Cities 143 (2023) 104584 Contents lists available at ScienceDirect Cities journal homepage: www.elsevier.com/locate/cities Strangers in a strange land: Mapping household and neighbourhood associations with improved wellbeing outcomes in Accra, Ghana Alicia C. Cavanaugh a, Jill C. Baumgartner b, Honor Bixby b, Alexandra M. Schmidt b, Samuel Agyei-Mensah c, Samuel K. Annim d, Jacqueline Anum d, Raphael Arku e, James Bennett f, Frans Berkhout g, Majid Ezzati f, Samilia E. Mintah d, George Owusu c, Jacob Doku Tetteh c, Brian E. Robinson a,* a Department of Geography, McGill University, Montreal, QC, Canada b Department of Epidemiology, Biostatistics and Occupational Health, McGill University, Montreal, QC, Canada c Department of Geography and Resource Development, University of Ghana, Legon, Ghana d Ghana Statistical Service, Accra, Ghana e Department of Environmental Health Sciences, University of Massachusetts Amherst, United States of America f Department of Epidemiology and Biostatistics, Imperial College London, London, UK g Department of Geography, King's College London, London, UK A R T I C L E I N F O A B S T R A C T Keywords: Urban poverty is not limited to informal settlements, rather it extends throughout cities, with the poor and Inequality affluent often living in close proximity. Using a novel dataset derived from the full Ghanaian Census, we Poverty investigate how neighbourhood versus household socio-economic status (SES) relates to a set of household Africa development outcomes (related to housing quality, energy, water and sanitation, and information technology) in Neighbourhood effects Segregation Accra, Ghana. We then assess “stranger” households' outcomes within neighbourhoods: do poor households fare Well-being better in affluent neighbourhoods, and are affluent households negatively impacted by being in poor neigh- bourhoods? Through a simple generalized linear model we estimate the variance components associated with household and neighbourhood status for our outcome measures. Household SES is more closely associated with 13 of the 16 outcomes assessed compared to the neighbourhood average SES. For 9 outcomes poor households in affluent areas fair better, and the affluent in poor areas are worse off. For two outcomes, poor households have worse outcomes in affluent areas, and the affluent have better outcomes in poor areas, on average. For three outcomes “stranger” households do worse in strange neighbourhoods. We discuss implications for mixed development and how to direct resources through households versus location-based targets. 1. Introduction overall well-being, but households with limited means may also be forced to make trade-offs in meeting basic needs and making such in- The global development agenda has focused on poverty reduction vestments. A dominant focus of development policy and the current since its inception. Adequate financial resources allow households to Sustainable Development Goals (i.e., SDG 10) has centred around how to protect themselves from risks and make investment choices that capi- support such households and break cycles of poverty and reduce overall talize on endowments, skills, or other natural advantages and make inequalities, particularly in Lower Middle-Income Countries (LMIC). choices, sometimes implying trade-offs, to maximize utility given their In addition to this focus on household-level poverty, a broad litera- budget constraints. These may include investments in education, asset ture across a myriad of disciplines now also investigates how one's accumulation, or other forms of capital that help buffer risk and improve location and the neighbourhood environment can have dramatic and * Corresponding author. E-mail addresses: alicia.cavanaugh@mcgill.ca (A.C. Cavanaugh), jill.baumgartner@mcgill.ca (J.C. Baumgartner), honor.bixby@essex.ac.uk (H. Bixby), alexandra. schmidt@mcgill.ca (A.M. Schmidt), sagyei-mensah@ug.edu.gh (S. Agyei-Mensah), samuel.annim@statsghana.gov.gh (S.K. Annim), jacqueline.anum@statsghana. gov.gh (J. Anum), rarku@umass.edu (R. Arku), umahx99@imperial.ac.uk (J. Bennett), frans.berkhout@kcl.ac.uk (F. Berkhout), majid.ezzati@imperial.ac.uk (M. Ezzati), similia.mintah@statsghana.gov.gh (S.E. Mintah), gowusu@ug.edu.gh (G. Owusu), brian.e.robinson@mcgill.ca (B.E. Robinson). https://doi.org/10.1016/j.cities.2023.104584 Received 23 January 2023; Received in revised form 28 July 2023; Accepted 21 September 2023 Available online 28 September 2023 0264-2751/© 2023 Elsevier Ltd. All rights reserved. A.C. Cavanaugh et al. C i t i e s 143 (2023) 104584 independent effects on household living conditions, well-being, and these differences, we use a simple variance-components regression health outcomes (Ansong et al., 2015; Doiron et al., 2020; Fowler & framework to separate the independent associations between house- Kleit, 2015; Mah et al., 2022; Montgomery & Hewitt, 2015; Parks et al., holds versus neighbourhood SES on household-level living standard 2014; Reardon & Bischoff, 2011; Rothwell & Massey, 2015; Sampson, outcomes. We find a number of household conditions are primarily 2008, 2011). Recent reviews of ‘neighbourhood effects’ suggest a associated with household SES, while fewer are associated more closely household's location can impact outcomes through both physical and with location SES. However, a key contribution from this work is that social contextual factors beyond just income. The physical environment even for outcomes where SES matters more, location interactions with includes commons that everyone shares (e.g. air quality, soils, etc.), household SES can still have major implications for living conditions. general living and working conditions (e.g. quality of housing stock, This provides evidence about the heterogeneity of neighbourhoods in reliable and safe employment), or access to services (e.g. education, Accra, and the implications of the poor living in better neighbourhoods. transport, health, sanitation) (Fowler & Kleit, 2015; Wen et al., 2003). Given the government's push for mixed development to encourage the Social environments can indicate the cohesiveness of community net- creation of economically diverse communities, we seek to understand works, levels of crime, and how a neighbourhood is perceived by the whether this type of planning benefits all households or whether it will wider community (Macintyre et al., 2002; Rothwell & Massey, 2015). need to be paired with more targeted interventions (Joseph et al., 2022). These neighbourhood characteristics often accumulate through multiple individual choices that can sort individuals and households into 2. Case study: Accra, Ghana particular neighbourhoods (Brock & Durlauf, 2001; Salhab et al., 2018; Walks, 2014; Wessel, 2022). Accra, the political and economic capital of Ghana, is one of Africa's Yet the literature on poverty and neighbourhood effects remains fastest growing cities. Recent economic growth in the Accra Metropol- relatively siloed. For example, neighbourhood context has been shown itan Area (hereafter simply referred to as Accra) is driven by growth in to mediate access to jobs, earning potential, and education, or limit the service sector – both in high-income jobs in the finance, insurance, access to public and health services, but the influence of household SES and real estate (FIRE) and information and communication technology relative to neighbourhood conditions are not directly compared (Chetty (ICT) sectors, and low-income informal employment. Despite growth in et al., 2016; Ludwig et al., 2012; Massey & Denton, 1993; Rothwell & high-skill formal occupations, Ghana's economy remains quite informal Massey, 2015; Smets & Salman, 2008). Neighbourhoods with larger – in Greater Accra, informal labour accounts for 73 % of all workers middle- or upper-class populations are associated with greater access to employed in services (Aryeetey & Baah-Boateng, 2016). Growth in these material and social resources that support local institutions (Browning & sectors contributes to a widening wage structure, increasing metropol- Cagney, 2003; Wilson, 1987), implying poorer households located in itan inequality across the city (Aryeetey & Baah-Boateng, 2016; Borel- affluent communities may be better off than a poor household in a poor Saladin & Crankshaw, 2009). community, and affluent households may be worse off when they are in Some areas developed under colonial urban planning projects that a poor neighbourhood, though these are not directly assessed. Massey were originally intended for civil servants and lease to European busi- (2001, 46) succinctly suggested that “living in a neighbourhood of nesses later became home for Ghanaian civil servants post-independence concentrated poverty accentuates and exacerbates whatever disadvan- (Agyei-Mensah & Owusu, 2010). Other areas were more recently tages come from living in a poor family, and that living in a neigh- developed, with many high-end developments built upon efforts to bourhood of concentrated affluence reinforces and strengthens the stimulate domestic growth (Gaisie et al., 2019), though many neigh- advantages of coming from an economically privileged family.” bourhoods remain highly mixed (Asiedu & Arku, 2009; Gaisie et al., Moreover, the literature on spatial inequalities and segregation sheds 2019). little light on how and when the neighbourhood context matters relative In addition to planning and policy decisions operating at a macro to household factors like income (Sharkey & Faber, 2014). Segregation scale, the socio-spatial organization of Accra is also shaped by in- alone has received a great deal of attention in higher- and lower-income dividuals making choices to select into areas based on access to country (HIC and LMIC, respectively) cities (Van Ham et al., 2021), but employment, and social connections. A primary consideration for the literature on segregation is generally silent on interactions with households is the rising costs of land, construction, and financing. For household factors. Although individual, household, and neighbourhood some, this pushes them from the formal housing market into over- effects have been well-examined in hierarchical frameworks, these crowded, poorly serviced informal settlements (Boamah, 2010; Gaisie typically do not focus on interaction (Diez Roux, 2001; Sharkey & Faber, et al., 2019). In 2009, around 58 % of the population of Accra resided in 2014; Chetty et al., 2022). Of the few studies we find that do, they focus overcrowded informal settlements (UN-HABITAT, 2009). In these areas on HICs and examine health outcomes (e.g., Kim & Cubbin, 2020; housing is often lived in rent-free (usually unowned housing or living Rachele et al., 2019). Another study in Ghana used a variance compo- with family members), though formal ownership or renting of housing is nents model to explore the interaction between neighbourhood context not uncommon. In poor and informal settlements, households often do and household education in determining aspects of health knowledge not have access to common public services like water and sanitation (Andrzejewski et al., 2009). Still, understanding the magnitude and infrastructure, and generally cannot afford private substitutes (Boamah, extent to which household versus neighbourhood pathways interact and 2010; Obeng-Odoom, 2011). Still, informal settlements and poor living affect important development outcomes could have important implica- conditions are not synonymous. Neighbourhood characteristics in part tions for designing place-based or household targeted policies (Ludwig culminate from individual choices around housing materials, lighting, et al., 2012; Partridge & Rickman, 2008). fuel choices, and information technology, though these can also be In this paper we ask how much does neighbourhood versus house- constrained by cost, availability, access, and supply (Danso-Wiredu, hold socio-economic status (SES) affect household outcomes? We 2018; MacTavish et al., 2023). Additionally, informal jobs often emerge disentangle these effects through use of a dataset derived from the full in public spaces and along major transit corridors (Oosterbaan et al., 100 % Ghanaian census. In collaboration with the Ghana Statistical 2012), sometimes bringing informal housing settlements along with Service, our unique dataset allows us to estimate the full distribution of them. SES and at the Enumeration Area (EA) (on average about 10,000m2, Thus while areas of concentrated affluence and poverty surely exist similar to a US census block) for Accra, Ghana. We investigate outcomes in Accra, poor, middle-class, and affluent households often co-locate. related to living standards and access to information among “stranger” Even in exclusive residential areas (i.e., gated communities), less well- households within neighbourhoods: do poor households fare better in off residents can live nearby (Asiedu & Arku, 2009). Despite the gov- affluent neighbourhoods, and are affluent households negatively ernment's encouragement of mixed income development, it is unclear impacted by being in poor neighbourhoods? To descriptively assess whether living in an affluent neighbourhood will benefit households 2 A.C. Cavanaugh et al. C i t i e s 143 (2023) 104584 that are less well off or lead to worse outcomes for outgroup households. utilities (Fuseini & Kemp, 2015; Korah et al., 2019). By the same token, in poor neighbourhoods, it is uncertain how atypical affluence might buffer a better-off household from local conditions of 3. Methods poverty and deprivation. Spatially, the coastal center of Accra (Fig. 1) is a European-style To understand how household-level resources and neighbourhoods central business district (CBD) (Accra Planning and Development Pro- shape development outcomes in a dense urban environment, we gramme, 1990) that originally served the port. To the north by the Odaw examine differences in SES groups outcomes when they are in the “in- River is a traditional market district (Grant & Yankson, 2003), while a group,” or surrounded by like households, and when they are in the more globally-focused CBD emerged along major roads near the newly- “out-group” (i.e., the “stranger” group) when they are surrounded by developed Airport City and Accra Mall south to Osu (Gaisie et al., 2019; different SES households. Testing how these “stranger” groups fare Grant & Nijman, 2002). The government has encouraged mixed-use shows how or whether living in a poor area reinforces the disadvantages development in this district, in part due to the rigidities of the old of the poor, as well as whether the resources of the affluent can ensure CBD (Oosterbaan et al., 2012). The density of households and pop- they meet basic standards of living. In this section we first discuss how ulations varies dramatically across the city as well, with population we develop household SES categories and neighbourhood SES classes clustered between the Densu Delta and surrounding the Odaw River and then briefly introduce the outcomes of interest we track in our data. floodplain, with pockets of communities (e.g., Osu, La, Burma Camp, Finally, we discuss the statistical modelling strategy used to separately Legon) scattered in the lower densityeast. Of the population clusters assess household compared to neighbourhood associations with the shown (Fig. 1), Old Fadama (Agbogbloshie), Nima, James Town, outcomes of interest. Chorkor, Sabon Zongo, and La are major informal settlements. Accra is vulnerable to many of the issues that attend unplanned development such as increased urban poverty, rising distributional and 3.1. Defining household- and neighbourhood-based consumption spatial inequality, and environmental degradation (Awumbila et al., categories 2014). Uneven development in Accra has produced a fragmented urban landscape characterized by concentration of the poor in some locations, 3.1.1. Defining household-based consumption categories persistent pockets of affluence, and a patchwork of formal and informal Since the census does not include income information, we use small growth along with increasing congestion, worsening environmental area estimation methods (SAE) to “borrow strength” from Ghana Living conditions, and growing inequalities in access to essential services like Standards Survey (GLSS6) and predict consumption for census transportation, health facilities, educational institutions, and public enumerated households. In collaboration with the Ghana Statistical Service (GSS), we applied this method to the 100 % 2010 Population and Fig. 1. Housing density and accra key locations. Adapted from Gaisie et al. (2019) and Accra Planning and Development Programme (1990). 3 A.C. Cavanaugh et al. C i t i e s 143 (2023) 104584 Housing Census microdata and developed a dataset with poverty and 2019; Weststrate et al., 2019) (see Appendix Table A.1 for additional inequality measures spatially identifiable at the enumeration area level details). These metrics indicate whether a household has ownership or (see the Methods Appendix). Since the GLSS does not sample the un- access to certain materials or services, but cannot speak to reliability, housed nor school-, hospital-, or prison-based populations, these are not availability, quality, or cost. Housing quality and the other service included in this analysis. These household and area consumption metrics outcomes are related to defense mechanisms that protect residents from have been used in analyses elsewhere (Alli et al., 2023; Bixby et al., life- and health-threatening pollutants, pathogens, and other environ- 2022; Clark et al., 2022; Tetteh et al., 2022). For this paper, we use these mental and social risks (Songsore & McGranahan, 1993). ICT outcomes consumption estimates to determine the number of households that fall are related to health as increased access to information can have a into three SES categories within each EA in Accra: households in the positive effect on the usage of health services (Abekah-Nkrumah et al., bottom 20th percentile (poor), 21st-79th percentiles (middle class), or 2014). the upper 20th percentile (affluent). These categories represent the In the models presented below, the dependent variable is the per- relative number of resources that households use to meet their needs centage of households with improved outcomes within an SES category. (and wants). In the housing quality domain, we include dwelling type, and wall, roof We have three reasons for using relative as opposed to absolute type, and flooring material. For energy use we examine the use of measures of poverty and affluence. First, although the national poverty improved lighting sources and cooking fuels. Sanitation outcomes line is set based on those unable to meet their food and non-food needs, include use of improved toilets, liquid waste, and solid waste disposal. there is a general tendency to underestimate urban poverty since the Drinking water outcomes are disaggregated into percent of households high cost of living within the city is not factored into to the determi- with piped, vended, and other improved sources to show variations in nation of poverty levels (Owusu & Wrigley-Asante, 2020; Owusu & water use. ICT outcomes provide information on individuals with mobile Yankson, 2007; Songsore, 2008). Second, in a context that has been phones or access to the internet, and households that have access to improving absolute levels of poverty, we chose a relative assessment of home computers and fixed landlines. Access to improved outcomes for poverty to understand the ability of those at the bottom to meet their some of these variables are contingent on household purchasing power, basic needs relative to their neighbours. A focus on the lower 20th while others, such as those related to neighbourhood amenities are more percentile group (“poor”) indicates a population that must make choices related to decisions made by private services and public authorities. and trade-offs to prioritize certain needs over others. Put another way, households in this “poor” group consume at least 40 % less than the 3.3. Modelling strategy median level of consumption in Accra. Third, our estimates of con- sumption are modelled. Predicted consumption estimates are most Our goal is to model how differences in rates of improved household confidently interpreted in a relative sense, removing the need for cali- outcomes are explained by household SES versus neighbourhood SES. bration of the distribution of predicted values to an externally valid Notably, our approach aims to separate out whether better living con- dataset of absolute consumption, an exercise that is fraught with chal- ditions are associated with household or neighbourhood SES. It is lenges. We tested model sensitivity using alternative SES thresholds for beyond the scope of this paper to develop models that attempt to explain the top/bottom 10th and 30th percentiles, which did not change the associations with improved outcomes through various covariates, rather qualitative outcomes of our findings. we focus on independent associations with household and neighbour- hood SES, and the interaction between them. 3.1.2. Defining neighbourhood-based consumption categories Outcomes for each SES group are summarized at the EA-level as We define neighbourhood SES for each EA using the Index of Con- proportions. We follow Papke and Wooldridge's (1996) quasi-likelihood centration at the Extremes (ICE), a metric that measures how segregated approach to estimate a generalized linear model using a logit-link an area is as the degree to which an area's population is concentrated function which respects the (0, 1) range of the dependent variable. into extremes of poor or affluent (Krieger et al., 2017; Massey, 2001). ( )/ The expected value of the improved outcome E(y|x) is estimated using a The ICE metric is defined as ICEj = Hj − Lj Tj where Hj is the number logistic function, and then parameters are estimated using Bernoulli of people living in high-income (top 20th percentile) households in quasi-likelihood methods. The advantage of this approach is that it does enumeration area j, L j is the number of people in low-income (bottom not assume any underlying structure to obtain y, only requiring that the 20th percentile) households, and Tj is the total number of individuals in conditional mean is specified correctly to ensure the predictions are the area. ICE values range from − 1 to 1, with negative values indicating bound between 0 and 1. concentrated poverty, and higher values concentrations of affluence. Our core model is specified as follows: Values around zero suggest areas could have a more even mix of q households or a more homogenous middle class. ICE quantifies extreme Yij = β0 + β1Hij + β2Nj + β3HijNj + εij (1) concentrations of household types with one metric, identifying areas that are most polarized. Here we define neighbourhood-based SES using where the Yij is the proportion of households with improved outcome q ICE tertiles: EAs in the bottom third are considered poorer neighbour- for household SES category i in enumeration area j, Hij is the household's hoods, while EAs at the top third of the ICE distribution are classified as SES category and Nj denotes neighbourhood SES category as indicated affluent neighbourhoods. We tested model sensitivity to alternative ICE by its ICE value. β1 estimates the independent effect of households' SES categories defined by the top/bottom ICE deciles or top/bottom ICE on the proportion with an improved outcome; β2 accounts for the same quintiles as well. at the neighbourhood level. β3 estimates the effect of the interaction of household and neighbourhood SES (HijNj) to jointly account for the type of household living in a specific neighbourhood type. We estimate this 3.2. Defining dependent variables: improved outcomes model in STATA 16 (StataCorp, 2019) using the glm function with the binomial family, logit link, and a robust variance estimator. We evaluate 16 “improved” outcome metrics as dependent variables that represent 5 different domains: housing quality, energy, sanitation, 3.4. Interpreting model results water, and information and communication technology (ICT) use. Def- initions for “improved” versus “unimproved” come directly from UN's The model presented in Eq. (1) estimates household versus neigh- Sustainable Development Goal guidelines (ITU, 2023; UNESCO, 2021; bourhood effects on improved outcomes independently and jointly via WHO-UNICEF, 2017) which have been used in past literature (e.g., the interaction term. We use two prediction methods to estimate inde- MacTavish et al., 2023) although are not without criticism (cf. Herrera, pendent effects of household versus neighbourhood SES while also 4 A.C. Cavanaugh et al. C i t i e s 143 (2023) 104584 accounting for their joint interactions: average marginal effects (AMEs) shows the extent to which these neighbourhoods are polarized, with and average adjusted predictions (AAPs). orange areas showing areas that are dominated by poor households and AMEs help us assess whether household SES or neighbourhood dark green areas shows those that are highly affluent. This map reflects affluence is more strongly associated with improved outcomes. AMEs the geography of the distributional maps, however, there is greater around 0 suggest little difference relative to the middle-income group. polarization in poor areas than affluent areas. The beige EAs show areas AAPs are a regression-adjusted response variable, which allows us to where there are a high proportion of middle-income households or interpret model results for different scenarios. To obtain AAPs, we use relatively equal mixes of poor and affluent households. These include the fitted model to predict the margin (average improved outcome rate) middle SES areas like Burma Camp and Tema and neighbourhoods to the for each group of interest (SES x Location) by fixing values of the west of the Odaw, and mixed areas like those near the university and covariates (Williams, 2012). AAPs allow us to estimate expected values those on the border of affluent and poor clusters. Similar to other LMIC for an outcome for each household SES category for each neighbourhood countries, poverty is concentrated in core areas, and affluence is ICE tertile. We spatially map the outcomes of AAPs for poor and affluent concentrated in historically affluent neighbourhoods and recently household in poor and affluent neighbourhoods. developed areas (van Ham et al., 2021). Table 1 provides a description of poor and affluent households living 4. Results in poor versus affluent neighbourhoods (EAs). Several characteristics are notable. At the household level, while about one-third of all households We first present a description of the spatial distribution of poverty are female-headed, the highest rates are in poor households in poor and affluence across Accra and briefly describe spatial segregation in the areas (41 %). Poor households are also more likely to live rent-free (22 % city. We then present results by topical domain (housing, energy, sani- and 37 % in poor and affluent areas, respectively) (typically in informal tation, water, and ICT) using AMEs and AAPs to answer our research settlements or supported by family members), while affluent households questions. rent their homes at higher rates (51 % and 52 % in affluent and poor areas, respectively). In Accra, engagement in agriculture is rare except 4.1. Descriptive results for affluent households within poor EAs. Characteristics of individuals in these areas show that poor households have a greater number of chil- Our data contain 501,851 households and 1,776,839 people living in dren under the age of 14 (34–35 %) while affluent households have a 11 sub-metropolitan districts and 2136 EAs in Accra. Across Accra, greater proportion of working-age adults. There are educational differ- while poor and affluent households exist in most neighbourhoods, rates ences as well, such as poor households have greater proportions that are of poverty and affluence vary significantly (Fig. 2). There are pockets of uneducated (12–23 %) or have just a basic education (36–38 %), and poverty in the urban core, in and surrounding the traditional CBD, affluent households are much more likely to have a post-secondary ed- extending up through industrial areas along the Odaw River. There are ucation (30–39 %). However, we also see that while poor households in high levels of concentrated poverty along the coastline, particularly near poor EAs have lower rates of secondary education than other groups (37 the Densu Delta in the west. There are high levels of poverty near the %), poor households in affluent EAs keep up with affluent households University of Ghana, and in areas near the airport. While many EAs with (48 %). high rates of poverty are in historically vulnerable neighbourhoods, such Regardless of SES, poor neighbourhoods have far greater concen- as Nima, Agbogbloshie, and Chokor, many are outside of places tradi- trations of Muslim households (20–27 %), while affluent neighbour- tionally considered deprived. Clusters of affluence are often directly hoods are predominantly Christian (85–90 %) supporting the idea that adjacent to poor areas in the core. Prominent affluent EAs can be seen in other characteristics also influence residential self-selection. planned western neighbourhoods such as Dansoman, Mamprobi & Kaneshie. East of the Odaw, there are high rates of affluence in low 4.2. Household SES versus neighbourhood average SES density neighbourhoods in a corridor that spans from the government Ministries to old colonial-era planned neighbourhoods to the newly Fig. 4 presents average marginal effects (AMEs) from Eq. (1) for developed CBD near the airport. There are also high rates of affluence improved household outcomes (for tabular results see Table A.2). We near the university. The middle SES group is spread throughout Accra, use AMEs to assess whether household SES or neighbourhood SES is but is most prominent in unplanned communities near Nima and New more strongly associated with improved outcomes, while accounting for Town as well as the Burma Camp military installation. their joint interaction. From these we can see whether household SES or While some neighbourhoods show highly concentrated poverty or location-based concentrations of SES are more associated with improved affluence, neighbourhoods often contain a mix of classes (Fig. 3). Fig. 3 living conditions. A point on each figure represents the AME for Fig. 2. The percent of households in an EA that are (a) poor, (b) middle income, and (c) affluent across Accra. 5 A.C. Cavanaugh et al. C i t i e s 143 (2023) 104584 Fig. 3. Segregation across Accra EAs. Panel A shows the ICE classification of each EA, and Panel B shows the EA percentage share of each SES group by ICE value. Black lines indicate tertile threshold boundaries, with grey marking alternative thresholds tested for robustness. household SES (Hi) or neighbourhood SES (Nj) as they relate to Table 1 improved well-being outcomes grouped into five categories: housing Population characteristics. (A), fuel use (B), sanitation (C), water (D), and ICT use (E). Since each Affluent EA (712) Poor EA (712) outcome is calculated separately, axes are scaled to reflect the range of Affluent Poor hh Affluent Poor hh AMEs in that domain. hh hh Row A presents outcomes related to housing: improved dwellings, Household characteristics (57,631) (15,034) (11,275) (54,581) floors, walls, and roofs. All rates showing in Row A are small relative to (n) effects in other panels but, most noticeably, affluent (+) and poor (− ) Head of Household households are strongly associated with rates of improved dwellings Female-headed 32 % 36 % 30 % 41 % (relative to middle income households). Most location-based marginal household Housing Tenure effects are smaller. Row B shows AMEs for improved energy sources, in Owner occupied 34 % 25 % 34 % 36 % which household SES is closely associated with the use of improved Renting 51 % 32 % 52 % 40 % cooking fuels while location effects have a more limited relationship. Rent-free 15 % 37 % 14 % 22 % Row C presents AMEs for sanitation outcomes. Here we see that Other 1 % 5 % 1 % 3 % Agrarian household SES status is associated with improved toilets and liquid Engaged in agriculture 8 % 1 % 11 % 0 % waste disposal, but in contrast to many other domains, location effects Crops 68 % 63 % 63 % 80 % are of a very similar magnitude. Row D shows detailed results for piped, Trees 9 % 4 % 10 % 5 % vended, and other drinking water sources. Household SES effects are Livestock 22 % 33 % 26 % 15 % strongly negatively related for piped water sources, showing affluent Fish 1 % 1 % 1 % 1 % Non-agricultural 92 % 99 % 89 % 100 % households are much more likely to use vended (bottled or sachet- household packet) water sources. A similar though slightly smaller effect is seen Individual characteristics (229,785) (45,397) (42,517) (199,256) for EAs. Finally, Row E presents AMEs related to ICT use, where we see (n) strong household effects but virtually no relationship with one's neigh- Age 0–14 23 % 34 % 22 % 35 % bourhood. These relationships persisted when testing model sensitivity 15–64 72 % 61 % 74 % 60 % with different thresholds (10 % and 20 % vs 33 %; Fig. 2B). The 64 plus 5 % 5 % 4 % 5 % magnitude of class and neighbourhood effects increased with stricter Religion definitions, and while levels decreased with more lenient cut-offs. Muslim 6 % 9 % 20 % 27 % In summary, there are two main takeaways from the AME results. Christian 90 % 85 % 77 % 65 % Other religions 2 % 2 % 1 % 1 % First, household SES is closely associated with consumption of improved No religion 1 % 4 % 2 % 6 % housing materials, energy use, solid and liquid waste disposal services, Nationality drinking water, and ICT use – all things that are universally available Ghanaian 94 % 98 % 94 % 97 % and are aspects of living conditions over which households have a great Foreign 6 % 2 % 6 % 3 % Education degree of choice. Second, for other key domains, notably water and Never schooled 3 % 12 % 5 % 23 % sanitation (improved toilets, liquid waste disposal services, and wall Ever schooled 97 % 88 % 95 % 77 % materials), location is as or more important than SES since these rely to Basic education 17 % 36 % 18 % 38 % some extent on publicly supplied infrastructure or markets that might be Secondary 41 % 48 % 47 % 37 % concentrated in affluent areas (e.g., vended water suppliers). Post-secondary 39 % 4 % 30 % 2 % Employment Sector Main employment 53 % 52 % 55 % 52 % Primary 1 % 1 % 1 % 2 % 4.3. How strangers fare: interactions between household and Secondary 9 % 18 % 10 % 13 % neighbourhood SES Tertiary 42 % 33 % 43 % 36 % Unemployed 47 % 48 % 45 % 48 % We now turn to average adjusted predictions (AAPs) to see how Note: Bold indicates values that are notably larger or smaller than other groups households living in specific neighbourhood types might fare, presented within that domain. in Figs. 4-8 (see Appendix Table A.2 for regression estimates). In each figure, panel A shows a line graph of the average predicted percent of affluent (solid line) or poorer (dashed line) households with improved 6 A.C. Cavanaugh et al. C i t i e s 143 (2023) 104584 Fig. 5. Average adjusted predictions for improved housing. Panel A: The pre- dicted % of households with ‘improved’ measures for four housing-related variables for the affluent (solid line) vs poor (dashed lines) households in affluent, mixed, and poor EAs. Panel B: Maps A, B, C, and D show the spatial distribution associated with improved dwellings (points A, B, C, and D, respectively) from the line graph in Panel A. Fig. 4. Average Marginal Effects (AMEs) for five household outcome themes. Each column shows the independent effect of an improved outcome as associ- affluent households in poor EAs, as in point (C), above. The bottom row ated with household- or neighbourhood-level SES. shows outcomes for poor households in affluent (B) and poor neigh- bourhoods (D). outcomes across the three neighbourhood EA types. Panel B presents a Fig. 5 presents AAPs for improved housing-related outcomes, with four-quadrant map of how outcomes for these groups differ spatially for the maps highlighting improved dwellings specifically. AAPs show that one highlighted outcome (noted by the corresponding bolded line and across all household and neighbourhood types, more than 90 % of letter markers in the graph). Each map quadrant shows the local rate of households have improved outcomes for walls, floors, and roofs. For improved outcomes for a subset of the population. The top left row dwelling types, however, the poor have much worse outcomes in affluent shows the spatial distribution of outcomes for affluent households in EAs (76.01 %) relative to in poor EAs (87.95 %), suggesting that having affluent EAs – corresponding to the situation at point (A) in the line affluent neighbours is not related to improved dwelling conditions for graph above. Similarly, the top right row shows the distribution for the poor. Spatially, we see most affluent households (Fig. 5.A and C) have very high rates of improved dwellings, save for those who live in 7 A.C. Cavanaugh et al. C i t i e s 143 (2023) 104584 Fig. 6. Average adjusted predictions for improved energy. Panel A: The pre- Fig. 7. Average adjusted predictions for improved sanitation. Panel A: The dicted % of households with ‘improved’ measures for lighting and cooking fuel predicted % of households with ‘improved’ sanitation measures for the affluent for the affluent (solid line) vs poor (dashed lines) households in affluent, mixed, (solid line) vs poor (dashed lines) households in affluent, mixed, and poor EAs. and poor EAs. Panel B: Maps A, B, C, and D show the spatial distribution Panel B: Maps A, B, C, and D show the spatial distribution associated with associated with improved cooking fuel (points A, B, C, and D, respectively) from improved toilets (points A, B, C, and D, respectively) from the line graph in the line graph in Panel A. Panel A. industrial areas in the urban core. Poor households (Fig. 5.B and D) have here are small. Outcomes for lighting are a much more better as most lower rates of improved dwellings. The poor have very low rates in affluent (99 %) and poor households (83 %) have improved lighting. affluent areas around the AMA, particularly in planned EAs. In poor Again, however, the poor have better lighting outcomes in poor neigh- areas, low rates of improved dwellings are highly concentrated in the bourhoods (86 %) than affluent areas (81 %) Spatially, the maps high- urban core, and near newly developed areas, but most areas do very light that disparities in cooking fuel are strongly associated with well. Overall, these results suggest, perhaps counterintuitively, the poor household and not neighbourhood type, although affluent households in have on average better housing conditions in poorer neighbourhoods. poorer core EAs have poorer outcomes than elsewhere. Fig. 6 presents AAPs for improved energy use. Disparities are Fig. 7 presents AAPs for improved sanitation outcomes on three noticeably large in affluent versus poor households for use of improved metrics: solid waste disposal, liquid waste disposal, and improved toilet cooking fuel where, on average, 85.8 % of the affluent and only 8.4 % of access, with the maps highlighting access to improved toilets. While use the poor use improved fuels. Poor households also have worse outcomes of improved solid waste disposal methods are common (95 % of affluent in poor areas compared to affluent ones, but the neighbourhood effects and around 90 % of poor households), only 40 % and 53 % of households 8 A.C. Cavanaugh et al. C i t i e s 143 (2023) 104584 map focusing on vended sources. Piped drinking water use among affluent households is highest in poor EAs (65.9 %) and lowest in affluent EAs (54.9 %). Vended water fills in these gaps, with the highest rates in affluent EAs (43 %) and the lowest in poor EAs (32 %). Poor households rely on piped water to a much greater degree, ranging from 79.5 % in affluent EAs to 83 % in poor EAs. Spatially, in affluent neighbourhoods, there is greater use of vended water with affluent households than poor households indicating there is service but affluent households choose to purchase water. However, this is highly variable in outlying neighbourhoods. In poor EAs, particularly those near the coast, there are very low rates of households using vended water, however vended water rates increase in affluent households further out from the Fig. 8. Average adjusted predictions for improved water. Panel A: The pre- dicted % of households with water sources for the affluent (solid line) vs poor (dashed lines) households in affluent, mixed, and poor EAs. Panel B: Maps A, B, C, and D show the spatial distribution associated with vended water (points A, B, C, and D, respectively) from the line graph in Panel A. have improved liquid waste disposal and toilets in the AMA on average. These are strongly patterned by both household and neighbourhood SES. Improved toilet rates for the affluent are higher in affluent EAs (92 %) than in poor EAs (50 %). Similarly, the poor have higher improved toilet rates in affluent EAs (62.5 %) as compared to poor EAs (21 %). This is our only result where the poor living in an affluent area fare better than affluent households living in poor areas. The maps show that there is a distinct spatial component where there are low rates of improved toilet access in the urban core poor areas regardless of household SES. In affluent neighbourhoods, where the affluent have Fig. 9. Average adjusted predictions for ICT use. Panel A: The predicted % of very high access, there is greater variation in access to improved toilets households with ICT use for the affluent (solid line) vs poor (dashed lines) for the poor, with the lowest rates appear in EAs that border poor areas. households in affluent, mixed, and poor EAs. Panel B: Maps A, B, C, and D show Fig. 8 shows AAPs for sources of drinking water with the quadrant the spatial distribution associated with mobile phone use (points A, B, C, and D, respectively) from the line graph in Panel A. 9 A.C. Cavanaugh et al. C i t i e s 143 (2023) 104584 core. therefore choices they make, but neighbourhoods also shape the services Fig. 9 shows AAPs for information, communication and technology and markets to which households have access. Table 2 summarises the (ICT) outcomes regarding mobile phone, internet use, desktop owner- household and neighbourhood effects, with summary implications for ship, and presence of a fixed land lines. In general, affluent groups have development policy, taking into account different ways household and much better ICT outcomes than all SES groups, as also reflected in Fig. 4. neighbourhood status interact for measures of well-being. The resulting Around 70 % of affluent households have a mobile phone, regardless of outcomes fall into four general patterns. Case A covers most of the neighbourhood, while about 57 %, 49 % and 45 % of poor households outcomes measured – household SES has a positive association with have a mobile in affluent, middle, and poor neighbourhoods, respec- improved wellbeing metrics, and generally neighbourhoods pull tively. For other ICTs, affluent households in affluent EAs have the stranger's outcomes closer to their mean (poor households do better in highest rates, and declines in poor EAs. Among other ICTs, the poor have affluent areas, affluent households do worse in poor areas). For floors all have very low rates of use. Spatially, there is a lot of homogeneity in and piped water – case B – the poor in affluent neighbourhoods have affluent EAs, and the poor have the high rates of access near the airport worse outcomes and the affluent in poor neighbourhoods have better and university. In poor EAs, affluent households have high rates of outcomes. In case C, strangers always have worse outcomes in a stranger mobile use except in Agbogbloshie, while usage drops for poor house- neighbourhood. Case D represent outcomes that show no relationship holds in the urban core and the wester coastline. with household or neighbourhood SES. The key takeaway from AAP results is that, even for the areas where Understanding how household SES and neighbourhoods interact for SES matters more (i.e., as demonstrated in Fig. 5), location and house- improved outcomes is important for directing limited resources to hold interactions can have major effects on outcomes. Concentrated ameliorate deprivation in different areas of concern. When household areas of affluence or poverty can improve or worsen outcomes for SES dominates, policy can target households based on household SES or ‘stranger’ households, though it seems the direction of these effects are socio-demographic traits. For example, Ghana's Livelihood Empower- somewhat unpredictable. In some instances, living in an affluent EA is ment Against Poverty (LEAP) cash transfer program targets the most associated with greater rates of improved outcomes for the poor, but vulnerable of households (Wodon, 2012; Cuesta et al., 2021). Alterna- there are also cases where living in an affluent EA seems to worsen tively, when neighbourhoods relate more strongly to deprivation, policy outcomes for poor households. Conversely, there are situations where makers can direct resources and programming in a more location-based living in a poor EA can worsen circumstances for affluent households. way. We discuss these results and their implications for policy design in the Neighbourhood effects capture aspects of public services, markets, next section. and environmental quality to which nearby households have access, which are often unevenly distributed. Affluent, high-status areas can 5. Discussion attract investment for further development and services infrastructure, while older, poorer neighbourhoods more often repel capital flows 5.1. Household versus neighbourhood effects (Gottlieb, 1997; Krätke, 2014). Concentrations of poverty can reproduce and reinforce disadvantages, though policy might implement place- Household consumption levels influence household budgets and based programs that can target such marginalized areas (Barca et al., Table 2 How do stranger groups fare in strange neighbourhoods? Household Stranger effect Development SES effect Poor in affluent n'hood Affluent in poor n'hood implicaons A. Strangers’ outcomes are closer to group outcomes. The poor in affluent neighbourhoods have beer outcomes; the affluent in poor neighbourhoods have worse outcomes. Toilets Liquid Waste Walls Vended water Fuel Mixed development reduces inequality Mobiles Phone lines Desktops Internet B. Marginalizaon exacerbated. The poor in affluent neighbourhoods have worse outcomes; the affluent in poor neighbourhoods have beer outcomes. Floors Mixed development Piped Water exacerbates inequality C. Stranger disadvantage. Strangers do worse (or no beer) in stranger neighbourhoods. Dwelling Mixed development Lighng makes all worse off (only hh income Solid Waste maers) D. No relaon. SES and neighbourhood have no associaon with improved outcomes. Roofs No impact Other Water Notes: Green indicates a positive and statistically significant effect, red denotes a negative and statistically significant effect, grey denotes effects indistinguishable from zero (Figs. 5-9). Darker shading indicates which effect has a larger AME (Fig. 4). 10 A.C. Cavanaugh et al. C i t i e s 143 (2023) 104584 2012; Partridge & Rickman, 2008). To reduce disparities in domains 5.3. Implications for mixed development that are particularly affected by location (e.g., sanitation), improving infrastructure and capacities in the most deprived places may make for How well does Massey's (2001) assertion that ‘living in a poor area effective public investment (Barca et al., 2012). Other than public reinforces disadvantages of the poor, and living in affluent neighbour- infrastructure investment, location-oriented policies might include hoods strengthens the advantages of the affluent’ hold up in a LMIC regulating private operators to prevent or subsidize unmanageable price context? In many cases our results generally support the assertion. Poor increases for basic needs in poor communities or to ensure operators of in poor areas have worse outcomes in use of improved cooking fuel, critical services provide adequate and complete coverage over all toilets, vended water use, and all ICT metrics. Similarly, affluent neighbourhoods (Appiah-Effah et al., 2019; Oteng-Ababio et al., 2013). household living in affluent areas have better outcomes in dwelling types, cooking fuel, all sanitation outcomes, and vended water, and ICT 5.2. Application to drinking water outcomes (though mobile ownership is constant). However, Massey's assertion is far from a universal truth. Table 2 A complementary household- and neighbourhood-based strategy suggests that moving a poor household from a poor area to a middle would be useful in contending with Accra's water challenges. The two income or affluent neighbourhood only signals an improvement in 9 out dominant sources of drinking water are through the public piped of 16 cases, only slightly better than half of our wellbeing outcomes. In network and through vended (purchased) water outlets (Moulds et al., contrast, for a number of housing characteristics (dwelling type, floors, 2022; Tetteh et al., 2022). Affluent, planned communities such as roofs) as well as solid waste disposal and lighting, poor households in Airport Residential Areas, Ridges, and Cantonments often have better poor areas have better outcomes than they do in affluent areas. This may access to the public water supply network than poorer communities, indicate that “pockets” of poverty within affluent areas may be partic- which influences access and pricing (Mahama et al., 2014; Tetteh et al., ularly precarious or informal, or costs to improved housing and services 2022). While most households have access to improved drinking water, in affluent areas may be even more prohibitive than they are in poorer the Ghana Water Company has not been able to meet growing demand settlements. In these areas concentrated affluence worsens outcomes for and thus must ration water delivery at times (Stoler et al., 2013; Tetteh ‘stranger’ households since the poor may not have the same community et al., 2022). Further, rationing can be spatially inconsistent – some resources, social network, or access to services that often underpin neighbourhoods receive water every day while in other areas delivery is welfare and community health. On the other hand, in most cases being sporadic or non-existent (Dapaah & Harris, 2017). Inconsistency in affluent in a poor area is associated with poorer outcomes than they piped water delivery carries an additional risk since negative water would have in a affluent neighbourhood: moving to a poor area de- pressures risk seepage and cross contamination with raw sewage (Stoler creases affluent well-being in 11 out of 16 indicators. Overall promotion et al., 2013). Thus, piped water has become something of an inferior of mixed-income neighbourhoods is likely good for a number of out- good and vended water is perceived as the safer, healthier choice (Stoler comes, as shown in Table 2.A. Mixed development may be further et al., 2014). According to Moulds et al., 2022, over half of urban justified for a number of dynamic and intergenerational reasons (e.g., households in the region who use sachet water as their primary source Chetty et al., 2016, 2014, 2022), however, in some cases it can exac- also have a piped supply connection. erbate inequality (Table 2.B) or result in worse outcomes for stranger Our results indicate poor households are highly reliant on piped households (Table 2.C). Targeted policies to aid vulnerable commu- water, while affluent households are more likely to use vended water. nities, regardless of where they live, are still needed. For example, poor Previous work has found that when households turn to vended sources households in affluent areas where infrastructure is available would outside of the piped system due to rationing and health concerns, benefit from policies focused on providing assistance in accessing these affluent households are more likely to use bottled water, and poorer services. households are likely to purchase sachet water (Moulds et al., 2022; This study has some data limitations that also help put our results in Tetteh et al., 2022). In areas particularly plagued by unmaintained context. To model the aggregate effects of household versus neigh- infrastructure and rationing, poor households must choose to use ven- bourhood characteristics, we abstract from contextual factors that could ded water despite the high markup (Moulds et al., 2022; Stoler et al., be important determinants of the outcomes we investigate and operate 2014; Tetteh et al., 2022). However, there are many households for through the household or neighbourhood. We also do not take an explicit which that is not a feasible strategy as it would put too much pressure on spatial regression approach so we do not control for clustering of limited budgets. neighbourhood types. Instead, we account for space using random ef- To address these challenges programs could help make vended water fects. Finally, the patterns we assess are the product of individual affordable to vulnerable households in areas where infrastructure is household residential choices. As such, our results are indicative of limited, strengthen, and maintain the water supply system to ensure empirical patterns but are not causal estimates. Still, these patterns and safety and prevent loss of public potable water supply, or regulate pri- results help us see what outcomes are possible, and thus likely, when vate sector water provision to ensure equitable and safe service across promoting mixed-use development. the city. In areas where there are high rates of vended water use among the affluent, but not the poor, indicates there may be issues with the 6. Conclusion quality of piped water (allowing those who can afford to choose the better option to do so). Other areas, such as nearby the university, have Good quality housing is crucial to ensuring household welfare and low rates of vended water use among the affluent and point to trust in health; poor housing and environmental conditions put residents at risk the quality of water, and high vended water rates among the poor. This of health problems such as infectious diseases, stress, and depression indicates that access to connecting to piped-water infrastructure may be (Yakubu et al., 2014). Poor quality, overcrowded, and badly located a major barrier for poorer households. The former case may require housing not only influences physical health, but mental well-being investment in infrastructure to improve safety and quality of a more which affects workforce participation and educational attainment affordable water source. The latter example may need cash assistance or (Yakubu et al., 2014). Overall housing conditions are affected by type of expansion of the existing system to reach households and provide water dwelling, construction materials, household facilities and the coverage at a more affordable price. Appropriate targeting of such policies would of neighbourhood services (Gibson et al., 2011). ICT access has signifi- be best informed by understanding household and neighbourhood cant implications for potential educational attainment and access to interactions. formal banking services. Examining the disparities in housing quality and service provision can inform policy efforts in Accra. Identifying how household SES and neighbourhood effects jointly influence improved 11 A.C. Cavanaugh et al. C i t i e s 143 (2023) 104584 outcomes can help decision makers choose how to direct resources via Samuel Agyei-Mensah: Writing – Review & editing; area-based or population-based programs. While 2010 data reflects the Samuel K Annim: Resources, Data access Accra as it was ten years ago, this analysis serves as a baseline for future Jaqueline Anum: Resources, Data access analysis of the just completed 2021 census. Raphael Arku: Writing – Review & editing; We find household SES is more closely related to higher rates of James Bennett: Writing – Review & editing; improved housing materials, energy use, solid waste disposal services, Frans Berkhout: Writing – Review & editing; drinking water, and ICT use. In these cases, policy makers might be well Majid Ezzati: Writing – Review & editing; Project administration; served to focus on household-level barriers to consumption of these Funding acquisition; services. A household's location is as or more important than household Samilia E Mintah: Resources, Data access SES in the case of walls, improved toilet use, improved disposal of liquid George Owusu: Writing – Review & editing; waste, and vended water use. In these cases, policy makers may consider Jacob Doku Tetteh: Writing – Review & editing; an area-based approach that target particularly deprived areas. Finally, Brian E Robinson: Conceptualization; Methodology; Writing – Orig- our results suggest that location interacts with household status – often inal Draft; Supervision further advantaging affluent households in affluent areas or increase disadvantages experienced by poor households in poor areas and being a Declaration of competing interest ‘stranger’ in one location brings you closer to that location's mean. Still, we also find cases where being poor in an affluent neighbourhood is The authors declare that they have no known competing financial associated with larger disadvantages – namely for housing related out- interests or personal relationships that could have appeared to influence comes – suggesting pockets of poverty may present particularly isolated the work reported in this paper. and marginalized living conditions. Ultimately, both personal cost as well as infrastructural or neighbourhood-level barriers stymie efforts to Data availability address most environmental and health challenges. As such, it is important to identify where these issues are most critical between and Data will be made available on request. among household SES groups to give policy makers a more accurate portrait of disparities. Acknowledgements CRediT authorship contribution statement This work is supported by the Pathways to Equitable Healthy Cities grant from the Wellcome Trust [209376/Z/17/Z]. For the purpose of Alicia C. Cavanaugh: Conceptualization; Methodology; Analysis; Open Access, the author has applied a CC BY public copyright licence to Writing – Original Draft; Data curation; Visualization any Author Accepted Manuscript version arising from this submission. Jill C. Baumgartner: Writing – Review & editing; We additionally thank participants and residents of Accra, Ghana and Honor Bixby: Writing – Review & editing; participants at the Pathways Annual meetings for constructive feedback Alexandra Schmidt: Writing – Review & editing; and comments on oral versions of this manuscript. Appendix A. SAE methods description Our data are uniquely derived from the 100 % 2010 Population and Housing Census (PHC) collected by the GSS. This procedure is documented in detail in Cavanaugh et al., 2022. The Ghanian census is spatially identifiable at the EA-level (on average ~10,000 m2), however, the census does not contain income or consumption data that are typically used as an indicator of a household's socioeconomic status (SES). In the developing context, this is often due to expense of collecting data and difficulty obtaining answers. Even when surveys include this information, collected data is only available for larger geographical units or the sample size is not large enough to produce accurate estimates (Nyugen et al., 2017). Since we do not have in- formation on income, we measure consumption. While income is preferred when examining economic standards of living, the benefit of using consumption is that it is an appropriate measure of “someone's actual standard of living regardless of how it is attained” (Johnson et al., 2005). The GLSS6 provides information on many categories of household expenditures, and expenditures including rent is used as our measure for household consumption. To estimate consumption for households with the 100 % census, we first use another dataset – the Ghana Living Standards Survey (GLSS6), which is measured in detail – to develop a statistical relationship between consumption and common predictors that capture demographics, education, employment, and housing conditions (see below for additional detail on these). Notably, these predictors are also available in exactly the same format in the census. Following the small area estimation (SAE) literature (Elbers et al., 2003; Molina & Rao, 2010), we first fit a linear mixed model with area-level random intercepts to GLSS6 survey data. Our dependent variable is consumption divided by the square root of household size to account for household-level economies of scale (Buhmann et al., 1988), SAE methods that “borrow strength” from the detailed information in the GLSS6 and apply it to the more spatially detailed and representative census data Using the parameter estimates from the GLSS6 consumption model, we then predict consumption for the full census data. We simulate this model 100 times, drawing from variance around parameter estimates and assign the average of the simulated values as the consumption. Appendix References Buhmann, B., Rainwater, L., Schmaus, G., & Smeeding, T. M. (1988). Equivalence scales, well-being, inequality, and poverty: sensitivity estimates across ten countries using the Luxembourg Income Study (LIS) database. Review of income and wealth, 34(2), 115–142. XXXX [anonymized by the authors] (2022). Locating poverty and inequality: an application of small area estimation methods using survey and census data from Ghana. Manuscript in preparation. Elbers, C., Lanjouw, J. O., & Lanjouw, P. (2003). Micro-level estimation of poverty and inequality. Econometrica, 71(1), 355–364. Johnson, D. S., Smeeding, T. M., & Torrey, B. B. (2005). Economic inequality through the prisms of income and consumption. Monthly Lab. Rev., 12 A.C. Cavanaugh et al. C i t i e s 143 (2023) 104584 128, 11. Molina, I., & Rao, J. N. K. (2010). Small area estimation of poverty indicators. Canadian Journal of Statistics, 38 (3): 369–385. Nguyen, M. C., Corral, P., Azevedo, J. P., & Zhao, Q. (2017). Small area estimation: An extended ELL approach. World Bank. Appendix Table A.1 Improved/Unimproved household characteristic classification. Improved Unimproved Housing Dwelling type Compound house (rooms) Huts/Buildings (different compound) Flat/Apartment Huts/Buildings (same compound) Semi-detached house Improvised home (kiosk/container, etc.) Separate house Living quarters attached to office/shop Other Tent Uncompleted building Wall material Burnt bricks Bamboo Cement blocks/Concrete Mud brick/Earth Landcrete Other Metal sheet/Slate/Asbestos Palm leaf/Thatch (grass)/Raffia Stone Wood Floor material Burnt brick Earth/Mud Cement/Concrete Other Ceramic/Porcelain/Granite/Marble tiles Wood Stone Terrazzo/Terrazzo tiles Vinyl tiles Roof material Cement/Concrete Bamboo Metal sheet Mud/Mud bricks/Earth Roofing tile Other Slate/Asbestos Thatch/Palm leaf or Raffia Wood Energy Lighting Electricity (mains) Candle Electricity (private generator) Crop residue Solar energy Firewood Flashlight/Torch Gas lamp Kerosene lamp Other Cooking Fuel Electricity Animal waste Gas Charcoal Kerosene Crop residue None, no cooking Other Saw dust Wood Sanitation Toilet KVIP Bucket/Pan Pit latrine No facilities (bush/beach/field) W.C. Other Public toilet (WC, KVIP, Pit, Pan, etc.) Liquid waste disposal Through drainage into a pit (soak away) Other Through drainage system into a gutter Thrown into gutter Through the sewerage system Thrown onto compound Thrown onto the street/outside Solid waste disposal Collected Buried by household Public dump (container) Burned by household Public dump (open space) Dumped indiscriminately Other Drinking Water Source Pipe-borne Pipe-borne inside dwelling Dugout/Pond/Lake/Dam/Canal Pipe-borne outside dwelling Other Vendor Bottled water Rainwater Sachet water River/Stream Other Bore-hole/Pump/Tube well Tanker supply/Vendor provided Protected spring Unprotected spring Protected well Unprotected well Public tap/Standpipe ICT Mobile Phone Owns mobile phone Does not own mobile phone Internet Access Accesses the internet Does not access the internet Desktops Household has desktop or laptop Household has no desktop or laptop Fixed phone line Household has fixed phone line Household has no fixed phone line 13 A.C. Cavanaugh et al. C i t i e s 143 (2023) 104584 Appendix Table A.2 GLM regression results. obs. = 40,798 EA = 2136 Improved SES ICE SES#ICE Intercept AIC BIC Log living pseudolikelihood conditions Aff. Poor Aff. Poor Aff. Aff. Poor#Aff. Poor#Poor #Aff. #Poor Housing Dwelling 0.973* − 0.826* − 0.261* − 0.038 0.327* − 0.273* − 0.274* 0.338* 2.514* 0.487 − 421,432 − 9919.800 Wall 0.542* − 0.156* 0.488* − 0.407* − 0.120 − 0.214* − 0.056 0.017 3.120* 0.301 − 424,732 − 6120.713 Floor 0.371 − 0.278 − 0.138 0.208 0.121 − 0.213 − 0.047 0.065 2.986 0.33 − 424,120 − 6721.978 Roof 0.091 − 0.156 − 0.106 − 0.1 − 0.171 − 0.153 − 0.135 − 0.084 4.314 0.134 − 429,654 − 2730.524 Energy Lighting 1.577 − 1.473 − 0.034 0.16 0.344 − 0.645 − 0.111 0.064 3.099 0.340 − 425,231 − 6933.437 Fuel 2.350* − 2.123* 0.310* − 0.291* 0.097* − 0.487* 0.004 0.17* − 0.340* 0.886 − 413,765 − 18,053.303 Sanitation Solid Waste 0.255* − 1.235* − 0.654* 0.076 0.532* − 0.182 − 0.001 − 0.057 3.517* 0.324 − 424,154 − 6591.175 Liquid 0.787* − 0.160* 0.627* − 0.069* 0.132* − 0.273* − 0.215* 0.03 − 0.7* 1.101 − 404,741 − 22,443.625 Waste Toilet 1.104 − 0.290 1.227 − 0.975 0.181 − 0.111 − 0.375 0.011 − 0.047 0.959 − 412,273 − 19,545.729 Drinking Water Piped − 0.452* 0.694* − 0.2901* 0.08* − 0.042 0.048 − 0.028 − 0.157* 0.983* 0.991 − 407,377 − 20,214.71 Vendor 0.459* − 0.718* 0.286* − 0.072* 0.038 − 0.074 0.010 0.158* − 1.059* 0.969 − 407,669 − 19,746.87 Other − 0.054 − 0.279* 0.034 − 0.079 0.157 0.37 0.209 − 0.14 − 4.716* 0.087 − 430,738 − 1760.847 ICT Mobiles 0.577* − 0.512* 0.108* − 0.011 − 0.152* 0.017 0.168* − 0.147* 0.369* 0.919 − 427,772 − 18,730.41 Internet 1.523* − 1.77* 0.155* 0.009 − 0.093* − 0.139* 0.278* − 0.15* − 2.11* 0.550 − 426,206 − 11,216.269 Desktops 2.381* − 2.569* 0.111* − 0.093* 0.079* − 0.227* 0.287 0.179 − 2.256 0.537 − 422,470 − 10,938.709 Phone line 1.989* − 0.991* 0.820* − 0.183* − 0.274* − 0.119 0.168 − 0.032 − 3.863* 0.278 − 426,849 − 5655.583 Estimated in Stata using specification suggested by Baum (2008): glm Y i.SES##EA_SES, link(logit) family(binomial) vce(robust). * Denotes statistical significance at the 0.05 level. References Boamah, N. A. (2010). Housing affordability in Ghana: A focus on Kumasi and Tamale. Ethiopian journal of environmental studies and management, 3(3). Abekah-Nkrumah, G., Guerriero, M., & Purohit, P. (2014). ICTs and maternal healthcare Borel-Saladin, J., & Crankshaw, O. (2009). Social polarisation or professionalisation? Another look at theory and evidence on deindustrialisation and the rise of the service utilization. Evidence from Ghana. International Journal of Social Economics, 41(7), sector. Urban Studies, 46(3), 645–664. 518–541. Accra Planning and Development Programme. (1990). Strategic plan for the Greater Accra Brock, W. A., & Durlauf, S. N. (2001). Discrete choice with social interactions. The Review of Economic Studies, 68(2), 235–260. metropolitan area: Context report (Vol. 1). Ministry of Local Government (Ghana), Department of Town and Country Planning (Ghana), United Nations Development Browning, C. R., & Cagney, K. A. (2003). Moving beyond poverty: Neighbourhood structure, social processes, and health. Journal of Health and Social Behavior, Programme, United Nations Centre for Human Settlements. 552–571. Agyei-Mensah, S., & Owusu, G. (2010). Segregated by neighbourhoods? A portrait of ethnic diversity in the neighbourhoods of the Accra Metropolitan Area, Ghana. Chetty, R., Hendren, N., & Katz, L. F. (2016). The effects of exposure to better neighbourhoods on children: New evidence from the moving to opportunity Population, Space and Place, 16(6), 499–516. Alli, A. S., Clark, S. N., Wang, J., Bennett, J., Hughes, A. F., Ezzati, M., Brauer, M., experiment. American Economic Review, 106(4), 855–902. Chetty, R., Hendren, N., Kline, P., & Saez, E. (2014). Where is the land of opportunity? Nimo, J., Bedford-Moses, J., Baah, S., Cavanaugh, A., Agyei-Mensah, S., Owusu, G., Baumgartner, J., & Arku, R. E. (2023). High-resolution patterns and inequalities in The geography of intergenerational mobility in the United States. The Quarterly Journal of Economics, 129(4), 1553–1623. ambient fine particle mass (PM(2.5)) and black carbon (BC) in the Greater Accra Chetty, R., Jackson, M. O., Kuchler, T., Stroebel, J., Hendren, N., Fluegge, R. B., … Metropolis, Ghana. Science of the Total Environment, 875, 162582. Andrzejewski, C. S., Reed, H. E., & White, M. J. (2009). Does where you live influence Wernerfelt, N. (2022). Social capital I: Measurement and associations with economic mobility. Nature, 608, 108–121. what you know? Community effects on health knowledge in Ghana. Health & Place, 15(1), 228 238. Clark, S. N., Abosede, S. A., Ezzati, M., Brauer, M., Toledano, M. B., Nimo, J., … Arku, R. – (2022). Spatial modelling and inequalities in environmental noise in Accra, Ghana. Ansong, D., Ansong, E. K., Ampomah, A. O., & Adjabeng, B. K. (2015). Factors Environmental Research, 214, 113932. contributing to spatial inequality in academic achievement in Ghana: Analysis of district-level factors using geographically weighted regression. Applied Geography, Cuesta, J., Devereux, S., Abdulai, A. G., Gupte, J., Ragno, L. P., Roelen, K., … Spadafora, T. (2021). Urban social assistance: Evidence, challenges and the way 62, 136–146. Appiah-Effah, E., Duku, G. A., Azangbego, N. Y., Aggrey, R. K. A., Gyapong-Korsah, B., & forward, with application to Ghana. Development Policy Review, 39(3), 360–380. Danso-Wiredu, E. Y. (2018). Housing strategies in low income urban communities in Nyarko, K. B. (2019). Ghana’s post-MDGs sanitation situation: An overview. Journal of Water, Sanitation and Hygiene for Development, 9(3), 397 415. Accra, Ghana. GeoJournal, 83(4), 663–677. – Aryeetey, E., & Baah-Boateng, W. (2016). Understanding Ghana s growth success story Dapaah, E. K., & Harris, L. M. (2017). Framing community entitlements to water in ’ Accra, Ghana: A complex reality. Geoforum, 82, 26–39. and job creation challenges. In WIDER working paper no. 2015/140. Helsinki: The United Nations University world Institute for Development Economics Research (UNU- Diez Roux, A. V. (2001). Investigating neighborhood and area effects on health. American Journal of Public Health, 91(11), 1783–1789. WIDER). Asiedu, A. B., & Arku, G. (2009). The rise of gated housing estates in Ghana: Empirical Doiron, D., Setton, E. M., Shairsingh, K., Brauer, M., Hystad, P., Ross, N. A., & Brook, J. R. (2020). Healthy built environment: Spatial patterns and relationships of multiple insights from three communities in metropolitan Accra. Journal of Housing and the exposures and deprivation in Toronto, Montreal and Vancouver. Environment Built Environment, 24(3), 227–247. Awumbila, M., Owusu, G., & Teye, J. K. (2014). Can rural-urban migration into slums International, 143, 106003. Fowler, C. S., & Kleit, R. G. (2015). The effects of industrial clusters on the poverty rate. reduce poverty? Evidence from Ghana. In Migrating out of poverty research Programme consortium, working paper 13. Economic Geography, 90(2), 129–154. Fuseini, I., & Kemp, J. (2015). A review of spatial planning in Ghana’s socio-economic Barca, F., McCann, P., & Rodríguez-Pose, A. (2012). The case for regional development intervention: Place-based versus place-neutral approaches. Journal of Regional development trajectory: A sustainable development perspective. Land Use Policy, 47, Science, 52(1), 134 309–320. –152. Gaisie, E., Kim, H. M., & Han, S. S. (2019). Accra towards a city-region: Devolution, Bixby, H., Bennett, J. E., Bawah, A. A., Arku, R. E., Annim, S. K., Anum, J. D., … Baumgartner, J. (2022). Quantifying within-city inequalities in child mortality across spatial development and urban challenges. Cities, 95, 102398. Gibson, M., Petticrew, M., Bambra, C., Sowden, A. J., Wright, K. E., & Whitehead, M. neighbourhoods in Accra, Ghana, BMJ Open. (2011). Housing and health inequalities: A synthesis of systematic reviews of 14 A.C. Cavanaugh et al. C i t i e s 143 (2023) 104584 interventions aimed at different pathways linking housing and health. Health & Parks, M. J., Dodoo, F. N. A., & Ayernor, P. K. (2014). Applying neighbourhood-effect Place, 17(1), 175–184. research to a global south city: A case study of collective efficacy in Accra, Ghana’s Gottlieb, P. D. (1997). Neighbourhood development in the metropolitan economy: A low-income areas. The Global South, 8(2), 119–138. policy review. Journal of Urban Affairs, 19(2), 163–182. Partridge, M. D., & Rickman, D. S. (2008). Place-based policy and rural poverty: Insights Grant, R., & Nijman, J. (2002). Globalization and the corporate geography of cities in the from the urban spatial mismatch literature. Cambridge Journal of Regions, Economy less-developed world. Annals of the Association of American Geographers, 92(2), and Society, 1(1), 131–156. 320–340. Rachele, J. N., Schmid, C. J., Brown, W. J., Nathan, A., Kamphuis, C. B., & Turrell, G. Grant, R., & Yankson, P. (2003). Accra. Cities, 20(1), 65–74. (2019). A multilevel study of neighborhood disadvantage, individual socioeconomic Herrera, V. (2019). Reconciling global aspirations and local realities: Challenges facing position, and body mass index: Exploring cross-level interaction effects. Preventive the sustainable development goals for water and sanitation. World Development, 118, Medicine Reports, 14, 100844. 106–117. Reardon, S. F., & Bischoff, K. (2011). Income inequality and income segregation. International Telecommunication Union. (2023). Measuring digital development, facts and American Journal of Sociology, 116(4), 1092–1153. figures: Focus on least developed countries. Geneva, Switzerland: ITU. Rothwell, J. T., & Massey, D. S. (2015). Geographic effects on intergenerational income Joseph, M. L., Arthur, I. K., & Botchway, E. K. (2022). Promoting a geography of mobility. Economic Geography, 91(1), 83–106. opportunity in Accra, Ghana: Applying lessons from mixed-income development Salhab, R., Le Ny, J., & Malhamé, R. P. (2018). Dynamic collective choice: Social optima. successes and shortcomings. Journal of Urban Affairs, 44(3), 305–324. IEEE Transactions on Automatic Control, 63(10), 3487–3494. Kim, Y., & Cubbin, C. (2020). Neighborhood poverty histories and physical activity Sampson, R. J. (2008). Moving to inequality: Neighbourhood effects and experiments among children: Findings from the geographic research on wellbeing (GROW) study. meet social structure. American Journal of Sociology, 114(1), 189–231. American Journal of Health Promotion, 34(8), 876–885. Sampson, R. J. (2011). Neighbourhood effects, causal mechanisms and the social Korah, P. I., Matthews, T., & Tomerini, D. (2019). Characterising spatial and temporal structure of the city. In P. Demeulenaere (Ed.), Analytical sociology and social patterns of urban evolution in Sub-Saharan Africa: The case of Accra, Ghana. Land mechanisms (pp. 227–249). Cambridge, UK: Cambridge University Press. Use Policy, 87, 104049. Sharkey, P., & Faber, J. W. (2014). Where, when, why, and for whom do residential Krätke, S. (2014). Cities in contemporary capitalism. International Journal of Urban and contexts matter? Moving away from the dichotomous understanding of Regional Research, 38(5), 1660–1677. neighbourhood effects. Annual Review of Sociology, 40, 559–579. Krieger, N., Feldman, J. M., Waterman, P. D., Chen, J. T., Coull, B. A., & Hemenway, D. Smets, P., & Salman, T. (2008). Countering urban segregation: Theoretical and policy (2017). Local residential segregation matters: Stronger association of census tract innovations from around the globe. Urban Studies, 45(7), 1307–1332. compared to conventional city-level measures with fatal and non-fatal assaults (total Songsore, J. (2008). Environmental and structural inequalities in Greater Accra. Journal and firearm related), using the index of concentration at the extremes (ICE) for of the International Institute, 16(1). racial, economic, and racialized economic segregation, Massachusetts (US), Songsore, J., & McGranahan, G. (1993). Environment, wealth and health: Towards an 1995–2010. Journal of Urban Health, 94(2), 244–258. analysis of intra-urban differentials within the Greater Accra Metropolitan Area, Ludwig, J., Duncan, G. J., Gennetian, L. A., Katz, L. F., Kessler, R. C., Kling, J. R., & Ghana. Environment and Urbanization, 5(2), 10–34. Sanbonmatsu, L. (2012). Neighbourhood effects on the long-term well-being of low- StataCorp. (2019). Stata statistical software: Release 16. College Station, TX: StataCorp income adults. Science, 337(6101), 1505–1510. LLC. Macintyre, S., Ellaway, A., & Cummins, S. (2002). Place effects on health: How can we Stoler, J., Tutu, R. A., Ahmed, H., Frimpong, L. A., & Bello, M. (2014). Sachet water conceptualise, operationalise and measure them? Social Science & Medicine, 55(1), quality and brand reputation in two low-income urban communities in Greater 125–139. Accra, Ghana. The American Journal of Tropical Medicine and Hygiene, 90(2), 272. MacTavish, R., Bixby, H., Cavanaugh, A., Agyei-Mensah, S., Bawah, A., Owusu, G., … Stoler, J., Weeks, J. R., & Otoo, R. A. (2013). Drinking water in transition: A multilevel Baumgartner, J. (2023). Identifying deprived “slum” neighbourhoods in the Greater cross-sectional analysis of sachet water consumption in Accra. PLoS One, 8(6), Accra Metropolitan Area of Ghana using census and remote sensing data. World Article e67257. Development, 167, 106253. Tetteh, J. D., Templeton, M. R., Cavanaugh, A., Bixby, H., Owusu, G., Yidana, S. M., … Mah, S., Sanmartin, C., Riva, M., Dasgupta, K., & Ross, N. (2022). Associations of the Agyei-Mensah, S. (2022). Geographic inequalities in household drinking water use in active living environment, walking and premature mortality in Canada. Annals of the Greater Accra metropolitan area (GAMA), Ghana. Population and Environment, 44 Behavioral Medicine, 56(5), 523–532. (1–2), 46–76. Mahama, A. M., Anaman, K. A., & Osei-Akoto, I. (2014). Factors influencing UNESCO. (2021). Global education monitoring report. In 2: Non-state actors in education: householders’ access to improved water in low-income urban areas of Accra, Ghana. Who chooses? Who loses? Paris, France: UNESCO. Journal of Water and Health, 12(2), 318–331. UN-HABITAT. (2009). Ghana: Accra urban profile. Nairobi: UN-HABITAT. Massey, D. S. (2001). The prodigal paradigm returns: Ecology comes back to sociology. Van Ham, M., Tammaru, T., Ubarevičienė, R., & Janssen, H. (2021). Urban socio-economic In A. Booth, & A. C. Crouter (Eds.), Does it take village? (pp. 53–60). Hove, East segregation and income inequality: A global perspective. Springer Nature: Cham, Sussex, UK: Psychology Press. Switzerland. Massey, D. S., & Denton, N. A. (1993). American apartheid: Segregation and the making of Walks, A. (2014). From financialization to sociospatial polarization of the City? Evidence the underclass. Cambridge, MA: Harvard University Press. from Canada. Economic Geography, 90(1), 33–66. Montgomery, M., & Hewitt, P. C. (2015). Urban poverty and health in developing Wen, M., Browning, C. R., & Cagney, K. A. (2003). Poverty, affluence, and income countries: Household and neighbourhood effects. Demography, 42(3), 397–425. inequality: Neighbourhood economic structure and its implications for health. Social Moulds, S., Chan, A. C. H., Tetteh, J. D., Bixby, H., Owusu, G., Agyei-Mensah, S., … Science & Medicine, 57(5), 843–860. Templeton, M. R. (2022). Sachet water in Ghana: A spatiotemporal analysis of the Wessel, T. (2022). Business services, income inequality, and income segregation in recent upward trend in consumption and its relationship with changing household metropolitan areas: Direct and indirect links. Economic Geography, 98(5), 464–486. characteristics, 2010–2017. PLoS One, 17(5), Article e0265167. Weststrate, J., Dijkstra, G., Eshuis, J., Gianoli, A., & Rusca, M. (2019). The sustainable Obeng-Odoom, F. (2011). The informal sector in Ghana under siege. Journal of Developing development goal on water and sanitation: Learning from the millennium Societies, 27(3–4), 355–392. development goals. Social Indicators Research, 143, 795–810. Oosterbaan, C., Arku, G., & Asiedu, A. B. (2012). Conversion of residential units to Williams, R. (2012). Using the margins command to estimate and interpret adjusted commercial spaces in Accra, Ghana: A policy dilemma. International Planning Studies, predictions and marginal effects. The Stata Journal, 12(2), 308–331. 17(1), 45–66. Wilson, W. J. (1987). The truly disadvantaged: The inner city, the underclass, and public Oteng-Ababio, M., Arguello, J. E. M., & Gabbay, O. (2013). Solid waste management in policy. Chicago: University of Chicago Press. African cities: Sorting the facts from the fads in Accra, Ghana. Habitat International, Wodon, Q. (2012). Improving the targeting of social programs in Ghana. Washington, D.C.: 39, 96–104. World Bank. Owusu, G., & Wrigley-Asante, C. (2020). Urban inequality and poverty in the global World Health Organization- UNICEF. (2017). 2017 annual report WHO/UNICEF joint south. Equilibri, 24(speciale), 79–90. monitoring programme for water supply, sanitation and hygiene. In 2017 annual Owusu, G., & Yankson, P. W. K. (2007). 'Poverty in Ghana is basically a rural report WHO/UNICEF joint monitoring programme for water supply, sanitation and phenomenon': Are we underestimating urban poverty? Ghana Journal of Development hygiene (p. 20-20). Studies, 4(1), 87–105. Yakubu, I., Akaateba, M. A., & Akanbang, B. A. (2014). A study of housing conditions and Papke, L. E., & Wooldridge, J. M. (1996). Econometric methods for fractional response characteristics in the Tamale Metropolitan Area, Ghana. Habitat International, 44, variables with an application to 401 (k) plan participation rates. Journal of Applied 394–402. Econometrics, 11(6), 619–632. 15