12 months, it is likely that other contextual factors are playing a role in determining MAHFP scores than infrastructure access alone. That said, the regression analysis demonstrated a strong predictive relationship between infrastructure access and the MAHFP. Second, the comparative predictive strength of household income versus infrastructure access with the MAHFP was different from the relationships observed between these variables and the HFIAP. While the predictive strength of household income was the same in its relationship with both the MAHFP and the HFIAS, infrastructure access was a weaker predictor of the MAHFP than with the HFIAS (while still being a much stronger predictor of

10

B. Frayne, C. McCordic / Geoforum 65 (2015) 1–11

the MAHFP than household income). In other words, households which went without consistent infrastructure access were less likely to have months of inadequate food provisioning than they were to have limited food access. One potential reason for these differences may lie in the conceptualization of ‘‘access’’ which was shared by both the HFIAP and the LPI, indicating that the two measures may be underpin a common phenomenon. As mentioned earlier, the MAHFP is also a proxy indicator for household food stability, a much broader phenomenon which is potentially manipulated by other determinants. Similar to the HFIAP regression analyses, the regression models used to analyze the relationship between infrastructure access and the MAHFP did not determine whether these relationships were better account for by household income. As such, future regression models will need to include both household income and these infrastructure access variables to make a clear comparison of these variables. This investigation is also based on cross-sectional data, limiting the extent to which long-term food access can be determined by the independent variables in this investigation. Future research will need to establish these relationships in a longitudinal research design relying on a panel regression analysis.

6.3. Integration of findings with background literature Consistent access to all social and physical infrastructure resources and services was a much stronger predictor of household food security than the income category of a household alone. Across the analyses completed in this paper, it was evident that social infrastructure had a greater impact on household food access and the stability of that food access than physical infrastructure. Interestingly, access to water demonstrated the weakest relationship with the urban household food security measures than any of the other indicators included in this investigation, in spite of the intuitive importance of water for household food security. One potential explanation for this finding is that the sampled households may have multiple points of access to water (i.e. shallow wells or bottled water), indicating the options available to help households cope with inconsistent water access. These coping mechanisms may therefore obscure the impact of inconsistent water access on household food access. That said, access to electricity also had a relatively weak relationship, though a significant one, with the household food security measures. Access to cooking fuel was the only consistently strong predictor of the household food security measures among the physical infrastructure access variables (perhaps for obvious reasons). Access to a cash income and medical care were very strong predictors of household food security by far (exceeded only by access to cooking fuel). This finding seems to support Ogun’s (2010) computational simulations on the relative impact of social versus physical infrastructure on urban poverty. This relationship may also be explained by trade-offs. Specifically, the relative importance of each physical and social infrastructure resource or service may demonstrate trade-offs in household expenditure. For example, while a household may be willing to forego consistent clean water or electricity access for food access in household expenditures, that household might be less willing to forego needed medical care for food access. Some of these infrastructure services may also be necessary preconditions for household food security (i.e. access to a cash income). That said, while access to a cash income was undoubtedly a strong predictor of household food security, the level of household income did not appear to be as significant a predictor (when compared with the impact of total LPI scores on household food security). This is a distinction of overall consistent access to a monetary infrastructure (a necessary condition for urban livelihoods) versus the amount of income earned by a household.

As highlighted in the introduction to this paper, Cecilia Tacoli argued that while income is important, access to infrastructure is also a key factor that contributes to urban household food security (Tacoli et al., 2013). The key contribution of this paper is that it empirically demonstrates that this causal relationship identified by Tacoli between infrastructure and food security at the household level is indeed evident among the sampled households. The major conclusion of these findings is that there is great potential to improve household food access through social and physical infrastructural development in poorly serviced areas. Specifically, the significant association between social and physical infrastructure access and household food security demonstrates the need for holistic urban planning that focuses on infrastructure access. In addition to improving household food security, proactive urban planning and development can play a key role in urban poverty mitigation more broadly. 6.4. Limitations The sample for this investigation was exclusively drawn from poor urban neighborhoods in Southern Africa. Thus, the results of this investigation cannot be generalized beyond the geographic boundaries of the sample. In addition, this survey was completed in the end of 2008, at a time when there was international food price volatility and there is the potential that household food security scores recorded in this investigation are manipulated by seasonality effects. Further empirical analysis will be needed to determine whether the relationships observed in this investigation are replicable in other urban areas at other points in time. Given the potential conceptual associations between some of the variables included in these regression analyses, multicollinearity between variables might have been a possible confound. As such, all regression analyses were subjected to OLS linear regression in order to test for VIF and Tolerance values. Among all the variables in the analyses, the VIF values ranged from 1.094 and 1.636 while the Tolerance values range from .611 and .914. These values suggest that multicollinearity was not a significant confound in explaining the regression coefficient relationships with the dependent variable. While many of the regression models in this investigation demonstrated decent pseudo R2 values (for logistic regression models), only one model demonstrated an insignificant Hosmer and Lemeshow chi-square value. While the Hosmer and Lemeshow statistic alone should not be used to determine that any model is a poor fit for the data, these findings should still be taken into consideration when determining the quality of these models. As such, caution should be taken before attempting to build a regression equation which predicts household food security using these regression models. The objective of this investigation was not to build such models for the purpose of designing a regression equation to predict household food insecurity. Rather, the objective of this investigation is to demonstrate the strength of relationship between household infrastructure access and household food access. Given these limitations, future research will need to demonstrate whether these findings are applicable to other forms of infrastructure. In particular, it would be interesting to determine whether household access to sanitation has any impact on household food security (given the importance of adequate sanitation for other public health measures). It would also be interesting to determine the extent to which the relationship established in this investigation is generalizable across urban environments outside of Southern Africa. Future research could also determine the extent to which this relationship is spatially distributed. Infrastructure access is a particular challenge among informal settlements and it would be fascinating to determine whether the relationships observed here between infrastructure access and food access is

B. Frayne, C. McCordic / Geoforum 65 (2015) 1–11

better explained by location of a household within a formal or informal area. Acknowledgements The publication of this paper was made possible by funding from the Canadian Government through the Canadian International Development Agency (CIDA) under its University Partners in Cooperation and Development (UPCD) Tier One Program. We wish to thank our colleagues in AFSUN for their assistance. References Amis, P., 1995. Making sense of urban poverty. Environ. Urban. 7 (1), 145–158. http://dx.doi.org/10.1177/095624789500700102. Barrett, C.B., 2010. Measuring food insecurity. Science (New York, NY) 327 (5967), 825–828. http://dx.doi.org/10.1126/science.1182768. Battersby, J., Crush, J., 2014. Africa’s urban food deserts. Urban Forum 25 (2), 143– 151. Battersby, J., Peyton, S., 2014. The geography of supermarkets in Cape Town: supermarket expansion and food access. Urban Forum 25 (2), 153–164. Beall, J., Fox, S., 2009. Cities and Development. Routledge, New York. Burdett, R., Sudjic, D., 2010. The Endless City: The Urban Age Project by the London School of Economics and Deutsche Bank’s Alfred Herrhausen Society. Phaidon Press, London. Canning, D., Bennathan, E., 2000. The Social Rate of Return on Infrastructure Investments. World Bank Policy Research Working Paper No. 2390. World Bank, New York. Corder, G.W., Foreman, D.I., 2009. Nonparametric Statistics for Non-Statisticians. John Wiley and Sons, Toronto. Crush, J., Caesar, M., 2014. City without choice: urban food insecurity in Msunduzi, South Africa. Urban Forum 25 (2), 165–175. Crush, J., Frayne, B., 2009. The Invisible Crisis: Urban Food Security in Southern Africa. AFSUN Urban Food Security Series No. 1. Idasa Publishing, Cape Town. Crush, J., Frayne, B. (Eds.), 2010. Surviving on the Move: Migration, Poverty and Development in Southern Africa. Development Bank of Southern Africa and Idasa Publishing, Cape Town. Crush, J., Drimie, S., Frayne, B., Caesar, M., 2011. The HIV and urban food security nexus in Africa. Food Security 3 (3), 347–362. http://dx.doi.org/10.1007/ s12571-011-0137-0. Crush, J., Frayne, B., Pendleton, W., 2012. The crisis of food insecurity in African cities. J. Hunger Environ. Nutr. 7 (2–3), 271–292. Duminy, J., Andreasen, J., Lerise, F., Odendaal, N., Watson, V. (Eds.), 2014. Planning and the Case Study Method in Africa the Planner in Dirty Shoes. Palgrave MacMillan, London. Faber, M., Schwabe, C., Drimie, S., 2009. Dietary diversity in relation to other household food security indicators. Int. J. Food Saf., Nutr. Public Health 2 (1), 1– 15. Food and Agriculture Organization, 2006. Food Security. Policy Brief Issue No. 2. FOA Food Security Program, Rome. Food and Agriculture Organization, 2012. The State of Food Insecurity in the World 2012: Economic Growth is Necessary but not Sufficient to Accelerate Reduction of Hunger and Malnutrition. FAO, Rome. Frayne, B., 2004. Migration and urban survival strategies in Windhoek, Namibia. Geoforum 35, 489–505. http://dx.doi.org/10.1016/j.geoforum.2004.01.003. Frayne, B., Pendleton, W., Crush, J., Acquah, B., Battersby-lennard, J., Bras, E., Tawodzera, G., 2010. The State of Urban Food Insecurity in Southern Africa. Urban Food Security Series No. 2. AFSUN, Cape Town. Frayne, B., Crush, J., McLachlan, M., 2014. Urbanization, nutrition and development in Southern African cities. Food Security 6, 101–112. Glaeser, E., 2011. Triumph of the City: How Our Greatest Invention Makes Us Richer, Smarter, Greener, Healthier, and Happier. Penguin Press, London and New York, NY. Joseph, S.-L., 2009. Tackling informality: why HIV/AIDS needs to be a critical component of urban development policies. Urban Forum 21 (2), 85–105. http:// dx.doi.org/10.1007/s12132-009-9073-x. Krausmann, F., Gingrich, S., Eisenmenger, N., Erb, K.H., Haberl, H., Fischer-Kowalski, M., 2009. Growth in global materials use, GDP and population during the 20th century. Ecol. Econ. 68 (10), 2696–2705. McLachlan, M., Garrett, J., 2008. Nutrition change strategies: the new frontier. Public Health Nutr. 11 (10), 1063–1075. Mitlin, D., Satterthwaite, D., 2013. Urban Poverty in the Global South: Scale and Nature. Routledge, London.

11

Morgan, K., Sonnino, R., 2010. The urban foodscape: world cities and the new food equation. Cambridge J. Regions, Econ. Soc. 3 (2), 209–224. Neuwirth, R., 2005. Shadow Cities: How 600 Million Squatters are Creating the Cities of Tomorrow. Routledge, London. OANDA, 2013. Historical Exchange Rates (Data File). . Ogun, T.P., 2010. Infrastructure and Poverty Reduction: Implications for Urban Development in Nigeria. WIDER Working Paper No. 2010/43. World Institute for Development Economics Research, Helsinki. Owuor, S.O., 2006. Bridging the Urban–Rural Divide: Multi-Spatial Livelihoods in Nakuru Town, Kenya. African Studies Centre, Leiden. Parnell, S., Oldfield, S., 2014. The Routledge Handbook on Cities of the Global South. Routledge, London. Parnell, S., Pieterse, E. (Eds.), 2014. Africa’s Urban Revolution. Zed Books, New York. Pendleton, W., Crush, J., Campbell, E., Green, T., Simelane, H., Tevera, D., de Vletter, F., 2006. Migration, Remittances and Development in Southern Africa. Southern African Migration Project Migration Policy Series No. 44. Idasa Publishing, Cape Town. Pieterse, E., 2014. Filling the void: an agenda for tackling African urbanisation. In: Parnell, S., Pieterse, E. (Eds.), Africa’s Urban Revolution. Zed Books, New York. Pieterse, E., Simone, A., 2013. Rogue Urbanism: Emergent African Cities. Jacana Media, Johannesburg. Pothukuchi, K., Kaufman, J.L., 1999. Placing the food system on the urban agenda: the role of municipal institutions in food systems planning. Agric. Hum. Values 16, 213–224. Pugh, C. (Ed.), 2000. Sustainable Cities in Developing Countries: Theory and Practice at the Millennium. Earthscan, London. Rakodi, C., 1997. The Urban Challenge in Africa: Growth and Management of its Large Cities. United Nations University Press, New York. Ravallion, M., 2007. Urban poverty. Financ. Dev. 44, 15–17. Robinson, J., 2011. The spaces of circulating knowledge: city strategies and global urban governmentality. In: McCann, E., Ward, K. (Eds.), Mobile Urbanism: Cities and Policy-Making in the Global Age. Minnesota University Press, Minneapolis, MN. Rocha, C., Lessa, I., 2009. Urban governance for food security: the alternative food system in Belo Horizonte, Brazil. Int. Plan. Stud. 14 (4), 389–400. Ruel, M.T., Haddad, L., Garrett, J.L., 1999. Some Urban Facts of Life: Implications for Research and Policy. FCND Discussion Paper No. 64. IFPRI, Washington. Sassen, S., 2012. Cities in a World Economy, fourth ed. Sage Publications, London. Saunders, D., 2011. Arrival City: The Final Migration and Our Next World. Random House LLC, Toronto. Simone, A., 2004. For the City Yet to Come: Changing African Life in Four Cities. Duke University Press, Durham, NC. Sonnino, R., 2009. Feeding the city: towards a new research and planning agenda. Int. Plan. Stud. 14 (4), 425–435. Steel, C., 2008. Hungry City: How Food Shapes Our Lives. Vintage, London. Tacoli, C., Bukhari, B., Fisher, S., 2013. Urban Poverty, Food Security and Climate Change. Human Settlements Working Paper 37. IIED, London. Tibaijuka, A., 2005. Cities without slums. Our Planet 16 (1), 12–13. Turok, I., 2014. Linking urbanisation and development in Africa’s economic revival. In: Pieterse, E., Parnell, S. (Eds.), Africa’s Urban Revolution, first ed. Zed Books, London, pp. 60–81. UNDP, 2012. Africa Human Development Report 2012: Towards a Food Secure Future. UN-Habitat, 2008. State of the World’s Cities 2008/2009 – Harmonious Cities. UNHabitat, Nairobi. UN-Habitat, 2011. State of the World’s Cities 2010/2011 – Cities for All: Bridging the Urban Divide. UN-Habitat, Nairobi. UN-Habitat, 2014. State of African Cities 2014. UN-Habitat, Nairobi. United Nations, 2015. Millennium Development Goals and Beyond 2015. United Nations, New York. USAID, 2007. Household Food Insecurity Access Scale (HFIAS) for Measurement of Food Access: Indicator Guide (Version 3). FANTA, Washington. USAID, 2010. Months of Adequate Household Food Provisioning (MAHFP) for Measurement of Household Food Access: Indicator Guide (Version 4). FANTA, Washington. Von Braun, J., Mccomb, J., Fred-Mensah, B., Pandya-Lorch, R., 1993. Urban Food Insecurity and Malnutrition in Developing Countries: Trends, Policies and Research Implications. International Food Policy Research Institute, Washington, DC. Watson, V., 2009. Seeing from the South: refocusing urban planning on the globe’s central urban issues. Urban Stud. 46, 2259–2275. World Bank, 2009. World Development Report 2009: Reshaping Economic Geography. World Bank, Washington, DC. World Food Programme, 2008. Vulnerability and Food Insecurity in Urban Areas of Lesotho. WFP, Rome. Yu, B., You, L., Fan, S., 2010. Toward a Typology of Food Security in Developing Countries. IFPRI Discussion Paper 00945. IFPRI, Washington.