Policy Research Working Paper 10127
Putting a Price on Safety
A Hedonic Price Approach to Flood Risk
in African Cities
Alvina Erman
Ingrid Dallmann
Urban, Disaster Risk Management, Resilience and Land Global Practice
July 2022
Policy Research Working Paper 10127
Abstract
This paper uses a hedonic property price function to esti- variable bias. For example, only 12 percent of households
mate the relationship between flood risk and rents in four living in flood-prone areas were aware of the flood risk when
Sub-Saharan Africa cities: Accra, Antananarivo, Dar es they moved in. In Antananarivo, job density is associated
Salaam, and Addis Ababa. The analysis relies on household with higher rents while in Accra and Addis Ababa, higher
survey data collected after flood events in the cities. Flood job density is associated with lower rents. Results are nega-
risk is measured with self-reported data on past flood expo- tive but not significant in Dar es Salaam. When interacting
sure and perception of future risk of flooding of households. job density with flood risk for each city, the negative effect
The study finds that flood risk is associated with lower rents of job density on rents is higher (in absolute value) when
in Accra, Antananarivo, and Addis Ababa, ranging from flood risk is high in Accra and Addis Ababa, and the positive
14 to 56 percent lower. In contrast, risk is associated with effect of job density on rents becomes negative when flood
higher rent in Dar es Salaam, which could be potentially risk is high in Antananarivo. This relationship is not found
attributed to a combination of lack of awareness of flood in Dar es Salaam. The finding seems to suggest that access to
risk among renters, high transaction costs and omitted jobs is a factor driving people to settle in flood-prone areas.
This paper is a product of the Urban, Disaster Risk Management, Resilience and Land Global Practice. 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/prwp. The
authors may be contacted at aerman@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
Putting a Price on Safety – a Hedonic Price Approach to Flood
Risk in African Cities*
Alvina Erman and Ingrid Dallmann†
JEL Classification: O18, O55, O57, Q54, R20, R31
Keywords: natural disasters, disaster risk management, hedonic regression, poverty, resilience, urban
housing, sustainable cities, urbanization, urban floods, access to jobs, Africa, Ethiopia, Ghana,
Madagascar, Tanzania
* We thank Samira Barzin and Paolo Avner for data and codes on job density, and Nathalie Picarelli for data on job accessibil-
ity in Ghana. We are very grateful to Kirsten Hommann, Catherine Lynch, Shohei Nakamura, and Steven Rubinyi, for detailed
and invaluable feedback and suggestions. We also thank to participants in the GFDRR seminar analytic team.
†
GFDRR, World Bank, aerman@worldbank.org
1 Introduction
Urban risks related to climate change are increasing (IPCC, 2014). Over half of cities with more than
500,000 inhabitants are at high risk of exposure to at least one type of natural disaster. Countries in Sub-
Saharan Africa (SSA) have experienced high rates of urbanization in recent decades, and it is expected
that between 2018 and 2050 the SSA urban population will triple (UN, 2019). Unfortunately, the jour-
ney to urbanization in SSA has been accompanied by increasing numbers of informal settlements, poor
quality public services and infrastructure (Castells-Quintana and Wenban-Smith, 2020; Fay and Opal,
1999), and little or no preparedness for natural disasters. These conditions make SSA urban populations
extremely vulnerable to natural disasters and especially floods, which are frequent and often very se-
vere. In the absence of insurance markets and other risk transfer options, much of the impact of flooding
is absorbed by the households, resulting in enormous costs, which accumulate over time and pose a
threat to development gains (Erman et al., 2020). In a context of climate change, increasing urbanization
and a predicted “lost decade” of development due to the COVID-19 crisis (WB, 2021), it is important to
understand how disaster risk is impacting households in cities in SSA. Risk informed city planning and
significant investments in public services, such as solid waste management and drainage are needed to
meet the challenges of rapid urbanization, urban poverty and climate risk. Much of the responsibility
fall on city and local governments that have limited capacity and fiscal space to meet the investment
needs. Quantifying the cost of flood exposure for households opens opportunities to inform land-based
financing and disaster risk management policy. This study does this quantification by using a hedonic
price approach to evaluate the relationship between flood impact and rent pricing in Accra (Ghana), Ad-
dis Ababa (Ethiopia), Antananarivo (Madagascar), and Dar es Salaam (Tanzania). Although crucial for
SSA countries and for low- and middle- income countries more generally, this type of analysis is scarce
in the development economics literature.
Climate change is expected to increase the frequency and intensity of floods (IPCC, 2014). Severe
and moderate floods can cause population displacement, water-borne diseases (Bartlett, 2008), and food
insecurity; destruction or deterioration of assets, property, and public infrastructure; loss of income and
lives; disruption to economic activities, school attendance, and transport networks; electrical energy and
water outrages, etc. Other things being equal, the higher the concentration of population and assets, the
higher the value of the loss, which makes urban areas especially vulnerable.
Ability to quantifying the cost of avoiding flood would help policy makers to conduct cost-benefit
estimations of investment in flood resistant infrastructures and contingency plans. Reducing exposure
to disasters in a given area of a city shines a spotlight on the importance of location and can make certain
areas more attractive for new residents, businesses, and amenities – all dynamics that increase the value
of the land. This provides an opportunity for local authorities to capture the land value to pay for projects
that will increase resilience to natural disasters, or to finance other public priorities. However, we lack
the tools to forecast land value appreciation from disaster mitigation investments and studying whether
or not land pricing internalizes flood risks would facilitate development of such tools.
Hedonic property price models can be used to capture the value of avoiding flood exposure. They
enable estimation of property prices based on bundles of housing, amenities, and location characteristics
(Bishop et al., 2020; Brueckner, 2011; Rosen, 1974). In an efficient housing market, properties exposed to
2
flooding are expected to have a lower price tag than equivalent properties not exposed to flooding. This
price discount can be interpreted as the welfare cost of flood exposure, or in other words the willingness
to pay to avoid flooding. Numerous studies which focus on high- or middle- income countries use
hedonic price models to calculate the property price discount for location in an area exposed to flooding.1
This study is one of very few analyses of the relationship between flooding risk and rents in low- and
lower-middle income countries. Meta-analysis of 37 published papers and 364 point estimates shows
an et al., 2018). These meta-
that all of them focus high-income countries, mainly the United States (Beltr´
regressions show that properties located in flood plains which suffer from inland flooding are 4.6 percent
cheaper on average than a similar property not exposed to flood risk. However, the effects of flooding
an et al. (2018) vary between a 75.5 discount to a 61
on property prices in the 37 papers studied by Beltr´
percent price premium. Thus, their study finds very high levels of heterogeneity between the effects of
flooding on property prices even in an efficient housing market.
Examples of studies of the relationship between the housing market and flooding in upper-middle-
´
income economies include Rabassa and Zoloa (2016), and Alvarez and Resosudarmo (2019) who study
La Plata, Argentina, and Jakarta, Indonesia respectively. Rabassa and Zoloa (2016) estimate the benefits
of a new hydraulic infrastructure based on estimation of the willingness to pay to avoid flood exposure
using data on real estate prices and a flood risks map. They find that flood plain areas have a 3.5 percent
price discount compared to other areas, and that this discount can be five times higher in areas of higher
´
flood risk and less resilience. Alvarez and Resosudarmo (2019) measure the correlation between the
water level of floods in Jakarta in 2007 and housing rental prices and find that a 10 percent increase in
the flood depth is associated with 1.24 percent lower rents. Using the same data sources as this study,
Erman et al. (2020) and (WB, 2017) assess the relationship between rents and reported flood exposure
in Accra, Ghana and Antananarivo, Madagascar respectively. Erman et al. (2020) identify a 27 percent
price discount on rent among flood-affected households in Accra and (WB, 2017) find that flood-affected
households pay 15 percent lower rents in Antananarivo, Madagascar.
Most of the literature uses hazard maps to identify flood plains or high-risk locations. For instance,
Bin and Polasky (2004) look at both location within a floodplain and location outside a floodplain in Pitt
County, North Carolina, US which was severely impacted by the major floods that followed hurricane
Floyd in 1999. They find that properties located in floodplains were cheaper than equivalent proper-
ties located in other areas, and that the price discount in floodplains was significantly larger after the
hurricane.
This study uses self-reported data from household surveys to assess the willingness to pay to avoid
floods. The analysis relies on four household surveys conducted between 2016 and 2018 in four SSA
cities that are the most economically and politically relevant in their countries: Accra, Antananarivo, Dar
es Salaam, and Addis Ababa. Flood risk is captured in two ways: by having reported being affected by
a flood in recent years and respondents’ perceived risk of future flooding. The use of household survey
data provides an opportunity to expand the use of the hedonic pricing approach to evaluate the benefits
of urban resilience and infrastructure investment in countries that rely heavily on household survey data
and where hazard risk and property value data may not be readily available. The household survey data
are complemented by information on access to jobs.
1
an et al. (2018) for an extensive review of the effect on property prices of housing location in a flood plain.
See Beltr´
3
The results show that rents tend to be lower in areas exposed to flood risk in three of the four cities.
The relationship between job density and rents is less clear, which is most likely a result of the nature
of city shape and the role of informal job markets. However, job density seems to play a role in the
household decision to settle in high risk areas.
Overall, the study indicates that the drivers of rents in cities in SSA are still poorly understood and
that lack of data is the main barrier to a better understanding. This analysis shows that household
surveys that include rent or housing costs, housing characteristics, access to public services, tenure ar-
rangements, and household experiences with flooding or other natural hazards are an important source
of information in developing countries to value the willingness to pay to avoid hazard risk, and thus,
the benefits associated to risk reduction investments. In housing markets with no or limited access to in-
formation about flooding risks, surveys with information on the experience of households with flooding
and their expectation of future floods capture better the internalization of flooding risks by households
than modeled flood maps.
The remainder of the paper is organized as follows. Section 2 provides a general overview of the
flood and vulnerability contexts in the cities studied. Section 3 introduces the data and the empirical
estimation strategy. Section 4 presents the empirical results. Section 5 concludes the paper.
2 Urbanization and vulnerability to floods in four Sub-Saharan African cities
Urbanization is associated with development, higher incomes, better access to employment, better pub-
lic services, and better access to information. However, if the urbanization growth rate exceeds the rate
of development, or is accompanied by poor quality institutions, urbanization can increase population in-
equality. This results in poorer people being more exposed to poor infrastructure, low quality housing,
and natural hazards. In particular, some urban conditions such as high population and asset densities,
unplanned settlements, low quality drainage systems, unimproved solid waste management, poor hous-
ing conditions, and certain kinds of locations (coastal areas, lowland, near to rivers) increase the risk of
exposure to natural hazards.
The share of Africans living in urban areas is projected to grow from 42 percent in 2018 to 59 percent
in 2050 (UN, 2019). This share is the lowest among world regions. However, after 2020 the urbaniza-
tion growth rate in Africa is likely to be the highest in the world, overtaking rates in Asia. Among
the countries studied, Ghana has the highest urban population rate (57 percent) and has experienced the
highest urban population increase since the mid-1990s (+18 percentage points). It is expected that almost
three-quarters of Ghana’s population will be urban in 2050. Madagascar and Tanzania have experienced
similar urbanization rates, with urban population shares in 2020 of 38 percent and 35 percent respec-
tively, expected to increase to 58 percent and 55 percent by 2050. Ethiopia has experienced a slower pace
of urbanization and in 2020 had an urban population share of 22 percent although this is expected to
almost double by 2050 (see Figure 1 and Table 1).
This paper studies the largest and most important cities in each country: Accra, Antananarivo, Dar
es Salaam, and Addis Ababa. The cities make up between 15 (Accra) and 31 percent (Antananarivo) of
each country’s urban population (Table 1).
4
F IGURE 1: Urban population share evolution
Source: Authors’ calculation based on World Development Indicators
TABLE 1: Urbanization, poverty, and floods
Ethiopia Ghana Madagascar Tanzania
Country
total population, 2019 (millions)1 112.1 30.4 27.0 58.0
1
urban population share, 2020 (%) 21.7 57.3 38.5 35.2
2
projected urban population share in 2050 (%) 39.0 73.1 57.9 55.4
City
total population, 2018 (millions)2 4,400 2,440 3,058 6,048
2
% on total urban population, 2018 20 15 31 30
Poverty rate at international poverty line 30.8 12.7 78.8 49.4
3
population below $1.90 a day (%)
Floods 2013-20174
number of events 3 2 2 8
1
World Development Indicators.
2
UN (2019).
3
Reference years: Ethiopia: 2015; Ghana: 2016; Madagascar: 2012; Tanzania: 2017.
4
Dartmouth Flood Observatory. Floods are defined in three categories: class 1, large flood events,
significant damage to structures or agriculture, fatalities, and/or 1-2 decades-long reported interval
since the last similar event; class 2, very large flood events, more than 20 years but less than 100
year recurrence interval, and/or a local recurrence interval of at 10-20 years; class 3, extreme events,
with an estimated recurrence interval of more than 100 years.
5
The combination of rapid urbanization, urban poverty, lack of investments in infrastructure and pub-
lic services, as well as climate change, has made SSA cities increasingly vulnerable to impacts of flooding
and other natural hazards. Dartmouth Flood Observatory (DFO) data consider flooding as a major event
involving at least significant damage to structures and agriculture, fatalities, and displacements. In the
five years preceding the survey all four countries suffered at least two such flood events and Tanzania
suffered eight (Table 1). Major events such as those reported by DFO are well-documented and their
impacts are captured in official numbers reported by news and governments. In addition to larger flood
events, these cities are also affected by high-frequency, low-intensity flooding. The high frequency flood-
ings are less costly in terms of direct losses but are highly disruptive to the everyday lives of residents
affected by them. Floods affect people’s lives in many ways. They damage assets, such as furniture,
electronics, and vehicles with high replacement values for urban poor. Flooding can disrupt infrastruc-
ture, such as road networks, water, and energy services, which affects residents’ ability to access jobs,
school and operate an enterprise. Flooding also has health consequences and can lead to the spread of
water-borne diseases, urinary tract infections, skin fungus, etc. Although both poorer and richer areas
are affected by flooding, the vulnerability of housing structures and lack of access to insurance, savings
and other financial support, poorer households are worse off when affected by flooding. In many cities,
informal settlements also tend to be in flood plains because these areas were previously undeveloped
and perhaps offered other benefits, such as proximity to jobs and services.
Since climate change is likely to make flood events more frequent and intense and the urban poor are
likely to suffer disproportionately, poverty rates can be expected to increase as a result (Hallegatte et al.,
2020). This is particularly important to consider in the context of the current global pandemic which is
affecting urban populations disproportionately. These are important issues which need to be addressed
in combination.
3 Data and methodology
3.1 The surveys
The analysis relies on integrated and harmonized data from four surveys, covering the four cities of
Accra, Antananarivo, Dar Es Salaam, and Addis Ababa. For Accra, Dar Es Salaam and Addis Ababa the
Disaster-poverty survey data is used, which contain information on how households experience floods,
their perceived risk of future floods, and characteristics of household members, housing conditions and
public services, and proxy for household expenditure, rent and housing costs. For Antananarivo, a
Living Standards Measurement Survey (LSMS) is used, which covers similar variables and was therefore
added to the study. The surveys were conducted in 2016 (Antananarivo), 2017 (Accra, Dar es Salaam,
Addis Ababa), and 2018 (follow-up in Dar es Salaam).
To define flood exposure, self-reported information from the surveys is used. The Accra and Antana-
narivo surveys capture information on specific disasters (severe flooding in 2015 in Accra, and Cyclone
Chedza in 2015 in Antananarivo). In the other two cities, the surveys focused on frequent flood exposure
rather than a specific disaster.2 Households were asked also about their perceived risk of future flood
2
However, in Dar es Salaam, a follow up phone survey was conducted in 2018 to capture information on the flood experi-
enced in that year.
6
exposure.3 Both variables are used in the analysis to proxy exposure to flood risk by using two binary
variables: 1) if the household reports having been directly affected by floods, and 2) if the household per-
ceives the risk of future flood exposure as high or very high. Appendix Table A.1 provides the precise
questions and the definitions of the flood exposure variables for each city. Self-reported disaster data
have two main advantages. First, they capture actual household experiences and expectations rather
than modeled risk estimations. Flood impacts can differ from street to street depending on elevation,
housing quality, access to, and topography, etc., and flood models do a generally poor job of capturing
localized flood risk, especially in urban areas.4 Second, in the context where information on flood risk as
defined in flood maps is not readily available, up-to-date or well-known to the public, past experiences
of flood and the perception of risk can be better predictors of rent values. The use of self-reported prox-
ies for flood risk may therefore increase the accuracy of hedonic models when used in contexts where
buyers (or renters) are not fully informed.
In hedonic analyses, house-transaction prices are ideal to define housing value (Bishop et al., 2020).
However, they can be difficult to access even in developed countries. Bishop et al. (2020) provide a list of
the countries where microdata on housing transactions is available: it includes several European coun-
tries, the United States, Australia, Chile, China, Japan, and the Republic of Korea. Examples of studies
that use transaction data in middle-income countries are Argentina (Rabassa and Zoloa, 2016) and In-
´
donesia (Alvarez and Resosudarmo, 2019). Limited data availability explains why, while the hedonic
approach to estimate households’ willingness to pay for a change in an environmental amenity is used
extensively by researchers, it is difficult to find studies of African countries. Household survey data
overcome this limitation since surveys often capture information needed to construct a hedonic pricing
model, including housing characteristics, access to services and different determinants of housing costs.
The surveys exploited in this study include detailed questions on actual housing costs and rents and an
estimate of how much a tenant would pay in rent for the dwelling if rented, as estimated by the respon-
dent. Since only a subset of households are renters, exploiting the “self-estimated” rental values helps
increase the sample size. The analysis therefore combines reported rent values and the self-estimated
rent values5 to construct a rent value variable for all four cities. Observations with missing values for
both items were excluded. Comparison of the rents, self-estimated rents, and the combined variable dis-
tributions are depicted in Appendix Figure B.1. Measurement errors can affect rent and self-estimated
rent values. To reduce these errors, outliers were removed and replaced with estimated values.6 Values
are in local currency and transformed to 2017 PPP US$.
The final sample is based on observations without missing values for rents and self-estimated rents.
It includes 3,132 households in four cities and four countries: Accra (996 households), Antananarivo
(919 households), Dar es Salaam (466 households), and Addis Ababa (751 households). Table 2 presents
a summary of the different database characteristics. More detail on the survey methodology and char-
3
The questions related to flood perception were almost identical for all four cities although for Antananarivo the time
horizon was 5 rather than 2 years.
4
an et al., 2018) are large surfaces.
In most studies, flood models or risk area identifications (Beltr´
5
The question related to self-estimated rent is “If somebody else wants to rent a dwelling just like this today, how much
money would he/she have to pay monthly?”
6
Outliers are defined as observations that deviate from the mean by three standard deviations or more. Outliers were
replaced by estimated rents/self-estimated rents. Rents/self-estimated rents are estimated based on data on expenditure,
housing conditions, access to public services, and time to business center. Outliers represent 0.5% to 2% of observations,
depending on the city.
7
acteristics for Accra, Dar es Salaam, and Antananarivo can be find in Erman et al. (2020), Erman et al.
(2019), and WB (2017) respectively.
TABLE 2: Databases descriptions
City Database Area Date Flood/s HH selection method
Accra Disaster-Poverty 9 neighborhoods May-Jun/2017 2015 flood 4 strata: combination of high/low
household survey considered slum areas risk of flood and high/low level of poverty.
in the Odaw basin area Informal settlements.
Not representative at the city level.
Antananarivo Urban Living Greater Antananarivo Oct-Dec/2017 2015 flood 2 strata: risky and no risky areas, where
Standards “Chedza flood” risky are EAs affected by Chedza.
Representative at the city level.
Dar es Salaam Disaster-Poverty Dar es Salam Nov/2017, Most recent flood, 3 levels of flood risk (no risk, low risk and
household survey Sep/2018 2018 flood high risk). Representative at the city level.
Addis Ababa Disaster-Poverty Addis Ababa May-Jun/2017 Most recent flood Representative at the city level.
household survey
3.2 Accessibility to jobs
In an efficient housing market, property prices rise with increased access to jobs and amenities (Ahlfeldt,
2013), and in choosing where to live people make a trade-off between low housing value and proxim-
ity to jobs. To understand rents, we need to understand access to economic opportunities and how it
influences housing markets. However, there are two significant data constraints, common to develop-
ing country cities which make it difficult to capture access to economic opportunities, particularly for
SSA cities: 1) lack of information on job locations, and 2) lack of information on transit routes and travel
times (Peralta-Quiros et al., 2019). This study captures access to economic opportunities using the spatial
density of employment developed in Barzin et al. (2022). This measure is used to proxy for job density.
It was created using a machine learning based algorithm, and data from OpenStreetMap (OSM) and
Google Earth Engine (GEE). OSM data provide information on the location of different types of pub-
lic and private services and institutions (hospitals, pharmacies, banks, schools, etc.), amenities (rivers,
parks, etc.), and infrastructure (roads, airports, etc.). GEE data were used to add information on popu-
lation density, intensity of night lights, and land cover. This is a novel way to estimate job density. The
main advantage of this proxy is that it can be calculated for any city which has accurate and available
OSM and GEE data and where data on economic activity are scarce or missing. It allows comparison of
economic activity across cities, based on a unique measure. The job density proxy measures relative job
density where a low value denotes lower job density and a high value higher job density. The values
themselves cannot be interpreted directly. The maps of the four cities analyzed here, and with the spatial
distribution of jobs are provided in Appendix Figures D.1 to D.4.
3.3 Descriptive statistics
Thirteen percent of households in our sample had been affected directly by at least one flood event: 44
percent in Accra, 9 percent in Antananarivo, 20 percent in Dar es Salaam, and 8 percent of households
in Addis Ababa. In Accra and Dar es Salaam, the proportion of households that perceived risk of future
8
flooding to be high is similar to the proportion of households actually affected by flooding. In contrast,
the proportion of households in Antananarivo and Addis Ababa which perceived future flooding risk
as high was much higher than the proportion of households actually affected by floods (Table 3). This
suggests that these households are in flood prone areas but have not been affected by floods in recent
years.
The share of households in the sample that report owning their dwelling is 15 percent in Antana-
narivo, 20 percent in Accra, and 52 percent in Addis Ababa (Appendix Table D.1). Renters represent
44 percent in Accra, 34 percent in Antananarivo, 36 percent in Dar es Salaam, and 42 percent in Addis
Ababa of the total sample (Appendix Table D.1). All renters are considered in the sample used for analy-
sis. Since the variable used for analysis combines reported rent value and self-estimated rent, the sample
used for analysis also consists of households in other tenure arrangements. In the case of Dar es Salaam,
combining rents and self-estimated rents does not increase the number of observations since only renters
responded to the rent estimation question.
In SSA cities, ownership does not necessarily equate with tenure security. Families often consider
themselves owners although they may not have legally recognized rights to occupy the dwelling and/or
the land. In Accra 29 percent of households in the sample are “rent free”. The “rent-free” dwellings are
mainly owned by household relatives that are not household members (92 percent).
Access to public services is generally low. In Antananarivo, only 6 percent of households have access
to improved toilet facilities and 17 percent have access to improved drinking water. The dwellings have
an average of between 1.5 (Accra) and 3.1 (Addis Ababa) rooms. In terms of rents, households in Addis
Ababa pay the highest average annual rents in the sample, and households in Accra pay the lowest
(Table 3). In all the cities considered, there is a high concentration of low rents. However, Antananarivo
and Dar es Salaam have a flatter distribution of higher rents compared to Accra and Addis Ababa (Figure
2).
Among the surveyed households, those in Accra have the lowest estimated job density, and those in
Antananarivo and Dar es Salaam have the highest.
9
TABLE 3: Descriptive statistics by city
Accra Antananarivo Dar es Salaam Addis Ababa
Obs % Obs % Obs % Obs %
Exposure to floods
Households directly affected by flood 996 44.5 919 9.1 466 19.6 751 8.4
Households that perceive flood risk as high 996 40.2 919 51.1 466 17.7 751 31.2
Dwelling tenure 996 919 466 751
Owner 202 20.3 135 14.7 0 0 386 52.5
Rented 440 44.2 781 85.0 466 100 105 44.5
Free 293 29.4 0 0 0 0 15 2.6
Other 61 6.2 3 0.3 0 0 1 0.4
Housing conditions
10
Access to improved toilet facility 996 15.6 919 6.0 466 26.4 751 26.6
Access to improved drinking water 996 45.3 919 17.1 466 34.9 751 98.0
Access to collected solid waste 996 73.9 919 65.3 466 88.6 751 66.6
Lighting with electricity 996 96.7 768 99.2 466 91.9 751 99.5
Obs Mean Obs Mean Obs Mean Obs Mean
Household size 996 3.4 919 4.1 466 3.2 751 4.3
Number of rooms 996 1.5 919 1.7 370 2.6 751 3.1
Annual rent
Rent (2017 PPP US$) 482 463.6 785 895.8 465 1072.2 156 1871.9
Self-estimated rent (2017 PPP US$) 979 106.8 134 4081.6 466 1142.1 708 4727.9
Combined rent and self-estimated rent (2017 PPP US$) 996 258.3 919 1372.8 466 1077.9 751 4126.1
Job density 817 1.5 916 2.0 466 2.0 751 1.7
Notes: Figures are weighted.
F IGURE 2: Distribution of annual rent by flood exposure
Source: Authors’ calculation based on Disaster-Poverty Household Survey.
Notes: The rent variable is a combination of the rents declared by households and their estimations of rent values.
Rent is measured in 2017 PPP US$by year.
11
3.4 Empirical method
The aim of the study is to understand the willingness to pay to avoid floods among households in four
SSA cities. To this end, this study employs a hedonic pricing approach (Rosen, 1974) to assess drivers
of rent values. In a hedonic regression the price of a dwelling is determined by a bundle of internal
housing attributes, housing location, amenities, transport costs, and other determinants of life quality
(Bishop et al., 2020; Brueckner, 2011). The estimated coefficients in the hedonic regressions represent the
share of the price for each attribute. The hypothesis of this study is that dwellings that are exposed to
flooding have a lower price than dwellings located elsewhere. The hedonic model isolates the effect of
flood exposure on housing prices from other drivers and therefore reflects the price that residents assign
to safety from floods. The following econometric specification is used:
ln Renti = α0 + α1 f loodi + α2 Ti + α3 Hi + εi (1)
where i indexes the household, Rent is the rent value, f lood is the previous exposure to floods or
high perception of risk of future flooding, T is job density, H is a vector of housing characteristics, and
ε is the error term.7 If the housing market is efficient then all else being equal, we would expect α
ˆ1 < 0
ˆ 2 > 0.
and α
The hedonic model assumes perfect competition. This is a bold assumption considering the high level
of informality in the cities analyzed. For example, perfect competition in housing markets assumes that
clients are fully informed, the price is set by the market, and zero transaction costs. However housing
markets are usually not perfectly competitive, especially not in developing countries. For example,
information that can affect housing pricing may not be available or difficult to access. This is particularly
true regarding hazard risk information. As a result, unless an area has been recently flooded, residents
may not know that an area is flood-prone. Information on pricing may not be readily available either. In
more informal housing markets, where there is a lack of trust and legal protection, prices are often set
in informal networks, such as extended family or within religious or ethnic groups. Transaction costs
are not zero. Large down payments are common in informal housing markets, as a way for landlords
to protect themselves. Despite these challenges, the hedonic pricing model has been used extensively to
estimate the willingness to pay for changes in environmental amenities and to inform public and private
decision makers (Bishop et al., 2020). By controlling for other possible drivers of housing costs in the
context of African cities, the analysis reduces part of the potential bias. However, it is important to keep
these drawbacks in mind when interpreting the results of this study.
7
Bishop et al. (2020) recommend that in selecting an econometric specification 1) the price function should be assumed to
be nonlinear, and 2) the estimations should rely on robust standard error and be clustered at the spatial scale of the variable
of interest to account for heteroskedasticity and spatial autocorrelation. The analysis in this paper takes account of these
recommendations.
12
4 Results
4.1 Flood risk and rent
Appendix Tables C.1 to C.4 present the estimations of Equation (1) for previous flood exposure and
perception of future flood risk and for each city separately. Figures 3 and 4 illustrate the modeled impact
of flood exposure and perceived flood risk, respectively, on annual rent, based on estimations (7) and (8)
in Tables C.1 to C.4. The diamond represents the average marginal willingness to pay to avoid flood risk
as a percentage of the annual rents in each city,8 while the tails correspond to the confidence intervals.
ˆ 1 < 0 in Equation (1) holds true in Accra, Antananarivo, and Addis Ababa for
The hypothesis of α
both definitions of f lood. In Antananarivo and Addis Ababa it is significant when measured using
previous exposure to flooding and in Accra, the relationship is significant when flood risk is measured
using the perception of future flood risk.. Generally, the estimates for the marginal willingness to pay
to avoid flood risk is 14 percent (Accra), 14 percent (Antananarivo), and 56 percent (Addis Ababa) of
annual rent values.
The findings for Dar es Salaam are less intuitive. A higher perceived risk of future flooding is associ-
ated with a 28 percent higher rent and previous exposure to flooding is associated with 9 percent higher
rent (although the latter result is not statistically significant). Using the same dataset, Erman et al. (2019)
also found that past exposure to flooding is associated with higher reported rents, but they also found
that affected house owners consistently value their dwelling at a lower price than non-affected house-
holds (32 to 36 percent lower). The non-intuitive relationship between rent and flood risk found in this
study is most likely caused by a combination of 1) lack of transparency of what areas are flood-prone, 2)
high transaction costs for renters and 3) omitted variable bias. In terms of awareness about flood risk,
only 12 percent of households report having been aware of the flood risk when they moved into the
area. Among the 78 percent that provided a down payment to access the rental, 80 percent report that
it covered more than six months. As a result, many of the households may face barriers to leave the
dwelling even after they realize it is located in a flood-prone area. On omitted variable bias, it is possible
that flood risk is associated with other characteristics that can make an area more attractive, such as aes-
thetics, air quality, community engagement, proximity to green spaces, universities, or other amenities
that can influence rents and that the model is not controlling for. These characteristics may influence
rents more than exposure to flooding since a household may decide to make the trade-off of accessing
more amenities at the “expense” of facing higher flood risk, knowing that renting is temporary. Other
studies have found a positive relationship between flood risk and housing values. For instance, in Bin
and Kruse (2006) flood risk reduces property values in inland areas but increases values in coastal areas.
The results for Accra and Antananarivo are sensitive to the exclusion/inclusion of tenure status
(owner, renter, free housing, other) and tenure agreement (non-agreement, written and non-written
agreements). The explained rent variance (R2 ) increases by 47 percentage points and 23 percentage
points for Accra and Antananarivo respectively when including tenure variables.9 In the case of Accra,
Erman et al. (2020) show that exposure to floods is associated with 27 percent lower rents. The difference
8
The marginal effects of the dummy variables are calculated as (expβi -1), where βi is the estimated coefficient of the variable.
9
For Accra an increase of 27 percentage points is attributed to tenure type, and 20 percentage points to tenure agreement.
For Antananarivo, the total increase is attributed to the inclusion of tenure agreement.
13
between the results in this study and those in Erman et al. (2020) are that their specifications focus only
on renters and they did not account for tenure agreements. When not including tenure arrangements
(see columns (1) to (6) in Appendix Tables C.1 to C.4), estimates are more consistent to Erman et al.
(2020) finding that exposure to floods is associated with 21 percent lower rents and higher perceived
flood risk is associated with 27 percent lower rents (columns (1) and (2) in Appendix Table C.1). The
result emphasizes the close relationship between flood risk and tenure in these cities. The estimates for
Antananarivo are consistent with a previous study which used a similar approach with the same data
(WB, 2017), despite some specification differences.10
F IGURE 3: Marginal effect of flood exposure on annual rent
Notes: The diamonds denote the marginal effects of exposure to flood as percentages of annual rents (Appendix
Tables C.1 to C.4, column (7)). The thick line is 90% Confidence Intervale and the thin line is 95% Confidence
Intervale.
10
In WB (2017), the dependent variable is rent and is defined only for renters. Flood exposure is defined at the enumeration
area (EA) level. The definition of flood risk exposure combines information on EA identified as being “at risk of floods” if more
than 2.5% if its population were affected by the Chedza floods, and household in the EA that declared been affected by Chedza.
14
F IGURE 4: Marginal effect of flood perception on annual rent
Notes: The diamonds denote the marginal effects of higher perceived risk of future flooding as percentages of
annual rents (Appendix Tables C.1 to C.4, column (8)). The thick line is 90% Confidence Intervale and the thin line
is 95% Confidence Intervale.
15
4.2 Access to jobs and rents
A job density proxy was measured based on spatial data of the economic activity distribution (Barzin
et al., 2022) and added to the model to assess the relationship between rent and access to jobs. Better
access to jobs tends to be associated with higher rents and housing costs so it is therefore assumed that
ˆ 2 > 0 in Equation (1). Figure 5 presents the value of the marginal effects of job density on rents in
α
percentage, based on estimations including the variable exposure to actual floods (Appendix Tables C.1
to C.4, column (7)).
In Accra and Addis Ababa, areas with higher job density are associated with lower rents. These
results are contrary to our hypothesis. In Accra, while some expensive dwellings located in the north-
east of the Accra district, have relatively high job density, many households living in more expensive
dwellings in the south-west have relatively low access to jobs. These households in the south-west
could be driving the results (Appendix Figure D.1).
In the case of Addis Ababa, the highest job density is around the Ketema district, a populous area
with low rents and improvised settlements, that means rents are lower in the economic center, and
higher in the periphery. Two of the richest locations in Addis Ababa are close to the airport, and close to
the African Union, respectively two and one hours travel time from the economic center (see Appendix
Figure D.4).
In Dar es Salaam, the relationship between job density and rents is also negative but not statistically
significant.
In Antananarivo, areas with high job density have higher rents, on average. WB (2017) also found
that rents close to the city center were higher. The result is intuitive also since Antananarivo is a relatively
monocentric city, where jobs are concentrated in the center (see Figure D.2)
Since the variable we use to measure job density is non dimensional, it is difficult to interpret a
marginal increase, but the results show that the relative importance of job density on rents is highest in
Accra compared to the other three cities.
To check for a relationship between flood risk and access to jobs, we interact the proxy for job density
with the variables measuring flood risk for each city (Appendix Tables C.1 to C.4, columns (5) and (6)).
The negative effect of job density on rents is higher (in absolute value) when flood risk is high in Accra
and Addis Ababa, and the positive effect of job density on rents becomes negative when flood risk is
high in Antananarivo.11 The relationship is not found in Dar es Salaam. This finding seems to suggest
that access to jobs is an important factor driving people to settle in flood-prone areas since residents
in all but one city face a significant discount when renting dwellings in flood-prone areas that are also
close to job opportunities. Kocornik-Mina et al. (2020) analyze large urban floods in 40 countries, mostly
developing countries, and find that low elevation areas experience flooding more frequently and that
they also concentrate a higher density of economic activities.
However, the findings also show that job density is linked to lower rents in three of the four cities
studied (although in Dar es Salaam the result is not significant), indicating that the most economically
active areas are not necessarily the most attractive for households. While the job density proxy is based
on access to amenities, services, and infrastructure, it does not account for the quality of the same, which
11
A Wald test for the joint significance of the interaction term was conducted; the only non- statistically significant interaction
term was for Dar es Salaam, for the interaction between job density and actual flood exposure (Appendix Table C.3, column(5)).
16
F IGURE 5: Marginal effect of job density on annual rent
Notes: The diamonds denote the marginal effects of job density as percentages of annual rents (Appendix Tables
C.1 to C.4, column (7)). The thick line is 90% CI and the thin line is 95% CI.
could influence the result. There are also factors that could be associated with job density that may make
a place less attractive that we are not controlling for, such as crime rate, air and water pollution, and
access to recreational opportunities. So, while these factors could be driving rents down in economically
active areas, our analysis also shows that high flood risk is driving it down even more.
4.3 Other factors influencing rents
The results for housing conditions (Appendix Tables C.1 to C.4 and Figure 4) show that as expected,
in all four cities the higher the number of rooms (a proxy for housing size), the higher the rents, and
that this effect is decreasing with the number of rooms. Unimproved houses (such as like improvised
and/or shared/compound houses, tents, shacks) are associated with lower rents. Other factors such as
improved wall and floor materials, electric lighting, and toilet facilities are associated with higher rents.
In the case of access to public services, rents are, on average, higher for dwellings with access to solid
waste collection and improved water sources. There are some heterogeneities concerning the relevance
of each determinant across cities but with the exception of improved roofing materials, the results are
in line with expectations. These findings suggest that rent value, as measured in this study, is a good
predictor of housing quality, and that the hedonic approach can be considered appropriate to analyze
17
these SSA cities.
TABLE 4: Results on housing conditions
Accra Antananarivo Dar es Salaam Addis Ababa
Number of rooms (+) (+) (+) (+)
Number of rooms squared (-) (-) (-) (-)
Unimproved house (-) N.S. (-) (-)
Unimproved roof material (+) N.S. N.S. N.S.
Unimproved wall material (-) (-) N.S. N.A.
Unimproved floor material N.S. N.S. (-) (-)
Unimproved drinking water source N.S. (-) (-) N.S.
Unimproved toilet facility N.S. N.S. (-) (-)
Not collected solid waste N.S. (-) (-) N.S.
Electric lighting (+) (+) (+) N.S.
Notes: The direction of the correlation between rent and housing characteristics (Appendix Tables C.1
to C.4)is denoted by positive (+) and negative (-) signs for statistically significant results (at the 10%
level). Rela- tionships in the expected direction are in blue and in red otherwise. NS is not statistically
significant at the 10% level, NA is not applicable i.e. the variable is not included in the estimations.
4.4 Discussion of the main challenges
There are some major challenges related to assessing the relationship between flood risk and rents in
the context of African cities. Some of the specifications in Appendix Tables C.1 to C.4 show a non-
significant relationship but this does not mean that people are not willing to pay to avoid exposure to
floods. Rather it highlights some of the remaining issues and knowledge gaps which make it difficult to
capture the influence of disaster risk on housing markets in African cities.
First, tenure matters because it is linked to both housing costs and flood risk and violates the as-
sumption of perfect markets in a hedonic regression. Unfortunately for researchers, tenure is not a
binary concept but rather a range. In Accra 80 percent of owners have no tenancy/ownership agree-
ment, but not all of them fear eviction. Fifteen percent of households report fearing eviction in the next
12 months. Informal housing markets in Accra rely on social networks. For example, for 46 percent of
households, non-household relatives own the land that the dwelling is housed on. In Dar es Salaam,
74 percent of renters provided down payment. In 80 percent of cases the down payment covers more
than six months. Being locked into a rental agreement with huge upfront payments increases transaction
costs significantly. Tenure insecurity itself also drives vulnerability since it can discourage investment in
housing quality (Nakamura, 2017), which could increase resilience to floods.
Another problem is the lack of transparency related to flood risk. As reported, in Dar es Salaam
for example, only 12 percent of households directly affected by floods reported having been aware of
the flood risk when they moved into their homes. Flood maps and information on flood risk are not
readily available, and landlords can choose to withhold information on historic flood experiences to
new tenants. In contexts where deals are made in social networks and rental properties are not being
widely marketed, it becomes difficult for consumers to access information on what rents should be in
different neighborhoods. As such, dysfunctional and informal housing market results in factors that
18
would affect prices in a more functional market, such as flood risk, not being internalized.
5 Conclusions
This paper investigates the relationship between flood exposure and rents in Accra, Addis Ababa, An-
tananarivo, and Dar es Salaam, using a hedonic property price approach with four cross-sectional house-
hold surveys complemented with data on job density. Flood exposure is measured using self-reported
information from households on whether they have been affected by flooding. Flood risk is measured
with self-reported information on the perception of the risk of future flood exposure.
The results suggest that flood exposure/risk is associated with lower rents in three of the four cities
studied. The willingness to pay to avoid flood exposure is between 14 percent (in Accra and Antana-
narivo) and 56 percent (in Addis Ababa) of average rents. In contrast, in Dar es Salaam, flood expo-
sure/risk is associated with higher rents. The relationship between flood exposure and rents is made
more complex by the close correlation between tenure and flood exposure, the lack of public awareness
on flood risk, high transaction costs of housing and the importance of omitted variables, such as cultural
and historical importance, presence of community networks, access to green spaces and other amenities
which potentially could encourage people to move to areas with high flood risk.
The paper uses a novel way to measure access to jobs, based on data on the spatial distribution of
amenities and economic activities. In Accra and Addis Ababa, rents are negatively correlated with job
density, which is an unexpected finding for an efficient housing market. In Antananarivo, a more mono-
centric city compared with the other three cities, areas with high job density are more expensive, on
average. No relationship between job density and rent is found for Dar es Salaam. When interacting
job density with flood exposure/risk for each city, the negative effect of job density on rents is higher
when flood exposure/risk is high in Accra and Addis Ababa, and the positive effect of job density on
rents becomes negative when flood exposure/risk is high in Antananarivo. The result indicates that job
access is a driving factor for households settling in flood prone areas. In addition to flood exposure/risk,
it is found that housing conditions and public services are important determinants of rents. It is inher-
ently difficult to measure flood risk using self-reported data. For example, if a binary variable is used
to distinguish affected from non-affected populations, no differentiation is made between households
that have experienced a one-off flood event and those that experience frequent. However, self-reported
data on flood exposure have two advantages compared to frequently used flood risk maps: 1) house-
hold survey data are more often available than flood risk maps particularly in the case of data scarce
environments, and 2) information on actually affected households and their expectation about the risk
of future flooding are more likely to influence the value households assign to housing than data that is
rarely known to the public.
The different results for different cities indicate that the ability of people to internalize flood risks
through rents is context specific. This study contributes to our understanding of this issue for four cities
in Africa. The findings for the relationship between flood risk and rents are economically significant
despite the limits related to measuring flood risk, property pricing, and accessibility to jobs. They pro-
vide some important insights on how households assign value to amenities including flood protection,
in low- income, highly informal contexts. They can help policy makers estimate the potential value that
19
can be created by investment in flood protection measures even in the poorest areas. The results are
relevant also for local governments since property taxation is usually their main source of fiscal revenue.
This study sheds light on the poorly understood relationship between flood risk and housing costs
in a developing country context. It highlights drivers of housing pricing in cities that have highly infor-
mal and inefficient housing markets. As investments in urban resilience and climate change adaptation
increase, the need for more research on the value of disaster risk reduction also increases. Measuring the
willingness to pay of local residents to avoid flood risk is a novel and useful way to achieve this. Along-
side investments in risk reduction, this study also shows that it is important to improve awareness of
flood risk in cities and strengthen zoning and land use enforcement to avoid having households settle
in high-risk areas. Meanwhile, it is also important to assure the availability of affordable housing with
good job accessibility.
20
References
A HLFELDT, G. M. (2013): “If we build it, will they pay? Predicting property price effects of transport
innovations,” Environment and Planning A, 45, 1977–1994.
´ LVAREZ , J. A. C. AND B. P. R ESOSUDARMO (2019): “The cost of floods in developing countries′ megac-
A
ities: a hedonic price analysis of the Jakarta housing market, Indonesia,” Environmental Economics and
Policy Studies, 21, 555–577.
B ARTLETT, S. (2008): “Climate change and urban children: impacts and implications for adaptation in
low-and middle-income countries,” Environment and Urbanization, 20, 501–519.
B ARZIN , S., P. AVNER , J. R ENTSCHLER , AND N. O′ C LERYA (2022): “Where Are All the Jobs? A Machine
Learning Approach for High Resolution Urban Employment Prediction in Developing Countries,”
World Bank, Washington, DC, Policy Research working paper, no. WPS 9979.
´ , A., D. M ADDISON , AND R. J. E LLIOTT (2018): “Is flood risk capitalised into property values?”
B ELTR AN
Ecological Economics, 146, 668–685.
B IN , O. AND J. B. K RUSE (2006): “Real estate market response to coastal flood hazards,” Natural Hazards
Review, 7, 137–144.
B IN , O. AND S. P OLASKY (2004): “Effects of flood hazards on property values: evidence before and after
Hurricane Floyd,” Land Economics, 80, 490–500.
B ISHOP, K. C., N. V. K UMINOFF , H. S. B ANZHAF, K. J. B OYLE , K. VON G RAVENITZ , J. C. P OPE , V. K.
S MITH , AND C. D. T IMMINS (2020): “Best practices for using hedonic property value models to mea-
sure willingness to pay for environmental quality,” Review of Environmental Economics and Policy, 14,
260–281.
B RUECKNER , J. K. (2011): “Lectures on urban economics,” MIT press.
C ASTELLS -Q UINTANA , D. AND H. W ENBAN -S MITH (2020): “Population Dynamics, Urbanisation with-
out Growth, and the Rise of Megacities,” The Journal of Development Studies, 56, 1663–1682.
E RMAN , A., E. M OTTE , R. G OYAL , A. A SARE , S. TAKAMATSU , X. C HEN , S. M ALGIOGLIO , A. S KINNER ,
N. Y OSHIDA , AND S. H ALLEGATTE (2020): “The Road to Recovery the Role of Poverty in the Exposure,
Vulnerability and Resilience to Floods in Accra,” EconDisCliCha, 171–193.
E RMAN , A., M. TARIVERDI , M. O BOLENSKY, X. C HEN , R. C. V INCENT, S. M ALGIOGLIO ,
J. R ENTSCHLER , S. H ALLEGATTE , AND N. Y OSHIDA (2019): “Wading out the storm: The role of
poverty in exposure, vulnerability and resilience to floods in Dar Es Salaam,” The World Bank, Wash-
ington, DC, Policy Research working paper, no. WPS 8976.
FAY, M. AND C. O PAL (1999): “Urbanization without growth: a not-so-uncommon phenomenon,” The
World Bank, Washington, DC, Policy Research working paper, no. WPS 2412.
21
H ALLEGATTE , S., A. V OGT-S CHILB , J. R OZENBERG , M. B ANGALORE , AND C. B EAUDET (2020): “From
poverty to disaster and back: A review of the literature,” Economics of Disasters and Climate Change, 4,
223–247.
IPCC (2014): “Climate Change 2014: Impacts, Adaptation, and Vulnerability. Part A: Global and Sectoral
Aspects. Contribution of Working Group II to the Fifth Assessment Report of the Intergovernmental
Panel on Climate Change,” Tech. rep., IPCC, Cambridge University Press, Cambridge, United King-
dom and New York, NY, USA.
K OCORNIK -M INA , A., T. K. M C D ERMOTT, G. M ICHAELS , AND F. R AUCH (2020): “Flooded cities,”
American Economic Journal: Applied Economics, 12, 35–66.
N AKAMURA , S. (2017): “Tenure security premium in informal housing markets: A spatial hedonic anal-
ysis,” World Development, 89, 184–198.
P ERALTA -Q UIROS , T., T. K ERZHNER , AND P. AVNER (2019): “Exploring Accessibility to Employment
Opportunities in African Cities: A First Benchmark,” The World Bank, Washington, DC, Policy Research
working paper, no. WPS 8971.
R ABASSA , M. J. AND J. I. Z OLOA (2016): “Flooding risks and housing markets: a spatial hedonic analysis
for La Plata City,” Environment and development economics, 21, 464–489.
R OSEN , S. (1974): “Hedonic prices and implicit markets: product differentiation in pure competition,”
Journal of political economy, 82, 34–55.
UN (2019): “World Urbanization Prospects: The 2018 Revision (ST/ESA/SER.A/420),” United Nations,
Department of Economic and Social Affairs, Population Division.
WB (2017): “Greater Antananarivo: Urban poverty and resiliance study,” The World Bank, unpublished
manuscript.
——— (2021): “Global Economic Prospects, January 2021,” The World Bank.
22
Appendix
A Flood exposure and perception
TABLE A.1: Definition of being affected by flood
City Question/s Definition
Flood exposure
Accra Were you, your household directly Directly affected: Water in the house, water and electricity services affected,
exposed to 2015 flood? loss of valuable assets, loss of income, illness, schooling of children affected,
changes in food intake
Antananarivo Was your household affected by the Answer = Yes
flooding/landslides in 2014/2015?
Dar es Salaam Has water ever entered your house as Directly affected: Water in the house, loss of valuable assets
a consequence of heavy rainfall/flood
or have your house been damaged by
flood/heavy rainfalls?
Addis Ababa Have you been directly exposed to flood; If shock = flood: Water in the house, water and electricity services affected,
landslide; or fire? loss of valuable assets, loss of income, illness, schooling of children affected,
changes in food intake
Flood perception
Accra In your opinion, what is the likelihood Answer = Very likely or Likely
that the area where you live will be
flooded anytime between now and in
the next couple of years?
Antananarivo How likely do you think it is that your Answer = Extremely likely or Very likely
neighborhood will experience an episode
of extreme flooding in the next 5 years
Dar es Salaam In your opinion, what is the likelihood Answer = Very likely or Likely
that your household will be affected by
floods/heavy rains in the next 2 years?
Addis Ababa In your opinion, what is the likelihood Answer = Very likely or Likely
that the area where you live will be
flooded anytime between now and in
the next couple of years?
23
B Rent definition
F IGURE B.1: Distribution of rent, rent self-estimated, and combined self estimated rent
C Rent estimations
24
TABLE C.1: Estimation of rent, ACCRA
Simple Job density Interaction Tenure
(1) (2) (3) (4) (5) (6) (7) (8)
Directly affected by flood -0.215* -0.198* -0.042
(0.119) (0.112) (0.061)
Flood risk perception -0.274** -0.253** -0.149**
(0.111) (0.106) (0.073)
Job density -0.814*** -0.789*** -0.564 -0.406 -0.511*** -0.479***
(0.274) (0.276) (0.356) (0.300) (0.137) (0.138)
Number of rooms 0.420*** 0.396*** 0.391*** 0.369*** 0.384*** 0.345*** 0.409*** 0.397***
(0.121) (0.118) (0.114) (0.111) (0.112) (0.110) (0.069) (0.072)
Number of rooms squared -0.052*** -0.048*** -0.048*** -0.045*** -0.048*** -0.042*** -0.038*** -0.036***
(0.013) (0.013) (0.013) (0.012) (0.013) (0.012) (0.007) (0.007)
Unimproved house -0.390** -0.378** -0.310* -0.302* -0.311** -0.323** -0.463*** -0.461***
25
(0.184) (0.175) (0.157) (0.153) (0.153) (0.153) (0.106) (0.109)
Unimproved roof material 0.443*** 0.438*** 0.487*** 0.481*** 0.485*** 0.471*** 0.140*** 0.140***
(0.099) (0.099) (0.096) (0.097) (0.095) (0.098) (0.046) (0.045)
Unimproved wall material -0.357*** -0.356*** -0.337*** -0.337*** -0.325*** -0.329*** -0.223*** -0.215***
(0.118) (0.118) (0.118) (0.118) (0.118) (0.117) (0.047) (0.051)
Unimproved floor material 0.053 -0.018 -0.044 -0.107 -0.089 -0.155 -0.136 -0.153
(0.279) (0.297) (0.234) (0.253) (0.220) (0.246) (0.095) (0.113)
Unimproved drinking water source -0.110 -0.114 -0.076 -0.081 -0.082 -0.092 -0.051 -0.043
(0.097) (0.093) (0.097) (0.093) (0.097) (0.093) (0.048) (0.049)
Unimproved toilet facility -0.027 -0.039 -0.078 -0.088 -0.077 -0.077 -0.085 -0.071
(0.110) (0.097) (0.111) (0.100) (0.109) (0.095) (0.075) (0.075)
Not collected solid waste 0.121 0.097 0.097 0.076 0.103 0.094 -0.017 -0.022
(0.110) (0.104) (0.110) (0.108) (0.111) (0.106) (0.058) (0.059)
Continued on next page
Table C.1 – Continued from previous page
(1) (2) (3) (4) (5) (6) (7) (8)
Electric lighting 0.656*** 0.645*** 0.581*** 0.574*** 0.584*** 0.588*** 0.210*** 0.217***
(0.178) (0.192) (0.172) (0.185) (0.174) (0.190) (0.067) (0.073)
Dwelling tenure agreement
No Agreement ref. ref.
Written Agreement 0.334*** 0.337***
(0.072) (0.075)
Non-written Agreement 0.028 0.045
(0.080) (0.075)
Dwelling tenure status
Owner ref. ref.
Rented 1.308*** 1.290***
(0.099) (0.098)
26
Free 0.101* 0.092*
(0.059) (0.055)
Other -0.056 -0.075
(0.178) (0.175)
Affected by flood 0.625
(0.599)
Affected by flood × Job density -0.552
(0.413)
Flood event is likely 1.273*
(0.648)
Flood event is likely × Job density -1.020**
(0.448)
Observations 817 817 817 817 817 817 817 817
R2 0.17 0.18 0.19 0.20 0.19 0.20 0.67 0.67
Note: The dependent variable is the log of rent (combined rent paid and self-estimated rent), measured in 2017 PPP US$by year.
Robust standard errors are clustered at the enumeration area level. *** p<0.01, ** p<0.05, and * p<0.1.
27
TABLE C.2: Estimation of rent, ANTANANARIVO
Simple Job density Interaction Tenure
(1) (2) (3) (4) (5) (6) (7) (8)
Directly affected by flood -0.137 -0.140 -0.152*
(0.107) (0.107) (0.088)
Flood risk perception -0.043 -0.044 -0.059
(0.064) (0.064) (0.058)
Job density 0.164*** 0.164*** 0.183*** 0.100 0.172*** 0.172***
(0.057) (0.057) (0.061) (0.067) (0.056) (0.056)
Number of rooms 0.639*** 0.639*** 0.640*** 0.640*** 0.636*** 0.640*** 0.497*** 0.497***
(0.054) (0.055) (0.052) (0.052) (0.052) (0.051) (0.055) (0.054)
Number of rooms squared -0.039*** -0.039*** -0.038*** -0.038*** -0.038*** -0.038*** -0.032*** -0.032***
(0.007) (0.007) (0.007) (0.006) (0.007) (0.006) (0.005) (0.005)
Unimproved house -0.103 -0.100 -0.117 -0.114 -0.120 -0.105 -0.069 -0.066
28
(0.078) (0.078) (0.077) (0.077) (0.076) (0.075) (0.072) (0.073)
Unimproved roof material 0.000 -0.001 0.002 0.001 -0.007 0.009 0.094 0.095
(0.088) (0.087) (0.088) (0.087) (0.089) (0.087) (0.088) (0.086)
Unimproved wall material -0.519*** -0.520*** -0.536*** -0.537*** -0.548*** -0.520*** -0.398*** -0.401***
(0.152) (0.148) (0.146) (0.142) (0.145) (0.133) (0.138) (0.134)
Unimproved floor material 0.012 0.008 -0.006 -0.009 0.002 -0.015 -0.033 -0.039
(0.079) (0.079) (0.081) (0.081) (0.079) (0.080) (0.064) (0.064)
Unimproved drinking water source -0.625*** -0.629*** -0.556*** -0.561*** -0.558*** -0.570*** -0.582*** -0.588***
(0.084) (0.085) (0.085) (0.085) (0.086) (0.084) (0.072) (0.073)
Unimproved toilet facility -0.057 -0.056 -0.078 -0.077 -0.085 -0.061 -0.068 -0.066
(0.090) (0.092) (0.094) (0.095) (0.094) (0.093) (0.083) (0.085)
Not collected solid waste -0.154** -0.166*** -0.139** -0.152** -0.132** -0.148** -0.132** -0.147***
(0.061) (0.059) (0.061) (0.059) (0.063) (0.059) (0.055) (0.055)
Continued on next page
Table C.2 – Continued from previous page
(1) (2) (3) (4) (5) (6) (7) (8)
Access to electricity 0.477*** 0.481*** 0.428*** 0.433*** 0.424*** 0.446*** 0.369*** 0.374***
(0.071) (0.071) (0.075) (0.076) (0.076) (0.078) (0.065) (0.066)
Dwelling tenure agreement
No Agreement ref. ref.
Written Agreement 0.142 0.116
(0.091) (0.087)
Non-written Agreement 0.021 0.010
(0.068) (0.067)
Not Applicable 0.000 0.000
(.) (.)
Dwelling tenure status
Owner ref. ref.
29
Rented -1.074*** -1.061***
(0.124) (0.124)
Free 0.000 0.000
(.) (.)
Other 0.685*** 0.727***
(0.110) (0.112)
Urban district 0.234*** 0.228*** 0.120 0.115 0.115 0.122 0.085 0.081
(0.084) (0.080) (0.090) (0.086) (0.090) (0.086) (0.090) (0.091)
Affected by flood 0.513
(0.423)
Affected by flood × Job density -0.304
(0.209)
Flood event is likely -0.316
(0.193)
Continued on next page
Table C.2 – Continued from previous page
(1) (2) (3) (4) (5) (6) (7) (8)
Flood event is likely × Job density 0.132
(0.086)
Observations 916 916 916 916 916 916 916 916
R2 0.47 0.47 0.48 0.48 0.48 0.48 0.60 0.60
Note: The dependent variable is the log of rent (combined rent paid and self-estimated rent), measured in 2017 PPP US$by year.
Robust standard errors are clustered at the enumeration area level. The variable Dwelling tenure status it is not added because
is collinear with tenure agreement, for the observations of Antananarivo. *** p<0.01, ** p<0.05, and * p<0.1.
30
TABLE C.3: Estimation of rent, DAR ES SALAAM
Simple Job density Interaction Tenure
(1) (2) (3) (4) (5) (6) (7) (8)
Directly affected by flood 0.077 0.084 0.086
(0.080) (0.079) (0.078)
Flood risk perception 0.242*** 0.251*** 0.244***
(0.091) (0.091) (0.092)
Job density -0.070 -0.083 -0.120 -0.114 -0.064 -0.076
(0.116) (0.115) (0.120) (0.110) (0.117) (0.116)
Unimproved house -0.111 -0.143* -0.111 -0.144* -0.119 -0.137 -0.116 -0.149*
(0.081) (0.085) (0.080) (0.085) (0.080) (0.085) (0.081) (0.085)
Unimproved roof material -0.088 -0.086 -0.096 -0.096 -0.104 -0.119 -0.084 -0.087
(0.163) (0.154) (0.163) (0.153) (0.163) (0.152) (0.163) (0.153)
Unimproved wall material 0.124 0.111 0.126 0.114 0.124 0.113 0.132 0.120
31
(0.085) (0.090) (0.087) (0.092) (0.087) (0.091) (0.086) (0.092)
Unimproved floor material -0.217* -0.242* -0.214* -0.239* -0.219* -0.208* -0.162 -0.196
(0.117) (0.132) (0.119) (0.136) (0.120) (0.126) (0.124) (0.142)
Unimproved drinking water source -0.233** -0.232** -0.227** -0.225** -0.234** -0.224** -0.224** -0.223**
(0.101) (0.098) (0.098) (0.095) (0.098) (0.095) (0.099) (0.096)
Unimproved toilet facility -0.685*** -0.700*** -0.681*** -0.695*** -0.669*** -0.702*** -0.678*** -0.693***
(0.114) (0.113) (0.112) (0.111) (0.112) (0.110) (0.113) (0.112)
Not collected solid waste -0.253** -0.224** -0.299** -0.277** -0.309** -0.299** -0.278** -0.259**
(0.098) (0.093) (0.123) (0.119) (0.124) (0.116) (0.128) (0.125)
Electric lighting 0.496*** 0.485*** 0.495*** 0.482*** 0.504*** 0.470*** 0.479*** 0.469***
(0.084) (0.080) (0.086) (0.081) (0.088) (0.082) (0.086) (0.082)
Dwelling tenure agreement
No Agreement ref. ref.
Continued on next page
Table C.3 – Continued from previous page
(1) (2) (3) (4) (5) (6) (7) (8)
Written Agreement -0.303*** -0.358***
(0.104) (0.107)
Non-written Agreement -0.440*** -0.470***
(0.143) (0.142)
Affected by flood -0.562
(0.387)
Affected by flood × Job density 0.315
(0.200)
Flood event is likely -0.383
(0.662)
Flood event is likely × Job density 0.299
(0.316)
32
Observations 466 466 466 466 466 466 466 466
R2 0.32 0.34 0.33 0.34 0.33 0.34 0.33 0.34
Note: The dependent variable is the log of rent (combined rent paid and self-estimated rent), measured in 2017 PPP US$by year.
The variables Number of rooms and Number of rooms squared are added only in estimations (7) and (8) because of missing values
reducing the total number of observations. The variable Dwelling tenure status is not added because all the observations for rent
are only for renters. Robust standard errors are clustered at the enumeration area level. *** p<0.01, ** p<0.05, and * p<0.1.
TABLE C.4: Estimation of rent, ADDIS ABABA
Simple Job density Interaction Tenure
(1) (2) (3) (4) (5) (6) (7) (8)
Directly affected by flood -1.054** -0.863** -0.822**
(0.391) (0.361) (0.335)
Flood risk perception -0.249 -0.269 -0.197
(0.257) (0.227) (0.226)
Job density -0.744*** -0.773*** -0.687** -0.706** -0.464** -0.492**
(0.250) (0.255) (0.284) (0.304) (0.192) (0.196)
Number of rooms 1.044*** 1.046*** 0.919*** 0.916*** 0.919*** 0.923*** 0.813*** 0.812***
(0.205) (0.214) (0.140) (0.146) (0.138) (0.145) (0.142) (0.140)
Number of rooms squared -0.081*** -0.082*** -0.070*** -0.070*** -0.070*** -0.071*** -0.060*** -0.060***
(0.018) (0.019) (0.013) (0.013) (0.013) (0.013) (0.013) (0.013)
Unimproved house -1.102*** -1.195*** -0.712** -0.776*** -0.696** -0.771*** -0.600** -0.668**
33
(0.326) (0.350) (0.263) (0.273) (0.260) (0.267) (0.270) (0.280)
Unimproved roof material 0.002 0.092 0.030 0.120 0.081 0.135 0.066 0.141
(0.236) (0.253) (0.231) (0.260) (0.249) (0.264) (0.200) (0.229)
Unimproved floor material -0.548* -0.608** -0.619** -0.664** -0.606** -0.661** -0.529* -0.573*
(0.268) (0.283) (0.267) (0.277) (0.267) (0.276) (0.282) (0.294)
Unimproved drinking water source 0.938** 0.577 0.592 0.294 0.347 0.233 0.554 0.271
(0.381) (0.370) (0.409) (0.505) (0.493) (0.525) (0.336) (0.402)
Unimproved toilet facility -0.555*** -0.618*** -0.511** -0.558*** -0.527*** -0.564*** -0.409*** -0.463***
(0.161) (0.162) (0.186) (0.187) (0.177) (0.185) (0.121) (0.124)
Not collected solid waste -0.361 -0.367 -0.287 -0.300 -0.305 -0.318 -0.493 -0.482
(0.409) (0.437) (0.379) (0.389) (0.377) (0.385) (0.446) (0.456)
Electric lighting -1.153 -1.347 -0.497 -0.614 -0.567 -0.701 0.154 0.018
(0.896) (0.991) (1.002) (1.088) (0.978) (1.119) (1.039) (1.107)
Continued on next page
Table C.4 – Continued from previous page
(1) (2) (3) (4) (5) (6) (7) (8)
Dwelling tenure agreement
No Agreement ref. ref.
Written Agreement -0.789*** -0.720***
(0.206) (0.209)
Non-written Agreement 0.143 0.268
(0.227) (0.233)
Dwelling tenure status
Owner ref. ref.
Rented -1.049*** -1.049***
(0.339) (0.343)
Free -0.120 -0.131
(0.170) (0.175)
34
Other -0.598 -0.543
(0.552) (0.560)
Affected by flood -0.054
(0.579)
Affected by flood × Job density -0.397
(0.372)
Flood event is likely 0.060
(0.504)
Flood event is likely × Job density -0.195
(0.337)
Observations 751 751 751 751 751 751 736 736
R2 0.36 0.35 0.43 0.42 0.43 0.42 0.49 0.48
Note: The dependent variable is the log of rent (combined rent paid and self-estimated rent), measured in 2017 PPP US$by year.
Robust standard errors are clustered at the enumeration area level. *** p<0.01, ** p<0.05, and * p<0.1.
D Additional estimations and analysis
35
TABLE D.1: Descriptive statistics by city, total sample
Accra Antananarivo Dar es Salaam Addis Ababa
Obs % Obs % Obs % Obs %
Exposure to floods
Households directly affected by flood 1,006 44.6 2,272 8.1 1,335 22.3 810 7.9
Households that perceive flood risk as high 1,006 40.3 2,272 47.2 1,335 20.7 810 31.3
Dwelling tenure 1,006 2,272 1,335 795
Owner 204 20.3 988 43.5 783 56.7 435 54.7
Rented 442 44.0 775 34.1 483 36.2 333 41.9
Free 296 29.4 382 16.8 64 4.8 21 2.7
36
Other 63 6.3 128 5.6 5 0.4 5 0.7
Housing conditions
Access to improved toilet facility 1,006 15.6 2,272 5.2 1,335 31.3 810 28.7
Access to improved drinking water 1,006 45.3 2,272 18.7 1,335 42.7 810 98.2
Access to collected solid waste 1,006 74.0 2,272 59.2 1,335 83.9 810 66.6
Lighting with electricity 1,006 96.7 1,863 99.3 1,335 90.4 810 99.5
Obs Mean Obs Mean Obs Mean Obs Mean
Household size 1,006 3.4 2,272 4.2 1,335 4.1 810 4.3
Number of rooms 1,006 1.5 2,272 2.1 1,053 3.7 810 3.2
Job density 820 1.5 2,261 1.9 1,335 2.0 810 1.7
Notes: Figures are weighted.
TABLE D.2: Comparison of rent variables, ACCRA
Rent value Self estimated rent Combined self estimated rent
(1) (2) (3) (4) (5) (6)
Directly affected by flood -0.024 -0.033 -0.198*
(0.100) (0.048) (0.112)
Flood risk perception -0.150 -0.052 -0.253**
(0.139) (0.060) (0.106)
Job density -0.264* -0.257* -0.285** -0.278** -0.814*** -0.789***
(0.151) (0.153) (0.123) (0.123) (0.274) (0.276)
Number of rooms 0.641*** 0.600*** 0.485*** 0.480*** 0.391*** 0.369***
(0.123) (0.110) (0.070) (0.073) (0.114) (0.111)
Number of rooms squared -0.108*** -0.099*** -0.047*** -0.047*** -0.048*** -0.045***
(0.017) (0.015) (0.008) (0.008) (0.013) (0.012)
Unimproved house -0.461** -0.458** -0.313*** -0.312*** -0.310* -0.302*
37
(0.182) (0.177) (0.070) (0.071) (0.157) (0.153)
Unimproved roof material 0.208*** 0.206*** 0.142*** 0.141*** 0.487*** 0.481***
(0.077) (0.075) (0.045) (0.045) (0.096) (0.097)
Unimproved wall material -0.200** -0.203** -0.259*** -0.258*** -0.337*** -0.337***
(0.092) (0.097) (0.043) (0.045) (0.118) (0.118)
Unimproved floor material -0.037 -0.018 -0.102 -0.113 -0.044 -0.107
(0.145) (0.165) (0.105) (0.108) (0.234) (0.253)
Unimproved drinking water source -0.051 -0.036 -0.071 -0.071 -0.076 -0.081
(0.070) (0.079) (0.046) (0.046) (0.097) (0.093)
Unimproved toilet facility 0.175 0.184 -0.248*** -0.248*** -0.078 -0.088
(0.219) (0.210) (0.047) (0.046) (0.111) (0.100)
Not collected solid waste 0.034 0.035 0.045 0.041 0.097 0.076
(0.107) (0.103) (0.044) (0.045) (0.110) (0.108)
Continued on next page
Table D.2 – Continued from previous page
(1) (2) (3) (4) (5) (6)
Electric lighting -0.085 -0.134 0.262*** 0.261*** 0.581*** 0.574***
(0.119) (0.140) (0.064) (0.064) (0.172) (0.185)
Observations 375 375 815 815 817 817
R2 0.11 0.12 0.38 0.38 0.19 0.20
Notes: The dependent variables are in log, measured in 2017 PPP US$by year. Robust standard errors are clustered
at the enumeration area level. *** p<0.01, ** p<0.05, and * p<0.1.
38
TABLE D.3: Comparison of rent variables, ANTANANARIVO
Rent value Self estimated rent Combined self estimated rent
(1) (2) (3) (4) (5) (6)
Directly affected by flood -0.156 -0.058 -0.140
(0.100) (0.249) (0.107)
Flood risk perception -0.067 -0.092 -0.044
(0.062) (0.233) (0.064)
Job density 0.174*** 0.171*** 0.236 0.245* 0.164*** 0.164***
(0.057) (0.056) (0.148) (0.147) (0.057) (0.057)
Number of rooms 0.567*** 0.555*** 0.505*** 0.502*** 0.640*** 0.640***
(0.107) (0.106) (0.169) (0.166) (0.052) (0.052)
Number of rooms squared -0.048** -0.046** -0.030** -0.030** -0.038*** -0.038***
(0.021) (0.021) (0.012) (0.011) (0.007) (0.006)
Unimproved house -0.088 -0.082 -0.052 -0.044 -0.117 -0.114
39
(0.064) (0.065) (0.193) (0.203) (0.077) (0.077)
Unimproved roof material 0.110 0.116 0.075 0.078 0.002 0.001
(0.102) (0.100) (0.174) (0.171) (0.088) (0.087)
Unimproved wall material -0.423*** -0.419*** -0.223 -0.262 -0.536*** -0.537***
(0.149) (0.144) (0.418) (0.391) (0.146) (0.142)
Unimproved floor material -0.083 -0.087 0.160 0.143 -0.006 -0.009
(0.062) (0.063) (0.245) (0.239) (0.081) (0.081)
Unimproved drinking water source -0.565*** -0.568*** -0.712*** -0.710*** -0.556*** -0.561***
(0.079) (0.079) (0.199) (0.202) (0.085) (0.085)
Unimproved toilet facility -0.051 -0.044 -0.213 -0.218 -0.078 -0.077
(0.094) (0.097) (0.239) (0.231) (0.094) (0.095)
Not collected solid waste -0.179*** -0.192*** 0.116 0.120 -0.139** -0.152**
(0.054) (0.052) (0.228) (0.243) (0.061) (0.059)
Continued on next page
Table D.3 – Continued from previous page
(1) (2) (3) (4) (5) (6)
Access to electricity 0.353*** 0.362*** 0.360 0.353 0.428*** 0.433***
(0.065) (0.066) (0.314) (0.317) (0.075) (0.076)
Urban district 0.050 0.047 0.206 0.233 0.120 0.115
(0.096) (0.092) (0.252) (0.299) (0.090) (0.086)
Observations 784 784 132 132 916 916
R2 0.49 0.49 0.38 0.38 0.48 0.48
Notes: The dependent variables are in log, measured in 2017 PPP US$by year. Robust standard errors are clustered
at the enumeration area level. *** p<0.01, ** p<0.05, and * p<0.1.
40
TABLE D.4: Comparison of rent variables, DAR ES SALAAM
Rent value Self estimated rent Combined self estimated rent
(1) (2) (3) (4) (5) (6)
Directly affected by flood 0.087 0.047 0.084
(0.079) (0.081) (0.079)
Flood risk perception 0.240*** 0.268*** 0.251***
(0.090) (0.095) (0.091)
Job density -0.074 -0.085 0.072 0.054 -0.070 -0.083
(0.116) (0.115) (0.111) (0.110) (0.116) (0.115)
Unimproved house -0.109 -0.140 -0.110 -0.147 -0.111 -0.144*
(0.081) (0.085) (0.091) (0.091) (0.080) (0.085)
Unimproved roof material -0.102 -0.102 -0.323* -0.320* -0.096 -0.096
(0.163) (0.153) (0.179) (0.174) (0.163) (0.153)
Unimproved wall material 0.129 0.119 0.160* 0.140 0.126 0.114
41
(0.087) (0.091) (0.090) (0.099) (0.087) (0.092)
Unimproved floor material -0.210* -0.235* -0.272** -0.296** -0.214* -0.239*
(0.120) (0.136) (0.131) (0.146) (0.119) (0.136)
Unimproved drinking water source -0.219** -0.217** -0.115 -0.112 -0.227** -0.225**
(0.096) (0.094) (0.108) (0.105) (0.098) (0.095)
Unimproved toilet facility -0.687*** -0.699*** -0.762*** -0.778*** -0.681*** -0.695***
(0.112) (0.111) (0.115) (0.113) (0.112) (0.111)
Not collected solid waste -0.297** -0.276** -0.209 -0.192 -0.299** -0.277**
(0.123) (0.119) (0.129) (0.124) (0.123) (0.119)
Electric lighting 0.493*** 0.481*** 0.403*** 0.391*** 0.495*** 0.482***
(0.086) (0.082) (0.126) (0.119) (0.086) (0.081)
Observations 465 465 466 466 466 466
R2 0.33 0.34 0.29 0.31 0.33 0.34
Notes: The dependent variables are in log, measured in 2017 PPP US$by year. Robust standard errors are clustered
at the enumeration area level. *** p<0.01, ** p<0.05, and * p<0.1.
42
TABLE D.5: Comparison of rent variables, ADDIS ABABA
Rent value Self estimated rent Combined self estimated rent
(1) (2) (3) (4) (5) (6)
Directly affected by flood -0.260 -0.147 -0.863**
(0.694) (0.239) (0.361)
Flood risk perception 0.017 -0.030 -0.273
(0.529) (0.109) (0.226)
Job density -1.476*** -1.501*** 0.192*** 0.189*** -0.744*** -0.778***
(0.136) (0.112) (0.058) (0.058) (0.250) (0.258)
Number of rooms 1.045** 1.056** 0.562*** 0.561*** 0.919*** 0.907***
(0.451) (0.453) (0.076) (0.076) (0.140) (0.145)
Number of rooms squared -0.117* -0.118* -0.039*** -0.039*** -0.070*** -0.069***
(0.059) (0.060) (0.008) (0.008) (0.013) (0.013)
Unimproved house 0.303 0.292 -0.281 -0.289 -0.712** -0.766**
43
(0.385) (0.370) (0.172) (0.170) (0.263) (0.273)
Unimproved roof material 0.015 -0.003 0.162 0.176 0.030 0.124
(0.708) (0.671) (0.174) (0.177) (0.231) (0.257)
Unimproved floor material -0.987 -1.030 -0.526*** -0.535*** -0.619** -0.664**
(0.637) (0.675) (0.138) (0.135) (0.267) (0.278)
Unimproved drinking water source -1.076 -1.294 0.275 0.223* 0.592 0.326
(1.130) (0.912) (0.168) (0.118) (0.409) (0.515)
Unimproved toilet facility -1.291 -1.314 -0.259** -0.267** -0.511** -0.558***
(0.874) (0.814) (0.105) (0.105) (0.186) (0.187)
Not collected solid waste -0.716 -0.681 -0.251 -0.253 -0.287 -0.305
(0.446) (0.430) (0.171) (0.169) (0.379) (0.387)
Electric lighting -0.055 -0.091 0.257 0.235 -0.497
(1.300) (1.260) (0.666) (0.698) (1.002)
Continued on next page
Table D.5 – Continued from previous page
(1) (2) (3) (4) (5) (6)
Observations 156 156 708 708 751 751
R2 0.42 0.41 0.47 0.47 0.43 0.42
Notes: The dependent variables are in log, measured in 2017 PPP US$by year. Robust standard errors are clustered
at the enumeration area level. *** p<0.01, ** p<0.05, and * p<0.1.
44
F IGURE D.1: Rent in surveyed households by exposure to floods, Accra
Source: Authors’ calculation based on Disaster-Poverty Household Survey and Barzin et al. (2022) data.
45
F IGURE D.2: Rent in surveyed households by exposure to floods, Antananarivo
Source: Authors’ calculation based on Disaster-Poverty Household Survey and Barzin et al. (2022) data.
46
F IGURE D.3: Rent in surveyed households by exposure to floods, Dar es Salaam
Source: Authors’ calculation based on Disaster-Poverty Household Survey and Barzin et al. (2022) data.
47
F IGURE D.4: Rent in surveyed households by exposure to floods, Addis Ababa
Source: Authors’ calculation based on Disaster-Poverty Household Survey and Barzin et al. (2022) data.
48