Poverty & Equity Global Practice Working Paper 156 THE ROAD TO RECOVERY THE ROLE OF POVERTY IN THE EXPOSURE, VULNERABILITY AND RESILIENCE TO FLOODS IN ACCRA Alvina Erman Elliot Motte Radhika Goyal Akosua Asare Shinya Takamatsu Xiaomeng Chen Silvia Malgioglio Alexander Skinner Nobuo Yoshida Stephane Hallegatte June 2018 Poverty & Equity Global Practice Working Paper 156 ABSTRACT In June 2015, about 53,000 people were affected by unusually severe floods in the Greater Accra Metropolitan Area, Ghana. The real impact of such a disaster is a product of exposure (“Who was affected?”), vulnerability (“How much did the affected households lose?”), and socioeconomic resilience (“What was their ability to cope and recover?”). This study explores these three dimensions to assess whether poor people were disproportionally affected by the 2015 floods. It reaches four main conclusions. (1) In the studied area, there is no difference in annual expenditures between the households who were affected and those who were not affected by the flood. (2) Poorer households lost less than their richer neighbors in absolute terms, but more when compared with their annual expenditure level, and poorer households are over-represented among the most severely affected households. (3) More than 30 percent of the affected households report not having recovered two years after the shock, and the ability of households to recover was driven by the magnitude of their losses, sources of income, and access to coping mechanisms, but not by their poverty, as measured by the annual expenditure level. (4) There is a measurable effect of the flood on behaviors, under-mining savings and investment in enterprises. The study concludes with two policy implications. First, flood management could be considered as a component of the poverty-reduction strategy in the city. Second, building resilience is not only about increasing income. It also requires providing the population with coping and recovery mechanisms such as financial instruments. A flood management program needs to be designed to target low-resilience households, such as those with little access to coping and recovery mechanisms, even those who are not living in poverty before the shock. This paper is a product of the Poverty and Equity Global Practice Group. It is part of a larger effort by the World Bank to provide open access to its research and contribute to development policy discussions around the world. The authors may be contacted at aerman@worldbank.org. The Poverty & Equity Global Practice 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. ‒ Poverty & Equity Global Practice Knowledge Management & Learning Team This paper is co-published with the World Bank Policy Research Working Papers. The Road to Recovery The Role of Poverty in the Exposure, Vulnerability and Resilience to Floods in Accra Alvina Erman, Elliot Motte, Radhika Goyal, Akosua Asare, Shinya Takamatsu, Xia- omeng Chen, Silvia Malgioglio, Alexander Skinner, Nobuo Yoshida, Stephane Hal- legatte 1. Introduction and summary On June 3, 2015, Accra was hit by a flood that claimed at least 152 lives and caused around US$100 mil- lion in asset losses. It was the most significant disaster to affect the city in recent times. An important fraction of the city was affected, and the impact on livelihoods and well-being was very large. Beyond the loss of life and direct impacts of the flood, a particular concern is the longer-term impact on the poorest and most vulnerable people in the city, who are likely to be less able to cope with and recover from a flood than the rest of the population. There is little information on the relationship between poverty and flood risk in Accra. The CityStrength diagnostics study conducted after the 2015 flood identified knowledge gaps, including the need for (i) a systematic study on the geographic location of poor people in hazard prone areas and (ii) a better under- standing of how floods affect poor households and the coping mechanisms they implement to deal with floods (World Bank, 2017). A previous study (Rain et al., 2011) assessed areas affected by floods in the Odaw River catchment and found that out of 172,000 people exposed to floods, approximately 20% lived in areas with the highest slum index, suggesting linkages between vulnerability to floods and poverty. This study aims at complementing this work, through a household survey focusing on the 2015 event. The Ghana Poverty-DRM study was designed to assess the relationship between poverty and disaster risk in areas identified as informal settlements in the Greater Accra Metropolitan Area (GAMA). The analysis builds on the review of previous survey exercises conducted in a few post-disaster locations and reviewed in Hallegatte et al. (2017). Following these authors, we use a framework separating the hazard (“What are the characteristics of the flood?”), the exposure of the population and assets (“Who is affected by the flood?”), the vulnerability of the population and assets (“How much did the affected people lose?”), and the socio-economic resilience of the population (“Was the affected population capable of coping with and recovering from the losses?”); see Figure 1. The study also investigates the effects of the flood – and of the risk of future floods – on households’ be- haviors, and especially on decisions regarding investment in housing and businesses. The effect of risk in general on individuals’ or firms’ savings and investment decisions has been documented elsewhere, but never in an urban flood context (Elbers et al., 2007; Hallegatte et al., 2016; ODI and GFDRR, 2015). 2 Figure 1: Risk assessment framework to estimate well-being losses; source: Hallegatte, et al., 2017 Our survey includes 1,006 households living in Accra’s informal settlements, chosen to cover a range of income level and flood impacts. The questionnaire includes questions related to their living conditions and household characteristics, an assessment of their annual expenditure level (based on the SWIFT methodology, see below), and detailed modules on the 2015 flood and its impacts. The range of expenditure levels in our survey matches the range in the full GAMA area in the last house- hold survey in 2012/2013, and the poverty rate in the surveyed households is below 2%. Compared with the rest of the city, they tend to lack access to infrastructure services. In particular, 70% rely on sachet water and 67% on public toilets. 74% report having their waste collected, a rate comparable to the rest of the city. Almost half of the interviewed households reported being directly affected by the 2015 floods, either though damages to their house, loss of assets, or other channels. The first three sections of this report describe the exposure, vulnerability, and resilience of the surveyed population, focusing on the relationship of these factors with poverty and wealth. Then, a fourth section discusses the impact of the floods – and, more generally, of risk perceptions – on savings and investment behaviors. The analysis leads to four main messages: Exposure: In the studied area, there is no measurable difference in annual expenditures between the households who were affected and non-affected by the 2015 floods. Various theories could explain why relatively rich households tolerate their exposure to flooding instead of moving to safer places, including the fact that the 2015 flood was exceptional in magnitude, mobility constraints linked to tenure arrange- ments, and potential benefits from living in the flood-prone areas. In particular, rents (or housing costs) are about 30% lower for affected households, creating a strong incentive for them to move into or stay in the flood-prone area. A large survey would be required to explore whether this result is valid in the whole city, beyond the informal settlements explored in this study. Vulnerability: Poorer households lost less than their richer neighbors in absolute terms, but more as a fraction of their total annual expenditures, and poorer households are over-represented among the most 3 severely affected households. Household losses were almost entirely composed of asset losses and hous- ing repairs, with indirect losses through health impacts or missed days of work contributing only margin- ally to total losses. On average, affected households lost 509 cedis due to the flood, representing about 4% of the value of the average annual household expenditures. But these average numbers belie large hetero- geneities in the affected population: only 54% of affected households lost more than 1% of total annual household expenditure. The proportions of affected households that lost more than 5% and 10% are re- spectively 21% and 10%. Poorer households were more vulnerable to the 2015 floods, in the sense that they lost a larger fraction of their annual expenditure than their richer neighbors. For instance, households from the poorest quartile were 60% more likely than the average household to lose more than 10% of their income. The large heterogeneity in losses and higher vulnerability of poorer people means that the average asset loss per household is a very poor indicator of the well-being impact on households. Socio-economic resilience: 31% of all affected households report not having recovered two years after the shock. The ability of households to recover from the 2015 flood was driven by the magnitude of their losses, their source of income, and their access to coping mechanisms, such as borrowing and remittances. In the case of Accra slums, the households’ ability to recover was not affected by their poverty, as meas- ured by the annual expenditure level and when controlling for other factors. In other terms, poorer house- holds struggle more to recover, but mostly because they have lower access to coping mechanisms and ex- perience larger (relative) losses, not because their income is lower. Households that lost more than 5% of their annual expenditures were 40% less likely to recover within two years, as compared to households that experienced lower relative losses. Few households seem to have relied on negative coping strategies such as reducing food intake below basic needs. Formal assistance played a very minor role. Behaviors: There is a measurable effect of the 2015 flood on economic behaviors, undermining savings and enterprise investment. Affected households are found to prioritize investments in their house at the expense of their business, which suggests that flood risks could represent an obstacle to economic activity and development, beyond the direct effects through asset and income losses. Flood management could thus generate more benefits than usually estimated, by promoting investments in addition to avoiding losses. The Triple Dividend report refers to this benefit as the “second dividend of disaster risk reduction” (ODI and GFDRR, 2015).1 However, these results remain preliminary and further research – including through dedicated surveys – would be needed to reach more conclusive results on this issue. This study focused on the 2015 floods, and it remains challenging to generalize the findings, or derive general conclusions for flood management and the provision of post-flood assistance. However, these four findings translate into two policy implications. First, relatively poor people suffer more than their neighbors from flooding, which suggests that flood management can be particularly beneficial for poorer households and thus could be considered as a com- ponent of the poverty-reduction strategy in the city. Further, the higher impact of the 2015 flood on poorer people seems to be mostly due to their higher vulnerability, linked to the nature and quality of their assets. In addition to usual flood management actions aiming at reducing flood frequency and magnitude 1 The first dividend is that disaster losses can be avoided, and the third dividend refers to co-benefits such as when a water retention area can also be used as a recreation park or a dike is combined with a road. 4 (e.g., with improved drainage), a poverty-focused flood management policy should therefore focus on the building stock (to increase the housing quality for poorer people) and financial inclusion (to help people save in a form that is not vulnerable to floods). Second, building resilience is not only about increasing income: it also requires providing the population with access to coping and recovery mechanisms, such as assistance and financial tools, and more stable income sources. It means that traditional poverty reduction instruments –with targeting based on poverty indicators – may not cover all the households who suffer from long-term consequences of floods. A resili- ence-building program needs (1) to target low-resilience households, such as those with little access to coping and recovery mechanisms, including those who are not living in poverty; and (2) to provide house- holds with coping mechanisms like savings opportunities and access to borrowing, while promoting the development of stable sources of income. 2. Contribution to the literature It is well documented that shocks represent a major cause of falling into poverty (Hulme and Shepherd, 2003; Quisumbing, 2007; Moser, 2008; Baulch, 2011). And shocks can have long lasting impacts, espe- cially on children, in the presence of negative coping strategies such as selling off assets, reduction in ca- loric consumption, or lower investment in education (Carter, et al., 2007; Skoufias, 2003). The risk of a ‘disaster-poverty trap’ – or at least long-term impacts – is exacerbated if the affected households have limited external support and lack access to coping strategies such as savings or insurance (Carter et al., 2007; Dang, et al., 2014; McCarthy, et al., 2017; Arouri, et al., 2015; Hallegatte et al., 2017). For these reasons, it is particularly important to better identify the population affected by disasters, understand how various people suffer from different types and magnitudes of losses, and assess the ability of different populations to cope with and recover from natural disasters. The scarce (but growing) literature on the relationship between poverty and the exposure and vulnerabil- ity to natural disasters suggests that these dynamics are highly context specific. This study on Accra and the 2015 floods contributes to the body of evidence on this issue, by focusing on a major urban flood in an African city. 2 It is interesting to ask whether poor people are overrepresented in the population affected by the 2015 flood because there are few surveys considering this question, and they suggest that poor people are often but not always overrepresented among the people affected by disasters (Hallegatte, et al., 2017). In Tu- valu, for example, Taupo et al. (2017) found that poorer households were more likely to reside in areas highly exposed to disasters. And in Vietnam, Narloch and Bangalore (2016) find that poor people in ur- ban areas were significantly more likely to live in areas with a higher risk of floods. But Noy and Patel (2014) found that non-poor households were more exposed to the 2011 flood in Thailand. In Kenya, Opundo (2013) did not find a relationship between income and exposure for floods in the Bunyala Dis- trict. This diversity of findings is consistent with the findings of Winsemius et al. (2017), who conclude that only in some countries are poorer people more likely to live in a flood zone than their richer neigh- bors. Our study confirms that – in contrast to what is often considered obvious – there are cases where 2 A World Bank report on urban floods in Antananarivo using a similar methodology is forthcoming. 5 poorer and richer households are equally likely to be directly affected; of course, this does not mean that poorer and richer households are affected equally. There are even fewer studies looking at the monetary losses of disaster-affected households and asking whether poorer people lose more or less than their richer neighbors (i.e. whether they are more or less vul- nerable). Hallegatte et al. (2017) reviewed five such case studies on household vulnerability, with three in Bangladesh. Other studies have been published, such as Taupo et al. (2016), looking at the impact of trop- ical cyclone Pam on Tuvalu. These studies consistently find that poorer households lose more in relation to their income when affected by a disaster (Hallegatte, et al., 2017). Our results confirm in the case of Accra the larger vulnerability of poorer people. There is an even more limited body of evidence on what determines socioeconomic resilience at the household level, and on the role of poverty and income. Arouri et al. (2015) find that households in Vi- etnam residing in communes with higher mean expenditures were more resilient to natural disasters. Ak- ter and Mallick (2013), on the other hand, find that poorer households have a better ability to respond to and recover from tropical cyclones in Bangladesh compared to their non-poor neighbors, despite being more vulnerable to shocks. Some studies have focused on the effectiveness of coping mechanisms as a way of measuring resilience on a household level. Looking at the effects of drought in rural Kenya, Wine- man et al. (2017) find that credit availability and access to different sources of income reduced house- holds’ chances of falling into poverty after a low-rainfall shock. Arouri et al. (2015) find similar results in Vietnam, showing that greater credit availability enabled households to better cope with the effects of nat- ural disasters. On the other hand, McCarthy et al. (2017) assess the impact of rural floods in Malawi and find that coping strategies such as holding a savings account and having access to non-agricultural income sources were mostly ineffective in mitigating the negative impacts of floods. On this dimension, our study concludes that the resilience of households – here defined as their ability to recover – depends much more on the availability of effective coping mechanisms than on the income level of the households. Poorer households are more likely to struggle to recover, but mostly because they tend to lose a large fraction of their income in the floods and have lower access to coping mechanisms – not because they have a lower income level. In other terms, the income level is an imperfect measure of the resilience of households to floods. 3. The Greater Accra Metropolitan Area and the 2015 flood The Greater Accra Metropolitan Area – or GAMA – hosts 20% of the country’s 25 million population, and contributes to about 25% of its GDP. Accra is located on the coast and lies within 0 to 144 meters above sea level (UNHABITAT, 2011). Although GAMA has the lowest poverty rates in the country (GLSS6, 2013), a significant share of its population lives in low-income communities and informal settlements. Slum dwellers constitute about 38.4% of the city’s population (UN-HABITAT, 2011), and most if not all of these are subject to at least one shelter deprivation in the form of lack of clean water and sanitation; insufficient living space; low quality, unaffordable housing structures; and/or no security of tenure (UN-HABITAT, 2008; Engstrom et al., 2017). 6 The Greater Accra Metropolitan Area is vulnerable to the consequences of perennial flooding (World Bank, forthcoming). The city’s rapid urbanization is characterized by a lack of urban planning and weak enforcement, which exacerbate its vulnerability to flooding (MESTI, 2016). Floods in the densely popu- lated areas of Accra are induced by heavy rainfall primarily during the rainy season (May-June). The Odaw River, within the Korle-Chemu catchment area, drains most parts of the built-up area in central Ac- cra. The river runs through the Odaw basin area, which covers 271 km2. The southern part of the basin is densely populated and includes the informal settlements Nima and Old Fadema, as well as the industrial and business areas in Kwame Nkrumah Circle and Kaneshie. The Odaw basin was the area most affected by the flood in Accra on June 3, 2015. Rainfall recordings in the southern part of the basin indicate a rainfall of 130 mm in 6 hours, equivalent to a return period of 10 years (Klopstra et al., forthcoming). Based on estimates from NADMO and World Bank, the flood caused damages around 100 million USD. Total rainfall, in combination with the inadequate discharge capacity of the lined Odaw drain, were the main reasons for the flood. Impacts were worsened by the gates of an inceptor weir that could not be opened at the time of the event. In addition, accumulated solid waste be- hind the weir and several bridges along Odaw also contributed to the rising water levels. This event turned particularly tragic when a fire broke out at a gasoline pump where people were seeking refuge from the waters, resulting in about 150 casualties. Although the 2015 flood corresponds to just a 10-year return period flood, it is remembered as an extraordinary event by Accra residents. 4. Survey design of Accra Poverty-DRM Survey To understand how flood risk and the level of poverty affect household livelihoods and behaviors, the sample selection followed a four-step process to stratify the targeted slums by flood proneness and the level of poverty. First, we selected areas that are considered as informal settlements in Accra. Second, we designed our sampling strategy to ensure that we have in our sample a diversity of flood risk levels, using elevation as a (very imperfect) proxy for flood risks. Third, we categorized areas as low poverty and high poverty by using a neighborhood level poverty estimate created by Engstrom et al. (2017). Fourth, we se- lected households from the four strata – low elevation and low poverty, low elevation and high poverty, high elevation and high poverty and high elevation and low poverty. The key four steps of the sampling procedure are described in more detail below. Note that the objective of the sampling was not to create a sample representative for the whole slum popu- lation of Accra, but to draw samples to contrast behavioral differences by the levels of poverty and flood proneness. Results can therefore be used to investigate the impact of poverty (and expenditure level) on the exposure, vulnerability, and socioeconomic resilience to the 2015 floods, but not to calculate total losses across the city or to map the risk in the city. 4.1. Sampling strategy The neighborhoods included in Table 1 are defined as slum areas based on the Accra Metropolitan As- sembly (AMA), UN-HABITAT (2011) and a slum index developed by Engstrom et al. (2017). The slum index is a formula based on key characteristics of Enumeration Areas (EAs) that are highly correlated with “typical” slums – as defined by experts, using machine learning techniques. The sample was selected from EAs that are (i) fully inside the informal settlements defined by AMA and UN Habitat (2011) and (ii) have a slum index higher than 0.7, which is significantly above the average slum index among all EAs 7 in the whole Accra (0.48) (Engstrom et al., 2017). In addition, some EAs are excluded from the sampling frame for political sensitivity reasons or security concerns for enumerators after consulting local experts.3 More details are provided in Appendix 1. Table 1: Profiles of EAs in the selected neighborhoods in flood risk and slum index mean elevation Share of EAs with Share of EAs with Neighborhood (meters) low elevation (%) high elevation (%) Gbegbeyise 5.55 100% 0% Korle Lagoon Area 7.35 100% 0% Jamestown 11.54 100% 0% Korle Dudor 12.01 100% 0% Pig Farm 21.09 42% 8% Banana Inn 24.37 0% 0% Nima 29.63 0% 31% Accra New Town 32.38 0% 34% Mamobi 33.34 0% 37% Abeka 33.57 0% 43% At the time of the sampling, there was no reliable flood map for the 2015 event that was representative on a city level for Accra. (Since then, such a map has been developed, see Klopstra et al., forthcoming.) As a result, we used elevation as a proxy for flood risk in our sampling strategy. Using this proxy was sup- ported by measurement of flood height data from the 2015 flood event that were collected by a group of researchers from TU Delft University after the 2015 flood event. Using these estimates, we classify EAs into high risk (elevation below 17.5 m), medium risk (between 17.5-34 m) and low risk (above 35 m). We then removed the EAs in the medium risk category, to obtain more contrasted results. We end up with 145 EAs across the 10 selected neighborhoods. The 145 EAs were the sample frame of this study, and the re- sults are thus applicable to the population living in those EAs. Since there are no data on poverty at the EA level, we used neighborhood level poverty estimates to clas- sify the EA as “high poverty” or “low poverty.” The neighborhood level poverty rates were estimated us- ing small-area poverty estimation methodology carried out by Engstrom et al. (2017) using GLSS6 and geospatial data and census data. EAs with a neighborhood poverty rate higher than the median (5.8%) are categorized as “high poverty areas” and those with a poverty rate lower than the median are categorized as “low poverty areas.” We selected 24 EAs from low elevation (12 “poor” and 12 “non-poor”) and 24 EAs from high elevation (12 “poor” and 12 “non-poor”) using Probability Proportional to Size (PPS). Subsequently, we randomly selected 20 households from each EA4 to arrive at a sample of 1,006 households living in areas officially defined as slums with different levels of flood risk exposure. The map in 3 Sabon Zongo and Gomorrah were excluded from the sampling frame for this reason. 4 The number of households was chosen to ensure that differences in key statistics across the four groups of EAs could be significant at a certain level of confidence (statistical power calculation approach). 8 Figure 2 shows the sampled EAs indicated in red, black, green and blue. Table 2: Final sample selection by elevation strata Elevation Neighborhoods Number of EAs Sample Design Interviewed Number of number of households households Low Elevation Jamestown 11 480 507 Korle Dudor 1 Gbegbeyise 8 Korle Lagoon Area 2 Pig Farm 2 High Elevation Nima 12 480 499 Abeka 3 Accra New Town 4 Mamobi 5 Total 48 960 1006 Figure 2: Map of sampled Enumeration Areas (EAs) in Accra __ High Poverty and __ High Poverty & __ Low Poverty & __ Low Poverty & High Flood Risk Low Flood Risk High Flood Risk Low Flood Risk 9 4.2. Questionnaire design, expenditure data collection using SWIFT and survey imple- mentation The questionnaire used in the survey was specifically developed for this study by the authors. It has four areas of focus: (i) socioeconomic characteristics, annual expenditure levels, and poverty; (ii) exposure to the 2015 floods; (iii) impacts of the 2015 floods; and (iv) behaviors associated with living in risk prone areas. Annual expenditure levels and poverty Collecting expenditure data is costly and increases significantly the duration of interviews. Beyond budg- etary and data processing issues, it also reduces the quality of the data by reducing the space available for the other questions in the survey and by increasing survey fatigue from respondents. To avoid these is- sues, the survey adopted the SWIFT approach to estimate household expenditures and poverty rates of sample areas from a few simple questions (Yoshida et al., 2015). More details on SWIFT are provided in Appendix 2. Instead of collecting household consumption expenditure data directly, the SWIFT approach uses the ex- penditure data of a subsample of households which is representative of the GAMA region from the na- tional household surveys (GLSS6, 2012/2013) to identify a set of around 20 simple questions that can best predict the expenditure level of any given household. The questions usually include demographics, education, and housing characteristics of households. Based on those, a statistical formula calibrated on the GAMA subsample of the national survey – here GLSS6 – provides an estimate of household expendi- tures. As the SWIFT methodology does not collect data on food consumption and other monetary varia- bles, these expenditure estimates reflect the socioeconomic profile of a household in the medium term, and do not account for short-term variations in consumption patterns. Exposure to the 2015 floods There is no unique definition of being “affected” by a flood. In the questionnaire, we define being af- fected as experiencing one of the following effects, which can be “localized” or “non-localized”:  Localized impacts (due to impacts on the household’s dwelling): o Asset losses; o Damages to dwelling (such as having water in the house); o Significant time (one day or more) spent cleaning the house, discharging water from house, or cleaning in and around the house due to flood water;  Non-localized impacts (can be due to impacts elsewhere): o Loss of water or electricity services; o Loss of significant share of income, including one day of work missed or more (due to transportation, illness, work place damages, or forced to stay home with dependents, or other reason); o Being fined or reprimanded due to late arrival at work; o Experience illness or other health effects; o Having children miss one day or more of schooling; o Having to change food intake (either quantity or quality). 10 Impacts of the floods The questionnaire is designed to estimate the monetary losses experienced by each household. We include housing repairs, asset losses, missed days of work and medical costs, but exclude non-monetary losses (e.g. deaths and injuries). Asset losses and labor income losses due to missed days of work were esti- mated combining household responses with complementary information, while housing repair and medi- cal costs were directly estimated by the households.  For asset losses, we collected price information on a number of commonly lost items from shops and households in the surveyed areas. We used this information to calculate asset losses per household. Since there is some variability in asset prices and richer households can be expected to own more expensive assets, asset losses for richer households are likely to be underestimated compared with poorer households’ asset losses.  For labor losses, we first calculated the average daily wage in different occupations, based on both annual household expenditures and household member occupation information. Then, we multiplied the estimated daily wage by the number of missed days due to the flood to generate a total value of losses caused by missed days of work (see details in Appendix 6).  Housing losses are calculated based on the cost of repairs, not actual damages. Since poorer households may not be in a financial position to (fully) repair the damages caused by the flood, their housing losses are likely to be underestimated compared with those of richer households.  Similar issues apply to medical costs, which are imperfect proxies for health impacts. The survey was conducted over a two-week period from the 27th of May to the 11th of June 2017 – two years after the 2015 flood. Field teams comprised of 22 enumerators and 4 supervisors administered questionnaires using Computer Assisted Personal Interviewing (CAPI). Each team had a tablet per enu- merator, a GPS device for supervisors, power banks, notepads and modems with internet bundles to ena- ble them to send data directly and regularly from the field. To learn more about the communities where the household questionnaire was administered, a community questionnaire was also submitted to local leaders. The responses from this questionnaire inform the analy- sis based on the household responses and help us better understand patterns we observed in the data. Con- clusions drawn from both the community and household analyses are incorporated into this report. 4.3. Sample Characteristics The SWIFT estimation shows that, in terms of expenditure, the Poverty-DRM sample covers all income groups of the city of Accra, despite the fact that the sample area only includes informal settlements. Aver- age poverty headcount rates in the Poverty-DRM survey appear lower than in GLSS6 data for GAMA, but the difference is not statistically significant. According to GLSS6 data, the poverty headcount rate of GAMA was 2.1% in 2012/13, compared with 1.4% in the Poverty-DRM survey. (Descriptive statistics for the surveyed population are provided in Appendix 3.) These results may seem surprising because the Poverty-DRM survey draws sample households from ur- ban slum areas only, but could be explained by economic growth between the two surveys. Also, GLSS6 data suggest that the difference in (monetary) poverty rates between non-slum and slum areas in Accra is quite small. Nevertheless, it remains important to note that the surveyed population includes households that are far from poverty, and has an expenditure distribution that is close to that of the full city. 11 Table 3: Distribution of annual household expenditures between GLSS6 (GAMA) and Poverty-DRM sur- veys in Ghana by quartile Quartile GLSS6 (GAMA) in cedis Poverty-DRM in cedis Q1 2,115 2,248 Q2 3,810 3,907 Q3 5,781 5,904 Q4 10,715 11,988 Source: Authors' estimation using GLSS6 and Poverty-DRM surveys Household heads in our sample are more likely to be women and have a lower education level than those of the GLSS6 sample for the full GAMA area. Further, households in our sample live in relatively smaller dwellings despite having the same number of household members, indicating that households are more densely crowded in the surveyed areas than in the rest of the city. Households surveyed appear to have a lower access to services than the rest of city. For instance, the use of public toilets is significantly higher for slum-dwellers than for Accra residents overall. While 32% of households in the GLSS6 survey reported using public toilets as their main toilet facility, the number for the DRM survey is 67%. The primary source of drinking water in the slum areas is sachet water, which is even more prevalent (at 70%) than in the rest of the city (54%). Most households (74%) reported having their waste collected, a high level for an informal settlement. Almost half of the households (44%) reported being directly affected by the 2015 flood. As expected, more households in the low elevation areas were affected by the flood. However, many households in the high elevation areas (29%) were also affected. These results confirm that elevation is not a good proxy for flood exposure in Accra. This is because flood risk is associated with characteristics other than elevation, such as insufficient drainage infrastructure, or poor waste management.5 5. Exposure: No visible difference between poorer and richer households The first question analyzed in this study is whether the affected population was poorer than the average. While it is often assumed to be the case, the review provided in Hallegatte et al. (2017) shows that this is far from universally true. For the 2015 flood in Accra, there is no significant difference in expenditure levels between the affected and non-affected households. The SWIFT-based estimates of household expenditures allow us to define four quartiles in terms of expenditure in the sample, and to explore how different quartiles experienced the flood. Applying these categories, we see an extremely small and statistically non-significant differ- ence between the likelihood of poorer and wealthier households being affected (see Table 17 in Appendix 5 Theoretically, it could also be linked to impacts on the workplace or school that may be located in a different (more flood-prone) area. But the analysis of losses below shows that this is not a significant explanation. 12 4). Figure 3 illustrates the distribution of per capita expenditure levels for affected and non-affected households. Figure 3: Income distribution by exposure Not affected Affected 0 5000 10000 15000 20000 25000 Annual expenditure in cedis While affected and non-affected households have indistinguishable levels of annual expenditures, they differ in several respects, including some usual proxies for non-monetary or asset-based poverty (See Ta- ble 18 in Appendix 4). In particular, people affected by the 2015 floods were more likely to have low- quality walls and roofs, and less likely to have piped water within their dwelling. However, the fact that the difference in these factors does not translate into a measurable difference in expenditures suggests that this difference remains moderate. Three factors may explain the finding that relatively well-off households were also affected by the 2015 flood. First, the relatively large intensity of the event may have led many better-off households to be affected as well, even if they live in places which are on an annual basis less exposed to flooding than poorer house- holds. Results would probably be different for low-intensity high-frequency floods, which would be ex- pected to affect mostly poor people who cannot move away. Second, a combination of tenure arrangements and housing costs may explain why even relatively better- off households stay in risk-prone areas. In our sample, only 4% of households exposed to the 2015 flood 13 moved after the flood.6 Even households with enough resources to move to another location did not do so, potentially explaining why there is no strong relationship between exposure to risk and poverty in our sample. Third, if flood prone areas are attractive because of lower housing costs or access to jobs, amenities, and services, then they may attract richer households in spite of the risk of floods. This is consistent with be- haviors observed in other contexts. For example, households in areas of Mumbai that flood regularly re- port that they are aware of the flood risks, but accept them because of the opportunities offered by the area such as access to jobs, schools, health care facilities, and social networks (Patankar 2015). In a study of Colombo, Sri Lanka, households explained that they decided not to move after the 2010 floods because they would not be comfortable living in another area (suggesting that social networks play a key role) and by listing access to services and jobs (Patankar 2017). When controlling for the neighborhood, rents for households affected by the 2015 floods are on average 250 cedis – or 27% – lower than for non-affected households (See regression results (2) in Table 19 in Appendix 5). Without controlling for the neighborhood, this effect is not visible. This suggests that there are relatively lower rents for households affected by flood within each neighborhood, but that the effect of floods is small when compared to other factors across neighborhoods. This is consistent with a locali- zation choice process in which people select first their neighborhoods based on amenities and access to jobs and services, and then pick the exact localization making trade-offs between flood risks and rent lev- els.7 It is important to note, however, that the effect is only weakly significant in the regression. When households that pay loans on purchased houses and rent-free households (cost set to zero) are in- cluded, the effect of exposure on housing costs becomes clearer, even without controlling for neighbor- hood fixed effects (see Table 20 in Appendix 5). Households affected in 2015 have housing costs that are 150 cedis lower that non-affected households (or 38% of average housing costs) and this difference is sig- nificant, controlling for distance to CBD, housing characteristics, expenditure levels, and with or without fixed effects.8 Differences between the behavior of rents and total housing costs suggests the existence of parallel markets with different actors and different levels of formalization, as identified in Mali in Du- rand-Lasserve et al. (2013). 6. Vulnerability: Poorer people are unambiguously more vulnerable than the rest of the population The second step in this analysis is to explore the vulnerability of households and their assets to the 2015 floods. Vulnerability can be defined as the losses that people experience, given that they have been af- fected by a flood. As a result of the structure of their portfolio (with a larger share in material form) and lower quality of their assets, past studies have systematically found that poorer people lose a larger share 6 We only have information on households who were living in the surveyed areas two years after the floods. We can- not say anything about households who have been affected but moved outside the surveyed areas after the shock. 7 Other determinants of cost of rent include dwelling type, roof material, size and type of water service. 8 Other determinants of total housing costs are wall material, type of water and sanitation services and distance to CBD. When elevation is introduced in the regression, however, the exposure to the 2015 flood does not have a sig- nificant impact on housing costs anymore. The correlation between elevation and exposure to the 2015 floods is likely to explain this result. 14 of their wealth when they are flooded (Hallegatte et al., 2017). This result is confirmed in the case of the 2015 flood in Accra. 6.1. Types of damages to households Among all affected households, 63.9% reported having their dwelling damaged by the flood and 52.7% reported asset losses/damages. Impacts through infrastructure services – primarily roads and electricity – were also commonly reported. Impacts on water and sanitation, work places, schools and health were not as common. Table 4: Types of damages reported by households exposed by the 2015 flood Total Dwelling 63.9% Assets 52.7% Paths or roads 21.3% Electricity 7.3% Water facilities (taps, tanks, pipes) 3.7% Sanitation Facilities 3.8% Work place 4.2% Schools 4.3% Illness in household 2.7% Observations 393 In the community survey, local leaders reported that when flooding happens, a wide range of facilities and services are generally affected. They indicated that schools tend to close during floods, water and electric- ity services are affected, roads become inaccessible, and some businesses are forced to close. However, few households were affected only indirectly: most people who report having been affected by the flood also reported having their homes flooded, or assets lost. Most of the reported damages were lo- calized (impacts to the household’s own dwelling) – Among affected households, 71% reported experi- encing localized damages only. Just 4.3% reported only non-localized damages (impacts elsewhere) while 24.7% reported both types. 6.2. Quantification of household impacts Cost associated with housing repairs and asset losses represented the largest share of total losses, averag- ing 67% and 29% of total losses, respectively. Labor losses incurred from missed work days and medical costs caused by the flood were less common, and totaled only 4% and 0.2% of total losses, respectively. Poorer households experienced relatively larger asset losses while richer households experienced rela- tively larger housing repair costs – most likely due to the higher capacity of richer households to repair housing damages. More information on the composition of type of losses can be found in Appendix 6. As a metric of disaster impacts, average losses per household is not only of limited value, but it can even be misleading, as losses are highly heterogeneous. Affected households lost approximately 509 cedis on average due to the flood, representing about 4% of the value of total annual household expenditures or 15 about 14 days of expenditure (Table 5). But these numbers hide a large diversity among the households’ experiences: only 54% of affected households lost more than 1% of total annual household expenditure. Again, among affected households, 21% lost more than 5% of total expenditures, and 10% lost more than 10%. The distribution of relative losses for households that have lost less than 50% of their annual in- come (i.e. approximately 95% of the affected population) is displayed in Figure 4. Table 5: Absolute and relative total loss, by quartile Absolute loss Proportion of annual expenditure Loss equivalent, in days (in Cedis) (Relative loss) of household expenditures All affected households 508.6 3.9% 14.1 Poorest quartile (Q1) 471.9 5.8% 21.2 Second quartile (Q2) 480.9 4.1% 14.9 Third quartile (Q3) 518.9 3.2% 11.8 Wealthiest quartile (Q4) 566.4 2.2% 8.1 Figure 4: Distribution of relative loss for households (restricted to those who lost less than 50% of annual income, i.e. 95% of the affected population). 0 10 20 30 40 50 Relative loss to income (%) On average, richer households lost more in absolute terms, but poorer households lost more in proportion to their annual expenditure level. Households in the poorest quartile lost around 472 cedis, compared to 566 cedis lost by households in the wealthiest quartile. This ranking changes if we consider relative losses: households in the 1st quartile lost on average around 6% of the value of annual expenditure com- 16 pared to around 2% for the 4th quartile households. This difference is not statistically significant. How- ever, the higher vulnerability of poorer households becomes more apparent when we focus on households that lost larger shares of their annual expenditure. Poorer households (Q1) are overrepresented among those who lost a larger share of their annual expendi- tures. Figure 5 displays the distribution of households by quartiles for affected households and households losing more than 1%, 5% or 10% of their total annual expenditure. For example, among the households that lost more than 5% of annual expenditure, 38% belong to the poorest quartile, and 15% belong to the richest quartile, and this difference is significant at the 1% level. For more extreme relative losses (e.g., larger than 10% of annual expenditures), the distribution becomes even more unbalanced. Many of these inter-quartile differences – but not all of them – are significant (see Table 24 in Appendix 7). For individual households, these differences translate into major risk differences. For example, house- holds in the poorest quartile were 52% more likely than the average household to experience losses larger than 5% of their annual expenditures.9 (Households from the richest quartile were 60% less likely than the average to do so.) Figure 5 Distribution of households over quartiles among entire population, among affected households, among households that lost more than 1%, 5% and 10% of their annual household expenditure 45% 41% 40% 38% 35% 30% 29% 30% 27% 28% 26% 25% 25% 25% 25% 25% 25% 24% 25% 24% 19% 19% 19% 20% 15% 15% 11% 10% 5% 0% Sample Total Affected Over 1% threshold Over 5% threshold Over 10% threshold Q1 Q2 Q3 Q4 Another way of looking at this issue is to consider the average annual expenditure of the households who experienced large losses: households who lost more than 1%, 5%, or 10% of their annual expenditure are significantly poorer than the rest of the population. The higher vulnerability of poorer households is illus- trated in Figure 6. Households that lost the most in relation to their annual expenditure have a signifi- cantly lower per capita consumption than the rest of the population. For example, average per capita con- sumption among households who lost more than 5% of annual expenditures is 4,772 cedis, while for the 9 Households from the first quartile represent 38% of the households losing more than 5% of annual expenditure, while they represent 25% of the population – the ratio 38/25 = 1.52. 17 rest of the population it is 6,404 cedis. The difference is statistically significant at the 5% confidence level for the households who lost more than 1% and 5% of their annual expenditures, and at the 1% confidence level for the households who lost more than 10% of their annual expenditures. Figure 6 Average per capita expenditure level of full sample, affected households and households losing more than 1%, 5% and 10% of annual household expenditure in comparison to the rest of the population 8000 6866 7000 6404 6124 6061 6175 6250 6000 5374 4772 5000 4285 4000 3000 2000 1000 0 Sample Total Affected Over 1% Over 5% Over 10% threshold threshold threshold Yes No In the literature, two mechanisms are generally invoked to explain the higher vulnerability of poor people (Hallegatte et al., 2017). First, poorer households tend to have a larger share of their assets and wealth in material form, and limited financial savings. The poorest urban dwellers tend to have most of their wealth in the form of their dwelling (Moser and Felton 2007). This means that most of their wealth is vulnerable to floods, compared with richer households with financial assets. It also means that financial inclusion and innovative savings instruments can be powerful tools to reduce poor people’s vulnerability. Second, the material assets of poor people tend to be of lower quality – and thus higher vulnerability – than the material assets of richer people. (See Akter and Mallick, 2013, for an illustration in Bangladesh and Hallegatte et al., 2017, for a global analysis.) This mechanism seems to play a key role in Accra since housing charac- teristics, like the materials used for the roof and wall, are strong determinants of household losses ( Table 25 in Appendix 7). A consequence of this concentration of losses within poorer households is that the impact on poverty is much larger than average losses suggest. On average, affected households lost 3.6% of their total annual household expenditure level. While this average number may seem low, removing households’ losses from their annual expenditure – assuming they cannot use their savings to smooth the impact – would be enough to increase the poverty rate in the affected population from 1.6% to 2.5%, a 50% increase.10 If we assume that these households were forced to compensate for the costs within a month and had no savings 10 We use the official Ghana national poverty rate of 1,314 cedis. 18 to smooth consumption, then the poverty level during that month would jump from 1.6% to 18% – a ma- jor jump, albeit for a very short period of time. While these numbers are illustrative only, they demon- strate how misleading average loss per household can be. These results are correlations, and cannot support definitive conclusions regarding causality. However, at least two insights suggest that these correlations stem from the fact that poorer households were more vul- nerable to the 2015 flood, rather than that the differences in annual expenditures are due to the flood it- self. First, most households report having fully recovered from the shock at the time of the survey, and the results are unchanged if the sample is restricted to the households that report having fully recovered. Sec- ond, the observed differences in annual expenditures represented in Figure 6 are much larger than what would be expected from the impact of the floods, considering the reported losses. However, it is possible that the correlation between poverty and flood vulnerability identified here is at least partly due to the cumulative effects of multiple floods and an amplifying feedback loop between poverty and vulnerability (Hallegatte et al., 2017). Our findings on the impact of risk perceptions on in- vestment behaviors provide some evidence that flood risks can have a long-term effect on poverty that goes beyond what asset losses suggest (see Section 0). A firmer conclusion on this causality question would require survey data – if possible a panel – conducted before and after the event, or even the track- ing of households over long periods of time. 7. Socioeconomic resilience: Boosting resilience requires more than increasing income The last element in our framework (see Figure 1) is socio-economic resilience. We define socioeconomic resilience as the ability of affected populations to cope with their losses — in this case, to recover from the impacts of floods without experiencing large well-being losses or long-term impacts (Hallegatte et al., 2017). Two years after the floods, 69% of affected households reported having fully recovered. Among these households, 54% reported having done so in less than one month. However, a significant fraction of the affected population took much longer to recover, or had not recovered after two years, making it im- portant to understand the resources people have to cope with the losses and the tools they have to recover and rebuild. 7.1. The fundamental role of savings The main coping mechanisms used after the 2015 flood were tapping into savings and reducing non-es- sential consumption. Table 6 summarizes the primary coping strategy that households employed to re- cover from the 2015 flood. While around 11% of affected households reduced food consumption, the fact that these households tend to be richer than the average suggests that this strategy did not include a reduc- tion in consumption of essential goods. Overall, these results suggest that direct and indirect losses from the flood did not threaten the basic needs of most affected households. The ability of households to use savings to cope with the flood seems to be independent of the level of expenditure. Rather, we find that resorting to savings depends on the source of income and on the gender of the household’s head (Table 7). Households with less stable sources of income, such as casual labor, are less likely to use savings. Households that have a stable business as a main source of income are less likely to use savings than employed workers. Female-headed households were significantly less likely than male-headed households to use saving as a coping strategy, as shown in Table 8. (A regression table on the use of savings as a primary coping strategy is provided in Table 26 in Appendix 8.) 19 Table 6: Coping strategies employed by households affected by the 2015 flood Share of af- fected house- Household expendi- holds ture (cedis/month) Used savings 42.6% 5,548 Reduction in food consumption/expenses 10.6% 7,059 Reduction of non-food consumption/expenses 18.1% 5,688 Received assistance 17.5% 5,571 Moved* 8.8% 9,447 Other 2.6% 4,841 N = 393 * Including temporary relocation Table 7: Savings used as primary coping strategy by household main source of income Used savings Did not use savings Sample size Monthly salary 59.2% 40.8% 221 Casual Labor 15.9% 84.1% 103 Hawking 57.0% 43.0% 53 Remittances 44.8% 55.2% 57 Safety nets, cash transfers 0.0% 100.0% 3 Stable business 41.6% 58.4% 532 Public works 38.7% 61.3% 20 No income 46.2% 53.8% 19 Table 8: Saving used as primary coping strategy by gender of household head Used saving Did not use saving Female headed 36.9% 63.1% Male headed 47.3% 52.7% 7.2. Assistance is important, and mostly from informal sources Households received assistance primarily after the flood receded, although some households received as- sistance during the flood event. Overall, 37% of households reported receiving assistance, with 1% re- ceiving help before the flood, 13% during the flood, and 23% after the event. Figure 7 shows the results per expenditure quartile, showing that poorer households seem to be more likely to receive support. Fur- ther, there is some evidence that assistance went toward the most affected: 40% of households that experi- enced losses larger than 5% of their annual experience received assistance, compared with only 29% of those that experienced smaller (relative) losses. However, due to the low number of observations, we are unable to establish statistical significance of these results. 20 Figure 7: Assistance received by affected households before, during or after the floods by expenditure quartiles Q4 25.8% Q3 29.5% Q2 34.7% Q1 35.4% 0.0% 10.0% 20.0% 30.0% 40.0% Most of the assistance was informal. Of the 59 households that received support during the flood, 55 (93%) received it from friends, neighbors and relatives. Similarly, of the 91 households that received sup- port after the flood, 69 indicated having received this support from informal sources. Consequently, the community seems to play a key role in helping households recover. Interestingly, this community support seems to target primarily poorer households – as evident from Figure 7 – though the survey does not give details on the magnitude of the support. The government also provided limited support: after the flood, 17 households reported having received support from NADMO, 11 from the local government, and one household from the central government. 7.3. Ability to recover from the shock About 31% of affected households reported not having fully recovered at the time of the interview (ap- proximately two years after the event), and these households have distinct characteristics. We report these differences in Table 27 in Appendix 8. Owning the dwelling, receiving remittances, and having access to borrowing are robust factors of resilience and ability to recover. In parallel, households with casual labor as their main source of income have more trouble recovering than the rest of the population, which is con- sistent with the findings in Section 7.1 on the availability and use of savings. Interestingly, differences in annual expenditures or education are not significant, and the sign of some of the differences is unexpected. For example, people who did not recover reported a higher level of ex- penditure than the others, though this difference is not significant. Households that have been in the same dwelling for more than 20 years are less likely to recover. This is a peculiar result, since one might as- sume long-term residents in high-risk areas to be more resilient to shocks. However, length of stay be- yond 20 years may also be associated with changing exposure to risk, given the rapidly urbanizing envi- ronment, and other characteristics such as an inability to move due to tenure issues or seniority. To explore the impact of the magnitude of the losses on households’ probabilities of recovery, we run three logistic regressions of the fact of having recovered on having lost at least a certain fraction of annual 21 income. The first regression tests for the causality of the 1% loss threshold on recovery, the second for the 5% loss threshold and the third for the 10% loss threshold. Each regression set accounts for control varia- bles in additional specifications. These include real per capita expenditure and other socio-demographic characteristics, and access to certain services. One of the challenges of such an analysis is the relatively small fraction of households experiencing large losses, which leads to an imbalance in the sample. For instance, there are only 91 households that lost more than 5% of their annual income (which we refer to as “treated” observations) and 302 households that were affected but lost less than 5% of their annual income (“control” observations). To reduce sample imbalance between treated and control observations, which is one of the main drivers of model dependency, we proceed to a set of matches following the Coarsened Exact Matching (CEM) method provided by Iacus, King and Porro (2011), and k-to-k matching. Appendix 9 provides details on the methodologies, and comparisons of regressions with no matching, and with CEM and k-to-k. The analysis leads to two conclusions: First, and perhaps unsurprisingly, large losses – here measured by having lost a large fraction of annual expenditure – impedes recovery. Table 9 illustrates this result with the impact of the loss of more than 5% of annual expenditure, with CEM matching using the age and gender of the household heads. Losing more than 5% of annual expenditures reduces the odds of recovering in less than two years by 40%, rela- tive to households that experienced lower relative losses. An even stronger relationship is found for larger losses (Table 30 in Appendix 9): households losing more than 10% of their annual expenditure are 60% less likely to recover in less than two years than other affected households. Confidence in these results is increased by the fact that the three approaches provide consistent results, even though the magnitude of the effects can vary (see all results in Appendix 9). Without matching, effects are not significant. The re- lationship is significant only with matching, either through CEM or k-to-k. Second, income source matters and income level does not. Households that derive income from casual labor find it more difficult to recover from the flood. Conversely, access to borrowing and remittances facilitates recovery. However, the level of annual expenditure does not affect the ability to recover, con- trolling for the magnitude of relative losses and access to coping mechanisms such as borrowing and re- mittances. It is important to note that this result does not imply that poorer households are as able to re- cover as richer people. Poorer people are less able to recover, but only because they experience higher rel- ative losses in the floods and they have lower access to coping mechanisms. This result is consistent with the findings of Noy and Patel (2014), who found that the negative impact for labor markets after the 2011 flood in Thailand was driven by the lack of job security for low-skilled work- ers. The fact that we do not find any significant relationship between underlying sociodemographic char- acteristics of the household (age, sex or education level of household head, percentage of employed indi- viduals in household) and recovery is also consistent with Jones et. al. (2018), who argue that sociodemo- graphic characteristics do not drive household resilience as measured subjectively. However, it is not con- sistent with the findings of Akter and Mallick (2013), who found that the households involved in more temporary and less formal employment were less likely to suffer negative income effects after a disaster since the flexible nature allowed them to switch to a sector in which demand was high. 22 Table 9: Probability of having recovered after two years, as a function of various housing characteristics, magnitude of losses, and coping mechanisms. N.B. Logistic regression is done after CEM. Details can be found in the Appendix. (1) (2) (3) (4) (5) Did not lose more than 5% of household ex- penditure due to the flood Lost more than 5% of HH expenditure due to the flood -0.541* -0.592* -0.620* -0.655** -0.748** (0.316) (0.312) (0.326) (0.323) (0.323) Log of real per capita expenditure -0.182 -0.131 -0.134 (0.197) (0.210) (0.223) Main income source: Monthly salary Main income source: Casual labor -1.026*** -1.016*** -0.917** (0.382) (0.386) (0.396) Main income source: Hawking 0.124 0.0994 0.0874 (0.725) (0.726) (0.754) Main income source: Remittances 0.192 0.174 -0.451 (0.541) (0.543) (0.611) Main income source: Stable business -0.182 -0.194 -0.212 (0.289) (0.290) (0.298) Main income source: Other -0.169 -0.182 -0.598 (0.619) (0.618) (0.657) Does not anyone to borrow money from in case of emergency Can borrow money from at least one or mul- tiple persons in case of emergency 0.473* (0.256) Did not receive remittances in the past year Received remittances in the past year 0.989*** (0.304) Constant 0.858*** 2.422 1.106*** 2.236 1.730 (0.139) (1.682) (0.270) (1.810) (1.933) Observations 393 393 393 393 393 Prob > F 0.0886 0.142 0.0515 0.0893 0.00263 These results show that building resilience is not only about increasing income (and expenditure levels), and that monetary poverty and resilience are different dimensions that are not perfectly correlated.11 There 11 With a broader definition of poverty that would include financial inclusion, social capital, and stability of income, poverty would affect the ability to recover. Here, we define poverty only through the level of annual expenditure. 23 are other determinants of recovery, namely social networks – measured through the ability to rely on oth- ers for financial support, either directly or through remittances – and certain socioeconomic characteris- tics such as income sources. Policies to increase the resilience of the population need not only to reduce poverty, but also to provide households with coping and recovery mechanisms, such as financial assis- tance and financial tools. 8. The hidden cost of risk: Risk perceptions affect behaviors and investment choices The main determinant of perceptions of future flood risks is the fact of having been affected by the flood. Households affected by the 2015 flood are more likely to expect to be affected in the future (see Figure 8 (a)). While people living in low elevation areas also have higher risk perception, the effect of elevation is much smaller than that of previous flood experience (Figure 8 (b)). The importance of past flood experi- ence as a driver of flood risk perceptions is further supported by regression results in Table 33, see Ap- pendix 10). Since risk perceptions may affect behaviors – constructively or negatively – it is interesting to investigate the relationship with investment decisions. Figure 8: Perception of likelihood of exposure to flood in next couple of years by exposure to 2015 flood (a) and area (high or low elevation) (b) (a) (b) 90% 84% 80% 70% 80% 70% 70% 70% 60% 54% 60% 46% 50% 50% 40% 40% 30% 30% 30% 30% 20% 16% 20% 10% 10% 0% 0% Unaffected Affected High elevation Low elevation Likely Unlikely/Indifferent Likely Unlikely/Indifferent 8.1. Impacts on investments Two different mechanisms could connect flood experience and risk perception to household decision making in investments. First, the households affected by the 2015 flood may have less financial capacity to carry out investments, after having used savings and current income to cover costs associated with flood impacts. This budget constraint could reduce investment for affected households. Second, affected households may adjust their behavior in response to perceived risk: for example, deeming investments in their home or business too risky to carry out. 24 Households affected in 2015 are more likely to have invested in housing in the last year than the non-af- fected (Figure 9).12 This result still holds when we control for expenditure levels, flood risk perception, as well as tenure arrangement of dwelling as displayed in the regression results in Table 34 in Appendix 10. The result is consistent with Noy and Patel (2014) who identified an increase in housing investments among households affected by the 2011 flood in Thailand. Figure 9: Fraction of households who made investment in housing in the past year by different subgroups 45% 39% 40% 35% 35% 27% 29% 30% 26% 24% 25% 20% 20% 17% 15% 10% 5% 0% Households affected in 2015 are also more likely to take flood risk into account when making decisions on housing improvement investments (see Table 10). And households that have been affected or live in low elevation areas are more likely to have avoided making housing investments in the past year due to the risk of floods than other households. This seemingly paradoxical observation can most likely be ex- plained by the fact that many families whose dwelling has been damaged in the flood have invested in re- construction and reinforcing the house for future events instead of prioritizing other housing investments. Table 10: Perception of risk and behaviors by exposure to the 2015 flood and location (high and low ele- vation area) Not af- High Eleva- Low Eleva- fected Affected tion tion Take flood risk into account when making decision on housing improvement 67.3% 85.3% 74.0% 76.1% Avoided home improvements due to flood risk 4.2% 18.0% 5.6% 13.2% Among households involved in enterprises (614 households), risk seems to play a role in household in- vestment decisions. The propensity to invest in enterprises seems to be related to having been affected by 12 Housing investments include a wide variety of actions such as expanding the dwelling, upgrading roof, wall or floor material, adding or heightening the floor, upgrading the windows or adding toilets or even blocking the walls to prevent flooding. 25 the 2015 flood, and to the annual expenditure level (see Figure 10), but these differences are not statisti- cally significant. However, the findings call for further investigation: do impacted business-owning households face a trade-off between housing repairs and investing more productively in their business? Figure 10: Fraction of households who made investment in enterprise in the last year by different subgroups 35% 29% 29% 30% 26% 25% 25% 21% 20% 20% 18% 17% 15% 10% 5% 0% 8.2. Prioritization of investments Results suggest that business-owning households that were affected by the flood are more likely to priori- tize investments in their house over investments in their enterprise. In Table 11, we include results from business-owning households only, which show that affected households are significantly more likely to invest in their house and significantly less likely to invest in their business. This clearly supports the case that due to the flood, business-owning households allocate more resources to their house in order to im- prove and/or repair it than pursuing more productive enterprise investments. These results still hold when controlling for per capita expenditure and for flood risk perception as proven through a multinomial logit regression. The results of this are displayed in Table 37 in Appendix 10. This finding suggests that impacts of floods on investment behaviors in income generating activities may have long term welfare implications on the household, and more generally on poverty reduction and even macroeconomic growth. Table 11: Choice between housing or enterprise investment by exposure Affected No Yes Difference House/Enterprise investment No investment 65% 57% -8% Investment in house but not in enterprise 10% 25% 14% *** Investment in enterprise but not in house 17% 8% -10% *** Investment in both house and enterprise 8% 11% 3% Observations 375 212 587 26 9. Discussion and policy implications This research supports the idea that poor people suffer disproportionally from floods, and therefore that flood management can be particularly beneficial for them. Flood management could be considered as a component of the poverty-reduction strategy in the city of Accra. This is particularly true because the im- pacts of floods seem to go beyond asset losses to affect behavior, potentially slowing down asset accumu- lation and poverty reduction among affected households (ODI and GFDRR 2015). But most importantly for policy design, it shows that building resilience is not only about increasing in- come: monetary poverty (defined by low annual expenditures) does not appear as a strong driver of the ability to recover from the losses due to the floods. Instead, access to coping and recovery mechanisms – such as assistance and financial tools – seems more important. This means that traditional poverty reduc- tion instruments – such as cash transfers with targeting based on poverty indicators – may not be able to prevent all long-term impacts of natural disasters. Flood management programs need to be designed to target low-resilience households, such as those with little access to coping and recovery mechanisms, even if they are not living in poverty before the shock. Third, the large heterogeneity of vulnerability and resilience across households makes the targeting of flood risk and impact mitigation and post-flood support particularly challenging. With constrained budg- ets, local or national authorities may want to target interventions to minimize the risk of floods toward the households who are the most vulnerable, i.e. who would be losing the most if they were affected in the future, or toward the households who are the least resilient, i.e. who would struggle to recover from a flood. Could the insights from the 2015 flood guide how to perform such a targeting? A first obstacle is the fact that our survey applies to a single event, which may not be representative of flood risks in the entire city. Also, we have focused our data collection in informal settlements, and could not perform a city-wide representative sampling. While focusing on informal settlements makes sense due to the high vulnerability of these areas, the limited sampling makes it impossible to draw conclusions re- garding the full distributional impact of floods in the city. An extension of the survey to the rest of the city would of course be an obvious next step in this research activity. But even assuming that the 2015 event is representative of flood risks in Accra and that households’ vul- nerability and resilience to future floods is equivalent to the vulnerability observed in 2015, using the sur- vey results to target disaster risk management interventions would be challenging. Our results suggest that it will be difficult to “predict” which households are most likely to lose a large fraction of their annual ex- penditures or to struggle to recover after a shock, based on the characteristics available before the event (for instance based on Census data). Even though having low annual expenditures makes it more likely for a household to lose a large share of its annual expenditure, many other observed and unobserved fac- tors contribute. These households cannot be easily identified based on their characteristics like access to services, housing quality and characteristics, type of toilet or waste collection. Available household char- acteristics only explain a small fraction of the variance of flood losses across households. In the face of these difficulties, one option that merits more in-depth analysis is the use of self-targeting instruments, such as providing access to loans (with or without subsidies) to affected households who may not have access to borrowing, or public work programs that can offer an alternative source of income to affected people. 27 This analysis is the first application of the Poverty-DRM household survey designed to investigate the in- teractions between poverty and flood risks. Variations of this survey are now being implemented in other cities, including Addis-Abeba and Dire Dawa in Ethiopia, Dar-es-Salam in Tanzania, and Porto Alegre in Brazil. Hopefully, comparison of the results across cities and for floods of various intensities will help us understand better how to mitigate the impact of flood risks on poverty and to prioritize the interventions that are the most likely to contribute to the long-term eradication of extreme poverty. 10. Acknowledgments This report was written by a team composed of Alvina Erman, Elliot Motte, Radhika Goyal, Akosua As- are, Shinya Takamatsu, Xiaomeng Chen, Silvia Malgioglio, Alexander Skinner, and Nobuo Yoshida, and led by Stephane Hallegatte. It benefited from contributions by Kirsten Hommann, Kathleen G. Beegle, Tomomi Tanaka, Ryan Engstrom, Dan Pavelesku, Yan F. Zhang, Shohei Nakamura, and Brian Walsh. The core team received invaluable support in Ghana from Rachel Annan, Frederick Addison, Akosua As- are and Charlotte Hayfron from the World Bank and Dr. Clement Adamba and Prof. Robert Osei from ISSER, University of Ghana. We would like to thank the Accra Metropolitan Assembly (AMA) for sup- porting this work and a special thank you to Lydia Addy and her team for providing guidance of the local context. We would also like to thank the Sub-Metro Directors and their teams for supporting enumerators during data collection in areas covered by the survey. This report and study is the result of a collaborative effort between the Global Facility for Disaster Reduc- tion and Recovery (GFDRR) and two World Bank Global Practices (Social, Urban, Rural and Resilience; and Poverty), initiated by Niels B. Holm-Nielson with the support of Francis Ghesquiere and Pierella Paci. A special thanks to Marianne Fay, Chief Economist for Sustainable Development, and Henry G. R. Kerali, Country Director for Ghana, for chairing the internal review and providing their guidance on the project. Invaluable contributions were provided by the report’s peer reviewers: Kirsten Hommann, Oscar Ishizawa, Emmanuel Skoufias and Sarah Coll-Black. For important contributions and advice, the team thanks Asmita Tiwari, Oleksiy Ivaschenko, Carl Chris- tian Dingel, Nancy Lozano Garcia, Yohannes Yemane Kesete, Edward Charles Anderson, Keren Carla Charles, Sajid Anwar, Julie Rozenberg, Eric Dickson, Monica Yanez Pagans, Tiguist Fisseha, Oscar Ishi- zawa, Emmanuel Skoufias, Pauline Cazaubon, Frederico Ferreira Fonse Pedroso, Jonas Ingemann Parby, Emilie Bernadette Perge, Claudia Soto, Ivo Imparato, Beatrix Allah-Mensah, Carlos Silva-Jauregui The report was sponsored by the Global Facility for Disaster Reduction and Recovery (GFDRR) with ad- ditional support from the Research Support Budget (RSB). 11. References Abeka, Emmanuel Anyang (2014). Annepu R, Themelis N. (2013) Analysis of Waste Management in Accra, Ghana and Recommendations for further improvement. Earth Engineering Center, Columbia University and Zoomlion Ghana. Akter, S., and B. Mallick. (2013). “The Poverty–Vulnerability–Resilience Nexus: Evidence from Bangladesh.” Eco- logical Economics 96: 114–24. doi:10.1016/j.ecolecon.2013.10.008. Alderman, H., Gentilini, U., Gilligan, D., Hoddinott, J. and Karachiwalla, N. (2014) “Designing and Implementing Urban Safety Nets: A Review of Selected Issues and Practices”. World Bank and IFPRI. Washington DC. Mimeo 28 Anomanyo E (2004). Integration of Municipal Solid Waste Management in Accra (Ghana): Bioreactor treatment technology as an integral part of the management process. Master’s thesis. Lund University. Arouri, M., Nguyen, C., and Youssef, A.B. (2015). “Natural disasters, household welfare, and resilience: evidence from rural Vietnam.” World Development, 70, pp. 59-77. Carter, M. R., P. D. Little, T. Mogues, and W. Negatu. (2007). “Poverty Traps and Natural Disasters in Ethiopia and Honduras. World Development 35: 835–56. doi:10.1016/j.worlddev.2006.09.010. Dang, Hai-Anh H.; Lanjouw, Peter F.; Swinkels, Robertus A. (2014). Who remained in poverty, who moved up, and who fell down? an investigation of poverty dynamics in Senegal in the late 2000s. Policy Research working pa- per; no. WPS 7141; Washington, DC: World Bank Group. Durand-Lasserve, Alain, Harris Selod, and Maylis Durand-Lasserve. (2013) "A systemic analysis of land markets and land institutions in West African cities: rules and practices-the case of Bamako, Mali." Policy Research working paper;Washington, DC: World Bank Group. Engstrom, R., D. Pavelesku, T. Tanaka and A. Wambile (2017). Monetary and non-monetary poverty in urban slums in Accra: Combining geospatial data and machine learning to study urban poverty, World Bank. Elbers, C., J. W. Gunning, and B. Kinsey. 2007. “Growth and Risk: Methodology and Micro Evidence.” World Bank Economic Review 21: 1–20. doi:10.1093/wber/lhl008. Friedman, J. H., and N. I. Fisher (1999), Bump Hunting in High-Dimensional Data, Statistics and Computing, 9, 123-143. Gentilini, Ugo. (2015). “Safety Nets in Urban Areas: Emerging Issues, Evidence and Practices.” Draft Report. World Bank, Washington, DC. Ghana Statistical Service (2012). 2010 Population & Housing Census: Summary Report of Final Results. Accra, Ghana: Ghana Statistical Service. Ghana Statistical Service (2013). 2010 Population & Housing Census: National Analytical Report. Accra, Ghana: Ghana Statistical Service. Ghana Statistical Service (2014). Ghana Living Standards Survey Round 6: Main Report. Accra, Ghana: Ghana Sta- tistical Service Hallegatte, S., M. Bangalore, and M. A. Jouanjean. (2016). “Higher Losses and Slower Development in the Absence of Disaster Risk Management Investments.” Policy Research Working Paper, World Bank, Washington, DC. Hallegatte, S., M. Bangalore, L. Bonzanigo, M. Fay, T. Kane, U. Narloch, J. Rozenberg et al. (2016). Shock Waves: Managing the Impacts of Climate Change on Poverty. Climate Change and Development Series. Washington, DC: World Bank. Hallegatte, S., Vogt-Schilb, A., Bangalore, M., & Rozenberg, J. (2017). Unbreakable: building the resilience of the poor in the face of natural disasters. World Bank Publications. Iacus, Stefano M., Gary King, and Giuseppe Porro. (2012) "Causal inference without balance checking: Coarsened exact matching." Political analysis 20.1:1-24. Jankowska M; Weeks J.R; Engstrom R (Kasanga K, Kotey N.A (Korboe D (1992) Rent Free Tenure in Ghana. Jones, L., E. Samman, and P. Vinck. 2018. Subjective measures of household resilience to climate variability and change: insights from a nationally representative survey of Tanzania. Ecology and Society 23(1):9. https://doi.org/10.5751/ES-09840-230109 Kasanga K, Kotey N.A (2001) Land Management in Ghana: Building on Tradition and Modernity. Rusell Press Nottingham UK. Klopstra, D., Udo, J., Gamadeku, R, van Bork, G., van Hussen, K., Berentsen, K,. Slabbers, S., Ahlijah, O. (forth- coming) Greater Accra Climate Risk Mitigation Strategy, draft report. HKV MESTI Ministry of Environment, Science, Technology and Innovation (2016). June 3 2015 floods in Accra, Assess- ment Summary Ministry of Lands and Forestry (2003) Emerging Land Tenure Issues. Ministry of Lands and Forestry, Ghana. Available at: http://www.hubrural.org/IMG/pdf/cilss_praia9_ghana_rapport_national.pdf 29 McCarthy, Nancy; Kilic, Talip; De La Fuente, Alejandro; Brubaker, Josh. (2017). Shelter from the storm ? house- hold-level impacts of, and responses to, the 2015 floods in Malawi. Policy Research working paper; no. WPS 8189. Washington, D.C. : World Bank Group. Moser, Caroline ON, ed. (2008) Reducing global poverty: The case for asset accumulation. Brookings Institution- Press. Narloch, Ulf; Bangalore, Mook. (2016). Environmental Risks and Poverty : Analyzing Geo-Spatial and Household Data from Vietnam. Policy Research Working Paper; No. 7763. World Bank, Washington, DC. World Bank. Noy, I., and P. Patel. (2014). “Floods and Spillovers: Households after the 2011 Great Flood in Thailand.” Working Paper Series No. 3609, School of Economics and Finance, Victoria University of Wellington. ODI (Overseas Development Institute) and GFDRR (Global Facility for Disaster Reduction and Recovery). (2015). “Unlocking the Triple Dividend of ResilienceWhy Investing in DRM Pays Off.” http://www.odi.org/tripledi- vidend. Opondo, D. O. (2013). “Erosive Coping after the 2011 Floods in Kenya.” International Journal of Global Warming 5: 452–66. doi:10.1504/IJGW.2013.057285. Patankar, A. (2015). “The Exposure, Vulnerability and Adaptive Capacity of Households to Floods in Mumbai.” Policy Research Working Paper 7481, World Bank, Washington, DC. Patankar, Archana Mahesh. (2017). “Colombo: exposure, vulnerability, and ability to respond to floods”. Policy Re- search working paper; no. WPS 8084. Washington, D.C.: World Bank Group. Peprah D, Baker K, Moe C, Robb K, Wellington N, Yakubu H, Null C. (2015) Public Toilet and their customers in Low Income Accra. Environment and Urbanization. Rain D; Engstrom, R; Ludlow, C; Antos S (2011). Accra Ghana: A City Vulnerable to Flooding and Drought in- duced migration. UN Habitat Cities and Climate Change: Global Report on Human Settlements 2011. Availa- ble online at: https://unhabitat.org/wp-content/uploads/2012/06/GRHS2011CaseStudyChapter04Accra.pdf Skoufias, E. (2003). Economic crises and natural disasters: Coping strategies and policy implications. World Devel- opment, 31(7), 1087-1102. Songsore, Jacob. (2003). Towards a better understanding of urban change: Urbanization, national development and inequality in Ghana. Accra, Ghana: Ghana Universities Press Stoler J, Weeks J, Fink G (2012) Sachet drinking water in Ghana’s Accra-Tema Metropolitan area: past, present and future. Available at https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3842094/pdf/nihms527642.pdf Taupo, T., Cuffe, H. & Noy, I. (2018). Household vulnerability on the frontline of climate change: the Pacific atoll nation of Tuvalu. Environ Econ Policy Stud. 20(7): 1-35 The Ghanaian Times (2014) The Plastic Waste Business in Ghana, How far? Ghanaian Times published on February 17, 2014. Available at http://www.ghanaiantimes.com.gh/plastic-waste-business-ghana-far/ The Earth Institute, University of Ghana (2013). Report to the Accra Metropolitan Assembly on Solid Waste Com- position in Aryee Diki electoral area Ayawaso sub metro, Accra New Town http://mci.ei.colum- bia.edu/files/2013/03/Accra-MCI-solid-waste-report-FINAL-DRAFT-2010.pdf UNHABITAT (2009). Ghana: Accra Urban Profile. Nairobi, Kenya. Available online at: https://urban- healthupdates.files.wordpress.com/2009/10/habitat-ghana_urban_profile2009.pdf UNHABITAT; AMA (2011). Participatory Slum Upgrading and Prevention Millennium City of Accra, Ghana. UNHABITAT (2008). UN-HABITAT: Ghana - Overview of the current Housing Rights situation and related activi- ties. Available online at http://lib.ohchr.org/HRBodies/UPR/Documents/Session2/GH/UNHABI- TAT_GHA_UPR_S2_2008_UnitedNationsHABITAT_uprsubmission.pdf Wineman, A., Mason, N. M., Ochieng, J., Kirimi, L., (2017) Weather extremes and household welfare in rural Kenya. Food Security. 9:281–300 World Bank (2017). ENHANCING URBAN RESILIENCE IN THE GREATER ACCRA METROPOLITAN AREA. World Bank Group. World Bank (2015). Ghana Rising Through cities. World Bank Group. 30 World Bank. 2017. “Climate Vulnerability Assessment: Making Fiji Climate Resilient.” 120756. The World Bank. http://documents.worldbank.org/curated/en/163081509454340771/Climate-vulnerability-assessment-making- Fiji-climate-resilient. World Bank (forthcoming). Ghana Multi-Sectoral Investment Framework for Climate and Disaster Risk Manage- ment. WSMP (2008) Use of Toilet Facilities in Ghana. WSMP brief Yoshida, N., R. Munoz, A. Skinner, C. Kyung-eun Lee, M. Brataj, W. Durbin, D. Sharma, and C. Wieser. (2015). SWIFT Data Collection Guidelines version 2. The World Bank. Appendix 1 – Sample selection methodology We have four different categories for the sampling – (i) low elevation and high poverty incidence; (ii) high elevation and high poverty incidence; (iii) low elevation and low poverty incidence; (iv) high eleva- tion and low poverty incidence. Table 12 shows the number of EAs that belong to each category. Table 12: Strata and Number of EAs Strata Total number of EAs (i) High Poverty Incidence & High flood risk 62 (ii) High Poverty Incidence & Low flood risk 21 (iii) Low Poverty Incidence & High flood risk 31 (iv) Low Poverty Incidence & Low flood risk 31 Total 145 Source: Authors' calculation using Engstrom et al. (2017). The sample size is determined using the power calculation, which is to find a minimum sample size under which differences in key statistics between the groups of interest – in our case the four strata – are signifi- cant at a certain level of precision. For this study, since we aim to find systematic differences in house- hold characteristics and behavior by flood risk or poverty incidence, we adopt the power calculation ap- proach for determining the sample size. Table 13 summarizes the results of the power calculations which concluded that 240 households were needed from each stratum – adding up to a total of 960. Additional households were added to the final sample to mitigate risks of having to drop observations due to data quality issues, resulting in a total sample of 1,006 households. Table 13: Final sample by strata Strata Neighborhoods Number of EAs Sample Design Interviewed num- Number of house- ber of households holds (i) High Poverty Inci- Jamestown 11 251 dence & Low Ele- 240 vation Korle Dudor 1 (ii) High Poverty Inci- dence & High Ele- Nima 12 240 250 vation 31 (iii) Gbegbeyise 8 Low Poverty Inci- 256 Korle Lagoon dence & Low Ele- 2 240 Area vation Pig Farm 2 (iv) Abeka 3 Low Poverty Inci- 240 249 Accra New dence & High Ele- 4 Town vation Mamobi 5 Total 48 960 1006 Appendix 2 – SWIFT methodology for estimating household consumption expenditure SWIFT (Survey of Well-being via Instant, Frequent Tracking) is a rapid poverty assessment tool. Devel- oped in-house at the Poverty and Equity Global Practice of the World Bank. It can produce accurate pov- erty data through household expenditure and poverty data in a very timely, cost-effective and user- friendly manner. It has also been used to improve availability and frequency of official poverty statistics. Compared with a typical household consumption data collection, SWIFT is much faster and more cost- effective for producing consumption or income data and poverty statistics. This is because instead of col- lecting primary household consumption or income data, SWIFT collects only 10 to 30 questions on pov- erty-correlated variables, then projects household income or expenditure from them using a custom-built model, and estimates poverty and inequality statistics from the projected income or expenditure data. The poverty correlates typically include variables such as household size, household head’s educational attain- ment, household head’s employment status, ownership of consumer durables and housing conditions. To collect responses to the select questions from a household, we usually need only 7 to 10 minutes. This is much faster than a typical household consumption or income data collection, which takes at least one hour. Furthermore, the SWIFT approach is very quick to estimate poverty and inequality statistics from data collected – in 1 minute or less. This is in contrast with traditional methods that often require over one year to process consumption data collected by an official household survey and estimate poverty and ine- quality statistics. Basics and Assumptions SWIFT collects only 10 to 30 questions on poverty correlates, projects household income or expenditure from them using a model, and estimates poverty and inequality statistics from the projected income or ex- penditure data. The poverty correlates usually include household size, household head’s educational at- tainment, household head’s employment status, ownership of consumer durables, housing conditions, etc. To do this accurately, model development is critical. 32 The model is developed assuming the relationship between household income or expenditure and poverty correlates is linear and that there is an error in projection.13 Equation (a.1) shows this relationship: ln ′ (a.1) where ln refers to a natural logarithm of household income or expenditure of household h, is a 1 vector of poverty correlates of household h, is a 1 vector of coefficients of poverty corre- lates, is the number of variables, and is a projection error. In principle, SWIFT estimates the linear formula by regressing the natural logarithm of household income or expenditure on a set of poverty corre- lates in household survey data that include both household income/expenditure and poverty correlates. The regression model becomes a formula, with which household expenditure or income will be projected into a data set that has only poverty correlates. The latter data set will be collected by a SWIFT survey. A SWIFT survey collects the poverty correlates. To improve accuracy of projections, SWIFT adopts ap- proaches used in machine learning, poverty mapping, and multiple imputation. More details are available in the annex of the guidelines for SWIFT (Yoshida, et al., 2015).14 The SWIFT modeling process includes multiple steps to improve the ability of the formula to project household income or expenditures by adjusting the coefficients ( and estimating the distributions of both the coefficients and the projection errors. No formula is perfect, so inclusion of the projection error is essential. Indeed, estimating the distribution of the projection error is key for estimating poverty rates and their standard errors. Cross Validation Since consumption patterns can differ significantly across areas and population groups, the SWIFT team makes efforts to create a model that is specific to the areas and population groups of interest. Such an ad- justment is good to create the model tailored to the needs, but can cause potentially large bias in poverty estimates because the sample used for creating a model declines by focusing on the specific group of pop- ulation. “Over-fitting” is one of such problems. The over-fitting problem means that while a model can perform well within the sample developed for the model, it can perform badly outside the data set. In a sense, the model over-fits the data set used to develop it. To detect this problem, the SWIFT team con- ducts a cross-validation analysis. The cross-validation approach separates data used for developing the model from those used for evaluating the model fitness. More specifically, a household survey data set is split randomly into 10 subsamples. Each of these sub- samples is called a “fold.” A consumption model is estimated from nine folds by running a stepwise Ordi- nary Least Square (OLS) regression.15 The stepwise OLS regression means that a statistical package searches for an OLS regression model where all variables are statistically significant, at a given p-value level. We use STATA and its stepwise selection model. The nine folds used for developing a model are known as “Training Data”. 13 This does not mean SWIFT does not use a non-linear model, but it means that SWIFT’s formula is linear in varia- bles created in the data set. Since some variables can be squares of other variables, SWIFT’s formula can be non- linear. One of the typical examples is that SWIFT uses household size and household size squared in a formula. 14 Yoshida, N., R. Munoz, A. Skinner, C. Kyung-eun Lee, M. Brataj, W. Durbin, D. Sharma, and C. Wieser. (2015). SWIFT Data Collection Guidelines version 2. The World Bank. 15 Or weighted least squares. 33 After a model is selected, household expenditure or income data are projected using the model in the re- maining fold, and a poverty rate and mean squared errors (MSEs) are estimated with the projected data. At the cross-validation stage, we project household expenditure or income data assuming the error term and regression coefficients follow normal distributions. More specifically, suppose is a vector of estimated coefficients and is an OLS estimator of error var- iance. We first draw a random value from a chi distribution with a degree of freedom, , where N refers to the total sample size and k refers to the number of variables selected by the stepwise regression procedure, and calculate / . We then draw from a normal distribution of , where X is a matrix of ( , … , , … , ′ . Finally, we draw a simulated household expenditure or income for household h, ln , from a normal distribution of , ∗ where ∗ refers to an identity matrix. This simulation process is repeated for all households, typically 20 times.16 A poverty headcount rate is calculated by comparing the simulated household expenditure or income with a poverty line for each of the 20 simulation rounds. The average poverty rate of the simulations is used as a poverty estimate. MSE is calculated in testing data by taking the average of the sum of squared differ- ences between and ∗ . This analysis is repeated 10 times, each of which uses a different fold as testing data to test the perfor- mance in terms of mean squared errors and the absolute value of the difference between the projected and actual poverty rates. This test detects the over-fitting problem because all testing statistics are calculated from out-of-sample. Figure 11 shows an illustration of a three-fold cross validation exercise. Figure 11 Illustration of 3-Fold Cross-Validation Step 1:Randomly split data into three folds (C refers to consumption; X refers to non-consumptiondata) Household Survey data Randomly Split by three Step 2: Select two folds as training data, develop a model there, and test model performance in the test- ing data Household Survey Data Training Data Testing Data modeling Compare ) Step 3: Repeat the above procedure three times by changing the testing data 16 This process can be done using STATA’s command “mi impute regress”, or STATA Corp LP (2013). 34 Training Data Testing Data This cross-validation exercise is conducted to determine the optimal threshold of the p-value for the step- wise regressions. For a specific p-value, the cross-validation exercise is done and produces the two testing statistics. The exercise is repeated for different levels of p-value, usually between 0.1% and 10%. The op- timal p-value is the value where the absolute value of the difference between the actual and the projected poverty rates is minimized. The mean squared error is also examined to check whether the over-fitting problem occurs. If the mean squared error is minimized at a level of p that is smaller than the value where the absolute difference between the actual and the projected poverty rates is minimized, then the former value is chosen as the optimal number. Figure 12 shows results of cross validation analysis using the Ghana Living Standard Survey (GLSS) 2012/13 data. The average MSE continues to decline as the threshold of the p-value for the stepwise re- gression increases. If MSEs are calculated in the same sample as where a model is developed, MSEs tend to decline as the p-value increases because the number of variables in a model tends to increase and the model fitness improves as the p-value increases. However, this is not always the case if we calculate MSEs out of sample because of the over-fitting problem. In the case of a cross-validation analysis for GLSS 2012/13 data, we did not see that, but we did see it in the other data set. This suggests that there is no over-fitting problem in the modeling in GLSS 2012/13 for the range of p-values we investigated. The average absolute values of the difference between actual and projected poverty rates show a different trend. Although the numbers fluctuate, the difference starts increasing once the p-value reaches 6%. Be- low 6%, the value fluctuates, but it is never below the value at the p-value of 6%. Therefore, we choose 6% as the optimal threshold of the p-value for the stepwise regression procedure. 35 Figure 12 Typical Results of Cross Validation Analysis for Ghana 2012/13 data Average absolute values of differences between Average MSE actual poverty rates and projected poverty rates .263 .013 .0125 .2625 .012 absdiff mmse .262 .0115 .2615 .011 .0105 .261 0 .02 .04 .06 .08 .1 0 .02 .04 .06 .08 .1 pe pe Source: Results of cross validation analysis using GLSS 2012/13 data. Finalizing the Model After the optimal p-value is selected, a stepwise OLS regression procedure is carried out with a full sam- ple of data to estimate a model. To ensure that the coefficients are stable, an OLS regression with the set of variables is carried out for all 10 testing data sets to see whether the coefficients of the select variables do not change signs or are dropped due to collinearity. If some variables are dropped due to collinearity or some signs of the coefficients change, then these variables will be dropped from the final model. After dropping these variables, an OLS regression is carried out to estimate the coefficients and variance of the coefficients and error terms. In addition to the statistical tests, it is recommended to check whether the signs and values of all estimated coefficients make sense to those who know a country very well. If a sign of a variable is the opposite of an expert’s intuition, this can be an indicator of multicollinearity and can be very unstable; therefore, it is strongly recommended to reconsider inclusion of such variables. Simulation and Estimation of Poverty Rates The final model is used to project household expenditure or income for all households 20 times following the procedure presented above. Poverty rates are estimated for each round of simulation and the average is taken as the estimate of the poverty rate. The variance of the poverty estimate is calculated using the following formula (Rubin, 1987 and Schafer, 1999): ∗ ∗ 1 ∑ ∑ (a.2) where m refers to the number of simulations, refers to the poverty estimate in round l of the simula- ∗ tion, refers to a mean of and the final estimate of the poverty headcount rate, and is an es- timate of the variance of the poverty estimate in round l of simulation. The first bracket presents the be- tween simulation variance, while the second squared bracket presents the within simulation variance. Consequently, the variance of the final poverty estimate is a weighted average of the within and between simulation variances. 36 Appendix 3 – Household characteristics and comparison with GLSS6 The GLSS6 was collected in 2012/2013 while the poverty-DRM data was collected in 2017. The GLSS6 data is representative for the population of GAMA, which not only includes the city of Accra but also its surrounding areas. Conversely, the poverty-DRM survey data was collected from only specific urban slum areas in GAMA and was not designed to be representative for either GAMA or urban slum areas in GAMA. As explained previously, the objective of the poverty-DRM survey is to examine whether there are any systematic differences in behaviors and livelihoods of slum population depending on the level of flood proneness or the level of poverty. Average household expenditures for quartile of the Poverty-DRM survey are very similar to those of the GLSS6 (GAMA) (Table 14). This result may seem surprising because the Poverty-DRM survey draws sample households from urban slum areas only, not the full city, and expenditure levels could be expected to be lower. There are two possible reasons that may explain why this is not the case. First, GDP per cap- ita in Ghana still grew 9.5% between 2012 and 2017, and expenditures may have increased faster than the predictors used in SWIFT, weakening the model stability. Second, a preliminary assessment using GLSS6 data shows that the difference in (monetary) poverty rates between non-slum and slum areas in Accra ap- pears quite small. This could mean that the key challenge faced by urban slum residents is not monetary poverty but concerns other dimensions of living standard (such as lack to access to services or poor hous- ing quality). Table 14: Changes in distribution of household expenditures between GLSS6 (GAMA) and Poverty-DRM surveys in Ghana Quartile GLSS6 (GAMA) Poverty-DRM Q1 2,115 2,248 Q2 3,810 3,907 Q3 5,781 5,904 Q4 10,715 11,988 Source: Authors' estimation using GLSS6 and Poverty-DRM surveys Household heads in the poverty-DRM sample are more likely to be women and have a lower education level than those of the GLSS6 sample. Further, in the poverty-DRM survey, households live in relatively smaller dwellings despite having the same amount of household members indicating that households are more crowded in the poverty-DRM sample. The proportion of household members employed is higher in the poverty-DRM data. To further explore potential vulnerabilities of households living in slum areas we will look closer at access to services and tenure. 37 Table 15: Descriptive statistics of Poverty-DRM survey and GLSS6 (GAMA) Household characteristics DRM GLS HH size 3.44 3.41 Age of head 45.60 41.98 Number of rooms 1.53 2.03 Female household head 44.6% 31.3% Highest level of schooling completed by the household head None 13.5% 8.0% Less than primary 8.9% 3.7% Primary completed 10.4% 6.8% JSS/JHS completed 20.6% 18.9% Middle school completed 17.4% 20.3% SSS/SHS completed or above 29.3% 42.4% Proportion of household members employed (question asked 64.9% 57.5% to members aged 5 and over) The use of public toilets is significantly higher for slum-dwellers than Accra residents. While 32% of households in the GLSS6 survey reported using public toilets as their main toilet facility, the number for the DRM survey is 67% (see Figure 14). Public toilets are shared pay-per-use toilet facilities within com- munities. Public toilets are ideally constructed for transient populations and for areas with heavy public activity (WSMP,2008). However, they are also common in slums where households do not have toilet fa- cilities within their housing structure. In Engstrom et al. (2017), the use of public toilets is highly corre- lated with slum status. We asked community leaders about the most important challenges associated with providing safe sanitation services and they mentioned overcrowding, lack of space in houses to install toi- lets and affordability. They mentioned examples of landlords making space meant for installing toilets into living areas in order to fit more people in the dwelling. From a service provider’s perspective, they mentioned that the community is planned in a way which makes it difficult to dislodge sewage and other effluent. The primary source of drinking water in the slum areas is sachet water (Figure 13). Sachet water is puri- fied water purchased by households in small plastic bags and is generally considered a clean and safe source of drinking water. It is however a significant waste management problem due to the amount of plastic waste it generates. This is reflected in the responses from community leaders who also highlight that sachet water is the least affordable option for accessing drinking water in Accra. They also say that even though pipe-borne water is available to households in slum areas it does not always flow. The main challenges in accessing safe pipe-borne water inside the units for households in slums are the nature of structure making separate access difficult. This forces households to share utility bills. They also mention lack of space in the communities to install water pipes. Despite these difficulties highlighted by the local leaders, pipe-borne water sources are reported frequently in the DRM survey. 38 Figure 13 Main source of drinking water in DRM Figure 14 - Access to sanitation service in DRM and and GLSS6 GLSS6 GLS DRM GLS DRM Public toilet and other 32% 70% 67% Sachet water 54% KVIP 20% 14% Pipeborne outside 17% dwelling Pit latrine 10% 22% 3% WC 34% 16% Pipeborne inside 9% dwelling 12% No facility 4% 0% 0% 20% 40% 60% 80% 0% 20% 40% 60% 80% Figure 13 Garbage disposal system for DRM and GLSS6 GLS DRM Dumped indiscriminately 1% 2% Public dump 16% 23% Burned by household 17% 1% Collected 66% 74% 0% 20% 40% 60% 80% In the DRM survey, more households report having their waste collected and using a public dump than the GLSS6 (see Figure 13). Most households – 74%, reported having their waste collected. Various waste collection options are available to households in slum settlements. In Accra, the mode of waste collection, at the household level, depends on the neighborhood. In high income neighborhoods, a franchise system is operated where private companies collect waste from house to house on a weekly basis and charge for their services. In slum settlements, results from the community survey report different types of waste 39 management services available for residents. In all communities included in the sample, city authorities provide central communal containers. Households transport their household waste to the disposal point and pay a user fee. In the community survey, the skip containers are reported being emptied twice a week or more. There are also formal and informal garbage pick-up options for residents of slum areas. In Ma- mobi, Nima and Pigfarm, formal service providers such as Zoom Allianz and ABC provide households with dustbins and collect refuse directly from households. Informal door-to-door waste collection is re- ported in Nima, Gbegbeyise, Jamestown and Mamobi and is carried out by tricycle and donkey operators. In the case of Jamestown, it is reported that informal collectors dump waste in the Lagoon. In Nima and Mamobi, city authorities discourage the use of informal operators and instead encourages households to register with the Assembly contracted private providers previously mentioned. In addition, most neigh- borhoods report having community cleaning days with regularity. Households in the Poverty-DRM survey tend to have less secure tenure arrangements than in the GLSS6. Among the households in the Poverty-DRM survey, fewer households claim ownership of their dwelling unit. Renting and “rent free” are more common arrangements and are closely related to land ownership, which will be discussed in more detail below. In Accra, “rent free” housing is commonly referred to as family house (Korboe, 1992) and it is a right to a house/room that one derives from being a member of a family. Such right is not always limited to the immediate family but also accommodates external family members as well. Family houses are safeguards against homelessness. As poor family members who may not be able to afford rent can always fall back on the family housing for free accommodation. Table 16: Tenure arrangement Poverty-DRM survey and GLSS6 (GAMA) Ownership of a housing unit DRM GLS Owning 20.8% 35.3% Renting 44.2% 41.0% Rent free 35.0% 23.8% There are three main types of tenure arrangements observed in the Poverty-DRM survey results – (i) own- ers of both land and dwelling, (ii) renters of dwellings on land owned by private individual and (iii) rent- free households living in dwelling and on land owned by non-household relative. A few households also report owning the dwelling on land owned by a non-household relative and a small group of households report paying rent to non-household members owning the land. The community survey results show that traditional leaders are the predominant owner of land inside the communities. It is reported that the land is mainly obtained through purchase and inheritance. Tenure structure differs markedly across neighbor- hood, due to the different histories of these places. Jamestown and Nima, for example, have existed since colonial times and are very established neighborhoods in Accra. Community leaders are not reporting on any recent push for relocation of slum dwellers in the areas included in the survey. 40 Appendix 4 – Descriptive statistics and regressions table for section 5 on Exposure Table 17: Flood exposure by (1) expenditure, (2) expenditure and elevation. (1) (2) Expenditure -5.19e-06 1.03e-05 (1.91e-05) (1.94e-05) High elevation (base) ref. Low elevation 1.072*** (0.363) Constant -0.191 -0.976*** (0.220) (0.293) Observations 1,008 1,008 Prob > F 0.788 0.0459 Table 18: Descriptive statistics for households affected or non-affected by the floods, and significant dif- ferences. Unaffected Affected Difference Real per capita annual expenditure 6,175 6,061 -114.7 Age 46.1 45.0 -1.1 Household size 3.2 3.7 0.5 *** Male-headed household 56.2% 54.4% -1.8% % of households in low elevation areas 52.0% 75.7% 23.7% *** Tenure situation Owner 11.0% 21.3% 10.3% *** Own dwelling but not land 3.9% 5.3% 1.4% Rent from private person 45.9% 33.3% -12.6% *** Rent from relative 2.4% 5.0% 2.7% * Rent free from relative 31.6% 27.3% -4.3% Other 5.2% 7.9% 2.7% Dwelling Type Separate house 1.8% 0.3% -1.4% ** Semi-detached house 5.5% 5.6% 0.0% Flat/apartment 1.3% 1.1% -0.1% Compound house 90.9% 92.9% 1.9% Huts 0.6% 0.1% -0.4% * Roof material Metal sheet 48.0% 44.0% -4.0% Slate/asbestos 47.5% 52.2% 4.7% Cement/concrete 4.5% 3.8% -0.7% Wall material Mud bricks/earth 2.8% 9.6% 6.8% *** Wood 3.4% 5.5% 2.1% Metal sheet 1.9% 1.1% -0.8% Cement/concrete 91.9% 83.9% -8.1% *** Floor material 41 Earth/mud 0.4% 2.4% 2.0% ** Cement/concrete 93.6% 93.3% -0.2% Vinyl tiles 1.2% 1.8% 0.6% Other 4.9% 2.4% -2.5% * Source of drinking water Pipe-borne inside dwelling 16.7% 6.4% -10.3% *** Pipe-borne outside dwelling 25.7% 17.6% -8.1% *** Public tap 10.7% 11.3% 0.6% Sachet water 45.8% 64.1% 18.4% *** Other 1.1% 0.5% -0.5% Toilet type W.C. 22.5% 7.5% -15.0% *** Pit latrine 3.2% 3.1% -0.1% KVIP 16.7% 9.6% -7.1% *** Public toilet 57.0% 78.8% 21.8% *** Other 0.7% 1.0% 0.3% Main income source of household Monthly salary 23.2% 20.3% -2.9% Casual labor 8.3% 12.7% 4.4% ** Hawking 6.5% 3.8% -2.6% * Remittances 5.9% 5.3% -0.6% Safety nets 0.2% 0.3% 0.1% Stable business 51.8% 54.0% 2.2% Cash transfers 1.9% 2.0% 0.1% Other 2.2% 1.5% -0.7% Duration of stay in household Less than five years 22.0% 16.7% -5.3% Five to ten years 12.5% 18.6% 6.0% * Ten to twenty years 26.2% 22.8% -3.5% More than twenty years 24.3% 27.5% 3.2% Received remittances 30.6% 27.7% -2.9% Able to save in the past month 73.2% 77.5% 4.3% Owns a bank account 71.6% 66.8% -4.7% Can borrow money from someone if needed 76.2% 73.3% -2.8% Appendix 5 – Regression tables related to rents and housing costs Table 19: Annual rent amount explained by household and location characteristics (1) (2) (3) (4) (5) Unaffected ref. ref. ref. ref. ref. Affected -176.2 -255.9* -209.2 -208.0 -175.3 (110.9) (149.0) (137.8) (125.6) (112.4) High-elevation ref. Low-elevation 156.6 (173.5) 42 Distance to CBD 27.48 (31.19) Dwelling type: Separate house ref. ref. ref. ref. ref. Semi-detached house -1,556*** -1,900*** -1,634*** -1,665*** -1,557*** (260.9) (331.3) (311.7) (297.2) (318.7) Flat/apartment 1,287*** 1,219*** 1,242*** 1,250*** 1,325*** (114.8) (137.4) (140.1) (141.9) (276.2) Compound house (rooms) -1,467*** -1,751*** -1,531*** -1,587*** -1,460*** (183.6) (189.5) (228.0) (196.7) (260.2) Huts/building (same compound) -1,616*** -1,770*** -1,640*** -1,753*** -1,606*** (230.9) (235.8) (235.0) (254.5) (294.3) Wall material: Mud bricks/earth/landcrete ref. ref. ref. ref. ref. Wood, Metal sheet/slate/asbestos 40.66 -12.69 -11.12 24.96 17.58 (120.1) (132.7) (116.9) (167.8) (132.3) Cement blocks/concrete, stone, burnt bricks 274.9** 251.7** 236.6** 268.1* 256.1** (107.2) (107.4) (100.5) (147.1) (117.3) Roof material: Mud/Mud bricks/earth ref. ref. ref. ref. ref. Metal sheet -757.7*** -818.6*** -659.0*** -671.8*** (145.7) (136.1) (158.5) (233.2) Slate/asbestos -699.8*** -740.0*** -659.9** 66.38 -613.2* (245.9) (213.4) (269.3) (230.5) (323.1) Cement/concrete -882.5*** -890.6*** -795.5*** 21.80 -810.7*** (180.2) (170.4) (209.3) (141.6) (245.0) Floor material: Earth ref. ref. ref. ref. ref. Cement/concrete, stone, burnt bricks 40.94 90.84 42.24 117.6 104.2 (143.7) (190.7) (139.5) (198.8) (198.1) Wood, vinyl tiles and other types of tiles 131.1 18.81 110.0 143.0 172.9 (173.0) (195.3) (169.8) (178.4) (206.7) Number of rooms 373.6** 388.6** 383.9** 405.2** 365.2** (140.4) (146.0) (146.9) (175.3) (137.3) Access to water: Pipe-borne inside dwelling ref. ref. ref. ref. ref. Pipe-borne outside dwelling 195.8 243.0 208.3 198.3 185.0 (132.3) (148.3) (136.6) (172.1) (135.2) Public tap/standpipe 834.1* 841.9* 854.1* 1,035* 822.7* (441.2) (451.0) (456.6) (560.9) (439.5) Sachet water 262.8** 274.6** 250.7** 278.0 241.1** (107.1) (124.9) (107.4) (191.2) (110.2) Other 299.1*** 452.1*** 333.6** 280.3** 250.3* (108.7) (164.1) (125.2) (125.7) (137.6) Toilet type: W.C. ref. ref. ref. ref. ref. Pit latrine 97.43 103.8 131.1 173.0 105.4 43 (316.9) (380.9) (325.7) (386.2) (323.6) KVIP 342.5 319.2 350.6 488.0 345.2 (261.5) (259.5) (263.0) (327.6) (256.6) Bucket/pan 239.9 30.39 270.6 244.7 217.8 (174.9) (191.9) (192.4) (207.2) (202.6) Public toilet -68.77 -74.11 -98.57 30.20 -54.14 (93.18) (113.5) (96.52) (105.0) (105.3) Waste disposal: Collected ref. ref. ref. ref. ref. Burnt by household -346.9 -570.0 -397.6 -591.5 -361.6 (565.8) (561.0) (611.7) (628.9) (565.7) Public dump 126.7 149.7 134.5 207.0 109.5 (206.5) (196.0) (209.6) (279.2) (201.2) Dumped indiscriminately -648.7* -797.0* -627.8 -839.7* -621.5 (382.8) (407.3) (380.7) (431.0) (386.7) Expenditure quartiles: Q1 ref. Q2 -16.01 (240.3) Q3 51.85 (246.8) Q4 120.5 (255.7) NEIGHBORHOOD FIXED EFFECTS NO YES NO NO NO Constant 2,008*** 2,077*** 1,971*** 1,024 1,849*** (363.0) (411.4) (384.8) (610.7) (538.7) Observations 484 484 484 376 484 Table 20: Total annual housing costs17 by household and neighborhood characteristics. (1) (2) (3) (4) (5) Unaffected ref. ref. ref. ref. ref. Affected -151.7** -141.0* -129.3 -158.5** -151.9** (65.03) (77.09) (77.98) (71.53) (63.76) High-elevation ref. Low-elevation -147.8 (111.5) 17 Total annual housing costs include annual rent for households that pay some form of rent, annual fees payed to traditional chiefs for those who pay fees to these entities, annual pay of construction loan (this concerns very few households). Households that occupy their dwelling without paying any rent or costs (occupation of a dwelling owned by family, etc.) have a total annual housing cost equal to 0. Households that own their dwelling have been ruled out of the analysis. 44 Distance to CBD 64.82*** (15.98) Dwelling type: Separate house ref. ref. ref. ref. ref. Semi-detached house -168.5 -166.2 -151.1 -257.1 -208.1 (329.6) (331.2) (323.0) (359.9) (321.8) Flat/apartment -43.65 19.20 -20.62 -61.77 -63.10 (492.2) (495.6) (481.7) (526.0) (493.7) Compound house (rooms) -76.88 -13.41 -62.91 -107.1 -104.6 (300.4) (308.0) (291.9) (330.9) (297.4) Huts/building (same compound) -236.3 -138.0 -252.6 -270.3 -246.0 (314.3) (332.5) (306.3) (335.0) (312.1) Wall material: Mud bricks/earth/landcrete ref. ref. ref. ref. ref. Wood, Metal sheet/slate/asbestos 136.5* 97.14 156.3* 73.53 111.2 (77.46) (89.76) (77.91) (66.00) (87.48) Cement blocks/concrete, stone, burnt bricks 236.7*** 197.6*** 250.4*** 187.8*** 220.8*** (43.29) (51.00) (42.25) (47.76) (47.34) Roof material: Mud/Mud bricks/earth ref. ref. ref. ref. ref. Metal sheet -243.9 -384.3 -285.4 123.3 -166.6 (341.1) (344.2) (345.9) (88.37) (341.2) Slate/asbestos -415.5 -449.2 -402.0 -15.24 -334.5 (353.9) (351.9) (356.0) (154.8) (357.9) Cement/concrete -339.4 -444.3 -368.3 75.38 -288.0 (360.3) (380.7) (367.3) (122.4) (359.3) Floor material: Earth ref. ref. ref. ref. ref. Cement/concrete, stone, burnt bricks -131.9 90.97 -115.8 123.4 -67.62 (101.6) (127.8) (101.1) (80.97) (111.1) Wood, vinyl tiles and other types of tiles -228.6* -29.71 -195.5 16.49 -195.3 (121.4) (134.3) (127.0) (84.40) (126.1) Number of rooms -4.053 -21.45 -13.10 -18.46 -13.38 (34.11) (33.66) (37.67) (37.63) (32.94) Access to water: Pipe-borne inside dwelling ref. ref. ref. ref. ref. Pipe-borne outside dwelling 94.74 91.71 84.71 83.80 89.50 (69.29) (68.69) (67.16) (72.67) (68.98) Public tap/standpipe 617.2** 535.5** 594.0** 627.4** 615.7** (257.8) (263.7) (267.7) (309.5) (258.4) Sachet water 113.2** 160.5*** 121.7*** 144.3** 94.17* (47.54) (53.73) (44.17) (63.75) (48.49) Other 298.2** 302.3*** 287.0** 284.5*** 254.0* (134.6) (101.9) (122.1) (101.9) (139.2) Toilet type: W.C. ref. ref. ref. ref. ref. Pit latrine -41.06 -68.55 -81.61 -90.02 -22.97 45 (75.12) (101.9) (78.79) (76.92) (85.06) KVIP 130.1 66.01 113.8 123.3 138.4 (133.6) (121.0) (129.0) (150.1) (133.5) Bucket/pan 468.6*** 283.6*** 419.4*** 455.4*** 431.9*** (99.81) (96.92) (113.9) (96.07) (104.2) Public toilet -70.26 5.108 -38.93 9.324 -42.96 (60.09) (68.16) (60.24) (68.14) (59.05) Waste disposal: Collected ref. ref. ref. ref. ref. Burnt by household -264.4 -285.2 -209.4 -278.2 -239.0 (255.9) (238.2) (247.4) (292.9) (248.8) Public dump 96.71 44.96 81.63 72.89 67.29 (113.8) (98.42) (123.7) (134.9) (108.7) Dumped indiscriminately -269.6* -408.5*** -281.9** -359.7** -237.7 (142.3) (132.4) (139.1) (151.2) (149.2) Expenditure quartiles: Q1 ref. Q2 42.94 (110.7) Q3 131.6 (124.4) Q4 226.4* (134.5) NEIGHBORHOOD FIXED EFFECTS NO YES NO NO NO Constant 670.2 566.4 722.3 -151.6 486.3 (515.7) (561.4) (526.4) (461.9) (545.6) Observations 1,002 1,002 1,002 816 1,002 Appendix 6 - Details on loss calculations As for loss due to missed days of work, we combined information from the flood impact module and household member roster to assess the total labor loss. We initially computed two different values based on two strategies. The first strategy divides total annual household expenditure with number of economi- cally active household members (above age of 5) and then we divide that number by number of work days in a year based on 20 monthly work days. This gives us a rough estimate of the value of one missed work day provided all economically active households members work full time and that all the value generated to be able to sustain their expenditure level is generated by labor income. The second strategy we used is to compute estimated salaries by occupation by regressing number of household members in each occupa- tion group on total household expenditure. This strategy estimates the attribution of having a household member working in a specific occupation on the overall household expenditure. Table 21 contains the result of this regression and the values can be used as estimates of wages for differ- ent occupations – assuming a straightforward relationship between labor and expenditure. For the first 46 strategy we multiplied the number of days missed by each household member by total household expendi- ture level divided by the number of workers. For this strategy we assume that each working household member brings to the household the same amount of money due to their work. As this is quite unrealistic, we turn to a second strategy consisting in multiplying the number of missed work days of a household member by the average daily wage of his/her occupation based on occupation information in Table 21. It should be noted that results derived from the first or second strategy are very similar. We choose to focus on those obtained through the second strategy only because of more realistic assumptions. Table 21: Estimated annual wage for different occupations (in Cedis) Occupations Estimated wage (annual) Managers 19276 Professionals 11632 T&A professionals 10648 Clerical support workers 8939 Service/sales workers 9799 Skilled agricultural workers 7618 Craft and trade 9876 Manufacturing 7222 Elementary Occupation 7274 Retiree 7715 The largest losses occur through housing repair costs and assets, and medical costs and missed work days only play a minor role. Total loss can be decomposed in four different sources (asset loss, repair costs, income loss due to missed days of work, and medical costs) which provides additional information on how households were impacted by the flood. Asset losses and housing repairs make up a large share of total losses (49% each). For the households affected in our sample, labor and medical related costs were marginal in comparison to asset losses and housing repairs. Among the 15% of affected households that missed days of work due to the flood, service workers stand out as the most affected occupation group. Service workers missed on average 10 days of work, which is significant. Other categories report longer interruption durations, but with sample sizes that are too small to draw any conclusion. Missed days of work account for a small portion of total losses. Table 22 - Missed days of work caused by being affected by flood by occupation Occupation Average number of missed days Standard deviation Sample Size Professionals 4 (1.780) 4 T&A Professionals 3 (0.307) 6 Clerical support workers 1 (0) 1 Sales/service workers 10 (2.453) 46 Skilled Agricultural workers 13 (4) 2 Craft & Trade 14 (9.313) 4 Manufacturing 14 (0) 1 Elementary Occupation 5 (1.936) 4 47 Table 23: Decomposition of total loss Loss sources Repair costs Asset loss Missed work days Medical costs Total of affected households 67% 29% 4% 0.2% Q1 67% 27% 5% 0.1% Q2 65% 32% 3% 0.2% Q3 67% 28% 4% 0.2% Q4 69% 29% 2% 0.4% Appendix 7 - Regressions tables for section 6 on Vulnerability Table 24: Significance of the across-quartile differences in Figure 7. Note: F-statistics are computed ac- cordingly to multiple imputation theory (see Appendix 2 – SWIFT methodology for estimating household consumption expenditure). Affected 1% threshold 5% threshold 10% threshold Prob(Q1=Q2) > F 0.867 0.635 0.356 0.589 Prob(Q2=Q3) > F 0.88 0.637 0.354 0.505 Prob(Q3=Q4) > F 0.927 0.475 0.616 0.526 Prob(Q1=Q3) > F 0.735 0.334 0.0856 0.25 Prob(Q2=Q4) > F 0.811 0.202 0.174 0.263 Prob(Q1=Q4) > F 0.669 0.109 0.00933 0.0562 Table 25: Log of total loss of affected households by (1) housing material, (2) housing material, expenditure and elevation, (3) housing material, expenditure and elevation, gender and if the household took preventive measures. (1) (2) (3) Roof material: Mud/earth/palm/bamboo Roof material: Wood, slate/asbestos, roof- ing tile -1.725*** -1.616*** -1.110** (0.338) (0.412) (0.463) Roof material: Metal sheet -1.505*** -1.576*** -1.030*** (0.357) (0.431) (0.364) Roof material: Concrete/Other -2.431*** -2.395*** -1.874*** (0.390) (0.422) (0.422) Wall material: Mud bricks/earth/landcrete Wall material: Wood, Metal sheet/slate/as- bestos, Bamboo, Palm leaves/thatch(grass/ruffian) 1.014*** 1.061*** 0.644* 48 (0.362) (0.312) (0.366) Wall material: Stone, Burnt bricks, Cement blocks/concrete, Other 0.954*** 0.995*** 0.683** (0.264) (0.223) (0.300) Floor material: Earth Floor material: Cement/concrete, stone, burnt bricks -0.00460 -0.00310 0.0620 (0.247) (0.275) (0.311) Floor material: Wood, vinyl tiles, ce- ramic/porcelain/granite/marble tiles, ter- razo/terrazo tiles, other 0.0441 0.102 0.0783 (0.543) (0.548) (0.657) Expenditure 1.01e-05 2.35e-06 (1.99e-05) (1.86e-05) High elevation Low elevation -0.403 -0.168 (0.307) (0.257) Female Male 0.326** (0.133) No preventive measure taken At least one preventive measure taken 0.671*** (0.247) Constant 6.616*** 6.784*** 5.688*** (0.308) (0.542) (0.490) Observations 299 299 292 Prob > F 2.38e-05 5.66e-05 2.65e-09 Appendix 8 – Descriptive statistics and regression tables for section 7 on Resilience Table 26: Used savings as a coping mechanism by expenditure, main source of expenditure and gender of household head. Used savings Expenditure -3.96e-05 (3.28e-05) Female ref. Male 0.478** (0.222) Main income source: Monthly Salary ref. Main income source: Casual labor -1.952*** 49 (0.516) Main income source: Hawking 0.117 (0.662) Main income source: Remittances -0.418 (0.569) Main income source: Safety nets, cash transfers omitted. omitted. Main income source: Stable business -0.589** (0.284) Main income source: Public work -0.764 (0.867) Main income source: No income -0.475 (0.871) Observations 392 Prob > F 0.000114 Table 27: Descriptive statistics of households that have recovered after the flood and those that have not yet recovered two years after the event. Did not recover Recovered Difference Real per capita annual expenditure 6,324 5,941 -383.2 Age of household head 44.9 45.1 0.2 Household size 3.8 3.7 -0.1 Percent of employed individuals in household 60.7% 62.1% 1.4% Male-headed household 60.2% 51.8% -8.4% Education level of household head None 11.7% 10.2% -1.6% Less than primary 5.7% 10.1% 4.4% Primary completed 12.5% 11.4% -1.1% JSS/JHS completed 27.1% 22.6% -4.5% Middle school completed 15.9% 17.0% 1.1% SSS/SHS completed or above 27.1% 28.7% 1.6% % of households in low elevation ar- eas 77.2% 75.1% -2.1% Tenure situation Owner 14.1% 24.6% 10.5% ** Own dwelling but not land 3.3% 6.2% 2.8% Rent from private person 38.1% 31.1% -7.0% Rent from relative 8.5% 3.4% -5.1% Rent free from relative 27.4% 27.2% -0.2% Other 8.7% 7.6% -1.1% Dwelling Type Separate house 0.0% 0.5% 0.5% Semi-detached house 2.5% 6.9% 4.4% ** 50 Flat/apartment 1.9% 0.8% -1.1% Compound house 95.4% 91.7% -3.7% Huts 0.2% 0.1% -0.1% Roof material Metal sheet 55.9% 38.7% -17.2% *** Slate/asbestos 39.6% 57.8% 18.2% *** Cement/concrete 4.6% 3.5% -1.1% Wall material Mud bricks/earth 12.2% 8.3% -3.9% Wood 5.3% 5.6% 0.3% Metal sheet 1.1% 1.2% 0.1% Cement/concrete 81.4% 85.0% 3.5% Floor material Earth/mud 5.8% 0.9% -4.9% ** Cement/concrete 91.0% 94.4% 3.5% Vinyl tiles 0.8% 2.3% 1.5% Other 2.5% 2.4% -0.1% Source of drinking water Pipe-borne inside dwelling 4.7% 7.2% 2.6% Pipe-borne outside dwelling 17.4% 17.7% 0.3% Public tap 11.0% 11.5% 0.5% Sachet water 66.7% 63.0% -3.7% Other 0.3% 0.6% 0.3% Toilet type W.C. 9.5% 6.5% -2.9% Pit latrine 2.4% 3.5% 1.1% KVIP 7.1% 10.7% 3.6% Public toilet 80.8% 78.0% -2.8% Other 0.3% 1.3% 1.0% Main income source of household Monthly salary 17.3% 21.7% 4.4% Casual labor 21.7% 8.5% -13.2% *** Hawking 3.4% 4.0% 0.7% Remittances 4.7% 5.6% 0.9% Safety nets 0.0% 0.5% 0.5% Stable business 49.1% 56.2% 7.1% Cash transfers 2.1% 2.0% -0.1% Other 1.6% 1.4% -0.2% Duration of stay in household Less than five years 11.2% 19.9% 8.7% Five to ten years 16.4% 19.8% 3.4% Ten to twenty years 22.5% 22.9% 0.5% More than twenty years 36.9% 22.2% -14.7% ** 51 Received remittances 16.5% 32.8% 16.3% *** Able to save in the past month 72.7% 79.7% 7.0% Owns a bank account 70.0% 65.4% -4.6% Can borrow money from someone if needed 62.4% 78.4% 16.0% *** Appendix 9 – The use of Coarsened Exact Matching and k-to-k matching to estimate the impact of vulnerability on resilience and recovery According to Iacus et. al. (2012), the goal of matching techniques is to study the “pure” effect of a treat- ment intervention (in our case losing more than a certain threshold of annual expenditures) on a phenome- non we want to explain (here, the ability of households to recover from the flood in less than two years). Matching is an extremely useful method when the original sample suffers from imbalance between the treatment group and the control group. There are two aspects of imbalance: (i) the treatment group is small in comparison to the control group; and (ii) potential explanatory variables of the phenomenon we want to explain have very different distributions within the treatment and the control groups. It has been proven that sample imbalance often leads to biased and unreliable results when doing statistical inference. More specifically, it induces model dependency which can be defined as obtaining substantially different results in inference estimates through small changes in a model’s specification. Matching helps to limit model dependency by rebalancing the sample. There are numerous matching techniques to strike balance within a sample amongst which the most popular are propensity score match- ing, matching based on distances, and coarsened exact matching. Following Iacus et. al. who have re- viewed the properties and efficiencies of these different matching techniques, we choose to adopt CEM to our data in order to obtain better balance in our sample. The process of matching consists in finding the most similar control observation for each treated observa- tion. This is done by designing segments (or grids) in the sample based on variables that are thought to be associated with the treatment. Once the entire sample is segmented, the matching process prunes all the control observations which are in segments with no treatment observations (i.e., they do not share com- mon characteristics with at least one treated observation). Segments with at least one treatment and one control observation are kept and form a matched subsample of the data. We can then perform standard causal inference on this subsample to estimate the “pure” treatment effect which we have detailed in the body of the report. We report the results obtained when proceeding to CEM after detailing those obtained with a more stand- ard procedure, i.e. logistic regressions of recovery on a set of household characteristics without matching data. As such we can compare regression results with and without the use of CEM. Table 28 displays the results for a logistic regression of recovery on losing more than 5% of annual expenditure and additional control variables. We find that without matching, the coefficient associated to the binary variable of los- ing more than the 5% threshold is not significant. 52 Table 28: Probability of having recovered after two years, as a function of various housing characteristics, magnitude of losses. N.B. Logistic regression is run with the original sample of affected households (no matching). (1) (2) (3) (4) (5) (6) Did not lose more than 5% of house- hold expenditure due to the flood Lost more than 5% of household ex- penditure due to the flood -0.354 -0.360 -0.433 -0.436 -0.458 -0.527 (0.376) (0.379) (0.385) (0.386) (0.382) (0.381) Log of real per capita expenditure -0.130 -0.0363 -0.0851 (0.236) (0.227) (0.237) Main income source: Monthly salary Main income source: Casual labor -1.244** -1.239** -1.126** (0.493) (0.492) (0.502) Main income source: Hawking -0.0798 -0.0900 -0.165 (0.516) (0.520) (0.598) Main income source: Remittances -0.0635 -0.0681 -0.678 (0.418) (0.418) (0.657) Main income source: Stable business -0.159 -0.165 -0.146 (0.292) (0.292) (0.298) Main income source: Other -0.216 -0.224 -0.592 (0.856) (0.856) (0.809) Does not anyone to borrow money from in case of emergency Can borrow money from at least one or multiple persons in case of emer- gency 0.675** 0.551* (0.304) (0.306) Did not receive remittances in the past year Received remittances in the past year 0.883*** 1.000** (0.318) (0.431) Constant 0.875*** 1.983 1.173*** 1.485 0.935 0.593 (0.160) (1.995) (0.322) (1.946) (2.043) (0.381) Observations 393 393 393 393 393 393 Prob > F 0.354 0.549 0.252 0.368 0.00269 0.0119 53 After matching the data with the CEM algorithm on two household characteristics (age and sex of house- hold head), we find that losing more than 5% of annual income significantly decreases the chance of re- covering from the flood. This is true whether we control for other household characteristics such as real per capita expenditure or if we estimate the “pure” treatment effect through a univariate logistic regres- sion. In additional regression sets, we also include important household characteristics such as the main source of income and the density of social networks to determine the impacts of these variables. Table 9 in the main text displays the full results of this regression. Results for the effect of having lost more than 10% of annual expenditure on recovery, with the addition of covariates, are presented in Table 29 below. For this regression, the impact of the CEM are even more tangible. Indeed, without proceeding to CEM we find no statistically significant relationship between rel- ative loss and recovery whereas after matching the net treatment effect (i.e. without even adding controls) is negative and highly significant. This can be viewed by comparing the coefficients in Table 29 and in Table 30. Table 29: Probability of having recovered after two years, as a function of various housing characteristics and magnitude of losses. N.B. Logistic regression is run with the original sample of affected households (no matching). (1) (2) (3) (4) (5) (6) Did not lose more than 10% of household expenditure due to the flood Lost more than 10% of household ex- penditure due to the flood -0.693 -0.701 -0.747 -0.754 -0.772 -0.810 (0.491) (0.495) (0.518) (0.519) (0.493) (0.505) Log of real per capita expenditure -0.132 -0.0364 -0.0836 (0.238) (0.230) (0.239) Main income source: Monthly salary Main income source: Casual labor -1.226** -1.222** -1.097** (0.500) (0.500) (0.511) Main income source: Hawking -0.0844 -0.0946 -0.177 (0.529) (0.534) (0.595) Main income source: Remittances -0.0313 -0.0361 -0.628 (0.430) (0.431) (0.655) Main income source: Stable business -0.151 -0.157 -0.127 (0.304) (0.304) (0.310) Main income source: Other -0.231 -0.239 -0.597 (0.866) (0.865) (0.823) Does not anyone to borrow money from in case of emergency 54 Can borrow money from at least one or multiple persons in case of emer- gency 0.668** 0.548* (0.302) (0.303) Did not receive remittances in the past year Received remittances in the past year 0.883*** 0.988** (0.316) (0.418) Constant 0.886*** 2.005 1.163*** 1.477 0.923 0.566 (0.169) (2.012) (0.340) (1.968) (2.056) (0.385) Observations 393 393 393 393 393 393 Prob > F 0.170 0.297 0.201 0.293 0.00376 0.0168 Table 30: Probability of having recovered after two years, as a function of various housing characteristics and magnitude of losses. N.B. Logistic regression is run after matching the data with CEM. (1) (2) (3) (4) (5) Did not lose more than 10% of household expenditure due to the flood Lost more than 10% of household ex- penditure due to the flood -0.954*** -1.014*** -1.002*** -1.041*** -1.106*** (0.367) (0.366) (0.381) (0.379) (0.386) Log of real per capita expenditure -0.176 -0.113 -0.107 (0.195) (0.205) (0.217) Main income source: Monthly salary Main income source: Casual labor -1.014*** -0.997*** -0.912** (0.380) (0.383) (0.394) Main income source: Hawking 0.0395 0.0160 -0.0101 (0.668) (0.669) (0.690) Main income source: Remittances 0.125 0.111 -0.516 (0.548) (0.549) (0.614) Main income source: Stable business -0.150 -0.161 -0.184 (0.285) (0.285) (0.291) Main income source: Other -0.217 -0.226 -0.646 (0.618) (0.618) (0.654) Does not anyone to borrow money from in case of emergency 55 Can borrow money from at least one or multiple persons in case of emer- gency 0.388 (0.250) Did not receive remittances in the past year Received remittances in the past year 0.956*** (0.299) Constant 0.810*** 2.312 1.029*** 2.005 1.467 (0.117) (1.674) (0.249) (1.771) (1.879) Observations 393 393 393 393 393 Prob > F 0.00957 0.0278 0.0140 0.0298 0.00136 One caveat of this method is the fact that estimation results depend on how the matching is specified be- forehand. When including more variables to match treated and control units, the estimate of the treatment effect sometimes changes, both in value and significance. We perform a k-to-k match, i.e. keeping only one control observation for each treated observation. Sam- ple size for the matched sample when balancing on the 5% loss threshold dropped to 182 observations (91 pairs or matches) and to 78 observations when balancing the 10% loss threshold (39 pairs or matches). After proceeding to a k-to-k match in the sample of affected households, and running identical regres- sions, the significant negative relationship between the fact of having lost more than 5% (resp. 10%) re- mains, even when we do not control for other variables. This is independent of the matching specification done prior to the logistic regression. Table 31: Probability of having recovered after two years, as a function of various housing characteristics and magnitude of losses. N.B. Logistic regression is run after matching the data with CEM on a k-to-k match. (1) (2) (3) (4) (5) Did not lose more than 5% of household expenditure due to the flood Lost more than 5% of household expendi- ture due to the flood -0.501 -0.543* -0.564* -0.589* -0.664** (0.309) (0.307) (0.318) (0.317) (0.318) Log of real per capita expenditure -0.168 -0.102 -0.0980 (0.186) (0.197) (0.209) Main income source: Monthly salary - Main income source: Casual labor -1.145*** -1.129*** 1.048*** (0.393) (0.396) (0.406) Main income source: Hawking -0.128 -0.150 -0.249 56 (0.655) (0.656) (0.682) Main income source: Remittances 0.0547 0.0426 -0.546 (0.544) (0.546) (0.605) Main income source: Stable business -0.164 -0.174 -0.196 (0.287) (0.288) (0.295) Main income source: Other -0.186 -0.197 -0.603 (0.605) (0.605) (0.645) Does not anyone to borrow money from in case of emergency Can borrow money from at least one or multiple persons in case of emergency 0.435* (0.253) Did not receive remittances in the past year Received remittances in the past year 0.905*** (0.298) Constant 0.817*** 2.260 1.071*** 1.951 1.425 (0.133) (1.595) (0.267) (1.708) (1.815) Observations 393 393 393 393 393 Prob > F 0.107 0.164 0.0447 0.0762 0.00335 Table 32: Probability of having recovered after two years, as a function of various housing characteristics and magnitude of losses. N.B. Logistic regression is run after matching the data with CEM on a k-to-k match. (1) (2) (3) (4) (5) Did not lose more than 10% of household expenditure due to the flood Lost more than 10% of household expendi- ture due to the flood -0.951*** -1.011*** -0.998*** -1.037*** -1.094*** (0.367) (0.367) (0.382) (0.381) (0.389) Log of real per capita expenditure -0.187 -0.119 -0.111 (0.193) (0.202) (0.212) Main income source: Monthly salary Main income source: Casual labor -1.125*** -1.103*** -1.014** (0.394) (0.397) (0.407) Main income source: Hawking -0.151 -0.175 -0.274 (0.658) (0.659) (0.683) Main income source: Remittances 0.0870 0.0726 -0.510 (0.553) (0.555) (0.614) Main income source: Stable business -0.175 -0.186 -0.200 (0.288) (0.289) (0.294) Main income source: Other -0.214 -0.226 -0.625 (0.614) (0.614) (0.654) 57 Does not anyone to borrow money from in case of emergency Can borrow money from at least one or mul- tiple persons in case of emergency 0.422* (0.255) Did not receive remittances in the past year Received remittances in the past year 0.901*** (0.300) Constant 0.807*** 2.404 1.055*** 2.079 1.510 (0.117) (1.649) (0.254) (1.742) (1.844) Observations 393 393 393 393 393 Prob > F 0.00975 0.0242 0.0103 0.0212 0.00115 Appendix 10 – Risk perceptions and investment behaviors Table 33: Flood risk perception by (1) flood exposure and location (low or high elevation), (2) flood expo- sure, location and expenditure. Dependent variable: very unlikely, unlikely, indifferent (0) VS likely, very likely (1). (1) (2) Affected: No Affected: Yes 2.444*** 2.469*** (0.233) (0.238) Location: High Elevation Location: Low Elevation 0.155 0.101 (0.330) (0.336) Income -3.77e-05 (2.51e-05) Constant -1.717*** -1.466*** (0.337) (0.374) Table 34: Propensity to invest in housing by (1) actual flood exposure in 2015, (2) location, (3) flood ex- posure and expenditure, (4) flood exposure, expenditure and flood risk perception, (5) flood exposure, ex- penditures and tenure arrangement. (1) (2) (3) (4) (5) Unaffected Affected 1.106*** 1.122*** 0.747*** 1.091*** (0.209) (0.209) (0.245) (0.207) High elevation Low elevation 0.0367 58 (0.314) Expenditure 4.54e-05** 5.13e-05** 4.25e-05** (2.13e-05) (2.14e-05) (2.00e-05) Flood risk perception: Unlikely/Indifferent Flood risk perception: Likely 0.722*** (0.211) Tenure: Owner Tenure: Own dwelling but not land -0.470 (0.604) Tenure: Rent from private person -0.0927 (0.375) Tenure: Rent from relative 0.779 (0.633) Tenure: Rent-free with relative -0.248 (0.387) Tenure: Other -0.000226 (0.440) Constant -1.563*** -1.025*** -1.857*** -2.037*** -1.728*** (0.169) (0.259) (0.223) (0.215) (0.384) Observations 1,008 1,008 1,008 1,008 1,008 Prob > F 3.50e-06 0.908 1.71e-05 2.39e-07 4.06e-05 Table 35: Propensity to invest in housing by (1) actual flood exposure and amount of rent payed (in cedis/month), (2) exposure, expenditure and amount of rent payed, (3) exposure, expenditure, amount of rent payed and flood risk perception. (1) (2) (3) Unaffected ref. ref. ref. Affected 1.089*** 1.080*** 0.883** (0.324) (0.323) (0.387) Monthly rent -0.00483 -0.00522 -0.00477 (0.00308) (0.00333) (0.00334) Expenditure 1.99e-05 2.34e-05 (2.41e-05) (2.42e-05) Flood risk perception: Un- likely/Indifferent ref. Flood risk perception: Likely 0.445 (0.288) Constant -1.158*** -1.270*** -1.408*** (0.201) (0.231) (0.266) 59 Observations 486 486 486 Prob > F 0.00539 0.0223 0.000834 Table 36: Has carried out enterprise investment in the last 12 months by flood exposure Affected No Yes Difference Significance level Enterprise investment 27.0% 18.6% -8.3% Observations 375 212 Table 37 includes results from a multinomial logit regression to assess further the relationship between investments and risk. In this regression model the base outcome has been set to households that invest solely in their house to ease interpretation for our purposes. As such, being affected has a negative impact on enterprise investment compared to the base outcome of sole housing investment. Meaning, affected business-owning households are less likely to invest in their enterprise and more likely to invest in their house than business-owning households that were not affected. Furthermore, the regression table below shows other interesting findings, namely the fact that when a business-owning household has been af- fected by the flood it chooses housing investment not only over enterprise investment but also over inac- tivity (no investment) and investment in both house and enterprise. This highlights the urge affected households feel to improve and/or repair their house after the flooding event. Finally, the role of expendi- ture is noteworthy as it positively impacts the choice to invest/repair in the house rather than not conduct- ing any type of investment. This is in line with our previous findings concerning the role of expenditure in foregoing housing investments (cf. supra and regression results in Table 34). However, we must be care- ful in interpreting these results as numbers are low for the concerned outcomes. Table 37 Multinomial logit regression results for trade-off between housing and enterprise investments for affected business-owning households Investment in house Investment in en- Invest in both house No investment only (base outcome) terprise only and enterprise Unaffected Affected -0.892*** ref. -1.750*** -0.544 (0.23) ref. (0.605) (0.391) Expenditure -4.93E-05 ref. -1.73E-05 1.41E-05 (0.0000335) ref. (0.0000386) (0.0000288) 60 Poverty & Equity Global Practice Working Papers (Since July 2014) The Poverty & Equity Global Practice 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. This series is co‐published with the World Bank Policy Research Working Papers (DECOS). It is part of a larger effort by the World Bank to provide open access to its research and contribute to development policy discussions around the world. For the latest paper, visit our GP’s intranet at http://POVERTY. 1 Estimating poverty in the absence of consumption data: the case of Liberia Dabalen, A. L., Graham, E., Himelein, K., Mungai, R., September 2014 2 Female labor participation in the Arab world: some evidence from panel data in Morocco Barry, A. G., Guennouni, J., Verme, P., September 2014 3 Should income inequality be reduced and who should benefit? redistributive preferences in Europe and Central Asia Cojocaru, A., Diagne, M. F., November 2014 4 Rent imputation for welfare measurement: a review of methodologies and empirical findings Balcazar Salazar, C. F., Ceriani, L., Olivieri, S., Ranzani, M., November 2014 5 Can agricultural households farm their way out of poverty? Oseni, G., McGee, K., Dabalen, A., November 2014 6 Durable goods and poverty measurement Amendola, N., Vecchi, G., November 2014 7 Inequality stagnation in Latin America in the aftermath of the global financial crisis Cord, L., Barriga Cabanillas, O., Lucchetti, L., Rodriguez‐Castelan, C., Sousa, L. D., Valderrama, D. December 2014 8 Born with a silver spoon: inequality in educational achievement across the world Balcazar Salazar, C. F., Narayan, A., Tiwari, S., January 2015 Updated on June 2018 by POV GP KL Team | 1 9 Long‐run effects of democracy on income inequality: evidence from repeated cross‐sections Balcazar Salazar,C. F., January 2015 10 Living on the edge: vulnerability to poverty and public transfers in Mexico Ortiz‐Juarez, E., Rodriguez‐Castelan, C., De La Fuente, A., January 2015 11 Moldova: a story of upward economic mobility Davalos, M. E., Meyer, M., January 2015 12 Broken gears: the value added of higher education on teachers' academic achievement Balcazar Salazar, C. F., Nopo, H., January 2015 13 Can we measure resilience? a proposed method and evidence from countries in the Sahel Alfani, F., Dabalen, A. L., Fisker, P., Molini, V., January 2015 14 Vulnerability to malnutrition in the West African Sahel Alfani, F., Dabalen, A. L., Fisker, P., Molini, V., January 2015 15 Economic mobility in Europe and Central Asia: exploring patterns and uncovering puzzles Cancho, C., Davalos, M. E., Demarchi, G., Meyer, M., Sanchez Paramo, C., January 2015 16 Managing risk with insurance and savings: experimental evidence for male and female farm managers in the Sahel Delavallade, C., Dizon, F., Hill, R., Petraud, J. P., el., January 2015 17 Gone with the storm: rainfall shocks and household well‐being in Guatemala Baez, J. E., Lucchetti, L., Genoni, M. E., Salazar, M., January 2015 18 Handling the weather: insurance, savings, and credit in West Africa De Nicola, F., February 2015 19 The distributional impact of fiscal policy in South Africa Inchauste Comboni, M. G., Lustig, N., Maboshe, M., Purfield, C., Woolard, I., March 2015 20 Interviewer effects in subjective survey questions: evidence from Timor‐Leste Himelein, K., March 2015 21 No condition is permanent: middle class in Nigeria in the last decade Corral Rodas, P. A., Molini, V., Oseni, G. O., March 2015 22 An evaluation of the 2014 subsidy reforms in Morocco and a simulation of further reforms Verme, P., El Massnaoui, K., March 2015 Updated on June 2018 by POV GP KL Team | 2 23 The quest for subsidy reforms in Libya Araar, A., Choueiri, N., Verme, P., March 2015 24 The (non‐) effect of violence on education: evidence from the "war on drugs" in Mexico Márquez‐Padilla, F., Pérez‐Arce, F., Rodriguez Castelan, C., April 2015 25 “Missing girls” in the south Caucasus countries: trends, possible causes, and policy options Das Gupta, M., April 2015 26 Measuring inequality from top to bottom Diaz Bazan, T. V., April 2015 27 Are we confusing poverty with preferences? Van Den Boom, B., Halsema, A., Molini, V., April 2015 28 Socioeconomic impact of the crisis in north Mali on displaced people (Available in French) Etang Ndip, A., Hoogeveen, J. G., Lendorfer, J., June 2015 29 Data deprivation: another deprivation to end Serajuddin, U., Uematsu, H., Wieser, C., Yoshida, N., Dabalen, A., April 2015 30 The local socioeconomic effects of gold mining: evidence from Ghana Chuhan-Pole, P., Dabalen, A., Kotsadam, A., Sanoh, A., Tolonen, A.K., April 2015 31 Inequality of outcomes and inequality of opportunity in Tanzania Belghith, N. B. H., Zeufack, A. G., May 2015 32 How unfair is the inequality of wage earnings in Russia? estimates from panel data Tiwari, S., Lara Ibarra, G., Narayan, A., June 2015 33 Fertility transition in Turkey—who is most at risk of deciding against child arrival? Greulich, A., Dasre, A., Inan, C., June 2015 34 The socioeconomic impacts of energy reform in Tunisia: a simulation approach Cuesta Leiva, J. A., El Lahga, A., Lara Ibarra, G., June 2015 35 Energy subsidies reform in Jordan: welfare implications of different scenarios Atamanov, A., Jellema, J. R., Serajuddin, U., June 2015 36 How costly are labor gender gaps? estimates for the Balkans and Turkey Cuberes, D., Teignier, M., June 2015 37 Subjective well‐being across the lifespan in Europe and Central Asia Bauer, J. M., Munoz Boudet, A. M., Levin, V., Nie, P., Sousa‐Poza, A., July 2015 Updated on June 2018 by POV GP KL Team | 3 38 Lower bounds on inequality of opportunity and measurement error Balcazar Salazar, C. F., July 2015 39 A decade of declining earnings inequality in the Russian Federation Posadas, J., Calvo, P. A., Lopez‐Calva, L.‐F., August 2015 40 Gender gap in pay in the Russian Federation: twenty years later, still a concern Atencio, A., Posadas, J., August 2015 41 Job opportunities along the rural‐urban gradation and female labor force participation in India Chatterjee, U., Rama, M. G., Murgai, R., September 2015 42 Multidimensional poverty in Ethiopia: changes in overlapping deprivations Yigezu, B., Ambel, A. A., Mehta, P. A., September 2015 43 Are public libraries improving quality of education? when the provision of public goods is not enough Rodriguez Lesmes, P. A., Valderrama Gonzalez, D., Trujillo, J. D., September 2015 44 Understanding poverty reduction in Sri Lanka: evidence from 2002 to 2012/13 Inchauste Comboni, M. G., Ceriani, L., Olivieri, S. D., October 2015 45 A global count of the extreme poor in 2012: data issues, methodology and initial results Ferreira, F.H.G., Chen, S., Dabalen, A. L., Dikhanov, Y. M., Hamadeh, N., Jolliffe, D. M., Narayan, A., Prydz, E. B., Revenga, A. L., Sangraula, P., Serajuddin, U., Yoshida, N., October 2015 46 Exploring the sources of downward bias in measuring inequality of opportunity Lara Ibarra, G., Martinez Cruz, A. L., October 2015 47 Women’s police stations and domestic violence: evidence from Brazil Perova, E., Reynolds, S., November 2015 48 From demographic dividend to demographic burden? regional trends of population aging in Russia Matytsin, M., Moorty, L. M., Richter, K., November 2015 49 Hub‐periphery development pattern and inclusive growth: case study of Guangdong province Luo, X., Zhu, N., December 2015 50 Unpacking the MPI: a decomposition approach of changes in multidimensional poverty headcounts Rodriguez Castelan, C., Trujillo, J. D., Pérez Pérez, J. E., Valderrama, D., December 2015 51 The poverty effects of market concentration Rodriguez Castelan, C., December 2015 52 Can a small social pension promote labor force participation? evidence from the Colombia Mayor program Pfutze, T., Rodriguez Castelan, C., December 2015 Updated on June 2018 by POV GP KL Team | 4 53 Why so gloomy? perceptions of economic mobility in Europe and Central Asia Davalos, M. E., Cancho, C. A., Sanchez, C., December 2015 54 Tenure security premium in informal housing markets: a spatial hedonic analysis Nakamura, S., December 2015 55 Earnings premiums and penalties for self‐employment and informal employees around the world Newhouse, D. L., Mossaad, N., Gindling, T. H., January 2016 56 How equitable is access to finance in turkey? evidence from the latest global FINDEX Yang, J., Azevedo, J. P. W. D., Inan, O. K., January 2016 57 What are the impacts of Syrian refugees on host community welfare in Turkey? a subnational poverty analysis Yang, J., Azevedo, J. P. W. D., Inan, O. K., January 2016 58 Declining wages for college‐educated workers in Mexico: are younger or older cohorts hurt the most? Lustig, N., Campos‐Vazquez, R. M., Lopez‐Calva, L.‐F., January 2016 59 Sifting through the Data: labor markets in Haiti through a turbulent decade (2001‐2012) Rodella, A.‐S., Scot, T., February 2016 60 Drought and retribution: evidence from a large‐scale rainfall‐indexed insurance program in Mexico Fuchs Tarlovsky, Alan., Wolff, H., February 2016 61 Prices and welfare Verme, P., Araar, A., February 2016 62 Losing the gains of the past: the welfare and distributional impacts of the twin crises in Iraq 2014 Olivieri, S. D., Krishnan, N., February 2016 63 Growth, urbanization, and poverty reduction in India Ravallion, M., Murgai, R., Datt, G., February 2016 64 Why did poverty decline in India? a nonparametric decomposition exercise Murgai, R., Balcazar Salazar, C. F., Narayan, A., Desai, S., March 2016 65 Robustness of shared prosperity estimates: how different methodological choices matter Uematsu, H., Atamanov, A., Dewina, R., Nguyen, M. C., Azevedo, J. P. W. D., Wieser, C., Yoshida, N., March 2016 66 Is random forest a superior methodology for predicting poverty? an empirical assessment Stender, N., Pave Sohnesen, T., March 2016 67 When do gender wage differences emerge? a study of Azerbaijan's labor market Tiongson, E. H. R., Pastore, F., Sattar, S., March 2016 Updated on June 2018 by POV GP KL Team | 5 68 Second‐stage sampling for conflict areas: methods and implications Eckman, S., Murray, S., Himelein, K., Bauer, J., March 2016 69 Measuring poverty in Latin America and the Caribbean: methodological considerations when estimating an empirical regional poverty line Gasparini, L. C., April 2016 70 Looking back on two decades of poverty and well‐being in India Murgai, R., Narayan, A., April 2016 71 Is living in African cities expensive? Yamanaka, M., Dikhanov, Y. M., Rissanen, M. O., Harati, R., Nakamura, S., Lall, S. V., Hamadeh, N., Vigil Oliver, W., April 2016 72 Ageing and family solidarity in Europe: patterns and driving factors of intergenerational support Albertini, M., Sinha, N., May 2016 73 Crime and persistent punishment: a long‐run perspective on the links between violence and chronic poverty in Mexico Rodriguez Castelan, C., Martinez‐Cruz, A. L., Lucchetti, L. R., Valderrama Gonzalez, D., Castaneda Aguilar, R. A., Garriga, S., June 2016 74 Should I stay or should I go? internal migration and household welfare in Ghana Molini, V., Pavelesku, D., Ranzani, M., July 2016 75 Subsidy reforms in the Middle East and North Africa Region: a review Verme, P., July 2016 76 A comparative analysis of subsidy reforms in the Middle East and North Africa Region Verme, P., Araar, A., July 2016 77 All that glitters is not gold: polarization amid poverty reduction in Ghana Clementi, F., Molini, V., Schettino, F., July 2016 78 Vulnerability to Poverty in rural Malawi Mccarthy, N., Brubaker, J., De La Fuente, A., July 2016 79 The distributional impact of taxes and transfers in Poland Goraus Tanska, K. M., Inchauste Comboni, M. G., August 2016 80 Estimating poverty rates in target populations: an assessment of the simple poverty scorecard and alternative approaches Vinha, K., Rebolledo Dellepiane, M. A., Skoufias, E., Diamond, A., Gill, M., Xu, Y., August 2016 Updated on June 2018 by POV GP KL Team | 6 81 Synergies in child nutrition: interactions of food security, health and environment, and child care Skoufias, E., August 2016 82 Understanding the dynamics of labor income inequality in Latin America Rodriguez Castelan, C., Lustig, N., Valderrama, D., Lopez‐Calva, L.‐F., August 2016 83 Mobility and pathways to the middle class in Nepal Tiwari, S., Balcazar Salazar, C. F., Shidiq, A. R., September 2016 84 Constructing robust poverty trends in the Islamic Republic of Iran: 2008‐14 Salehi Isfahani, D., Atamanov, A., Mostafavi, M.‐H., Vishwanath, T., September 2016 85 Who are the poor in the developing world? Newhouse, D. L., Uematsu, H., Doan, D. T. T., Nguyen, M. C., Azevedo, J. P. W. D., Castaneda Aguilar, R. A., October 2016 86 New estimates of extreme poverty for children Newhouse, D. L., Suarez Becerra, P., Evans, M. C., October 2016 87 Shedding light: understanding energy efficiency and electricity reliability Carranza, E., Meeks, R., November 2016 88 Heterogeneous returns to income diversification: evidence from Nigeria Siwatu, G. O., Corral Rodas, P. A., Bertoni, E., Molini, V., November 2016 89 How liberal is Nepal's liberal grade promotion policy? Sharma, D., November 2016 90 Pro-growth equity: a policy framework for the twin goals Lopez-Calva, L. F., Rodriguez Castelan, C., November 2016 91 CPI bias and its implications for poverty reduction in Africa Dabalen, A. L., Gaddis, I., Nguyen, N. T. V., December 2016 92 Building an ex ante simulation model for estimating the capacity impact, benefit incidence, and cost effectiveness of child care subsidies: an application using provider‐level data from Turkey Aran, M. A., Munoz Boudet, A., Aktakke, N., December 2016 93 Vulnerability to drought and food price shocks: evidence from Ethiopia Porter, C., Hill, R., December 2016 94 Job quality and poverty in Latin America Rodriguez Castelan, C., Mann, C. R., Brummund, P., December 2016 95 With a little help: shocks, agricultural income, and welfare in Uganda Mejia‐Mantilla, C., Hill, R., January 2017 Updated on June 2018 by POV GP KL Team | 7 96 The impact of fiscal policy on inequality and poverty in Chile Martinez Aguilar, S. N., Fuchs Tarlovsky, A., Ortiz‐Juarez, E., Del Carmen Hasbun, G. E., January 2017 97 Conditionality as targeting? participation and distributional effects of conditional cash transfers Rodriguez Castelan, C., January 2017 98 How is the slowdown affecting households in Latin America and the Caribbean? Reyes, G. J., Calvo‐Gonzalez, O., Sousa, L. D. C., Castaneda Aguilar, R. A., Farfan Bertran, M. G., January 2017 99 Are tobacco taxes really regressive? evidence from Chile Fuchs Tarlovsky, A., Meneses, F. J., March 2017 100 Design of a multi‐stage stratified sample for poverty and welfare monitoring with multiple objectives: a Bangladesh case study Yanez Pagans, M., Roy, D., Yoshida, N., Ahmed, F., March 2017 101 For India's rural poor, growing towns matter more than growing cities Murgai, R., Ravallion, M., Datt, G., Gibson, J., March 2017 102 Leaving, staying, or coming back? migration decisions during the northern Mali conflict Hoogeveen, J. G., Sansone, D., Rossi, M., March 2017 103 Arithmetics and Politics of Domestic Resource Mobilization Bolch, K. B., Ceriani, L., Lopez‐Calva, L.‐F., April 2017 104 Can Public Works Programs Reduce Youth Crime? Evidence from Papua New Guinea’s Urban Youth Employment Project Oleksiy I., Darian N., David N., Sonya S., April 2017 105 Is Poverty in Africa Mostly Chronic or Transient? Evidence from Synthetic Panel Data Dang, H.‐A. H., Dabalen, A. L., April 2017 106 To Sew or Not to Sew? Assessing the Welfare Effects of the Garment Industry in Cambodia Mejía‐Mantilla, C., Woldemichael, M. T., May 2017 107 Perceptions of distributive justice in Latin America during a period of falling inequality Reyes, G. J., Gasparini, L. C., May 2017 108 How do women fare in rural non‐farm economy? Fuje, H. N., May 2017 109 Rural Non‐Farm Employment and Household Welfare: Evidence from Malawi Adjognon, G. S., Liverpool‐Tasie, S. L., De La Fuente, A., Benfica, R. M., May 2017 Updated on June 2018 by POV GP KL Team | 8 110 Multidimensional Poverty in the Philippines, 2004‐13: Do Choices for Weighting, Identification and Aggregation Matter? Datt, G., June 2017 111 But … what is the poverty rate today? testing poverty nowcasting methods in Latin America and the Caribbean Caruso, G. D., Lucchetti, L. R., Malasquez, E., Scot, T., Castaneda, R. A., June 2017 112 Estimating the Welfare Costs of Reforming the Iraq Public Distribution System: A Mixed Demand Approach Krishnan, N., Olivieri, S., Ramadan, R., June 2017 113 Beyond Income Poverty: Nonmonetary Dimensions of Poverty in Uganda Etang Ndip, A., Tsimpo, C., June 2017 114 Education and Health Services in Uganda: Quality of Inputs, User Satisfaction, and Community Welfare Levels Tsimpo Nkengne, C., Etang Ndip, A., Wodon, Q. T., June 2017 115 Rental Regulation and Its Consequences on Measures of Well‐Being in the Arab Republic of Egypt Lara Ibarra, G., Mendiratta, V., Vishwanath, T., July 2017 116 The Poverty Implications of Alternative Tax Reforms: Results from a Numerical Application to Pakistan Feltenstein, A., Mejia‐Mantilla, C., Newhouse, D. L., Sedrakyan, G., August 2017 117 Tracing Back the Weather Origins of Human Welfare: Evidence from Mozambique? Baez Ramirez, J. E., Caruso, G. D., Niu, C., August 2017 118 Many Faces of Deprivation: A multidimensional approach to poverty in Armenia Martirosova, D., Inan, O. K., Meyer, M., Sinha, N., August 2017 119 Natural Disaster Damage Indices Based on Remotely Sensed Data: An Application to Indonesia Skoufias, E., Strobl, E., Tveit, T. B., September 2017 120 The Distributional Impact of Taxes and Social Spending in Croatia Inchauste Comboni, M. G., Rubil, I., October 2017 121 Regressive or Progressive? The Effect of Tobacco Taxes in Ukraine Fuchs, A., Meneses, F. September 2017 122 Fiscal Incidence in Belarus: A Commitment to Equity Analysis Bornukova, K., Shymanovich, G., Chubrik, A., October 2017 Updated on June 2018 by POV GP KL Team | 9 123 Who escaped poverty and who was left behind? a non‐parametric approach to explore welfare dynamics using cross‐sections Lucchetti, L. R., October 2017 124 Learning the impact of financial education when take-up is low Lara Ibarra, G., Mckenzie, D. J., Ruiz Ortega, C., November 2017 125 Putting Your Money Where Your Mouth Is Geographic Targeting of World Bank Projects to the Bottom 40 Percent Öhler, H., Negre, M., Smets, L., Massari, R., Bogetić, Z., November 2017 126 The impact of fiscal policy on inequality and poverty in Zambia De La Fuente, A., Rosales, M., Jellema, J. R., November 2017 127 The Whys of Social Exclusion: Insights from Behavioral Economics Hoff, K., Walsh, J. S., December 2017 128 Mission and the bottom line: performance incentives in a multi-goal organization Gine, X., Mansuri, G., Shrestha, S. A., December 2017 129 Mobile Infrastructure and Rural Business Enterprises Evidence from Sim Registration Mandate in Niger Annan, F., Sanoh, A., December 2017 130 Poverty from Space: Using High-Resolution Satellite Imagery for estimating Economic Well-Being Engstrom, R., Hersh, J., Newhouse, D., December 2017 131 Winners Never Quit, Quitters Never Grow: Using Text Mining to measure Policy Volatility and its Link with Long-Term Growth in Latin America Calvo-Gonzalez, O., Eizmendi, A., Reyes, G., January 2018 132 The Changing Way Governments talk about Poverty and Inequality: Evidence from two Centuries of Latin American Presidential Speeches Calvo-Gonzalez, O., Eizmendi, A., Reyes, G., January 2018 133 Tobacco Price Elasticity and Tax Progressivity In Moldova Fuchs, A., Meneses, F., February 2018 134 Informal Sector Heterogeneity and Income Inequality: Evidence from the Democratic Republic of Congo Adoho, F., Doumbia, D., February 2018 135 South Caucasus in Motion: Economic and Social Mobility in Armenia, Azerbaijan and Georgia Tiwari, S., Cancho, C., Meyer, M., February 2018 Updated on June 2018 by POV GP KL Team | 10 136 Human Capital Outflows: Selection into Migration from the Northern Triangle Del Carmen, G., Sousa, L., February 2018 137 Urban Transport Infrastructure and Household Welfare: Evidence from Colombia Pfutze, T., Rodriguez-Castelan, C., Valderrama-Gonzalez, D., February 2018 138 Hit and Run? Income Shocks and School Dropouts in Latin America Cerutti, P., Crivellaro, E., Reyes, G., Sousa, L., February 2018 139 Decentralization and Redistribution Irrigation Reform in Pakistan’s Indus Basin Jacoby, H.G., Mansuri, G., Fatima, F., February 2018 140 Governing the Commons? Water and Power in Pakistan’s Indus Basin Jacoby, H.G., Mansuri, G., February 2018 141 The State of Jobs in Post-Conflict Areas of Sri Lanka Newhouse, D., Silwal, A. R., February 2018 142 “If it’s already tough, imagine for me…” A Qualitative Perspective on Youth Out of School and Out of Work in Brazil Machado, A.L., Muller, M., March 2018 143 The reallocation of district-level spending and natural disasters: evidence from Indonesia Skoufias, E., Strobl, E., Tveit, T. B., March 2018 144 Gender Differences in Poverty and Household Composition through the Life-cycle A Global Perspective Munoz, A. M., Buitrago, P., Leroy de la Briere, B., Newhouse, D., Rubiano, E., Scott, K., Suarez-Becerra, P., March 2018 145 Analysis of the Mismatch between Tanzania Household Budget Survey and National Panel Survey Data in Poverty & Inequality Levels and Trends Fuchs, A., Del Carmen, G., Kechia Mukong, A., March 2018 146 Long-Run Impacts of Increasing Tobacco Taxes: Evidence from South Africa Hassine Belghith, N.B., Lopera, M. A., Etang Ndip, A., Karamba, W., March 2018 147 The Distributional Impact of the Fiscal System in Albania Davalos, M., Robayo-Abril, M., Shehaj, E., Gjika, A., March 2018 148 Analysis Growth, Safety Nets and Poverty: Assessing Progress in Ethiopia from 1996 to 2011 Vargas Hill, R., Tsehaye, E., March 2018 149 The Economics of the Gender Wage Gap in Armenia Rodriguez-Chamussy, L., Sinha, N., Atencio, A., April 2018 Updated on June 2018 by POV GP KL Team | 11 150 Do Demographics Matter for African Child Poverty? Batana, Y., Cockburn, J., May 2018 151 Household Expenditure and Poverty Measures in 60 Minutes: A New Approach with Results from Mogadishu Pape, U., Mistiaen, J., May 2018 152 Inequality of Opportunity in South Caucasus Fuchs, A., Tiwari, S., Rizal Shidiq, A., May 2018 153 Welfare Dynamics in Colombia: Results from Synthetic Panels Balcazar, C.F., Dang, H-A., Malasquez, E., Olivieri, S., Pico, J., May 2018 154 Social Protection in Niger: What Have Shocks and Time Got to Say? Annan, F., Sanoh, A., May 2018 155 Quantifying the impacts of capturing territory from the government in the Republic of Yemen Tandon, S., May 2018 156 The Road to Recovery: The Role of Poverty in the Exposure, Vulnerability and Resilience to Floods in Accra Erman, A., Motte, E., Goyal, R., Asare, A., Takamatsu, S., Chen, X., Malgioglio, S., Skinner, A., Yoshida, N., Hallegatte, S., June 2018 157 Small Area Estimation of Poverty under Structural Change Lange, S., Pape, U., Pütz, P., June 2018 For the latest and sortable directory, available on the Poverty & Equity GP intranet site. http://POVERTY WWW.WORLDBANK.ORG/POVERTY Updated on June 2018 by POV GP KL Team | 12