WPS8654 Policy Research Working Paper 8654 Spatial Distributions of Job Accessibility, Housing Rents, and Poverty in Nairobi, Kenya Shohei Nakamura Paolo Avner Poverty and Equity Global Practice Social, Urban, Rural and Resilience Global Practice Global Facility for Disaster Reduction and Recovery November 2018 Policy Research Working Paper 8654 Abstract Whether individuals and job opportunities are well con- city: Nairobi residents can on average access fewer than 10 nected is a key determinant of productive urban labor percent of existing jobs by foot within an hour. Even using markets. The overall level of job accessibility in a city a minibus, they can reach only about a quarter of the jobs. depends on the locations of jobs and workers’ residences, This paper further shows that poorer households and/or as well as transport networks. Moreover, who has good those who live in informal settlements can reach a more access to job opportunities hinges on the trade-off faced limited number of jobs. Living closer to job opportunities is by households in their residential choices over job acces- indeed costly in Nairobi, not only because housing quality sibility, living conditions, and housing costs. This paper and living conditions tend to be better in such areas, but empirically analyzes the spatial distributions of job acces- also job accessibility itself is valued as a great amenity in the sibility, housing rents, and poverty in Nairobi, Kenya. It housing markets, which challenges low-income households’ finds that workers and jobs are not well connected in the residential location choice. This paper is a product of the Poverty and Equity Global Practice; the Social, Urban, Rural and Resilience Global Practice; and the Global Facility for Disaster Reduction and Recovery. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://www.worldbank.org/research. The authors may be contacted at snakamura2@worldbank.org. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team Spatial Distributions of Job Accessibility, Housing Rents, and Poverty in Nairobi, Kenya1 Shohei Nakamura Paolo Avner World Bank World Bank Keywords: job accessibility, urban poverty, slums, urban planning, housing rent JEL Classification: R14, R21, R31, R41, O18 1 This paper has been prepared for the World Bank Kenya Poverty and Gender Assessment by Shohei Nakamura (snakamura2@worldbank.org) and Paolo Avner (pavner@worldbank.org). We are grateful to Nancy Lozano Gracia and Johan Mistiaen for useful comments. We thank participants in the 2018 Jobs and Development Conference at Bogota, Colombia, and the 65th Annual North American Meetings of the Regional Science Association International at San Antonio, TX. We would also like to thank Japan International Cooperation Agency (JICA) for sharing the survey data. This work was financially supported by UKAID/DfID (Department for International Development) through the Multi-donor Trust Fund on Sustainable Urbanization. 1. Introduction Job accessibility is potentially a key determinant of not only the productivity of labor forces in cities but also their welfare and poverty. Job accessibility is commonly referred to as the number or share of existing job opportunities a city resident can reach within a given amount of time. At the level of an urban area, a low level of job accessibility for city residents can inhibit the productivity gains from agglomeration economies by constraining a better matching between jobs and job seekers (Combes and Gobillon 2015; Duranton and Puga 2004). More importantly, limited job accessibility may disproportionally affect the economic performance of disadvantaged workers. In many African cities, lacking affordable and reliable transport, many people are too poor to use any transport and just walk to work (Lall, Henderson, and Venables 2017).2 Low-income households that live farther away from clusters of jobs face higher commuting and job search costs. As the spatial mismatch hypothesis posits (see a review by Gobillon and Selod 2014), having such bad job accessibility could make them even poorer by limiting their employment outcomes. However, which city residents face limited access to job opportunities depends on, among other factors, the shape of the city, the clustering patterns of employment, the function of housing markets, and transport networks. For example, low-income households may live in inner-city informal settlements by prioritizing their proximity to job opportunities over more desirable living conditions and incur environmental and health risks due to overcrowding and inadequate access to basic services. Living near clusters of jobs may be particularly important for the urban poor, who often engage in multiple temporary and casual jobs. However, the long-term influence of unfavorable living conditions on their well-being is a concern. Thus, it is imperative to empirically assess who suffers from limited job accessibility and whether choosing to live in informal settlements offers relatively good job accessibility to low-income populations in exchange for giving up decent living conditions. In a worse case, informal settlements provide neither good access to jobs nor a healthy environment, instead creating poverty traps (Marx, Stoker, and Suri 2013; Turok and Borel-Saladin 2018). Unfortunately, empirical evidence on how job accessibility and poverty are linked in African cities is scarce. Such analysis requires the data on (a) the spatial distribution of jobs, (b) transport networks, and (c) housing rents and household consumption. Spatial distribution of jobs is hard to obtain for African cities, given the prevalence of informal jobs. Even more difficult is information about minibus networks, which is a common transport mode in many African cities. Although housing rents and household consumption are usually available in Living Standards Measurement Survey (LSMS) type household surveys, they are rarely representative at both the informal and formal residential areas. By addressing all these data challenges, this paper examines the link between job accessibility and urban poverty in an African city by assessing the spatial distributions of job accessibility, housing rents, and poverty. Our approach is threefold. First, we assess the overall level of job accessibility for a city. Our study focuses on Nairobi, Kenya. A relatively monocentric city (Antos, Lall, and Lozano-Gracia 2016; Henderson et al. 2016), Nairobi accommodates more than 1 million residents in informal settlements, most of whom are tenants (KNBS 2012). The severity of living conditions in Nairobi’s informal settlements (particularly Kibera) has been widely documented (for example, Bird, Montebruno, and Regan 2017). We measure job accessibility based on the number of jobs accessible within a certain range of travel time from a residence or neighborhood. Second, we investigate who has better or worse job accessibility in Nairobi, by focusing on poorer people 2 In the Unites States, the poor live in central cities with better access to public transport (Glaeser, Khan, & Rappaport 2008). 2 and/or residents of informal settlements. The analysis utilizes a recently collected household survey, the Cities Baseline Survey, which is representative at both the formal and informal residential areas in Nairobi. Finally, we analyze what trade-off Nairobi residents face in selecting residences by examining the link between job accessibility and housing rents in the city. When housing rents are determined in equilibrium in functioning (even informal) housing markets, the rents reflect job accessibility among other characteristics (Rosen 1974). We estimate hedonic regression models to examine the costs of better job accessibility in Nairobi. Our analysis illustrates how job accessibility is linked to the spatial distributions of housing rents and poverty in Nairobi. We find that Nairobi’s average job accessibility is indeed limited. By spending 30 minutes traveling, a Nairobi resident can on average reach 2 percent and 4 percent of jobs that exist in the city by foot and by minibus, respectively. She or he can reach 7 percent and 24 percent of jobs within an hour by foot and minibus, respectively. Given the fact that a large fraction of Nairobi residents—particularly the low-income—walk to work, this limited job accessibility must have created considerable inefficiencies in the labor market. Moreover, we find that the job accessibility of poorer households and/or residents of informal settlements is overall worse in Nairobi. Compared to richer households (in the fourth per-capita consumption quartile), poor households (in the first per-capita consumption quartile) can on average reach 20 percent fewer jobs by foot within 60 minutes. Similarly, the number of jobs that residents of informal settlements can reach by foot within 60 minutes is 30 percent smaller than for residents in formal residential areas. Our hedonic regression analysis demonstrates that it is costly to live in housing with better job accessibility in Nairobi, explaining in part why poorer households have such limited job accessibility. Our findings illustrate the need to assess and enhance access to economic opportunities for disadvantaged workers in African cities, and Nairobi is a case in point. While the poverty headcount ratio in Nairobi declined from 21 percent in 2005/06 to 17 percent in 2015/16, the number of poor slightly increased during the period (KNBS 2018). Despite the overall low poverty rate, nearly 30 percent of residents in informal settlements still remain in poverty (World Bank 2018). More than 20 percent of the poor in Nairobi are unemployed, and about 40 percent of the poor are casual workers. The 2015/16 Kenya Integrated Household Budget Survey (KIHBS) also shows that only 15 percent of the poor use a minibus for commuting, while 70 percent of the poor walk to work. Despite the concentration of poverty, worse living conditions, and limited job accessibility in Nairobi’s informal settlements, it has been difficult for many residents to move out of informal settlements (Nakamura and Karasawa 2018). In such situations, it is critically important to enhance job accessibility to support the poor and/or residents of informal settlements to escape from poverty. Our research is related to three bodies of literature. First, our work is related to the literature focusing on job accessibility from a perspective of agglomeration economies. Facilitating better matching between firms and workers is, among other channels such as sharing and learning effects, a critical component of agglomeration economies (Duranton and Puga 2004). A body of empirical studies estimated the effects of city size and density for productivity, though most of them focus on developed countries (see Combes and Gobillon 2015 for a review). Limited job accessibility could, however, restrict such benefits of agglomeration economies. In view of cities as labor markets (Bertaud 2014), some studies assess accessibility, measured as the number of opportunities an individual can access within a given amount of time, for Buenos Aires (Quirós and Mehndiratta 2015), Dakar (Stokenberga 2017), and Nairobi (Avner and Lall 2016). We advance the Nairobi study by Avner and Lall (2016) by focusing on accessibility to jobs instead of opportunities proxied by land use. 3 A second line of literature our work is aligned with is that on spatial mismatch. Originally developed by Kain (1968) for the racial segregation context in the United States, the spatial mismatch literature argues that unskilled workers who are segregated away from job opportunities tend to be unemployed and have lower wages due to high commuting and job search costs (see a review by Gobillon and Selod 2014). This is relevant to African cities as well. In fact, a randomized control trial (RCT) recently conducted in Addis Ababa, Ethiopia, finds that providing a transport subsidy to disadvantaged job seekers increased their chance of finding better jobs (Abebe et al. 2017; Franklin forthcoming). Analyzing such causal impacts of job accessibility on labor market outcomes must follow detailed assessments of job accessibility for disadvantaged workers. Our study aims to provide such diagnostics of job accessibility among Nairobi residents by paying specific attention to informal settlements. Finally, our research is related to the literature on hedonic regression analysis for housing prices. Urban land use theory (Alonso 1964; Mills and Hamilton 1995; Muth 1968) explains how the rent gradient over distance from the central business district (CBD) is determined in a monocentric city. In reality, however, jobs are not necessarily concentrated in the CBD. We add to the hedonic regression literature by estimating the link between job accessibility and housing rents, by explicitly including housing in informal residential areas. A few studies have estimated hedonic regression for informal housing in the developing world, such as Gulyani, Bassett, and Talukdar (2012) and Marx, Stoker, and Suri (2017) for Nairobi and Nakamura (2017) for Pune, India. Another recent study by Atuesta et al. (2018) found a clear link between job accessibility and housing values in Mexico City. The rest of this paper is structured as follows: Section 2 describes the context about job accessibility challenges in Nairobi. Section 3 discusses the methodology by explaining various data sets and analytical methods employed in this study. Section 4 presents the results on the job accessibility calculated for the city, the levels of job accessibility by different household groups, and hedonic regressions. Section 5 concludes with a brief summary of findings. 2. Context: Spatial challenges in Nairobi Recently collected surveys help to grasp spatial challenges in Nairobi. Most Nairobi residents are rent-paying tenants, and low-cost rental units are mainly located in informal residential areas. According to the Cities Baseline Survey (explained in Section 3), the distribution of housing rents in Nairobi is skewed with a small group of very high rents (Panels A and B in Figure 1). The average housing rent among the surveyed rental units is K Sh 5,315.3, 4 A quarter of rental units have a rent of lower than K Sh 1,800; half the rental units have rents less than K Sh 2,500; and 75 percent of rental units have rents less than K Sh 4,400. About 10 percent of units have very high rents (more than K Sh 30,000). The mean rent value in informal settlements (K Sh 3,600) is less than half of that in formal residential areas (K Sh 9,000), and the majority of cheap rental units (less than K Sh 2,500) in the city are located in informal residential areas. 3 The housing rents used in this study include utilities. 4 K Sh 5,315 was roughly equivalent to US$61 as of 2013. 4 Figure 1. Distributions of housing rents in Nairobi (A) Housing rent (B) Housing rent (cumulative) (C) Distance from CBD (D) Distance from CBD (cumulative) Source: Cities Baseline Survey 2013. As reflected in the rent gap, there is a stark contrast in housing quality and living conditions between formal and informal residential areas. Most dwellings in Nairobi are structured with walls made of either stone/brick/block (52 percent) or corrugated iron sheet (35 percent) (Table 1 and Table A1 in Appendix). About 93 percent of dwellings in formal residential areas have walls made of stone/brick/block, while only 25 percent of informal dwellings have such walls. Nearly 20 percent of units in the city have no access to electricity, which is a very low standard even compared with other African cities.5 In fact, 94 percent of dwellings in formal residential areas have electricity in Nairobi, whereas only 70 percent of informal housing has electricity. Security is also a big concern in Nairobi: 40 percent of surveyed households report security as a problem in their neighborhoods. Residents in informal settlements are more likely to worry about security (50 percent) than residents of formal residential areas (30 percent). With respect to environmental hazards, 47 percent of households report a garbage dump in their neighborhood as a problem. As expected, a larger proportion of residents in informal settlements face such problems (60 percent). 5 For example, more than 95 percent of houses in slums of Greater Accra (Ghana) and Addis Ababa (Ethiopia) have access to electricity (Nakamura and Yoshida 2018). 5 Table 1. Household and housing characteristics in Nairobi (only tenants) Consumption quartile Informal area Q1 Q2 Q3 Q4 Yes No (1) (2) (3) (4) (5) (6) Household characteristics Living in informal neighborhood (%) 62.6 49.2 40.8 34.8 100.0 0.0 Per capita monthly expenditure (K Sh) 2351 4505 7419 18046 6560 9405 Age of household head 34.5 34.0 33.4 33.3 32.7 34.7 Years of household head’s education 11.1 11.4 12.4 14.8 11.0 13.7 Household size 3.9 3.1 2.6 2.2 2.8 3.1 Travel time to work (one way, in minutes) 31.7 30.7 33.7 29.0 28.6 33.6 Commute by foot (%) 39.3 29.8 31.9 28.2 40.9 24.8 Commute by minibus (%) 43.4 53.4 51.3 50.3 41.0 57.1 Housing characteristics Housing rent (K Sh) 2409 3391 5803 14342 3609 8995 Written tenancy agreement (%) 8.1 14.7 31.2 52.7 14.0 37.9 Floor area (m2) 16.36 16.57 22.06 24.35 16.71 22.49 Wall: stone/brick/block (%) 38.8 57.3 64.9 82.9 25.0 92.7 Wall: corrugated iron sheet (%) 47.6 32.2 29.2 11.6 56.9 6.1 Wall: mud/wood (%) 13.6 10.5 5.9 5.4 18.0 1.2 Roof: corrugated iron sheet (%) 92.0 89.1 82.3 66.0 92.9 73.2 Roof: clay tiles (%) 1.6 3.3 7.5 23.7 4.2 13.2 Roof: concrete (%) 6.5 7.6 10.1 10.3 2.9 13.6 Floor: earth/clay (%) 8.2 5.7 5.1 3.9 11.6 0.7 Floor: tiles (%) 0.0 1.6 5.4 28.2 4.8 12.2 Floor: cement (%) 91.8 92.7 89.5 68.0 83.5 87.1 Water: piped inside (%) 0.9 4.0 11.0 17.2 3.2 12.7 Water: piped outside (%) 1.7 3.4 6.5 10.2 3.7 6.9 Water: shared tap (%) 40.3 50.2 48.2 50.6 28.1 64.3 Water: other (%) 57.1 42.5 34.4 22.1 65.0 16.1 Toilet: flush inside (%) 2.3 11.8 25.4 51.6 9.3 34.6 Toilet: flush/VIP latrine outside (%) 27.8 35.8 29.1 20.6 16.7 38.5 Toilet: private pit latrine (%) 14.7 8.7 11.8 4.7 16.0 4.7 Toilet: shared toilet (%) 53.8 43.1 33.3 22.8 57.1 21.9 Toilet: other (%) 1.3 0.7 0.3 0.2 0.9 0.3 Electricity (%) 74.2 80.9 83.7 93.2 69.8 94.4 Garbage dump as problem (%) 52.1 47.4 42.9 27.7 60.6 26.7 Factory as problem (%) 12.4 5.9 10.1 9.8 14.0 5.5 Secure (%) 55.7 55.9 64.1 74.1 51.6 72.1 Bus stop within 500 meters (%) 71.1 75.6 77.5 83.2 65.8 85.9 Due to the limited availability of affordable transport, a large fraction of people—particularly in informal neighborhoods—walk to work in Nairobi. Most rental units are located within 3−12 km from the CBD with some agglomerations of informal settlements.6 Informal and formal housing are equally distributed within 5 km from the CBD, while a large informal area (Kibera slum) exists around 6−7 km from the CBD (Panels C and D in Figure 1). Most informal houses are located within 10 km from the CBD. Given such spatial distribution of residences, 35 percent of household heads walk to work, while 43 percent of household heads use minibuses for commuting.7 Only less than 10 percent of household heads commute using their own car in Nairobi. The share of household heads walking to work is larger in informal settlements (41 percent) than in formal 6 Distance from the CBD is measured based on Euclidian distance without considering road networks. We also carried out all the analyses using network distance from the CBD and obtained similar results (not reported). 7 According to the 2015/16 KIHBS, 39 percent of Nairobi residents walk to work, and 38 percent of workers use minibuses for commuting. Private cars account for only 5 percent. Walking is the main commuting mode for the poor (75 percent), followed by minibus (15 percent). 6 residential areas (25 percent), reflecting income gaps between those areas (Table 1).8 In addition, even getting to a bus stop itself takes more time in informal residential areas. A larger proportion of housing is located more than 500 meters away from the nearest bus stop in informal settlements (34 percent) than in formal residential areas (only 14 percent). To understand the trade-off faced by households in their residential choices, a detailed spatial diagnostic is necessary. Job accessibility for a housing depends on the distance from job clusters and the availability of transport to reach there. If housing rents are determined in the market by not only the dwelling and neighborhood characteristics but also the level of job accessibility, households face a trade-off over them in their residential choices. The existing stock of housing with good job accessibility may be scarce in a not-well-connected city like Nairobi, and thus it can be expensive to live in such housing. Some may have to sacrifice living conditions to live in proximity to job opportunities, given the importance of job accessibility in their labor performance. Low-income households could particularly face a challenge to access job opportunities if low-rent residences in informal settlements do not really offer reasonable job accessibility. Therefore, it is critically important to investigate the spatial distribution of job accessibility in the city and its relationship to spatial distribution of poverty (or households in different income groups) and housing rents. 3. Methodology 3-1. Data Transport To calculate job accessibility in Nairobi, it is necessary to account for various transportation modes: walking, driving and using the semi-formal minibus (matatus) network. Data for the ‘minibus network’ travel times are available online in the General Transit Feed Specification (GTFS) format (see Figure 2 for the minibus network).9 These data were compiled by the Digital Matatus consortium comprising University of Nairobi, Columbia University, Massachusetts Institute of Technology, and Groupshot (Klopp et al. 2015; Williams et al. 2015). The GTFS data were processed using a Dijkstra algorithm, which provided times to reach any minibus station from any other in the urban area.10 Travel times to reach the closest minibus station were computed based on a straight-line distance or ‘as the crow flies’ from the center of each grid cell. 8 The distributions of commuting time do not substantially differ between informal and formal residential areas. In Nairobi, most households spend less than 1 hour for commuting (one way). Half of the household heads in Nairobi spend 30 minutes or less getting to their workplaces. No substantial difference is observed in the mean commuting time among households with different consumption quartiles and within/outside informal residential areas. 9 GTFS data are a common format for public transportation schedules and associated geographic information. They are a collection of multiple comma-separated values (csv) files that describe transit routes, stops (including geographic coordinates), travel times between stops, and a number of other public transit properties. 10 Dijkstra’s algorithm is an algorithm for finding the shortest path between two nodes in a graph that may represent, for example, road networks. 7 Figure 2. Minibus transport network Source: Digital Matatus Project. The transport times for cars were computed through a network analysis using the road layers provided by OpenStreetMap (Figure 3). Three different travel speeds were assumed based on the road category. Assumptions were made that with congestion, first-class roads would carry vehicles traveling at 30 km/h, second-class roads at 16 km/h, and finally third-class roads at 12 km/h. These speed assumptions for the road network are broadly consistent with congestion figures reported in the Kenya Urbanization Review (World Bank 2016). Finally, walking times were calculated using the road network assuming an average walking speed of 4 km/h for all road links irrespective of their class. Figure 3. Road network and road hierarchy Source: OpenStreetMap. Jobs The information about the volume and distribution of jobs is extracted from the Japan International Cooperation Agency (JICA) personal travel survey conducted in 2013 (JICA 2013). The JICA personal trip survey asked 38,000 persons in 10,000 households a number of questions about their mobility behaviors and also about their employment status. Among these questions are the status of occupation of the respondent, the sector of employment, and the approximate location of the employment at the census sublocation level within Nairobi County. Population counts are also 8 provided through the personal travel survey at the census zone level. The job density pattern at the census zone level is represented in Figure 4. Figure 4. Job density in Nairobi Source: Authors’ calculation based on JICA (2013). Given that census zones are of varying sizes, with some as large as 58.5 km2 and others as small as 0.12 km2, and that in parallel the transportation data have a high resolution, an effort was made to achieve a more disaggregated distribution of jobs and people. The finer grain distribution was obtained by using a grid constituted of 1 km2 grid cells. Population and jobs were attributed to each grid cell in proportion of the land area that they intersect with census zones. The underlying assumption is that population and jobs are homogeneously distributed over census zones. Housing and households The data used for the locations and characteristics of housing and households are from the Cities Baseline Survey, which were collected in 15 Kenyan cities in 2013. The survey used a two- or three-stage stratified cluster sampling design intended to be representative of poor and non-poor households living in formal and informal settlements within each municipality.11 The first-stage sampling frame was based on the 2009 census frame of enumeration areas (EAs). In the census sample frame, EAs are identified as urban, peri-urban, or rural. EAs are further identified as containing formal or informal settlement types. For the first-stage sampling, EAs were selected from strata identified as informal, urban-formal, peri-urban-formal, and rural. The second stage involved a random selection of households from each selected EA. A fixed number of households from each selected EA was drawn, irrespective of EA size. Main data collection was conducted between July 17, 2012, and March 14, 2013. A total of 1,182 households (582 in informal areas and 600 in formal areas) were surveyed for Nairobi. Among them with geocoded information, 989 households are rent-paying tenants (551 tenants in informal areas and 438 tenants in formal areas). In the data, informal and formal residential areas were categorized adopting the 2009 census framework. The Kenya National Bureau of Statistics (KNBS) identified some 2009 EAs as informal areas when they are a “settlement characterized by at least two of the following: inadequate access to safe water; inadequate access to sanitation and other infrastructure; poor structural quality of housing; 11 Nairobi, Mombasa, Kisumu, Nakuru, Eldoret, Malindi, Naivasha, Kitui, Machakos, Thika, Nyeri, Garissa, Kericho, Kakamega, and Embu. See NORC (2013) and World Bank (2016) for details. 9 overcrowding and insecure residential status” (KNBS 2012, 4−5).12 Table 1 reports household and housing characteristics by per capita consumption quartiles and informal status of their neighborhoods (see Table A1 for detailed summary statistics). Table A2 in Appendix A reports summary statistics for rental units in the city. 3-2. Methods Job accessibility at the city level We use an urban accessibility metric to measure job accessibility in Nairobi (Avner and Lall 2016; Quirós and Mehndiratta 2015; Stewart and Zegras 2016). The model counts the number of reachable opportunities within a travel time isochrone. The most commonly used accessibility indicator is the share of opportunities that can be reached within a given time threshold. This normative choice has been recognized as meaningful (Bertaud 2014), although necessarily imperfect. The accessibility metrics, the job accessibility index (JAI), can be described by the following: ∑ ∗ ∑ (1), where is the accessibility indicator for location , represents the number of opportunities in location , is the travel time between location and location and is a time threshold. in the numerator is a test that takes the value of either 0 or 1. In the latter case, opportunities in location j can be reached and are added to the total of reachable opportunities. In the other case they are assigned the value 0. When averaged over the whole urban area , localized accessibilities are weighted by population , to account for the fact that higher population densities are generally found in areas that benefit from good accessibility. ∑ ∗ ∑ ∗ ̅ ∑ (2) ∗∑ We also calculate the JAI for different groups of households based on their neighborhood status (informal or formal) and consumption quartiles. It is not clear a priori who benefits from better job accessibility between poor and non-poor households. Richer households may indeed choose to live at a further distance from the CBD and most job clusters, to enjoy larger houses—as is often assumed in urban economic theory—in exchange for longer commutes. They could however compensate longer distances with higher commuting speeds if they are equipped with cars and benefit from both good accessibility and larger dwellings. Yet, because only a tiny fraction of people commutes by car in Nairobi, even richer people may prioritize living closer to jobs. Poorer people are often constrained to live in informal settlements; so, their job accessibility depends on (a) the overall level of job accessibility in informal residential areas compared to formal areas and (b) the variation of job accessibility across informal neighborhoods. Thus, it is worth examining the JAI for different groups. Hedonic models In the standard hedonic framework developed by Rosen (1974), the market prices of housing units represent the sum of expenditures in a bundle of characteristics that can be priced separately (see 12 The number of residents in Nairobi’s informal settlements varies depending on sources, and it is possible that the 2009 Census undercounted them. Even the population count of Kibera, the largest informal settlement in Nairobi, ranges from 170,070 to 270,000 (Lucci, Bhatkal, and Khan 2018). 10 Taylor [2008] for a general introduction of hedonic analysis). Let Z represent a housing bundle with characteristics Z = z1, z2, … zr. With the consumption on non-housing items C, the utility of household i is written as C, , , … ; (3), where di are the demographic characteristics of household i. The budget constraint for household i is given as Yi = C + P(Z), where Yi is the income of household i, and P is the price schedule for Z determined in an equilibrium. The household seeks to maximize the utility by choosing C and each element of Z such that the following marginal condition is satisfied for each zj: ⁄ ⁄ (4). Given the hedonic framework above, the base hedonic regression model used for this study is an ordinary least squares (OLS) model expressed as follows: ln rent JAI X (5), where the dependent variable is the natural logarithm of the monthly rent of the i-th housing; JAIi is the JAI for the i-th housing; Xi is a vector of structural characteristics, access to services, and neighborhood and environmental factors; and εi is a random error. The parameter estimate β1 indicates the degree of association between the JAI and housing rents after controlling for other observed characteristics in X. The coefficient estimate β1 can also be interpreted as the Marginal Willingness to Pay for living closer to jobs. We also add a squared term of the JAI to account for non-linearity. 4. Results Job accessibility at the city level Combining information about the distribution of jobs and travel times, we calculate the JAIs in Equation 2. These accessibility indicators were computed for (a) all jobs and each individual occupation and employment sector, (b) for each travel mode: car, minibus, and pedestrian travel, and (c) for three different time thresholds: 30 minutes, 45 minutes, and 60 minutes. The overall picture is one of higher accessibility levels in the more central areas of the city whatever the travel mode. Traveling by car provides significantly higher accessibility levels compared to using a minibus. Finally, traveling by foot considerably limits accessibility to jobs. Figure 5 displays the maps of accessibility levels to all jobs within 60 minutes when traveling by foot, minibus, and car, respectively. 11 Figure 5. Share of accessible jobs within 60 minutes (A) By foot (B) By minibus (C) By car Source: Authors’ calculation based on JICA (2013), Digital Matatus Project, and OpenStreetMaps. The average share of accessible opportunities for different time thresholds, and different travel modes is summarized in Table 2. The average accessibility drastically differs by transport mode and time thresholds. Using a minibus, the main form of motorized transport, on average a resident in Nairobi can reach 4 percent, 11 percent, and 25 percent of the jobs within 30 minutes, 45 minutes, and 60 minutes of travel, respectively. By comparison, in the metropolitan area of Buenos Aires, an urban area that has four times more population, accessibility figures using public transportation are 7 percent, 18 percent, and 34 percent for the same time thresholds (Quirós 2015). In addition, in Greater Dakar, an urban area roughly equivalent in size to Nairobi with a population above 3 million, the share of accessible jobs within 1 hour is 52 percent, more than twice the level in Nairobi (Stokenberga 2017). Moreover, average accessibility levels for pedestrians are much lower, standing at 7 percent of accessible opportunities. Given the limited use of private cars as a commuting mode in Nairobi, the most relevant average accessibility figures are those that concern pedestrians and minibus riders. Table 2. Average shares of accessible jobs in Nairobi Walking Minibus Cars (1) (2) (3) Within 30 minutes 1.8% 3.9% 43.7% Within 45 minutes 4.0% 10.8% 71.8% Within 60 minutes 7.3% 23.9% 88.7% Note: Numbers are the average share of jobs in the city that the Nairobi residents can reach by foot (column 1), minibus (column 2), and car (column 3) within 30, 45, and 60 minutes. 12 Job accessibility by household groups Our analysis confirms that residents in informal settlements have a lower level of job accessibility. Figure 6 shows the distributions of the JAIs for households in informal and formal areas (see Figure A1 for the distributions of the JAIs with different time thresholds, as well as the JAIs by car).13 The distribution of job accessibility for households in informal neighborhoods by foot within 60 minutes is low and is clustered within the 0 percent to 5 percent range (panels C and D). Except for residents in informal settlements living very close to the CBD, most households in informal areas can reach fewer than 10 percent of existing jobs by foot. The mean values of the JAIs for households in informal settlements (7.0 percent) are 30 percent lower than those of households in formal residential areas (10.0 percent). The share of accessible jobs by minibus in 45 minutes is distributed more widely (panels C and D). Almost all houses with a JAI of greater than 25 percent are located in formal residential areas. When measured for minibuses, the JAIs of households in informal neighborhoods (9.6 percent) are 44 percent lower than for households in formal areas (17.1 percent). Figure 6. Distribution of job accessibility in Nairobi (A) By foot (B) By foot (cumulative) (C) By minibus (D) By minibus (cumulative) Source: Authors’ calculation based on the Cities Baseline Survey, JICA (2013), Digital Matatus Project, and OpenStreetMap. 13 Figure A2 in Appendix A shows that the JAIs are not necessarily, and not linearly, related to the distance from the CBD, which motivates the use of the JAI over distance to the CBD. 13 Overall, richer households enjoy better job accessibility (Table 3). Regardless of the transport mode in measuring the JAIs, households in the first (that is, poorest) consumption quartile have the lowest scores, while households in the fourth (that is, richest) quartile have the highest scores. For example, households in the first consumption quartile can access 7.8 percent of Nairobi jobs by foot within 60 minutes, in contrast to the households in the fourth quartile (9.6 percent). By minibus, households in the first and fourth quartiles can access 11.8 percent and 15.7 percent of jobs within 45 minutes, respectively. Thus, the gap in the average job accessibility (either by foot or by minibus) between the poorest and richest groups of households is more than 20 percent. In other words, poor households can reach 20 percent fewer jobs than richer households in case of the same transport mode used for the same travel time. The gap becomes wider if we also consider the difference in their main transport modes (foot vs minibus/car). Table 3. JAI by consumption quartile and informal status Consumption quartile Informal area Q1 Q2 Q3 Q4 Yes No (1) (2) (3) (4) (5) (6) Distance from the CBD (km) 7.3 7.1 7.4 6.8 6.8 7.5 Share of accessible jobs by foot in 60 minutes (%) 7.8 8.7 8.5 9.6 7.0 10.0 Share of accessible jobs by minibus in 45 minutes (%) 11.8 13.3 13.6 15.7 9.6 17.1 The results also highlight how informal settlements are disconnected from the rest of the city. The gap in the JAIs between households in informal and formal areas (30 percent by foot and 40 percent by minibus) is wider than the gap between the poor and the rich (20 percent by both foot and minibus). Even using minibus, residents in informal settlements can on average reach only 10 percent of existing jobs in Nairobi. There are two possible interpretations for the fact that accessibility to jobs is low in informal neighborhoods. The first is that except for inner-city neighborhoods, informal settlements tend to be located in less desirable areas, typically those disconnected from employment opportunities. A second explanation however might reside in the structure of informal settlements themselves with high densities and limited rights of way which would constrain supply of public transport and force inhabitants to walk through often large informal areas to access jobs. In fact, a third of houses in informal settlements are located more than 500 meters away from the nearest bus stop, as opposed to 14 percent in formal residential areas. Nevertheless, lower-income households choose to live in informal neighborhoods because they cannot afford formal dwellings, but the question is how costly is it to live with better job accessibility in Nairobi? Trade-off over job accessibility, living conditions, and living costs Table 4 reports the estimation results of hedonic regression models in Equation 5. Columns 1 and 2 present the models with the distance from the CBD as a proxy of job accessibility. Columns 3 and 4 are the results with the share of accessible jobs by foot, while columns 5 and 6 report for the share of accessible jobs by minibus. The models overall fit well: the adjusted R-squared is around 0.78 in all models. The estimate of each variable is also reasonable; rents are higher for the units with written agreement, larger floor area, walls made of stone, brick, or block, water taps, flush toilets, and/or electricity. By contrast, the following characteristics are negatively correlated with housing rents: roof made of corrugated iron sheets, earth/clay floor, located near garbage dump sites, and/or being affected by pollution from factories. Interestingly, a dummy indicator of informal status is not clearly correlated with housing rents probably because other observed characteristics already capture the difference between informal and formal 14 dwellings/neighborhoods. Being located near a bus stop is also positively correlated with rent values.14 Table 4. Estimation results of hedonic regression models JAI = Share of JAI = Distance from JAI = Share of accessible jobs by CBD accessible jobs by foot minibus (1) (2) (3) (4) (5) (6) JAI -0.014** -0.060*** 0.008*** 0.004 -0.000 -0.009* (0.006) (0.019) (0.002) (0.011) (0.002) (0.004) JAI squared 0.002*** 0.000 0.000** (0.001) (0.000) (0.000) Informal neighborhood -0.033 -0.024 -0.024 -0.026 -0.049 -0.047 (0.047) (0.048) (0.051) (0.053) (0.051) (0.051) Written agreement 0.147** 0.154** 0.135** 0.136** 0.134** 0.139** (0.062) (0.062) (0.063) (0.063) (0.064) (0.064) Floor area 6.704*** 6.656*** 6.606*** 6.549*** 6.426*** 6.263*** (2.073) (2.056) (2.164) (2.161) (2.180) (2.176) Floor area squared -21.90*** -21.55*** -21.44*** -21.34*** -21.62*** -21.27*** (8.003) (7.934) (7.972) (7.983) (8.025) (8.016) Wall [base: Stone/Brick/Block] Corrugated iron sheet -0.195*** -0.197*** -0.185*** -0.185*** -0.183*** -0.179*** (0.048) (0.049) (0.049) (0.049) (0.048) (0.047) Mud/Wood -0.299*** -0.302*** -0.278*** -0.280*** -0.275*** -0.268*** (0.067) (0.069) (0.066) (0.066) (0.064) (0.064) Roof [base: Corrugated iron sheet] Clay tiles 0.194 0.180 0.194 0.191 0.196 0.205 (0.131) (0.133) (0.127) (0.126) (0.129) (0.128) Concrete 0.078 0.081 0.084 0.085 0.069 0.080 (0.068) (0.067) (0.070) (0.069) (0.071) (0.071) Floor [base: Earth/Clay] Tiles 1.134*** 1.158*** 1.191*** 1.195*** 1.180*** 1.159*** (0.146) (0.145) (0.158) (0.159) (0.157) (0.155) Cement 0.356*** 0.381*** 0.347*** 0.348*** 0.345*** 0.328*** (0.056) (0.057) (0.060) (0.060) (0.059) (0.061) Water [base: Other] Piped inside 0.125 0.137 0.142 0.147 0.128 0.140 (0.134) (0.134) (0.136) (0.132) (0.138) (0.136) Piped outside 0.325*** 0.342*** 0.311*** 0.313*** 0.298*** 0.294*** (0.113) (0.115) (0.114) (0.113) (0.115) (0.114) Shared tap 0.145*** 0.160*** 0.144*** 0.148*** 0.137*** 0.144*** (0.037) (0.037) (0.039) (0.040) (0.038) (0.038) Toilet [base: Private pit latrine] Flush toilet (inside) 0.835*** 0.856*** 0.864*** 0.867*** 0.844*** 0.848*** (0.120) (0.117) (0.124) (0.127) (0.132) (0.132) Flush toilet/VIP latrine (outside) 0.009 0.034 0.010 0.012 -0.004 0.007 (0.051) (0.051) (0.055) (0.055) (0.057) (0.057) Electricity 0.357*** 0.352*** 0.378*** 0.378*** 0.373*** 0.372*** (0.031) (0.031) (0.032) (0.032) (0.032) (0.032) Garbage dump -0.067* -0.063* -0.079** -0.076* -0.065* -0.058 (0.037) (0.037) (0.037) (0.039) (0.038) (0.038) Factory -0.112** -0.104* -0.088 -0.084 -0.098* -0.081 (0.054) (0.054) (0.055) (0.055) (0.055) (0.055) Secure 0.046 0.039 0.038 0.039 0.042 0.042 (0.030) (0.030) (0.031) (0.032) (0.032) (0.032) Bus stop within 500 meters 0.086** 0.077** 0.036 0.036 0.067 0.072* (0.037) (0.037) (0.042) (0.042) (0.041) (0.040) Adj-R2 0.778 0.780 0.781 0.781 0.777 0.778 Obs. 980 980 950 950 950 950 Note: Robust standard errors in parentheses. * p < 0.1, ** p < 0.05, *** p < 0.01. Dependent variable is the natural logarithm of monthly housing rents. Housing types and constant terms are not shown. VIP=Ventilated improved pit. 14 Some studies have found negative effects of being in proximity to bus stations on housing values due to, for example, increased exposure to crime. For instance, a recent study by Atuesta et al. (2018) estimated positive associations between job accessibility and housing values in Mexico City, while finding a negative linkage between proximity to bus stations and housing values. 15 The estimation results of hedonic regression models demonstrate that the JAIs—distance from CBD and share of accessible jobs by foot or minibus—are positively correlated with housing rents even after observed dwelling and neighborhood characteristics are controlled for. Monthly rents of housing units located 1 km closer to the CBD tend to be 1.4 percent higher than the other units with comparable characteristics (column 1). The relationship appears to be non-linear (column 2). In addition, the share of accessible jobs by foot is correlated with housing rents (columns 3 and 4). An additional 1 percentage point share of accessible jobs is related to 0.8 percent higher housing rents. Using the JAI calculated for different types of jobs (employers, employees, and own-account workers) does not substantially change the results (Table 5). Compared to the JAI based on foot, the share of accessible jobs by minibus is less clearly correlated with housing rents once observed characteristics are controlled for (columns 5 and 6). This is probably because minibuses are mainly accessible to neighborhoods with better quality of housing. Table 5. Estimation results of hedonic regression models with accessibility to different types of jobs JAI = Share of accessible jobs by foot JAI = Share of accessible jobs by minibus Own-account Own-account All Employer Employee worker All Employer Employee worker (1) (2) (3) (4) (5) (6) (7) (8) JAI 0.008*** 0.007*** 0.008*** 0.006** -0.009* -0.002 -0.009* -0.011* (0.002) (0.002) (0.002) (0.003) (0.004) (0.004) (0.005) (0.007) JAI squared 0.000** 0.000 0.000** 0.000 (0.000) (0.000) (0.000) (0.000) Note: Robust standard errors in parentheses. * p < 0.1, ** p < 0.05, *** p < 0.01. Dependent variable is the natural logarithm of monthly housing rents. Figure 7 visually illustrates the results. Without controlling for observed characteristics, housing rents are strongly correlated with the JAIs. Obviously, housing rents with better JAIs are higher not only due to the JAIs but also better structural and neighborhood quality. For example, the average monthly rents of housing located 2−3 km away from the CBD are 1.5 times as high as the average rents of units 6−7 km away from the CBD (panel A). The predicted rent curve over the distance from the CBD after controlling for observed characteristics is flatter but still positively correlated. Similarly, controlling for quality differences makes the predicted rent curve over the share of accessible jobs by foot in 60 minutes (panel B) flatter. In particular, rents become inelastic to the JAI between the 1 and 9 percentage point range, where informal dwellings are concentrated. In case of the minibus-based JAI (panel C), the result implies that housing rents with higher JAI have higher rents because they tend to have better quality. 16 Figure 7. JAI and rent (A) Distance from CBD (B) Share of accessible jobs by foot (C) Share of accessible jobs by minibus Source: Authors’ calculation based on the Cities Baseline Survey, JICA (2013), Digital Matatus Project, and OpenStreetMap. Note: Panels (A) to (C) are based on the estimation of hedonic regression models with the JAI (including its second- and third-order terms). The 90 percent confidence intervals are also shown. An additional analysis on the affordability of housing units with good job accessibility is presented in Table 6. Based on hedonic regressions, monthly rents are predicted for housing units with relatively good JAI (that is, 6 percent or more jobs can be reached by foot within 60 minutes) by different housing quality. Corresponding to the quartile of housing quality index, predicted housing rents increase from K Sh 1,580 to K Sh 2,181, K Sh 3,190, and K Sh 7,325.15 To live in housing of good quality and good job accessibility, poorer households in the first and second consumption quartiles need to spend about 50 percent and 28 percent of their expenditures on housing rents, respectively. With the 5:1 expenditure and rent ratio, only 42 percent of households can afford such housing. An additional 22 percent of households can afford housing with good job accessibility by tolerating bad housing quality. Given that food expenditures account for a large share of the budgets of low-income households, it is quite costly to live in proximity to job opportunities. 15 The housing quality index is calculated by principal component analysis based on the characteristics of dwelling, access to services, and neighborhoods. 17 Table 6. Affordability of housing units with good JAI (> 6% by foot in 60 minutes) Expenditure share by consumption Share of households that Housing Predicted monthly quartile (%) can afford quality index rent (K sh) 1 2 3 4 <20% <30% 1 [very bad] 1580 24.7 13.8 8.8 3.0 81.1 93.4 2 [bad] 2181 34.1 19.0 12.2 4.1 63.9 84.4 3 [good] 3190 49.8 27.9 17.8 6.0 42.2 65.2 4 [very good] 7325 114.5 64.0 40.9 13.7 13.8 23.3 Note: Housing quality index indicates the quartiles of scores calculated by principal component analysis based on the characteristics of dwelling structure, access to services, and neighborhoods. Housing rents are predicted based on a JAI (share of accessible jobs by foot within 60 minutes) and housing quality index. 5. Conclusion Limited access to job opportunities potentially constrains productivity gains from agglomeration economies by impeding a matching between workers and jobs. Moreover, there is a concern that disadvantaged workers could be disproportionally affected by living farther away from job opportunities. The overall level of job accessibility in a city depends on the locations of jobs and residences, as well as transport networks. Who actually has good access to job opportunities hinges on the trade-off faced by households in their residential choices over job accessibility, living conditions, and housing costs. To understand this, we empirically analyze the spatial distributions of job accessibility, housing rents, and poverty in Nairobi, Kenya. Overall, our analysis suggests that Nairobi residents—particularly the poor and/or residents in informal settlements—face a job accessibility challenge. We measure job accessibility based on the share of accessible jobs within a certain travel time threshold at the 1km by 1km grid level, by combining various data sets on the spatial distribution of jobs and transport networks in the city. Within 60 minutes on average, Nairobi residents can access fewer than 10 percent of existing jobs by foot and only about a quarter of jobs by minibus. We further show that many informal settlements are located in low accessibility areas, and this is what the poor can afford. It is indeed costly to live in neighborhoods with better job accessibility, since housing quality tends to be better in such places and thereby rents are higher. However, our hedonic regression analysis suggests that housing rents are higher in neighborhoods with better job accessibility even after other observed characteristics are controlled for, reflecting people’s willingness to pay for living closer to job opportunities. There are several policies potentially effective in improving job accessibility among low-income households. As a supply-side option, it is essential to promote the development of affordable transport networks in the long term. Facilitating the extension of minibus services, for example, to underserved areas where many low-income households reside would help reduce their job search and commuting costs. Large and highly dense informal settlements may not have enough space for transit. When slum upgrading projects are implemented in such areas, introducing a right-of-way for transport would support their economic integration with the rest of the city. Some demand-side policy, such as cash transfers can be effective in the short term. There is some evidence that demand-side transport subsidies can favor labor market access and employment outcomes (Franklin, forthcoming). Transport subsidies are, however, a complicated device to design effectively and can become a heavy burden for local finances. It should be carefully examined whether and how to implement them. 18 References Abebe, G., S. Caria, M. Fefchamps, P. Falco, S. Franklin, and S. Quinn. 2017. Anonymity or Distance? Job Search and Labor Market Exclusion in a Growing African City. GLM|LIC Working Paper No. 34. Alonso, W. 1964. Location and Land Use: Toward a General Theory of Land Rent. Cambridge, MA: Harvard University Press. Antos, S. E., S. V. Lall, and N. Lozano-Gracia. 2016. “The Morphology of African Cities.” World Bank Policy Research Working Paper 7911. Atuesta, L. H., J. E. Ibarra-Olivo, N. Lozano-Gracia, and U. Deichmann. 2018. “Access to Employment and Property Values in Mexico.” World Bank Policy Research Working Paper 8383. Avner, P., and S. Lall. 2016. “Matchmaking in Nairobi: The Role of Land Use.” World Bank Policy Research Working Paper 7904. Bertaud, A. 2014. Cities as Labor Markets. Marron Institute on Cities and the Urban Environment, New York University, Working Paper #2 (February). Bird, J., P. Montebruno, and T. Regan. 2017. “Life in a Slum: Understanding Living Conditions in Nairobi’s Slums across Time and Space.” Oxford Review of Economic Policy 33 (3): 496– 520. Combes, P. P., and L. Gobillon. 2015. “The Empirics of Agglomeration Economies.” In Duranton, G., J. V. Henderson, and W. C. Strange (eds.), Handbook of Regional and Urban Economics. Vol. 5. North-Holland, Amsterdam. Duranton, G., and D. Puga. 2004. “Micro-foundations of Urban Agglomeration Economies.” In Henderson, J. V., and J. F. Thisse (eds.), Handbook of Regional and Urban Economics. Vol. 4. North-Holland, Amsterdam. Franklin, S. “Location, Search Costs and Youth Unemployment: Experimental Evidence from Transport Subsidies.” The Economic Journal. Forthcoming. Glaeser, E. L., M. E. Khan, and J. Rappaport. 2008. “Why Do the Poor Live in Cities? The Role of Public Transportation.” Journal of Urban Economics 63: 1–24. Gobillon, L., and H. Selod. 2014. “Spatial Mismatch, Poverty, and Vulnerable Population.” In Fischer, M. M., and P. Nijkamp (eds.), Handbook of Regional Science. Springer: Heidelberg. Gulyani, S., E. M. Bassett, and D. Talukdar. 2012. “Living Conditions, Rents, and Their Determinants in the Slums of Nairobi and Dakar.” Land Economics 88 (2): 251–74. Henderson, V., A. J. Venables, T. Regan, and L. Samsonov. 2016. “Building Functional Cities.” Science 352 (6288): 946. JICA (Japan International Cooperation Agency). 2013. JICA Nairobi Personal Travel Survey. Kain, J. F. 1968. “Housing Segregation, Negro Employment, and Metropolitan Decentralization.” The Quarterly Journal of Economics 82 (2): 175–97. 19 Klopp, J., S. Williams, P. Waiganjo, D. Orwa, and A. White. 2015. “Leveraging Cellphones for Wayfinding and Journey Planning in Semi-Formal Bus Systems: Lessons from Digital Matatus in Nairobi.” In Geertman, Jr., S., F. Joseph, R. Goodspeed, and J. Stillwell (eds.) Planning Support Systems and Smart Cities (227–41). Cham: Springer International Publishing. http://dx.doi.org/10.1007/978-3-319-18368-8_12. KNBS (Kenya National Bureau of Statistics). 2012. 2009 Kenya Population and Housing Census: Counting Our People for Implementation of Vision 2030. Volume VIII Urbanization. ———. 2018. Basic Report on Well-being in Kenya: Based on the 2015/16 Kenya Integrated Household Budget Survey (KIHBS). Lall, S. V., J. V. Henderson, and A. J. Venables. 2017. Africa’s Cities: Opening Doors to the World. Washington, D.C.: World Bank. Lucci, P., T. Bhatkal, and A. Khan. 2018. “Are We Underestimating Urban Poverty?” World Development 103: 297–310. Marx, B., T. M. Stoker, and T. Suri. 2013. “The Economics of Slums in the Developing World.” Journal of Economic Perspectives 27 (4): 187–210. ———. 2017. There Is No Free House: Ethnic Patronage in a Kenyan Slum. Unpublished manuscript. Mills, E., and B. Hamilton. 1995. Urban Economics. Prentice Hall. Muth, R. 1968. Cities and Housing. Chicago: University of Chicago Press. Nakamura, S. 2017. “Tenure Security Premium in Informal Housing Markets: A Spatial Hedonic Analysis.” World Development 89: 184–98. Nakamura, S., and Y. Karasawa. 2018. Slums as Spatial Poverty Traps? Analysis of Intra-city Residential Mobility in Kenyan Cities. Unpublished manuscript. Nakamura, S., and N. Yoshida. 2018. Analysis of Housing in African Slums from a Poverty Perspective. Unpublished manuscript. NORC. 2013. Kenya State of the Cities Baseline Survey: Overview Report: Volume I. Quirós, T. Peralta. 2015. “Mobility for All: Getting the Right Urban Indicator.” Connections - Transport & ICT 25 (November). http://pubdocs.worldbank.org/pubdocs/publicdoc/2015/11/67931446737085683/TransportI CT-Newsletter-Note25a-Nov-highres.pdf. Quirós, T. Peralta, and S. Raj Mehndiratta. 2015. “Accessibility Analysis of Growth Patterns in Buenos Aires, Argentina: Density, Employment, and Spatial Form.” Transportation Research Record: Journal of the Transportation Research Board 2512 (November): 101– 9. https://doi.org/10.3141/2512-12. Rosen, S. 1974. “Hedonic Prices and Implicit Markets: Product Differentiation in Pure Competition.” Journal of Political Economy 82 (1): 666–77. Stewart, Anson F., and P. Christopher Zegras. 2016. “CoAXs: A Collaborative Accessibility-Based Stakeholder Engagement System for Communicating Transport Impacts.” Research in Transportation Economics 59: 423–33. 20 Stokenberga, A. 2017. Urban Accessibility Analysis for Dakar, Senegal. Unpublished manuscript. Taylor, L. O. 2008. “Theoretical Foundations and Empirical Developments in Hedonic Modeling.” In Baranzini, A., J. Ramirez, C. Schaerer, and P. Thalmann (Eds.), Hedonic Methods in Housing Markets: Pricing Environmental Amenities and Segregation (15–37). LLC, New York: Springer Science+Business Media. Turok, I., and J. Borel-Saladin. 2018. “The Theory and Reality of Urban Slums: Pathways-out-of- poverty or Cul-de-sacs?” Urban Studies 55 (4): 767–89. Williams, S., A. White, P. Waiganjo, D. Orwa, and J. Klopp. 2015. “The Digital Matatu Project: Using Cell Phones to Create an Open Source Data for Nairobi’s Semi-formal Bus System.” Journal of Transport Geography 49 (December): 39–51. World Bank. 2016. Kenya Urbanization Review. Washington, D.C.: World Bank. ———. 2018. Kenya Poverty and Gender Assessment. Washington, D.C.: World Bank. 21 Appendix A. Additional tables and figures Table A1. Summary statistics of households in Nairobi Count Mean SD Min Max Living in informal neighborhood (1 = yes; 0 = no) 1095 0.532 0.499 0.000 1.000 Household monthly expenditure (K Sh) 1089 20268 25104 1268 255600 Per capita monthly expenditure (K Sh) 1081 7888 10100 211.3 157000 Age of household head 1021 35.04 10.85 18.00 87.00 Years of household head’s education 1015 12.15 4.185 0.000 21.00 Household size 1086 3.080 1.734 1.000 13.00 Travel time to work (one way, in minutes) 905 30.54 21.40 1.000 120.0 Commute by foot (1 = yes; 0 = no) 1095 0.354 0.479 0.000 1.000 Commute by minibus (1 = yes; 0 = no) 1095 0.431 0.495 0.000 1.000 There is a bus stop within 500 meters (1 = yes; 0 = no) 1095 0.723 0.448 0.000 1.000 Distance from the CBD (km) 1095 7.155 3.406 0.301 20.46 Share of accessible jobs by foot in 45 minutes (%) 1095 4.311 4.562 0.093 28.21 Share of accessible jobs by minibus in 45 minutes (%) 1095 12.36 10.98 0.432 57.61 Share of accessible jobs by car in 45 minutes (%) 1095 75.28 15.40 12.74 94.77 Table A2. Summary statistics of rental units in Nairobi Count Mean SD Min Max Monthly rent (K Sh) 989 5315 9123 300.0 123300 Informal neighborhood (1 = yes; 0 = no) 989 0.557 0.497 0.000 1.000 Written tenancy agreement (1 = yes; 0 = no) 981 0.208 0.406 0.000 1.000 Type: single housing (1 = yes; 0 = no) 989 0.351 0.477 0.000 1.000 Type: single housing shared (1 = yes; 0 = no) 989 0.006 0.078 0.000 1.000 Type: single-story shared (1 = yes; 0 = no) 989 0.278 0.448 0.000 1.000 Type: a room in a house (1 = yes; 0 = no) 989 0.022 0.148 0.000 1.000 Type: shack (1 = yes; 0 = no) 989 0.002 0.045 0.000 1.000 Type: multistory (1 = yes; 0 = no) 989 0.335 0.472 0.000 1.000 Type: other (1 = yes; 0 = no) 989 0.006 0.078 0.000 1.000 Floor area (1/1000 m2) 960 0.018 0.024 0.001 0.336 Wall: stone/brick/block (1 = yes; 0 = no) 989 0.523 0.500 0.000 1.000 Wall: corrugated iron sheet (1 = yes; 0 = no) 989 0.350 0.477 0.000 1.000 Wall: mud/wood (1 = yes; 0 = no) 989 0.127 0.334 0.000 1.000 Roof: corrugated iron sheet (1 = yes; 0 = no) 989 0.869 0.338 0.000 1.000 Roof: clay tiles (1 = yes; 0 = no) 989 0.067 0.250 0.000 1.000 Roof: concrete (1 = yes; 0 = no) 989 0.065 0.246 0.000 1.000 Floor: earth/clay (1 = yes; 0 = no) 989 0.083 0.276 0.000 1.000 Floor: tiles (1 = yes; 0 = no) 989 0.062 0.241 0.000 1.000 Floor: cement (1 = yes; 0 = no) 989 0.855 0.352 0.000 1.000 Water: piped inside (1 = yes; 0 = no) 989 0.465 0.499 0.000 1.000 Water: piped outside (1 = yes; 0 = no) 989 0.067 0.250 0.000 1.000 Water: shared tap (1 = yes; 0 = no) 989 0.044 0.206 0.000 1.000 Water: other (1 = yes; 0 = no) 989 0.424 0.494 0.000 1.000 Toilet: flush inside (1 = yes; 0 = no) 989 0.173 0.378 0.000 1.000 Toilet: flush/VIP latrine outside (1 = yes; 0 = no) 989 0.268 0.443 0.000 1.000 Toilet: private pit latrine (1 = yes; 0 = no) 989 0.116 0.321 0.000 1.000 Toilet: shared toilet (1 = yes; 0 = no) 989 0.435 0.496 0.000 1.000 Toilet: other (1 = yes; 0 = no) 989 0.001 0.032 0.000 1.000 Toilet: none (1 = yes; 0 = no) 989 0.007 0.084 0.000 1.000 Electricity (1 = with access; 0 = no access) 989 0.806 0.396 0.000 1.000 Garbage dump as problem (1 = yes; 0 = no) 989 0.474 0.500 0.000 1.000 Factory as problem (1 = yes; 0 = no) 989 0.091 0.288 0.000 1.000 Secure (1 = yes; 0 = no) 987 0.590 0.492 0.000 1.000 22 Figure A1. Distribution of job accessibility in Nairobi (a) By foot in 30 minutes (b) By foot in 45 minutes (c) By foot in 60 minutes 20 30 15 15 20 10 Percent Percent Percent 10 10 5 5 0 0 0 0 5 10 15 20 0 10 20 30 0 10 20 30 40 Share of accessible jobs (%) Share of accessible jobs (%) Share of accessible jobs (%) (d) By minibus in 30 minutes (e) By minibus in 45 minutes (f) By minibus in 60 minutes 30 20 10 8 15 20 Percent Percent Percent 6 10 4 10 5 2 0 0 0 0 10 20 30 0 20 40 60 0 20 40 60 80 Share of accessible jobs (%) Share of accessible jobs (%) Share of accessible jobs (%) (g) By car in 30 minutes (h) By car in 45 minutes (i) By car in 60 minutes 15 10 20 8 15 10 Percent Percent Percent 6 10 4 5 5 2 0 0 0 0 20 40 60 80 20 40 60 80 100 50 60 70 80 90 100 Share of accessible jobs (%) Share of accessible jobs (%) Share of accessible jobs (%) Source: Authors’ calculation based on the Cities Baseline Survey, JICA (2013), Digital Matatus Project, and OpenStreetMap. Note: Panels (a) to (i) show the distribution of job accessibility indicators for tenants in the Cities Baseline Survey. 23 Figure A2. Distance to the CBD and job accessibility in Nairobi (a) By foot in 30 minutes (b) By foot in 45 minutes (c) By foot in 60 minutes Share of accessible jobs (%) Share of accessible jobs (%) Share of accessible jobs (%) 0 .05 .1 .15 .2 .25 .3 .3 .4 .2 .2 .1 .1 0 0 0 5 10 15 20 0 5 10 15 20 0 5 10 15 20 Distance from CBD (km) Distance from CBD (km) Distance from CBD (km) (d) By minibus in 30 minutes (d) By minibus in 45 minutes (e) By minibus in 60 minutes Share of accessible jobs (%) Share of accessible jobs (%) Share of accessible jobs (%) .6 .8 .3 .6 .4 .2 .4 .2 .1 .2 0 0 0 0 5 10 15 20 0 5 10 15 20 0 5 10 15 20 Distance from CBD (km) Distance from CBD (km) Distance from CBD (km) (f) By car within 30 minutes (g) By car in 45 minutes (h) By car in 60 minutes Share of accessible jobs (%) Share of accessible jobs (%) Share of accessible jobs (%) .8 1 .5 .6 .7 .8 .9 1 .8 .6 .6 .4 .4 .2 .2 0 0 5 10 15 20 0 5 10 15 20 0 5 10 15 20 Distance from CBD (km) Distance from CBD (km) Distance from CBD (km) Source: Authors’ calculation based on the Cities Baseline Survey, JICA (2013), Digital Matatus Project, and OpenStreetMap. Note: Markers indicate rental units in informal and formal residential areas based on the Nairobi survey. 24