Lake Chad Regional Economic Memorandum  |  Development for Peace Technical Paper 2. Climate Change, Rural Livelihoods, and Urbanization: Evidence from the Permanent Shrinking of Lake Chad Remi Jedwab (George Washington University), Federico Haslop (George Washington University), Takaaki Masaki (World Bank), and Carlos Rodríguez-Castelán (World Bank) 114 Technical Paper 2. Climate Change, Rural Livelihoods and Urbanization: Evidence from the Permanent Shrinking of Lake Chad 3.1 Introduction There is a vast economic literature studying the effects Other major examples of drying lakes include the Aral of climate change on long-run growth, migration, Sea (Kazakhstan and Uzbekistan), formerly the fourth urbanization and human capital, among several other largest lake in the world, and Lake Urmia (Iran), formerly outcomes. A sizable portion of this literature has been the largest lake of the Middle East. Both shrunk to less dedicated to the study of weather trends and shocks, and than 10 percent of their former size. However, the reasons droughts in particular—an area of research of increasing for their aridification differ. For Lake Chad, aridification importance in a world that is projected to become came from long-term climate change. For the two other considerably drier by the end of the century (see C.-E. lakes, human actions were responsible. First, the Aral Sea Park et al. 2018; S. Hsiang and Kopp 2018). dried up because its feeding rivers were diverted by Soviet irrigation projects. Second, Lake Urmia dried up because Nevertheless, very little attention has been paid in the its feeding rivers were dammed. In contrast, as explained literature to how aridification can impact livelihoods below, Lake Chad dried up for mostly geographical—and through the disappearance of lakes and other water locally exogenous—reasons. resources. Lake Chad, once the second largest wetland in Africa (Hutchinson et al. 1992), lost about 90 percent of Understanding the local economic effects of drying its surface water area—around 23,000 sq km—between lakes is not straightforward, as lake recessions can the mid-1960s and the mid-1980s. This is equivalent to have ambiguous effects. On the one hand, a receding the total area of 4,200 American football stadiums, in just lake frees up arable land that can be used for farming. 20 years. Alternatively, 23,000 sq km is about 10 percent On the other hand, a receding lake can negatively impact more than the total area of El Salvador, Israel or Slovenia. fishing communities, farmers that rely on the lake’s While its water level has been slightly recovering since the waters for their irrigation needs, and cattle herders who mid-1990s, it is still on average 80 percent less than in the need the lake’s waters and the vegetation around it so mid-1960s. The resultant increasingly harsh environment, that their cattle can drink and eat enough. In poor and in the absence of climate change adaptation measures, has poorly connected countries, urban communities may also led to the development of self-serving political elites in rely on the lake for transporting goods. Furthermore, as northern Nigeria and furthered the eruption of the Boko a lake keeps drying, the arable land that was originally Haram conflict (Onyia 2015). freed up may also aridify. All these factors may result into intensified competition over limited resources, potentially It is all the more important to study this research feeding into conflicts. At the same time, in the longer question as lakes are important economic assets for run, residents of lake shore areas can potentially adopt various developing regions and countries of the world, adaptation strategies that help them mitigate the impact such as the Caspian Sea in Eastern Europe, Central Asia, of lake recessions. The short and long run effects of lake and Western Asia, or Lake Victoria, Lake Tanganyika, recessions might thus differ. Lake Malawi, Lake Bangweulu and Lake Turkana in East Africa. Importantly, Africa has ten of the fifty largest lakes In this paper we analyze how the shrinking of Lake in the world.252 Before shrinking, Lake Chad was the 11th Chad affected local economic development—as proxied largest lake in the world and the 4th largest lake in Africa. by local population growth in the absence of better 252 1Other large lakes are usually found in North America, Russia or Central Asia. 3.1 Introduction 115 Lake Chad Regional Economic Memorandum  |  Development for Peace data—and urbanization—city population growth—in Figure 3.1:  Evolution of Lake Chad’s Total Surface Water Area (sq km), 1950–2020 the already vulnerable sub-Saharan African region of the Sahel, trying to understand the economic consequences Total surface water area (sq.km.), in thousands 30– Large Lake of a geographical phenomenon that may become more Chad Phase Shrinking Lake Chad Phase Recovered Lake Chad Phase and more common as the world becomes drier and drier. 25– We focus our analysis on three low-income countries 20– whose territory borders Lake Chad: Cameroon, Chad 15– and Niger  (see Map 3.1). In the three countries, the 1960s marked the beginning of a crisis for the lake. Figure 10– 3.1 presents the evolution of the lake’s total surface water 5– area from 1950 to 2020, showing an enormous decrease in size between 1965 and 1985 and partial recovery after 0– 1995. In our analysis, we thus divide the full period into 1950 1960 1970 1980 1990 2000 2010 2020 Notes: Total surface water area was obtained from the following sources: Olivry et al. 1996, three subperiods: pre-1965 (Large Lake Chad), 1965– Sédick n.d., FAO 2009, Comission du Bassin du Lac Tchad 2015, Okpara, Stringer, and Dougill 2016 and Ighobor 2019. 1995 (Shrinking Lake Chad), and post-1995 (Recovering Lake Chad). Next, we treat the shrinkage of the lake as Map 3.1: Location of Lake Chad and Subdistrict Boundaries for the Three Countries of Study Notes: This figure shows consistent reconstructed subdistrict boundaries for the three countries of study for the period circa 1950s–2010s. Cameroon, Chad and Niger are divided into 113, 138 and 119 subdistricts, respectively. Lake Chad is shown in the center of the map. We also show the location of the capital (and most populated city) of Niger (Niamey) and Chad (N’Djamena). For Cameroon, we show its capital city (Yaoundé) as well as its most populated city (Douala) today. Finally, we indicate the location of Lake Fitri (in Chad). 116 3.1 Introduction Technical Paper 2. Climate Change, Rural Livelihoods and Urbanization: Evidence from the Permanent Shrinking of Lake Chad an ”exogenous” shock, as the shrinkage was not driven by decline observed in the area. Cities might then have acted local economic or geographical conditions but reduced as a safety valve sector for economic refugees from the rainfall in a fourth country, the Central African Republic Lake Chad area. We observe different population growth (CAR) (see Map 3.1), thus assuaging reverse causality patterns for smaller (5K+) and larger (20K+) cities in the concerns. Indeed, the Logone and Chari rivers flow three countries, possibly due to different initial urban from the CAR through Chad into Lake Chad. Once one conditions. For example, in Niger, the disaster led to controls for proximity to these rivers, the shrinkage of the urban concentration in 20K+ cities. Finally, only in Niger lake could thus be considered “exogenous”. Alternatively, did the government disproportionately build higher- one can focus on Niger, the country most distant from quality roads towards Lake Chad. Yet, it does not appear the CAR. As such, the shrinkage of Lake Chad offers a that such investments mitigated the impact of the shock natural experiment that helps us examine how long- term on the total population. lake drying can affect both rural and urban communities. Our paper contributes to the literature on the socio- To conduct our analysis, we construct a novel data economic and demographic impact of climate change. set tracking total population patterns at a fine spatial Previous works have focused on the effects of rainfall or —113, 138 and 119 subdistricts in Cameroon, level droughts on a wide range of development outcomes, Chad and Niger, respectively—and city population such as migration (Gray and Mueller 2012; Rosenzweig patterns—166, 179 and 100 cities in Cameroon, Chad and Udry 2014), urbanization (Barrios, Bertinelli, and and Niger, respectively—from the 1950s to the 2010s. Strobl 2006; Henderson, Storeygard, and Deichmann We then use (relative) total population growth as our 2017), civil conflict (Harari and Ferrara 2018) and main outcome of interest, finding in a panel-difference- education (Maccini and D. Yang 2009; Shah and in-difference (DiD) framework for the three countries: Steinberg 2017). Other studies have also examined how (i) no differential effect of proximity to the lake before increases in temperature may drive rural-to-urban or 1965 (Large Lake Chad period); (ii) a substantial negative within-country migration (Bohra-Mishra, Oppenheimer, effect of proximity to the lake in 1965–1995 (Shrinking and S. M. Hsiang 2014; Partridge, Feng, and Rembert Lake Chad period); and (iii) an effect that remains 2017), international migration (Beine and Parsons strongly negative post-1995, despite the slow recovery 2015; Cattaneo and Peri 2016; Baez et al. 2017; Jessoe, of the lake’s water level (Recovering Lake Chad period). Manning, and Taylor 2017; Peri and Sasahara 2019), Our results suggest that fishing communities, farmers conflict (Eberle, Rohner, and Thoenig 2020), agricultural and cattle herders were negatively impacted by the lake output (Schlenker, Hanemann, and Anthony C. Fisher receding. As incomes probably decreased in the area, 2005; Schlenker, Hanemann, and Anthony C. Fisher households likely outmigrated to other areas. 2006; Deschénes and Greenstone 2007; Anthony C Fisher et al. 2012; Burke and Emerick 2016; Aragón, In addition, we study how the shrinkage of the lake Oteiza, and Rud 2021; S. Chen and Gong 2021; Steve impacted nearby urban communities. Using the same Miller et al. 2021), economic growth (Dell, Jones, and panel-DiD framework but studying city population Olken 2012), exports (Jones and Olken 2010; Kalemli- growth instead of total population growth, we find that Özcan, Nikolsko–Rzhevskyy, and Kwak 2020), mortality city population sizes increased (however, not significantly (Deschénes and Moretti 2009; Deschénes and Greenstone so) or remained stable in the long run. Hence, it is 2011; Barreca et al. 2015), and birth weight (Deschénes, suggested that (relative) rural population decline has been Greenstone, and Guryan 2009).253 the main component of the (relative) total population 253 See Tol 2009 and Dell, Jones, and Olken 2014 for a review of the literature on the impacts of climate change. 3.1 Introduction 117 Lake Chad Regional Economic Memorandum  |  Development for Peace Our paper is also related to the literature on natural The findings of our paper are also relevant to disasters and their impact on development. There understanding the economic effects of natural is already a well-established body of literature on resources. The literature has shown that the presence— how natural disasters may affect various development or discovery—of natural resources can be a blessing outcomes, including international migration (Mahajan (Aragón and Rud 2013; Arezki, Ramey, and Sheng 2016; and D. Yang 2020; Spitzer, Tortorici, and Zimran 2020; Allcott and Keniston 2017) or a curse for development Beine and Parsons 2015), domestic migration (J. J. Chen (Torvik 2002; Ploeg 2011; Venables 2016, Armand et et al. 2017; see Gröger and Zylberberg 2016 for Vietnam; al. 2020) depending on local contexts. Natural resources Bohra-Mishra, Oppenheimer, and S. M. Hsiang 2014, can also be a source of conflict and instability (Berman Kirchberger 2017, and Kleemans and Magruder 2017 for et al. 2017), even destabilizing the security situation of Indonesia; and Boustan, Kahn, and Rhode 2012 for the neighboring regions (Caselli, Morelli, and Rohner 2015; US), human capital outcomes (G. Caruso and Sebastian Adhvaryu et al. 2021). Our study sheds a new light on Miller 2015; G. D. Caruso 2017) and urban activity the nexus between natural resources (or lack thereof ) (Gallagher and Hartley 2017; Brooks and Donovan and development by studying how the withdrawal of 2020; Kocornik-Mina et al. 2020) among others. water resources due to lake shrinkage may disrupt local economies in areas near the lake and thereby hamper What sets our work apart from these existing studies urban growth. on the socio-economic impact of climate change or natural disasters is that we study the shrinkage of a Finally, we focus our analysis on three countries that lake as another important natural disaster shock are among the poorest in the world. Understanding the that explains long-term urbanization patterns in its effects of climate change and natural disasters in such neighboring regions. Lake disappearances are interesting, contexts is particularly important. Chad and Niger are and important, cases to study in and of themselves. Most then two Sahelian countries and are as such likely to be existing studies have examined the effects of climate very negatively impacted by climate change in the future, change by investigating the local economic effects of hence the need for more research on the effects of “past” weather-related shocks, in particular rainfall, temperature climate change events on their economies and societies.254 and humidity shocks. While weather-related shocks are most often locally exogenous, they are often temporary The paper is structured as follows: Section 3.2 dives shocks. It is much more difficult to find cases of permanent into some of the physical characteristics of Lake Chad shocks, such as a lake almost entirely drying over a period and its water sources. Section 3.3 introduces our novel of 20 years. To some extent, our shock resembles much data. Section 3.4 presents the hypothesis and empirical more the main object of such studies, i.e. climate change, strategy behind our analysis. Sections 3.5, 3.6 and 3.7 a permanent change in climate conditions. In addition, present results on total population, cities and roads, unlike existing studies on coastal flooding, which typically respectively. Finally, section 3.8 concludes. leads to crop losses and/or destruction in cities, we study the effects of lake recessions. Lake recessions have in theory more ambiguous effects, because some valuable land may become newly available. 254 The only few studies regarding the local effects of a smaller Lake Chad are (non-economics) articles that rely on contemporary small-sample village surveys to provide very detailed, but also very local, analyses of the situation (see, for example, Sarch and Charon Birkett 2000; Okpara, Stringer, and Dougill 2016, Luxereau, Genthon, and Karimou 2012). Despite the importance of these studies to understand how a smaller lake has affected households in the area, these analyses have many shortcomings that our paper addresses (despite the limitations of our own analysis). To the best of our knowledge, this is the first (economics) paper that identifies the short and long-term causal effects of a shrinking lake on local economic development, both in rural and urban settings. 118 3.1 Introduction Technical Paper 2. Climate Change, Rural Livelihoods and Urbanization: Evidence from the Permanent Shrinking of Lake Chad 3.2 Background: The Lake Chad and Its Tributaries About 90 percent of Lake Chad’s water comes from The fact that the Lake’s water level is primarily the Chari-Logone river system  (see Map 3.2). The determined by rainfall in another country south of our river system then primarily originates from rainfall in region of study provides reassurance that the results the mountainous areas of the Central African Republic will not be explained by reverse causality. Nonetheless, (CAR) (Hutchinson et al. 1992). Rainfall in the Adamawa this does not rule out other potential sources of bias. Highlands of Cameroon also somewhat contribute to the Because the Logone-Chari river system goes through the system. Because the water inflow of the lake depends territory of Cameroon and Chad, outcomes in the Lake’s almost exclusively on the Logone-Chari system, lack of surroundings may not be independent of outcomes rainfall over the CAR was by far the main reason behind upriver. Indeed, the same shock that affected the lake’s the large drop in water area observed after 1965 (Figure residents—dryer rivers due to lower rainfall in CAR— 3.1). may have also affected other households along the river Map 3.2: Major and Minor Rivers of the Chari-Logone River System Feeding Lake Chad Notes: The Chari River and its tributary, the Logone, provide almost all of Lake Chad’s water. The Chari River flows from the Central African Republic (shown in the map) through Chad into Lake Chad. We show in bold the main rivers of the Logone-Chari system (Shapefiles obtained from the Landscape Portal). In grey, we show other streams associated with the Logone-Chari system (Shapefiles obtained from FAO/GeoNetwork). We also show Lake Fitri. 3.2 Background: The Lake Chad and Its Tributaries 119 Lake Chad Regional Economic Memorandum  |  Development for Peace system. In that case, and in the case of Chad in particular, water (CM Birkett 2000). We will exploit this fact when areas farther away from the Lake are also directly affected. studying the effects for Chad, the only country where the If the Lake areas—i.e., the “more treated” group— two pools are present, thus expecting stronger effects for and the river areas—i.e., the “less treated” group—are the northern pool. similarly affected—either positively or negatively—then the magnitude of the estimated effect will likely be under- estimated. For these reasons, we will control for proximity to the Logone-Chari river system. In addition, the fact that the river system is not present in Niger implies that this country presents the “cleanest” environment to test our hypothesis, as the shock was more “exogenous” there than anywhere else. Furthermore, because the Logone- Chari system occupies a smaller share of Cameroon’s territory than Chad’s territory, Cameroon provides a more “exogenous” setting than Chad. However, we will find relatively similar results for the three countries, at least when studying total population levels. This gives us confidence that we are effectively controlling for any potential bias generated by the river system. This also ensures that we are not picking up an effect due to country-specific institutions or spatial policies. Finally, this reinforces the external validity of our results, especially considering that Cameroon is wealthier than Chad and Niger. Another characteristic of Lake Chad that must be considered is the heterogeneous degrees of dryness that were experienced over its different regions. Cutting Lake Chad in half lies what is called the Grande Barrie`re, an elevated area that in dry years divides the lake in a southern pool and a northern pool. Because the majority of the lake’s water enters through the south (via the Logone-Chari river system), it is only when the water level of the southern pool is high enough that water crosses the Grande Barrie`re to replenish the northern pool. During the Large Lake Chad era (pre-1965), this geological feature remained submerged, rendering it irrelevant. However, as the Logone-Chari river’s discharge rate declined, the Grande Barrie`re created a northern sink that dried almost completely in the 1980s (Okpara, Stringer, and Dougill 2016), and a southern sink that, although smaller in size, always retained an area of open 120 3.2 Background: The Lake Chad and Its Tributaries Technical Paper 2. Climate Change, Rural Livelihoods and Urbanization: Evidence from the Permanent Shrinking of Lake Chad 3.3 Data for the Reduced-Form Analyses For our analysis, we use subdistrict- and district-level (often administrative censuses). Values for the years 1988, data for Cameroon, Chad and Niger, some of the 2001 and 2012 come from the population censuses that poorest countries in the world. Unfortunately, data took place those years.257 availability is extremely scarce. However, any analysis of the impact of the lake shrinking demands localized data The Cameroon dataset includes the years 1963, 1967, for the period 1965–1995 (shrinking Lake Chad phase) 1976, 1987 and 2005. Information for 1963 and 1967 and the pre-1965 period (large Lake Chad phase). Due to comes from administrative sources. Population figures for this, total and urban population figures are the best (and the years 1976, 1987 and 2005 are based on population only) measures available.255 census counts. Unfortunately, no census has taken place since 2005. For example, the 2018 population census was postponed indefinitely. 3.3.1 Total Population Levels for the Subdistrict Samples The Chad dataset includes the years 1948, 1953, 1965, 1993 and 2009. Population measures for the years 1948, Few population censuses took place in the three 1953 and 1965 are based on administrative sources. countries and when population data is available, it is For the year “1965”, we use information from the 1962 not at a fine spatial level like say counties in the U.S. administrative census and 1964 demographic survey as Typically, the sources that we were able to get ahold of our baseline. When needed, we adjust the population report population data at the regional or district level, and levels that we obtain using information from the 1968 sometimes at the “subdistrict” level. However, subdistrict administrative census. We call this year “1965” because boundaries are rarely consistent across years. As such, we 1965 is the mid-year between 1962 and 1968. Lastly, had to reaggregate subdistricts in order to reconstruct a set we use census population figures for the years 1993 and of consistently defined subdistricts over periods spanning 2009. Next, we excluded Nigeria from our analysis due more than 50 years. Overall, our reconstructed subdistrict to a long history of disputed census results. In fact, the dataset contains 119 units for Niger (1951–2012), 1962 and 1973 results were never officially validated 113 for Cameroon (1963–2005), and 138 for Chad and published due to various controversies surrounding (1948–2009). These subdistricts correspond to third- their reliability and accusations of political manipulation level administrative units, in particular arrondissements (Ahonsi 1988). in Cameroon, sous-prefectures in Chad and communes in Niger. More details on the sources and the assumptions Finally, Map 3.1 shows the boundaries of the made can be found in the Web Data Appendix.256 In the reconstructed subdistricts. As seen, subdistricts are of a case of Niger, we have total population data for the years similar size across the three countries. Mean area is 9.1, 1951, 1956, 1959, 1962, 1969, 1988, 2001, 2012, 2013 10.6 and 4.2 thousand sq km in Cameroon, Chad and and 2017. Data for the years 1951, 1956, 1959, 1962, Niger, respectively. In comparison, the mean U.S. county 1969, 2013 and 2017 come from administrative sources is 2.8 sq km. 255 The Demographic and Health Surveys of USAID and national household or labor force surveys are typically not available before the 1990s. Likewise, only the 1976, 1987 and 2005 population censuses of Cameroon are available on the website of IPUMS International. By 1976, the lake’s level was already quite low. We are thus missing a year of data before the lake started shrinking. Finally, nighttime lights are only available from the year 1992. 256 The name of the third-level administrative units is also not constant over time in each country. The names referred here are the ones used by each country in the reports of their latest population census. 257 Administrative censuses are population counts that rely on official registers and other national and local files. 3.3 Data for the Reduced-Form Analyses 121 Lake Chad Regional Economic Memorandum  |  Development for Peace 3.3.2 District Samples 1960). Post-1968, we rely on reports of the population census (1977, 1988, 2001, 2012). Next, while we know In the case of Cameroon, we only have subdistrict the population size of almost all cities and for almost all population data for one year (1963) before the lake years in 1977–2012, information for the years 1900– started shrinking. As such, we cannot investigate whether 1968 is more patchy. In particular, when Niger was the parallel trends assumption holds for Cameroon. As a still a colony as well as in the early years of the post- solution, we verify that it holds if we use instead total independence period, no census was conducted. Instead, population data at the district level. More precisely, administrators would sequentially visit various regions of we reaggregate the 113 Cameroonian subdistricts into the country to proceed with administrative population 47 districts, which allows us to add one year of pre-1965 counts, as in 1955–1962 and 1965–1968. As such, for data (1956; source = administrative census). Next, in our 16 localities that already had more than 5,000 inhabitants econometric analysis we will also include district-specific before 1968, population is typically available for different linear trends. years for different cities. To create a consistent population series for the pre-1968 period, we use exponential While we use the same 47 districts for Cameroon, we interpolations. use 31 districts for Niger and 36 districts for Chad. Note that the distribution of districts does not reflect the There are then a few cities for which we know their distribution of districts in any particular year. Indeed, population before the 1940s and in the late 1950s for district boundaries and subdistrict boundaries to be but not in-between. In order to better predict their consistent, and in order to also preserve consistency over population circa 1950 (we indeed focus on the post-1950 time, some aggregations had to be made. However, our period in our analysis), we also consider their pre-1950 boundaries more or less correspond to district boundaries population. in the 1960s.258 Next, for later years, there are a few cities for which the first population estimate available exceeds 5,000 3.3.3 City Population Sizes by several thousands. As a result, these cities might have exceeded 5,000 in the previous years of data as well but we To study urbanization, we need a consistent definition cannot be sure. To allow for this possibility, and for each of cities across the three countries and for all years city without any early population estimate, we assume available. As in many studies in the urban literature, we that their 1945 population one inhabitant and then define as a city any locality with at least 5,000 inhabitants. use exponential interpolation to fill the missing years. We thus focus our data compiling efforts on localities that As such, this increases the likelihood that a city exceeds reached the threshold of 5,000 inhabitants at any point 5,000 if its value is well above 5,000 the following year during our period of study. of data. For Niger, 166 localities reached 5,000 inhabitants at Overall, for city-years where the obtained population least once in 1900–2012.259 For the pre-1968 period, we is not above 5,000, we are confident based on our rely on colonial and post-colonial administrative reports analysis that population is indeed below that number. of city population sizes (Niger became independent in 258 Districts correspond to departements or prefectures in the three countries 259 In particular, we have city population estimates for the following years: 1900, 1905, 1910, 1921, 1926, 1931, 1934, 1936, 1945, 1948, 1951, 1955–1962, 1965–1968, 1977, 1988, 2001, and 2012. 122 3.3 Data for the Reduced-Form Analyses Technical Paper 2. Climate Change, Rural Livelihoods and Urbanization: Evidence from the Permanent Shrinking of Lake Chad Our methodological choices should also not affect the administrative counts (Chad became independent in results as few city-years are ultimately concerned.260 1960). For the years 1993 and 2009, we use reports of the population census. For the years 1975 and 2000, we We then proceed similarly for both Cameroon and use administrative population count estimates provided Chad. For Cameroon, 186 localities reached 5,000 at by Chad’s Institute of Statistics. any point in 1932–2005.261 For the pre-1976 years, we use colonial and post- colonial administrative counts (Cameroon became independent in 1960–1961). For 3.3.4 Geographical Proximity to Lake the years 1976, 1987 and 2005, we use reports of the Chad population census. For Chad, 100 localities reached 5,000 at any point in the period 1937–2009.262 For We obtain from the RCMRD Geoportal of the World the pre-1968 years, we use colonial and post-colonial Bank a shapefile of the full (pre-shrinking) Lake Chad Map 3.3: Location of the Selected Country-Specific Centroids of Lake Chad Notes: This figure shows the centroids of Lake Chad considered for each country. 260 For a limited sample of the cities, we know their exact population when it is below 5,000. However, we do not make use of that information due to possible endogenous selection issues in why an estimate is available or not. 261 The data set covers the years 1932, 1939, 1941, 1945, 1950, 1953, 1956, 1958-1968, 1970, 1976, 1987, and 2005. 262 The data set covers the years 1937, 1939–1951, 1954–1956, 1961, 1964, 1968, 1975, 1993, 2000, and 2009. 3.3 Data for the Reduced-Form Analyses 123 Lake Chad Regional Economic Memorandum  |  Development for Peace area. We then construct for each subdistrict/district centroid the Euclidean distance to various “centroids” in the Lake Chad polygon. In Niger, residents only have access to the northern pool of the Lake. Thus, the centroid that we consider is the centroid of the section of the northern pool that is within the territory of Niger (see Map 3.3). In Cameroon, residents only have access to the southern pool of the Lake. Thus, the centroid that we consider is the centroid of the section of the southern pool that is within the territory of Cameroon (ibid.). In Chad, residents have access to both pools. We thus consider: (i) The centroid of the section of the northern pool that is within Chad’s territory; (ii) The centroid of the section of the southern pool that is within Chad; and (iii) The centroid of the section of the full Lake Chad area that is within Chad. Next, we also consider the centroid of the full Lake Chad area, thus abstracting from country boundaries. Finally, our main measure of proximity to Lake Chad is the negative of the logged Euclidean distances from these centroids to the subdistrict/district centroids. 124 3.3 Data for the Reduced-Form Analyses Technical Paper 2. Climate Change, Rural Livelihoods and Urbanization: Evidence from the Permanent Shrinking of Lake Chad 3.4 Main Hypothesis and Specification 3.4.1 Main Hypothesis “It’s difficult to determine boundaries on water, yet the gendarmes [from Cameroon and Chad] always come The historical drop in water levels starting circa 1965 after us and seize our fishing nets and traps and we have and the relative recovery of the lake after 1995 allow us to pay heavily to get them back.” (Murray 2007). to examine how the shrinking of a lake affects nearby communities. To do so, we exploit a simple difference- 3.4.2. Baseline Specification in-difference framework and study the effect of proximity to the lake on total population patterns. In particular, For subdistricts s and years t and each country at a we expect proximity to the lake to have no effect on time, we estimate the following model: (relative) population growth before the shock (pre-1965) and possibly some effects during the main shock period ln(Total Pop.)s,t = α + ∑ βv × Prox. Lakes × Iv=t v (1965–1995) and the slow recovery period (post-1995). + λs + θt + Ds × t + XsBs,t + μd,t (1) A priori, the effects of a lake shrinking on nearby  here ln(Total Pop.)s,t is the log of total population in w populations can be ambiguous. On the one hand, the subdistrict s in year t and our variables of interest are shrinking of Lake Chad made available arable land that the interactions between the (time-invariant) measure of was unclaimed before, allowing villagers to switch at least proximity to the lake and year dummies (we omit the first part of their activities from fishing to farming (Sarch and year of data so the effect is estimated relative to it). We Charon Birkett 2000). However, this coping strategy may add subdistrict (λs) and year (θt) fixed effects, as well as have not been made available to all villages. Furthermore, district-specific linear trends (Ds × t) to control for local as the lake kept shrinking, land that became available patterns of economic development at the district level in the early years of the lake shrinking became farther over time. To account for spatial auto-correlation, we use and farther away from the lake shore, which increasingly Conley standard errors (distance cut-off of 100 km).263 limited irrigation possibilities. A smaller lake also reduces incomes in fishing communities. It can also impact cattle Furthermore, our specification includes several time- herding, an important sector in the Lake Chad region invariant controls (XsBs,t) that we interact with year (herders typically sell their cattle to urban markets in effects to flexibly allow them to have a different effect Nigeria). Indeed, herders require the lake’s water and the over time. We first add the logged Euclidean distances vegetation that grows around it. Finally, conflict within to the largest city as well as the capital city and their and between villages may also be a negative consequence square.264 Doing this allows us to flexibly control for of a smaller lake: as the lake dried and people moved spatial patterns of economic development that may closer to its shores, increased competition for resources be related to economic or political centralization (or could have led to social conflict (Okpara, Stringer, and decentralization). This is important in the case of Chad, Dougill 2016). The fact that four different countries as N’Djamena, its capital and largest city, is near the Lake share ownership over portions of the lake makes things Chad area. even more complicated. As put by a local fisherman: 263 With few years of data pre-shrinking, subdistrict-specific linear trends ask too much of the data. 264 The largest city (Douala) is indeed not the capital city (Yaoundé) in Cameroon. 3.4 Main Hypothesis and Specification 125 Lake Chad Regional Economic Memorandum  |  Development for Peace For historical reasons, northern areas are less developed, and have been growing slower, than southern areas in the three countries. Geographical differences are also correlated with latitude, with declining vegetation density as one moves north and, in the case of Chad and Niger, desertification in the Sahel and Sahara zones. To control for this North-South gradient, we include the latitude of the subdistrict’s centroid which we interact with year fixed effects. We add two dummies for whether the subdistrict is crossed by a river of the main Logone- Chari river system or a river of the extended Logone-Chari river system, which we both interact with year fixed effects. Doing so controls for local effects of changes in the discharge rate of the Logone-Chari river system. As discussed previously, decreases in the discharge rate that eventually led to the shrinking of Lake Chad might have also led to differential patterns of development along the streams of the river system, in both Cameroon and Chad. The river flow may be associated with local economic development via changes in vegetation or irrigation. Finally, classical measurement error in the dependent variable, for example due to issues with the reporting of population levels in the original sources and/or the reaggregation process that we submit the underlying data to, should only affect standard errors. If anything, precision should be reduced, especially for earlier years where population data might be less reliable. 126 3.4 Main Hypothesis and Specification Technical Paper 2. Climate Change, Rural Livelihoods and Urbanization: Evidence from the Permanent Shrinking of Lake Chad 3.5 Reduced-Form Effects on Total Population We first study the effects of Lake Chad shrinking on data, especially pre-1965. More precisely, we have 119 subdistrict total population. As seen in Figure 3.1, the subdistricts x 17 years = 2,023 observations. lake’s water level dramatically dropped between 1965 and 1985. While residents might have initially expected Table 3.1 presents the results. In col. (1), Lake the “shock” to be temporary, it became clear over time proximity is defined using the centroid of the section that the shock was permanent. Residents’ economic of the northern pool that is contained with Niger’s and migratory responses to the aridification of the lake territory. We see no effect in 1956–1962 (the omitted may have thus evolved over time. Effects observed in year is 1951), suggesting parallel trends. In 1969, we the later period then capture the realization that the see a large negative effect of -0.31**. Relative to the year shock was permanent but also adaptation strategies 1962, the last year of available data before 1965, this that households may have adopted. In addition, post- effect is -0.23**. This implies that halving the distance 1995, the lake started recovering, albeit slowly (and from the lake is associated with a 23 percent relative erratically). Next, we assume that (relative) population decline in population. By then, the full Lake Chad, not growth is a good proxy for (relative) economic growth just its area contained within Niger, had shrunk by about patterns. Within a same country, and abstracting from 22 percent. This effects becomes even more negative in worker heterogeneity, the spatial equilibrium hypothesis 1988, at -0.41*** relative to the year 1962, implying that implies that spatial population growth patterns are halving the distance from the Lake is associated with a mainly explained by differential evolutions of nominal 41 percent relative decline in population. The Lake water wages, prices and quality-of-life amenities (Topel 1986, levels had then collapsed by 91 percent. Furthermore, in Gollin, Kirchberger, and Lagakos 2017 and Chauvin et terms of standardized effects, a one standard deviation in al. 2017). However, at (very) low income levels as in our proximity to the lake is associated with a 0.27 and 0.48 context, amenities not directly related to prolonging life standard deviation decrease in log population by 1969 expectancy should matter little (Duranton 2016, Chauvin and 1988, respectively. et al. 2017 and Jedwab and Vollrath 2019). Thus, by measuring population growth patterns, we should be The effects remain negative after 1995. However, one capturing mainly an effect on real wages. In other words, should be cautions when interpreting the effects after a (relative) population increase in one location should 2009, since that year marks the start of the Boko Haram indicate (relative) real wage growth. Population is then insurrection in Northeastern Nigeria. This could have the most reliable measure to capture economic growth (or affected local development. However, Blankespoor et al. decline) around Lake Chad, as consistent information on 2020 show that Boko Haram had an impact on night real wages or employment at a fine spatial level does not lights in 2009–2012 but did not affect population levels exist in our context for most of the period of study. by 2012. In addition, our 2001 and 2012 estimates are similar, suggesting a lack of population recovery after the lake regained some of its past water levels. In the post- 3.5.1 Effects on Total Population for 1995 period and relative to the year 1962, the relative Niger, 1951–2017 decrease in population was about 36 percent. In the same period, Lake Chad’s level was still, on average, about Niger possibly offers the best environment for our 20 percent of its pre-1965 level. analysis. Its territory does not contain any river belonging to the Logone-Chari system and we have more years of 3.5 Reduced-Form Effects on Total Population 127 Lake Chad Regional Economic Memorandum  |  Development for Peace Table 3.1: Effect of Proximity to the Lake, Total Population, Niger 1950s–2010s Dependent Variable: Log Subdistrict Population in Year t Lake Centroid: North NER Full Lake North NER Full Lake Omitted Year = 1951 (1) (2) (1) Cont’d. (2) Cont’d. -0.01 -0.02 -0.41* -0.64* Proximity to Lake (log) 1956 Proximity to Lake (log) 2012 [0.03] [0.04] [0.24] [0.36] -0.05 -0.07 -0.41* -0.65* Proximity to Lake (log) 1957 Proximity to Lake (log) 2013 [0.04] [0.05] [0.24] [0.36] -0.06 -0.09 -0.42* -0.66* Proximity to Lake (log) 1958 Proximity to Lake (log) 2014 [0.04] [0.06] [0.24] [0.37] -0.07 -0.11 -0.42* -0.67* Proximity to Lake (log) 1959 Proximity to Lake (log) 2015 [0.05] [0.07] [0.25] [0.37] -0.08 -0.12 -0.43* -0.68* Proximity to Lake (log) 1960 Proximity to Lake (log) 2016 [0.06] [0.08] [0.25] [0.38] -0.08 -0.13 -0.44* -0.69* Proximity to Lake (log) 1961 Proximity to Lake (log) 2017 [0.06] [0.09] [0.25] [0.38] -0.08 -0.13 -0.23** -0.29** Proximity to Lake (log) 1962 β31969–β31962 [0.07] [0.09] [0.09] [0.14] -0.31** -0.42** -0.41*** -0.59*** Proximity to Lake (log) 1969 β31988–β31962 [0.13] [0.19] [0.11] [0.17] -0.49*** -0.72*** -0.36* -0.56* Proximity to Lake (log) 1988 β32017–β31962 [0.15] [0.22] [0.21] [0.32] -0.40** -0.61** Subdistrict (119) FE, Year (17) FE Y Y Proximity to Lake (log) 2001 [0.20] [0.30] District (31) Trends; Controls Y Y Notes: 119 subdistricts x 17 years = 2,023 obs. “North NER” is the centroid of the Northern section of Lake Chad that is within the territory of Niger (NER). “Full Lake” is the centroid of the full Lake Chad area. Conley SEs (100 Km). Next, Figure 3.2a shows the effects when omitting Finally, in col. (2), we show the effects when proximity 1962 instead of 1951 (95 percent confidence intervals to the Lake is calculated using the Euclidean distance are also reported). As seen, while the effects are strong to the centroid of the whole Lake Chad area, not just in the sense that their magnitude is high, standard the area within Niger. If anything, the effects are now errors are high as well. If anything and relative to the stronger.265 Because the “Full Lake” centroid is located point estimates, the standard errors increase in the later more southern than the Niger-specific centroid, it gives period when population censuses became more, not more weight to subdistricts at the border with Nigeria. less, reliable. As such, the wide confidence intervals in One possibility is that the shrinkage of Lake Chad also the shrinking and post-shrinking periods imply that the had negative effects in Northeastern Nigeria (before Boko observed average effects hide heterogeneous effects across Haram), which then impacted subdistricts in Niger. subdistricts. If we study the effects in 1988, the average effect is about -40 percent and the corresponding lower- bound and upper-bound effects are about -20 percent and -60 percent respectively. This will led us to investigate subdistrict-specific factors that may have driven the heterogeneity in the effects. 265 In terms of standardized effects, a one standard deviation in proximity to the lake is associated with a 0.50 and 0.85 standard deviation decrease in log population by 1969 and 1988, respectively (vs. 0.37 and 0.58 in col. (1)). 128 3.5 Reduced-Form Effects on Total Population Technical Paper 2. Climate Change, Rural Livelihoods and Urbanization: Evidence from the Permanent Shrinking of Lake Chad Figure 3.2: Total Population Effect of Proximity to the Lake Chad, 1940s–2010s a. Niger Subdistricts (N=119) (1951–2017) b. Cameroon Subdistricts (N=113) (1963–2005) Relative effect in year t (omitted year=1962) Relative effect in year t (omitted year=1963) 0.2– 0.1– 0– 0– -0.1– -0.2– -0.2– -0.3– -0.4– -0.4– -0.5– -0.6– -0.6– -0.8– -0.7– 1950 1960 1970 1980 1990 2000 2010 2020 1950 1960 1970 1980 1990 2000 2010 2020 Q Districts (47) Q Subdistricts (N=113) c. Chad Subdistricts (N=138) (1948–2009) d. Water Loss and Post-1965 Effects (1967–2012) Water level loss of Lake Chad since 1965, Post-1965 effect Relative effect in year t (omitted year=1965) percent (percent, relative to latest pre-1965 year) 0– –-10 -20– -0.1– -0.2– –-20 -40– -0.3– –-30 -0.4– -0.5– -60– –-40 -0.6– -0.7– -80– –-50 -0.8– -0.9– -100– –-60 1950 1960 1970 1980 1990 2000 2010 2020 1970 1975 1980 1985 1990 1995 2000 2005 2010 Q Water level loss Q Effect in Niger ´ Cameroon J Tchad Notes: Subfigures (a)-(c) show for vairous samples the effects of proximity to Lake Chad (relative to the omitted year shown at left). For subfigures (a), (b) and (c), the specifications are similar to Table 1 col. (1), Table 2 col. (1) and Table 2 col. (3), and Table 3 col. (1), respectively. However, the omitted year is the latest year available before 1965 (incl.) instead of the first year available as in the tables. The dashed vertical lines show the years the lake started to decline (c. 1965) and recover (c. 1995). We report 95 percent confidence intervals. Conley SEs (100 Km). In subfigure (d), we plot the estimated water level loss ( percent) of Lake Chad relative to the year 1965 and the estimated post-1965 effects ( percent) for the three countries in 1967–2012 (relative to the latest year available in each country). 3.5.2 Effects on Total Population for In Col. (1), lake proximity is defined with respect to Cameroon, 1963–2005 the centroid of the section of the southern pool that is contained within Cameroon’s territory. In 1967, we For Cameroon, we have 113 subdistricts x 5 years = 565 observe a negative effect of -0.19**, implying that halving observations. The results are presented in Table 3.2. The the distance from the lake is associated with a 19 percent effects are estimated relative to the omitted year (1963). relative decline in population. By then, the lake had Unfortunately, the second year of data is 1967. We have shrunk by 22 percent. In 1976, the effect is -0.27**, and thus only one pre-shock year and cannot examine parallel the lake had shrunk by 77 percent. In 1987, when the trends. Lake’s size had collapsed to about 10 percent of its pre- drought size, the effect is -0.40***. Alternatively, a one 3.5 Reduced-Form Effects on Total Population 129 Lake Chad Regional Economic Memorandum  |  Development for Peace Table 3.2: Effect of Proximity to the Lake, Total Population, Cameroon 1960s–2010s Dependent Variable: (1)–(2): Log Subdistrict Population in Year t; (3)–(4): Log District Population in Year t South CMR South CMR South CMR Lake Centroid: Full Lake Full Lake Full Lake Omitted Year = 1963; 1956 (3) (4) (1) (2) (3) (4) Cont’d. Cont’d. 0.27* 0.38* Proximity to Lake (log) 1963 [0.14] [0.20] -0.19** -0.36*** 0.03 0.02 -0.24*** -0.36*** Proximity to Lake (log) 1967 β31967 –β31963 [0.08] [0.13] [0.15] [0.23] [0.09] [0.12] -0.27** -0.55** -0.08 -0.16 -0.35* -0.55** Proximity to Lake (log) 1976 β31976 –β31963 [0.12] [0.22] [0.22] [0.31] [0.20] [0.26] -0.40*** -0.80*** -0.12 -0.23 -0.39** -0.61** Proximity to Lake (log) 1987 β31987 –β31963 [0.14] [0.27] [0.16] [0.23] [0.19] [0.26] Proximity to Lake (log) 2005 -0.36** -0.72** [0.16] [0.34] Unit (113; 47) FE, Year (5; 6) FE Y Y Y Y District (47) Trends; Controls Y Y Y Y Notes: (1)–(2): 113 subdistricts*5 yrs = 565 obs. (3)–(4): 47 districts * 6 yrs = 282 obs. South CMR = centroid of the Southern section of the lake that is within Cameroon’s territory. Full Lake = full lake’s centroid. Conley SEs (100 Km). standard deviation in proximity to the lake is associated 1963 (relative to 1956). If we omit 1963 instead of 1956, with a 0.68 and 1.01 standard deviation decrease in log we then observe very negative and significant effects post- population in 1976 and 1987, respectively. 1965 (last two columns). The effects are similar to the effects at the subdistrict level (first two columns). Figure In the post-1995 period (year 2005), when the lake had 3.2b then shows these results graphically. partially recovered from its size reduction (82 percent of its pre-1965 size), the negative effect of being close the lake still exists (-36 percent). The results from 3.5.3 Effects on Total Population for Col. (1) can also be seen graphically in Figure 3.2b. As Chad, 1948-2009 seen, standard errors increase over time, which suggests heterogeneity in the effects. Col. (2) then presents the For Chad, we have 138 subdistricts x 5 years = 690 results when lake proximity is constructed using the observations (the omitted year is 1948). Chad contains Euclidean distance to the centroid of the whole lake area. in its territory portions of both the northern and southern Results are now stronger, for possibly the same reasons as pools of Lake Chad. Because of the presence of the Grand in Niger. Barrie`re, the northern pool was particularly vulnerable to droughts. We thus expect strong effects for areas Because we only have one pre-1965 year, we cannot test close to the northern pool. Furthermore, households for parallel trends in the subdistrict dataset. Relying on who relied on resources in the northern pool area could districts instead, we add the year 1956 to the analysis (47 have migrated south closer to the southern pool, whose districts x 6 years = 282 observations). We then use the more eastern areas in Chad were never completely dry. same specification as for the subdistrict analysis except Negative population growth effects in the southern pool the omitted year is now 1956. However, since district- area could be thus partially, or more than, compensated specific trends are included, one of the interacted effects by migration from the northern pool area. In contrast, in cannot not estimated (we omit the year 2005). As seen in Niger and Cameroon where residents only had access to cols. (3)–(4), a positive, not negative, effect is observed in one pool, between-pool migration was not possible. 130 3.5 Reduced-Form Effects on Total Population Technical Paper 2. Climate Change, Rural Livelihoods and Urbanization: Evidence from the Permanent Shrinking of Lake Chad Table 3.3: Effect of Proximity to the Lake, Total Population, Chad 1948–2009 Dependent Variable: Log Subdistrict Population in Year t Lake Centroid in Chad: North Full South North Lake Fitri South North Omitted Year = 1948 (1) (2) (3) (4) (5) -0.03 -0.01 -0.06 0.07 -0.15* 0.00 0.08** Proximity to Lake (log) * 1953 [0.04] [0.05] [0.07] [0.06] [0.08] [0.04] [0.04] 0.23*** 0.18** 0.25** 0.24* -0.03 0.19** -0.13 Proximity to Lake (log) * 1965 [0.09] [0.09] [0.12] [0.14] [0.15] [0.09] [0.09] -0.29*** -0.03 0.10 -0.46** 0.17 -0.28** 0.03 Proximity to Lake (log) * 1993 [0.09] [0.12] [0.09] [0.23] [0.23] [0.11] [0.15] -0.37*** -0.14 -0.09 -0.22 -0.30 -0.36*** 0.02 Proximity to Lake (log) * 2009 [0.12] [0.17] [0.18] [0.17] [0.27] [0.13] [0.18] -0.52*** -0.22 -0.15 -0.70** 0.21 -0.47*** 0.16 β1993 – β1965 [0.14] [0.19] [0.11] [0.32] [0.23] [0.15] [0.12] -0.60*** -0.32† -0.35** -0.46*** -0.27 -0.55*** 0.15 β2009 – β1965 [0.15] [0.21] [0.16] [0.12] [0.20] [0.15] [0.13] Subdistrict (138) FE, Year (4) FE Y Y Y Y Y District (36) Trends, Controls Y Y Y Y Y Notes: 138 subdistricts x 5 years = 690 obs. “North TCD” = centroid of the Northern section of Lake Chad that is within the territory of Chad. “Full TCD” = centroid of the Lake Chad area that is within the territory of Chad. “South TCD” = centroid of the Southern section of Lake Chad that is within the territory of Chad. “Lake Fitri” = centroid of Lake Fitri (fully contained within Chad). Conley SEs (100 Km). † p<0.15, * p<0.10, ** p<0.05, *** p<0.01. In col. (1) of Table 3.3, proximity to Lake Chad is in population. By 2009, this decline was 60 percent. Now, defined using the logged Euclidean distance to the in terms of standardized effects, a one standard deviation centroid of the northern pool area that is within in proximity to the lake is associated in 1993 with a 0.94 Chad’s territory. In col. (2) we use the centroid of the standard deviation in log population. Finally, the effects whole lake area that is contained within Chad’s territory. when omitting the year 1965 are represented visually in In col. (3) we use the centroid of the southern pool area Figure 3.2c. that is within Chad’s territory. Lastly, in col. (4), we simultaneously consider the northern pool centroid and If we consider instead the centroid of the whole lake the southern pool centroid. area within Chad (col. (2)), the post-1965 effects are still negative, but not significant. If we only consider As seen in col. (1), there is no effect in 1953 (relative the centroid of the southern pool area within Chad (col. to 1948) but there is a positive effect by 1965, thus (3)), only the 2009 effect is negative and significant indicating a positive pre-trend. If anything, households (relative to the year 1965). Lastly, if we simultaneously were disproportionately settling close to the lake before consider the northern and southern pool effects (col. (4)), it started shrinking.266 We then observe a strong negative we find very negative post-1965 effects for the northern effect in 1993 which became even more negative by 2009. pool and a positive, but not significant, effect for the In 1993 (when the lake was 92 percent smaller), the effect southern pool in 1993, suggesting that some areas closer is -0.29***. Relative to the year 1965, the effect is even to the southern pool may have indeed received migrants stronger, at -0.52***, implying that halving the distance from the northern pool. In 2009, both effects are negative to the lake is associated with a 52 percent relative decline (however, not significantly so for the southern pool), 266 Recall that population data circa the year 1965 uses population data from the years 1962–1964 as a baseline. For about half of the country, information from 1968 is also used, hence the need to always include as a control a dummy if 1968 information was ever used to recreate subdistrict population, which we interact with year fixed effects. 3.5 Reduced-Form Effects on Total Population 131 Lake Chad Regional Economic Memorandum  |  Development for Peace possibly because households realized that the southern the water level loss ( percent) of Lake Chad relative to pool was on its way to become as permanently affected as the year 1965 as well as the post-1965 effects ( percent) the northern pool. of the three countries in 1967–2012, the effects are also relatively similar between Cameroon and Niger. The long- Next we utilize Lake Fitri as a placebo check of our term effects are then stronger in Chad, which contains a analysis of the effects of Lake Chad shrinking. The larger share of the lake than the other two countries. In location of Lake Fitri can be seen in Map 3.1. According the late 1980s, and using 1990 country populations as to R. Hughes, J. Hughes, and Bernacsek 1992, the lake is weights, we find an average relative population decline located in a seasonally inundated plain that is fed by the of 43 percent. Circa 2010 (thus using 2010 populations Batha river that carries water all the way from the East as weights), the average decline is almost unchanged at of the country and the Ouaddai massif in particular. The 41 percent. size of Lake Fitri thus depends on rainfall at the border between Chad and Sudan. While Lake Fitri’s water levels have changed over time, it has not shrunk like Lake 3.5.4 Alternative Analysis Using Distance Chad. Because Lake Fitri provides rural households with Bins and Population Reallocation similar livelihood possibilities as Lake Chad does (e.g., fishing, farming, and cattle herding), it provides a good For a given country-year, our baseline specification placebo test of whether the effects observed in Lake Chad allows us to compare population growth patterns for are a consequence of changes in lake-related economic locations closer vs. farther away from the lake. As activities for the whole region instead of local economic such, it has the advantage of making us estimate only effects limited to the Lake Chad area. one coefficient per country-year, which facilitates the exposition of the results. However, it does not tell us how In col. (5), we report results when simultaneously the effect varies with proximity to the lake. In particular, including the Lake Chad variables (based on the we could imagine different scenarios with population centroid of the northern pool) and year fixed effects reallocating to non-shore areas located not too far from interacted with proximity to Lake Fitri (the negative of the lake or non-shore areas located far away from the lake. the logged Euclidean distance to Lake Fitri’s centroid). As seen, no effects are observed for Lake Fitri. If anything, We thus use the model of eq. (1) but instead of having a positive and significant effect is observed in 1948–1953, only one variable capturing proximity to the lake hinting that populations were moving closer to this lake we now use several dummies based on the Euclidean before 1953. However, the 1965, 1993 and 2009 effects distance between a subdistrict’s centroid and the are similar to the 1953 effect, indicating stable local selected lake centroid. More precisely, the mean land population patterns after 1953. Thus, the effects observed area of the 113 Cameroonese subdistricts, 138 Chadian for Lake Chad are specific to the Lake Chad region. subdistricts and 119 Nigerien subdistricts is about 9.1, 10.6 and 4.2 thousand sq km, respectively. Were these To summarize, we find strong negative effects of subdistricts shaped like a circle, their diameter would be proximity to the lake during the shrinking period 108, 116 and 72 km, respectively. Thus, the distance- and limited recovery post-shrinking. The effects appear based bins cannot be too small (e.g., 0–100 and 100– causal as no negative pre-trends are observed and no 200 km), otherwise each bin would include very few negative effects are found for Lake Fitri, another important subdistricts, which would lead to less precisely estimated lake. Finally, the effects are strong in the three countries, effects. At the same time, if the bins are too large (e.g., despite these countries having different geographies and 0–250 and 250–500 km), we could miss local effects of institutions, which strengthens the internal and external the lake shrinking. validity of our results. As seen in Figure 3.2d that shows 132 3.5 Reduced-Form Effects on Total Population Technical Paper 2. Climate Change, Rural Livelihoods and Urbanization: Evidence from the Permanent Shrinking of Lake Chad For each country, if we restrict the sample to and 2010 (2012, 2005 and 2009, respectively)—after subdistricts whose Euclidean distance to the lake is which the lake had started recovering. More precisely, for below the median, we find that the 5th percentile value subdistricts s and years t and each country at a time, the in the distance to the lake is 157, 196 and 125 km for model is as follows: Niger, Cameroon and Chad, respectively. We thus use bins of 150 km. More precisely, we create dummies if the ln(Total Pop.)s,t = α + ∑ ∑ βv ,s × Bins × Iv=t s v subdistrict is located within 0–150, 150–300 and 300– + λs + θt + Ds × t + XsBs,t + μd,t (2) 450 km from the lake and interact the dummies with the year fixed effects.267  here ln(Total Pop.)s,t is the log of total population in w subdistrict s in year t and our variables of interest are the We omit the last year of available data before 1965 interactions between the three distance bin dummies (0– (incl.), so 1962, 1963 and 1965 for Niger, Cameroon 150, 150–300 and 300–450 km) and the year dummies. and Chad, respectively. The effects are thus estimated As before, we add subdistrict (λs) and year (θt) fixed relative to the early 1960s, just before the lake began effects, as well as district-specific linear trends (Ds × t) and shrinking. In Table 3.4, we then only report the interacted several time-invariant controls (XsBs,t) interacted with effects for the years closest to 1990 (1988, 1987 and year effects. We then use Conley standard errors (cut-off 1993, respectively)—at the end of the shrinking period— of 100 km). Table 3.4: Effect of Proximity to the Lake, Total Population, Flexible Specification Dependent Variable: Log Subdistrict Population in Year t Niger (North) Cameroon (South) Chad (North) Country (Centroid): Omitted Yr = Early 60s (1) (2) (3) (4) (5) (6) (7) (8) (9) -0.36*** -0.46*** -0.64*** -0.43*** -0.45** -1.22*** -0.56* -0.80** -0.82* 0–150 Km*ca.1990 [0.05] [0.08] [0.08] [0.15] [0.20] [0.23] [0.31] [0.37] [0.44] -0.42*** -0.60*** -0.08 -0.85*** -0.18 -0.19 150–300 Km*ca.1990 [0.09] [0.08] [0.10] [0.13] [0.15] [0.27] -0.07 -0.88*** 0.00 300–450 Km*ca.1990 [0.05] [0.13] [0.16] -0.21** -0.31** -0.63*** -0.45** -0.35 -1.41*** -0.23 -0.98*** -0.99*** 0–150 Km*ca.2010 [0.09] [0.15] [0.15] [0.21] [0.27] [0.27] [0.21] [0.24] [0.29] -0.1 -0.42*** 0.09 -0.98*** -0.95*** -0.96*** 150–300 Km*ca.2010 [0.11] [0.09] [0.06] [0.09] [0.19] [0.24] -0.05 -1.08*** 0.02 300–450 Km*ca.2010 [0.08] [0.06] [0.02] Subdistrict FE, Year FE Y Y Y Y Y Y Y Y Y District Trends, Ctrls Y Y Y Y Y Y Y Y Y Notes: Obs.: Niger (1951–2017): 119 subdist. x 17 yrs = 2,023. Cameroon (1963–2005): 113 subdist. x 5 yrs = 565. Chad (1948–2009): 138 subdist. x 5 yrs = 690. Niger (North) = centroid of the Northern section of Lake Chad that is within the territory of Chad. Cameroon (South) = centroid of the Southern section of the lake that is within Cameroon’s territory. Chad (North): centroid of the Northern section of Lake Chad that is within the territory of Chad. For Niger, Cameroon and Chad, we omit 1962, 1963 and 1965, respectively. We interact the 0–150, 150–300 and 300–450 km dummies with the year fixed effects but only report the interacted effects for the years closest to 1990 (1988, 1987 and 1993, respectively) and 2010 (2012, 2005 and 2009, respectively). Conley SEs (100 Km). † p<0.15, * p<0.10, ** p<0.05, *** p<0.01. 267 In Niger, Cameroon and Chad, these bins correspond to (2,3,14), (1,8,4) and (6,14,10) subdistricts, respectively. 3.5 Reduced-Form Effects on Total Population 133 Lake Chad Regional Economic Memorandum  |  Development for Peace For Niger (cols. (1)–(3)), we find strong negative effects for the 0–150 km bin. Circa 1990, the effect for the 150–300 km bin is then as strong as the effect for the 0–150 km bin. However, in the long run (c. 2010), the 150–300 km effect is smaller than the 0–150 km effect. Therefore, recovery was only partial and only concerned the subdistricts located slightly farther away from the lake. For Cameroon and only including the 0–150 km and 150–300 km dummies ((4)–(5)), we find a strong negative effect for the 0–150 km bin. This effect is then weaker, and not significant, in the long run ((5)). However, if we also include the 300–450 km dummies, then the effects become very negative until 450 km (incl.), both in the short-run and the long-run ((6)). For Chad ((7)–(9)) and the medium run (c. 1990), we find a strong negative and significant effect for the 0–150 km bin only. The effect remains as strong in the long run. By then, the 150–300 km bin effect had also become very negative and significant.268 Lastly, the last spatial lags included do not have positive significant coefficients. Thus, populations did not necessarily “reallocate” to the vicinity of disaster- struck locations. Instead, the populations that would have stayed in/moved to the areas close to the lake in the absence of the shock might have stayed in/moved to other areas of the country somewhat proportionally. 268 For Niger and Chad where we have several years of data before 1965, we also verify that the coefficients of the pre-1965 interactions are never positive and significant, which would suggest negative pre-trends (not shown). Some of the interactions have a negative and significant coefficient, indicating positive pre- trends for a few distance bins, consistent with the pre-trends already observed when using (-) log distance as our measure of proximity to the lake. 134 3.5 Reduced-Form Effects on Total Population Technical Paper 2. Climate Change, Rural Livelihoods and Urbanization: Evidence from the Permanent Shrinking of Lake Chad 3.6 Effects on Cities and “Refugee” Urbanization In the previous sections, we showed how the shrinking For our city population analysis, we use the same bin of Lake Chad negatively impacted total population specification (2) as for our total population estimations levels in the areas close to the lake. Since rural sectors except the unit of analysis is a city, which we define were heavily reliant on the lake’s water level, and since as a locality of at least 5,000 inhabitants. As shown the regions surrounding the lake were little urbanized in by Jedwab and Vollrath 2015, the mean population 1965, our interpretation of the effects is that the effect threshold used in the world to define cities is 4,500. was driven by rural decline, not urban decline. Next, we restrict the analysis to the post-1950 period, to consider the same period as for the total population To test that more formally, we now examine how cities analysis. Finally, we consider three dependent variables: were impacted. In particular, cities in the area might (i) the log of (city population + 1) in year t (cols. (1)–(3) have been negatively affected either directly, due to the of Table 3.5);269 and (ii) two dummies equal to one if the fact that the lake was used for commerce in a context city had already reached 5K (cols. (4)–(6)) or 20K (cols. of high road-based transportation costs and non-existent (7)–(9)) by year t, respectively. Note that we consider railroads in the area, or indirectly, because of the impact 20K to study large cities separately. on the rural sector—via reduced fishing, farming and cattle herding—which then impacted the urban sector For Niger, between 1965 and 2012, the respective via rural—urban linkages. We call this scenario the “the number of 5K+ cities and 20K+ cities increased from rural disaster-led urban underdevelopment scenario.” 14 to 161 and from 4 to 26. For Cameroon and the years 1965 and 2005, the same numbers increased from 51 to At the same time, if individuals see reduced economic 173 and from 10 to 54. For Chad and the years 1964 and opportunities in the rural sector close to the lake, they 2009, the same numbers increased from 11 to 94 and may transition to the urban sector and thus migrate from 4 to 23. Unlike in developed countries, there are to cities, which could spur urbanization. In that case, relatively few 20K+ cities. Their emergence likely captures cities grow because of a natural disaster, not economic local economic development in a different way than 5K+ development per se. We call this scenario the “the rural cities. The growth of 5K+ cities is more likely connected disaster-led urban development scenario.” to rural economic development. First, as rural agro-towns grow, they are more likely to pass the 5,000 threshold. Therefore, the effect of the lake shrinking on city Second, the economic sectors of small cities depend population growth is ambiguous. It could be negative disproportionately more on economic development in (first scenario) or positive (second scenario), for example their surrounding rural areas. For example, fishermen, depending on rural-urban linkages (i.e. how city growth farmers and cattle herders may purchase goods and is affected by rural economic decline) and the absorptive services from local cities, and local cities may serve as capacity of cities there (i.e. how negatively wages respond trading stations for the goods they produce and sell to to increased migration flows and labor supply). larger cities farther away. 269 If the locality’s population is less than 5,000, we replace it by 0. 3.6 Effects on Cities and “Refugee” Urbanization 135 Lake Chad Regional Economic Memorandum  |  Development for Peace Table 3.5: Effect of Proximity to the Lake, City Population, Flexible Specification Log City Population in t Dummy City Pop. ≥5K t Dummy City Pop. ≥20K t Dependent Variable: Omitted Year=Early 60s (1) (2) (3) (4) (5) (6) (7) (8) (9) Panel A: Niger (166 Cities x 17 Years = 2,822 Obs.) 3.34** 3.52** 4.09** 0.35** 0.37** 0.44** 0.13 0.14 0.14 0–150 Km*ca.1990 [1.56] [1.67] [1.77] [0.17] [0.18] [0.19] [0.14] [0.15] [0.15] 1.85 2.37 0.21 0.28 0.14 0.14 150–300 Km*ca.1990 [2.94] [3.02] [0.33] [0.34] [0.15] [0.14] 1.78** 0.22** 0.01 300–450 Km*ca.1990 [0.83] [0.09] [0.02] 0.20 0.18 0.68 -0.07 -0.08 -0.01 0.75** 0.76** 0.71** 0–150 Km*ca.2010 [1.89] [2.04] [2.13] [0.21] [0.23] [0.24] [0.32] [0.35] [0.35] -0.23 0.23 -0.03 0.03 0.10 0.05 150–300 Km*ca.2010 [1.65] [1.74] [0.18] [0.19] [0.31] [0.31] 1.57 0.21* -0.18*** 300–450 Km*ca.2010 [1.03] [0.11] [0.06] Panel B: Cameroon (179 Cities x 18 Years = 3,222 Obs.) 3.43*** 2.94 3.65* 0.34** 0.30 0.42* 0.18* 0.17 0.21** 0–150 Km*ca.1990 [1.24] [2.20] [2.05] [0.13] [0.22] [0.23] [0.10] [0.21] [0.08] -0.15 0.79 -0.01 0.13 0.02 0.05 150–300 Km*ca.1990 [1.27] [1.68] [0.12] [0.20] [0.16] [0.12] 0.34 0.08 0.04 300–450 Km*ca.1990 [2.38] [0.26] [0.23] -0.60 -2.52 0.00 -0.14 -0.33 0.00 0.14 0.08 0.00 0–150 Km*ca.2010 [1.17] [2.47] [0.00] [0.13] [0.24] [0.00] [0.19] [0.43] [0.00] -2.17* 0.65 -0.22 0.14 -0.06 -0.13 150–300 Km*ca.2010 [1.30] [1.69] [0.14] [0.18] [0.29] [0.18] 2.03 0.27 -0.08 300–450 Km*ca.2010 [2.29] [0.23] [0.43] Panel C: Chad (100 Cities x 12 Years = 1,200 Obs.) -1.13 -2.29* -4.26*** -0.10 -0.25* -0.46*** 0.03 0.10 0.16 0–150 Km*ca.1990 [1.46] [1.35] [1.44] [0.15] [0.15] [0.15] [0.02] [0.16] [0.20] -1.48 -3.65* -0.19 -0.41** 0.04 0.12 150–300 Km*ca.1990 [1.78] [2.01] [0.20] [0.21] [0.15] [0.20] -2.55 -0.26 0.09 300–450 Km*ca.1990 [1.58] [0.17] [0.13] -1.28 -1.34 0.21 -0.09 -0.14 0.06 -0.20*** -0.04 -0.26 0–150 Km*ca.2010 [1.20] [2.12] [2.77] [0.13] [0.23] [0.30] [0.06] [0.33] [0.44] 1.39 3.27 0.12 0.35 0.18 -0.07 150–300 Km*ca.2010 [2.00] [3.04] [0.22] [0.33] [0.36] [0.48] 2.14 0.27 -0.29 300–450 Km*ca.2010 [2.36] [0.26] [0.23] Subdistrict FE, Year FE Y Y Y Y Y Y Y Y Y District Trends, Ctrls Y Y Y Y Y Y Y Y Y Notes: For Niger, we use the centroid of the Northern section of Lake Chad that is within the territory of Chad. For Cameroon, we use the centroid of the Southern section of the lake that is within Cameroon’s territory. For Chad, we use the centroid of the Northern section of Lake Chad that is within the territory of Chad. For Niger, Cameroon and Chad, we omit 1965, 1965 and 1964, respectively. We interact the 0-150, 150-300 and 300-450 km dummies with the year fixed effects but only report the interacted effects for the years closest to 1990 (1988, 1987 and 1993, respectively) and 2010 (2012, 2005 and 2009, respectively). Conley SEs (100 Km). † p<0.15, * p<0.10, ** p<0.05, *** p<0.01. 136 3.6 Effects on Cities and “Refugee” Urbanization Technical Paper 2. Climate Change, Rural Livelihoods and Urbanization: Evidence from the Permanent Shrinking of Lake Chad 3.6.1 Effects on City Population Sizes for in the longer run (c. 2010). However, the lack of long- Niger 1951–2012 run effects is misleading because almost all cities were above 5,000 then. Thus, with a dummy almost always For Niger, we have 166 cities x 17 years = 2,822 equal to one, the effect is identified off a very few cities observations. The effects are then estimated relative to only. Instead, it is more relevant to study the effect on 1965, the omitted year.270 As seen in cols. (1)–(3) of Panel 20K+ cities. Indeed, only 16 percent of the 166 cities had A of Table 3.5, positive, not negative, effects are observed reached that threshold by 2010 (in comparison, 8 percent on log city population in the shorter run (c. 1990). The of the 166 cities had already reached 5K by 1965). As short-run effects are especially strong close to the lake (0– seen in cols. (7)–(9), when the dependent variable is 150) and decrease with distance to it. In particular, while a dummy if the city is above 20,000, there is a strong we previously found that subdistricts close to the lake effect in the shorter run that becomes much stronger in grew about 60 percent slower than other locations (cols. the longer run. In particular, the coefficients suggest that, (1)–(3) of Table 3.4), these results suggest that cities close close to the lake, there is a 70 percent higher probability to the lake grew 400 percent faster than other cities on of cities reaching 20K. average. While 400 percent seems high, note that most cities were initially small or non-existent in our three To conclude, in Niger’s case, we find strong positive countries pre-1965. Consistent with an African context effects on urbanization during the shrinking period, of high migration rates and fast demographic growth which led in the longer run to the rise of larger cities (Jedwab, Christiaensen, and Gindelsky 2017), cities in in the area. Thus, in Niger, the results are consistent with our three countries then grew particularly fast during the rural disaster-led urban underdevelopment scenario. the post- 1965 period. For example, Niger’s total urban population had increased by 500 percent by 1988. Which cities in particular did benefit? This can be examined in Map 3.4. Circa 1965, there were no 5K+ In the longer run (c. 2010) and the specification with cities (grey circles) in the East, despite the existence of three distance bins, we still see positive effects but many small settlements close to the lake. The closest these are weaker and not significant. There are several 20K+ city was Zinder, 474 km from the lake. Roads in the possible interpretations for the reduction in the effects. East were then dirt roads. However, since it does not rain First, economic refugees from the Lake Chad area may much in the area, dirt roads are comparable to improved have only temporarily settled in the cities around the lake, (gravel) roads in the south-west of the country (Jedwab the time for them to find the resources to pay the costs of and Storeygard 2020). Assuming driving speeds of 40 km migration to other cities farther away. Second, economic per hour, the driving time from the lake to Zinder must decline in rural areas may have eventually impacted the have been at the very least 12 hours. Two days of driving urban sector in the area, thus causing cities in the area to were more likely, given roads are not straight lines and relatively lose inhabitants post-1990. road conditions more generally. In this context, given the initial lack of cities close to the lake, it is perhaps Now, are the effects driven by smaller or larger cities? If not surprising that two 20K+ cities appeared: Diffa and the dependent variable is a dummy if the city had already N’Guigmi. Diffa was a large village in 1965. Yet, by 2012, reached 5,000 by 1990 ((4)–(6)), we find strong effects it had become Niger’s 11th largest city. N’Guigmi was closer to the lake in the shorter run (c. 1990; a 40 percent historically located on the shore of the lake. As explained higher probability for the 0-150 km bin) and no effect by Geels 2006, the town was before 1965 a center for 270 Given the high number of controls and when estimating the model with several distance bin dummies interacted with year fixed effects, we need to omit one more year. We choose 1962, the second closest year before the year 1965. 3.6 Effects on Cities and “Refugee” Urbanization 137 Lake Chad Regional Economic Memorandum  |  Development for Peace Map 3.4: Evolution of City Population Sizes around Lake Chad, ca. 1965–ca. 2010 Notes: The map shows, for the Lake Chad area, the location of 5K+ and 20K+ urban settlements circa 1965, when the lake started shrinking, and circa 2010, at the end of our period of study. We also indicate regionally important (20K+) cities in the 1960s such as Diffa, N’Guigmi and Zinder in Niger, N’Djamena in Chad, and Maroua and Garoua in Cameroon. Finally, we show paved roads, improved roads, and earthen roads, all circa 1965. fishing communities. During the mid-1970s the lake’s 3.6.2 Effects on City Population Sizes for shore was 85 km away while it was 45 km away in the Cameroon 1951–2012 2000s. Thus, we might have expected N’Guigmi to have been negatively impacted by the shrinkage of the lake. Its For Cameroon, we have 179 cities x 18 years = 3,222 dramatic growth from 3000 people in 1962 to 25,000 observations. The effects are then estimated relative to today must have been driven by the locality functioning 1965, the omitted year. As seen in cols. (1)–(3) of Panel as a “refugee settlement” for individuals who lost their B in Table 3.5, we find in the shorter run (c. 1990) strong rural livelihoods. Finally, since we find significant long- positive effects close to the lake. In the specification with run effects on 20K+ cities but no significant long-run three distance bins (col. (3)), the effect for the 0–150 km effects on overall city population growth, it must be that bin is 3.65. Thus, cities closest to the lake grew 365 percent the shrinkage of the lake increased urban concentration faster than other cities on average. In the longer run in the area. (c. 2010), no clear effect is observed. If we rely on the specification with three distance bins (col. (3)), there is no effect at all for the 0–150 km bin, possibly because of the same reasons as for Niger. 138 3.6 Effects on Cities and “Refugee” Urbanization Technical Paper 2. Climate Change, Rural Livelihoods and Urbanization: Evidence from the Permanent Shrinking of Lake Chad Now, we find relatively similar patterns if the 3.6.3 Effects on City Population Sizes for dependent variable is a dummy equal to one if the city Chad 1950–2009 has reached 5K (cols. (4)–(6)). The shorter-run effects for 20K cities are then positive and relatively similar to For Chad, we have 100 cities x 12 years = 1,200 what we found for Niger (cols. (7)–(9)). However, in the observations. We use 1964, the first year before 1965, longer run, the effects have disappeared, unlike what we as the omitted year.272 As seen in Panel C of Table 3.5, found for Niger. To conclude, in Cameroon’s case, we also and unlike what we found for Niger and Cameroon, find positive effects on urbanization during the shrinking strong negative effects can be observed in the shorter run period. Thus, in Cameroon, the results are consistent with (c. 1990). However, these negative effects have disappeared the rural disaster-led urban underdevelopment scenario, in the longer run (c. 2010). In the specification with at least during the disaster period. However, in the longer three distance bins (col. (3)), no effect is observed within run, no clear effect is observed. 150 km and positive effects are observed for the 150–300 and 300–450 km bins (much like what we found for For the areas closest to the lake, why are the shorter- other two countries). run effects overall weaker in Cameroon than in Niger and why have these effects disappeared in the longer Likewise, in the shorter run, we find positive but run, unlike what we saw for Niger? These results make not significant effects if the dependent variable is a sense when visually inspecting how many 5K+ and dummy if the city has reached 20 km (cols. (7)–(9)). 20K+ cities existed in 1965 vs. emerged between 1965 In the longer run, we actually find negative effects but and circa 2010 (we use 2005 for Cameroon). As can be they are also not significant. Overall, in Chad and in the seen in Map 3.4, the Cameroonian areas close to the longer run, the shrinkage of the lake did not reduce city Lake had no cities in 1965. The closest city was Maroua, population sizes. However, given the possibly negative which was already larger than 20K in 1964. Maroua was extensive margin effects for larger cities, it must be that 252 km from the lake, so closer to the lake than Zinder small cities grew relatively fast. was in Niger. Garoua, farther away (407 km), was also larger than 20K in 1964.271Then, as can be seen, many The lack of a positive effect for 20K+ cities is not small cities eventually emerged close to the lake and a surprising given the presence of N’djamena, Chad’s few 20K+ cities also emerged close to Maroua. Thus, the largest city, 230 km away from the lake. As can be seen existence of Maroua and Garoua, two cities that were in Map 3.4, only one 20K+ city appeared in the vicinity already economically important in the 1960s, might have of the northern part of the lake. There were then more prevented other 20K+ cities from emerging closer to the 5K+ cities close to the former shore of the lake. Thus, it is lake (or at least at a higher rate than observed elsewhere). If possible that lake “refugees” with skills that allowed them anything, relying on the specification with three distance to be absorbed by more urban sectors disproportionately bins (col. (3)), we see positive (but not significant) effects went to the region of N’djamena instead of joining the for the 150–300 km bin (close to Maroua) and the 300– ranks of smaller cities that might have otherwise passed 450 km bin (Garoua). the 20K threshold. 271 Ndjamena, the capital and largest city of Chad, is not far, but on the other side of the border, and while borders are usually porous in the region, Cameroonians typically do not migrate to Chad, a significantly poorer country. 272 Given the high number of controls and when estimating the model with several distance bin dummies interacted with year fixed effects, we need to omit one more year. We choose 1961, the second closest year before the year 1965. 3.6 Effects on Cities and “Refugee” Urbanization 139 Lake Chad Regional Economic Memorandum  |  Development for Peace To conclude, across the three countries and focusing on the specification with three distance bins (col. (3)), the long-run effects on city population sizes are positive, however never significantly so. Therefore, some cities grew as a result of the shock and the observed (relative) population decline observed close to the lake must have been driven by rural population. We test this more formally by using the specification with three distance bins (eq. (2)) and studying how their shorter and longer run effects on log (total population) in year t vary when also controlling for the log of (total urban population + 1) in year t (total urban population is the total population of cities above 5K in t, which is sometimes equal to 0). The regressions are thus the same as in Table 3.4 except that we control for urban growth. When doing so, the effects are either unchanged or become even more negative (not shown, but available upon request), thus confirming that the observed relative decline in total population is driven by population decline. Depending on the initial economic geography of the country, we then find different responses for small cities vs. larger cities. In Niger, we find strong effects for 20K+ cities, hence urban concentration in the lake area. In both Cameroon and Chad, we find nil or negative (but not significant) effects for 20K+ cities. Already existing 20K+ cities such as Maroua and Garoua in Cameroon and N’djamena in Chad must have acted as a pull factor for lake refugees, thus preventing the emergence of medium-sized cities close to the lake. As a result, slower rural in-migration to the area or faster out-migration of rural residents from the area must have accelerated urbanization, in many cases away from the areas close to the shore of the lake. 140 3.6 Effects on Cities and “Refugee” Urbanization Technical Paper 2. Climate Change, Rural Livelihoods and Urbanization: Evidence from the Permanent Shrinking of Lake Chad 3.7 Governmental Responses to the Crisis One important question is whether governments, 4 times more “expensive” to build than improved roads, observing the increasingly negative impacts of the lake which are in turn 15 times more “expensive” to build during the shrinking period, used public infrastructure than earthen roads (Jedwab and Storeygard 2019). investments as a way to mitigate the effects of the shock or actually under-invested in the areas. Unfortunately, As seen when comparing Maps 3.4 and 3.5, Niger’s for most African countries, there is limited data on public government built between 1965 and 2014 paved investments over such a long period of time. roads all the way to the lake. The question is whether subdistricts close to the lake received relatively more road However, and for Kenya in 1964–2002 only, Burgess investments than other subdistricts, hence the need to et al. 2015 construct localized measures of road examine this econometrically. As seen in Panel A of Table investment that come from Michelin maps. They then 3.6, in Niger we see positive shorter run (c. 1990) and use this data to show how politically connected districts longer run (c. 2010) effects on paved road construction, disproportionately receive roads in more autocratic implying that lake areas indeed received more paved roads regimes.273 Jedwab and Storeygard 2020 also use the same than other areas on average (see cols. (1)–(3)). The paving type of data but for the whole continent and the period of roads nonetheless came at the expense of improved 1965–2014 to study the effects of roads on urbanization. roads (cols. (4)–(6)). However, lake areas received more We thus rely on their geospatialized data to obtain for paved roads than they “lost” improved roads (cols. (7)– each country-subdistrict and each year available the total (9)). Now, if we examine when lake areas experienced length (km) of paved roads, improved (laterite or gravel) more paved road building than other locations, we find roads and earthen roads.274 that it was in the mid-1970s (not shown), by which the the lake’s water level had already decline by 75 percent. We then use model (2) with the different distance bin The road investments were thus likely a response to dummies to study the effects of proximity to the lake the shrinkage of the lake. However, since the lake areas on road investment. The results are reported in Table experienced slower population growth, it must be that the 3.6. In cols. (1)–(3), (4)–(6) and (7)–(9), the dependent observed road investments had little impact on localized variable is the log of (total length of paved roads + 1), economic development (again, relative to other locations (total length of improved roads + 1) and (total length of in the country). We could then also imagine that roads paved or improved roads + 1), respectively. The effects lowered inter-regional migration costs for lake refugees, are estimated relative to the omitted year, 1965. In the thus accelerating outmigration to other areas in the rest table, we then only report the interacted effects for the of the country. years 1988—to mark the end of the shrinking period— and 2008—to capture the long-term effect of the lake In Cameroon, there were already a few improved roads shrinking.275 Finally, note that paved roads are on average (Map 3.4). By 2014, some of these in the area in 1965  273 As explained by Jedwab and Storeygard 2020, “Michelin uses four sources to create the maps: (i) the previous Michelin map, (ii) government road censuses/ maps, (iii) direct information from its tire stores across Africa, and (iv) correspondence from road users including truckers.” As such, the data is highly reliable for our purpose. 274 The available years are 1965, 1967, 1968, 1969, 1971, 1973, 1976, 1983, 1984, 1985, 1986, 1988, 1990, 1991, 1993, 1996, 1998, 2003, 2008 and 2014. No Michelin map was published for our countries of study before 1965. 275 Note that we have road data up to 2014 and also interact the distance bin dummies with the year 2014. However, to be consistent with our population estimations, we are more interested in seeing the effect circa 2010. 3.7 Governmental Responses to the Crisis 141 Lake Chad Regional Economic Memorandum  |  Development for Peace Table 3.6: Effect of Proximity to the Lake, Road Investments, Flexible Specification Log (Paved Km + 1) t Log (Improved Km + 1) t Log (Paved+Impr. Km + 1) t Dependent Variable: Omitted = 1965 (1) (2) (3) (4) (5) (6) (7) (8) (9) Panel A: Niger (166 Subdistricts x 20 Years = 2,380 Obs.) 0.14 2.75*** 2.33*** -0.06 -3.58*** -3.31*** 0.45 2.39*** 2.32*** 0–150 Km*ca.1990 [0.45] [0.48] [0.53] [0.47] [0.62] [0.73] [0.46] [0.50] [0.56] 3.32*** 2.97*** -3.48*** -3.26*** 2.90*** 2.86*** 150–300 Km*ca.1990 [0.45] [0.46] [0.55] [0.61] [0.49] [0.51] -0.94** 0.62 -0.07 300–450 Km*ca.1990 [0.41] [0.73] [0.55] -1.05** 3.57*** 3.22*** -0.18 -6.74*** -6.63*** -0.77 2.54*** 2.49*** 0–150 Km*ca.2010 [0.53] [0.69] [0.73] [0.67] [0.98] [0.98] [0.59] [0.69] [0.71] 4.12*** 3.82*** -6.41*** -6.32*** 3.04*** 3.00*** 150–300 Km*ca.2010 [0.75] [0.75] [0.87] [0.87] [0.76] [0.77] -0.90* 0.40 -0.00 300–450 Km*ca.2010 [0.47] [0.78] [0.61] Panel B: Cameroon (113 Subdistricts x 20 Years = 2,260 Obs.) -2.60*** -5.35*** -2.12*** 2.19*** 3.46*** -0.26 -0.61 -0.08 -1.19* 0–150 Km*ca.1990 [0.78] [0.76] [0.50] [0.69] [0.86] [0.65] [0.56] [0.77] [0.66] -2.98*** 0.18 1.19** -2.57*** 0.60 -0.58 150–300 Km*ca.1990 [0.33] [0.54] [0.47] [0.47] [0.44] [0.54] 2.66*** -4.13*** -1.70** 300–450 Km*ca.1990 [0.69] [0.53] [0.67] -2.42** -7.00*** 0.26 3.86*** 6.08*** -0.61*** 0.18 0.75 -0.44** 0–150 Km*ca.2010 [0.99] [0.96] [0.51] [0.73] [0.97] [0.22] [0.53] [0.71] [0.20] -4.68*** 2.55** 2.00*** -4.77*** 0.46* -0.78 150–300 Km*ca.2010 [0.44] [1.04] [0.34] [0.72] [0.25] [0.54] 7.03*** -7.30*** -1.70** 300–450 Km*ca.2010 [1.17] [0.89] [0.73] Panel C: Chad (138 Subdistricts x 20 Years = 2,760 Obs.) -0.23** -0.25* -0.23 0.10 0.19 0.25 0.06 0.13 0.15 0–150 Km*ca.1990 [0.11] [0.15] [0.17] [0.07] [0.17] [0.30] [0.06] [0.15] [0.27] -0.00 0.02 0.09 0.15 0.06 0.08 150–300 Km*ca.1990 [0.10] [0.13] [0.13] [0.27] [0.12] [0.24] 0.03 0.10 0.03 300–450 Km*ca.1990 [0.12] [0.24] [0.22] -0.19** -0.22 -0.21 0.11** 0.30 0.24 0.09 0.25 0.18 0–150 Km*ca.2010 [0.08] [0.16] [0.24] [0.05] [0.23] [0.30] [0.05] [0.21] [0.28] 0.02 0.02 0.18 0.13 0.17 0.10 150–300 Km*ca.2010 [0.12] [0.20] [0.17] [0.24] [0.16] [0.23] 0.01 -0.09 -0.09 300–450 Km*ca.2010 [0.09] [0.12] [0.13] Subdistrict FE, Year FE Y Y Y Y Y Y Y Y Y District Trends, Ctrls Y Y Y Y Y Y Y Y Y Notes: For Niger, we use the centroid of the Northern section of Lake Chad that is within the territory of Chad. For Cameroon, we use the centroid of the Southern section of the lake that is within Cameroon’s territory. For Chad, we use the centroid of the Northern section of Lake Chad that is within the territory of Chad. For Niger, Cameroon and Chad, we omit 1965. We interact the 0-150, 150-300 and 300-450 km dummies with the year fixed effects but only report the interacted effects for the years 1988 and 2008. Conley SEs (100 Km). † p<0.15, * p<0.10, ** p<0.05, *** p<0.01. 142 3.7 Governmental Responses to the Crisis Technical Paper 2. Climate Change, Rural Livelihoods and Urbanization: Evidence from the Permanent Shrinking of Lake Chad roads were paved and some earthen roads were improved, did not build better roads to the area. This is confirmed however not in the immediate vicinity of the lake (see econometrically in Panel C of Table 3.6. While the Map 3.5). Econometrically (see Panel of Table 3.6), we coefficients are not significant, we find, if anything, observe lower levels of road investment in the Lake Chad negative effects for paved road construction (col. (3)). area, except maybe farther away for the lake and for paved roads only (col. (3)). For the areas closest to the lake (0– Overall, Cameroon and Chad did not particularly 150 km), negative shorter run and longer run effects are respond to the crisis by building more roads to the observed when combining paved and improved roads Lake Chad area. Niger did respond by building more (col. (9)). Cameroon’s government thus did not appear roads but we know from our population estimations that to respond to the shrinkage of the lake by building more it did not prevent population decline in the area. We roads connecting the immediate lake areas to the rest of nonetheless observed urban concentration in Niger, with the country. a higher likelihood of having larger 20K+ cities. Thus, the roads might have contributed to lake refugees settling In Chad, the areas close to the lake did not have any in these better connected (and possibly diversified) cities paved or improved road in its vicinity in both 1965 rather than staying in smaller cities as in Cameroon or and 2014 (Maps 3.4 and 3.5). Chad’s government thus Chad. Map 3.5: Road Networks in the Lake Chad Area, ca. 2015 Notes: The figure shows for the year 2014 the location of paved roads, improved roads, and earthen roads. 3.7 Governmental Responses to the Crisis 143 Lake Chad Regional Economic Memorandum  |  Development for Peace 3.8 Conclusion and Policy Discussion Many of the world’s lakes are disappearing. Despite in roads that would connect these areas to less exposed an extensive literature on the economic consequences of locations. While roads might help the affected regions climate change, the economic effects of diminishing lakes cushion the shock by diversifying away from lake-related have not been widely investigated. We focused on Lake economic activities, they may also have no impact or even Chad, a vast African lake that lost about 90 percent of a detrimental local impact by accelerating outmigration. its surface area between 1965 and 1985, and recovered Unfortunately, the lack of data prevents us from better some of it post-1985. For Cameroon, Chad and Niger, analyzing which other public investments historically we constructed a novel data set tracking total and city took place in the region and what their mitigation effects population patterns at a fine spatial level from the 1950s were. to the 2010s. We then exploited a difference-in-difference strategy to estimate the effects of Lake Chad’s shrinking on nearby communities. We found relatively slower total population growth in the proximity of the lake, but only after the lake started shrinking. We did not find evidence for population recovery in the long run. We also found in many cases positive effects on the lake shrinking on city population growth nearby, which suggested that climate change might have induced “refugee” urbanization locally. Finally, we found that only Niger disproportionately built higher-quality roads to the Lake Chad area. However, comparing our different results, it did not appear to prevent population decline in the Lake Chad area. More generally, while our work cannot fully answer the question of how governments should respond to shrinking lakes, our results suggest that such natural disasters could have permanent negative localized economic effects in poor agrarian countries. In such countries, rural decline is likely to accelerate structural change and urbanization, in some cases locally. Unless governments and the international community find ways to reverse changes due to such natural disasters, which in this case could imply diverting other rivers regionally or stopping climate change globally, the shrinkage of lakes is likely to increase already existing pressures on cities. Factors that increase the absorptive capacity of their labor and housing markets might then help mitigate the economic impact of lake shrinkages. Finally, an important policy question is how much should governments invest in infrastructure in disaster-prone areas, for example 144 3.8 Conclusion and Policy Discussion Technical Paper 2. Climate Change, Rural Livelihoods and Urbanization: Evidence from the Permanent Shrinking of Lake Chad References Adhvaryu, Achyuta et al. (2021). “Resources, conflict, and economic development in Africa”. In: Journal of Development Economics 149, p. 102598. ISSN: 0304-3878. DOI: https://doi.org/10. 1016/j.jdeveco.2020.102598. Ahonsi, Babatunde A (1988). “Deliberate falsification and census data in Nigeria”. In: African affairs 87(349), pp. 553–562. Allcott, Hunt and Daniel Keniston (July 2017). “Dutch Disease or Agglomeration? The Local Economic Effects of Natural Resource Booms in Modern America”. In: The Review of Economic Studies 85(2), pp. 695–731. DOI: 10.1093/restud/rdx042. Aragón, Fernando M., Francisco Oteiza, and Juan Pablo Rud (Feb. 2021). “Climate Change and Agriculture: Subsistence Farmers’ Response to Extreme Heat”. In: American Economic Journal: Economic Policy 13(1), pp. 1–35. DOI: 10.1257/pol.20190316. Aragón, Fernando M. and Juan Pablo Rud (May 2013). “Natural Resources and Local Communities: Evidence from a Peruvian Gold Mine”. In: American Economic Journal: Economic Policy 5(2), pp. 1–25. DOI: 10.1257/ pol.5.2.1. Arezki, Rabah, Valerie A. Ramey, and Liugang Sheng (Nov. 2016). “News Shocks in Open Economies: Evidence from Giant Oil Discoveries*”. In: The Quarterly Journal of Economics 132(1), pp. 103–155. DOI: 10.1093/qje/ qjw030. Armand, Alex et al. (Nov. 2020). “Does Information Break the Political Resource Curse? Experimental Evidence from Mozambique”. In: American Economic Review 110(11), pp. 3431–53. DOI: 10.1257/aer.20190842. Baez, Javier et al. (May 2017). “Heat Exposure and Youth Migration in Central America and the Caribbean”. In: American Economic Review 107(5), pp. 446–50. DOI: 10.1257/aer.p20171053. Barreca, Alan et al. (May 2015). “Convergence in Adaptation to Climate Change: Evidence from High Temperatures and Mortality, 1900-2004”. In: American Economic Review 105(5), pp. 247–51. DOI: 10.1257/aer. p20151028. Barrios, Salvador, Luisito Bertinelli, and Eric Strobl (2006). “Climatic change and rural-urban migration: The case of sub-Saharan Africa”. In: Journal of Urban Economics. Beine, Michel and Christopher Parsons (2015). “Climatic factors as determinants of interna- tional migration”. In: The Scandinavian Journal of Economics 117(2), pp. 723–767. Berman, Nicolas et al. (June 2017). “This Mine Is Mine! How Minerals Fuel Conflicts in Africa”. In: American Economic Review 107(6), pp. 1564–1610. DOI: 10.1257/aer.20150774. Birkett, CM (2000). “Synergistic remote sensing of Lake Chad: Variability of basin inundation”. In: Remote sensing of environment 72(2), pp. 218–236. Blankespoor, Brian et al. (2020). “Spillover Effects of Foreign Conflict: Evidence from Boko Haram”. In: Working Paper. Bohra-Mishra, Pratikshya, Michael Oppenheimer, and Solomon M Hsiang (2014). “Nonlinear permanent migration response to climatic variations but minimal response to disasters”. In: Proceedings of the National Academy of Sciences 111(27), pp. 9780–9785. Boustan, Leah Platt, Matthew E. Kahn, and Paul W. Rhode (May 2012). “Moving to Higher Ground: Migration Response to Natural Disasters in the Early Twentieth Century”. In: American Economic Review 102(3), pp. 238–44. DOI: 10.1257/aer.102.3.238. References 145 Lake Chad Regional Economic Memorandum  |  Development for Peace Brooks, Wyatt and Kevin Donovan (2020). “Eliminating Uncertainty in Market Access: The Impact of New Bridges in Rural Nicaragua”. In: Econometrica 88(5), pp. 1965–1997. DOI: https://doi.org/10.3982/ECTA15828. Burgess, Robin et al. (June 2015). “The Value of Democracy: Evidence from Road Building in Kenya”. In: American Economic Review 105(6), pp. 1817–51. DOI: 10.1257/aer.20131031. Burke, Marshall and Kyle Emerick (Aug. 2016). “Adaptation to Climate Change: Evidence from US Agriculture”. In: American Economic Journal: Economic Policy 8(3), pp. 106–40. DOI: 10. 1257/pol.20130025. Caruso, Germán and Sebastian Miller (2015). “Long run effects and intergenerational transmission of natural disasters: A case study on the 1970 Ancash Earthquake”. In: Journal of Development Economics 117, pp. 134–150. ISSN: 0304-3878. DOI: https://doi.org/10.1016/ j.jdeveco.2015.07.012. Caruso, Germán Daniel (2017). “The legacy of natural disasters: The intergenerational impact of 100 years of disasters in Latin America”. In: Journal of Development Economics 127, pp. 209–233. ISSN: 0304-3878. DOI: https:// doi.org/10.1016/j.jdeveco.2017.03.007. Caselli, Francesco, Massimo Morelli, and Dominic Rohner (Feb. 2015). “The Geography of Interstate Resource Wars*”. In: The Quarterly Journal of Economics 130(1), pp. 267–315. DOI: 10.1093/qje/qju038. Cattaneo, Cristina and Giovanni Peri (2016). “The migration response to increasing temperatures”. In: Journal of Development Economics 122, pp. 127–146. Chauvin, Juan Pablo et al. (2017). “What is different about urbanization in rich and poor countries? Cities in Brazil, China, India and the United States”. In: Journal of Urban Economics 98, pp. 17–49. Chen, Joyce J. et al. (May 2017). “Validating Migration Responses to Flooding Using Satellite and Vital Registration Data”. In: American Economic Review 107(5), pp. 441–45. DOI: 10.1257/aer. p20171052. Chen, Shuai and Binlei Gong (2021). “Response and adaptation of agriculture to climate change: Evidence from China”. In: Journal of Development Economics 148, p. 102557. ISSN: 0304-3878. DOI: https://doi. org/10.1016/j.jdeveco.2020.102557. Comission du Bassin du Lac Tchad (2015). “Plan de Développement et d’Adaptation au Changement Climatique du Lac Tchad”. In: Dell, Melissa, Benjamin F Jones, and Benjamin A Olken (2012). “Temperature shocks and economic growth: Evidence from the last half century”. In: American Economic Journal: Macroeconomics 4(3), pp. 66–95. Dell, Melissa, Benjamin F Jones, and Benjamin A Olken (2014). “What do we learn from the weather? The new climate-economy literature”. In: Journal of Economic Literature 52(3), pp. 740–98. Deschênes, Olivier and Michael Greenstone (Mar. 2007). “The Economic Impacts of Climate Change: Evidence from Agricultural Output and Random Fluctuations in Weather”. In: American Economic Review 97(1), pp. 354–385. DOI: 10.1257/aer.97.1.354. Deschênes, Olivier and Michael Greenstone (2011). “Climate change, mortality, and adaptation: Evidence from annual fluctuations in weather in the US”. In: American Economic Journal: Applied Economics 3(4), pp. 152–85. Deschênes, Olivier, Michael Greenstone, and Jonathan Guryan (2009). “Climate change and birth weight”. In: American Economic Review 99(2), pp. 211–17. Deschênes, Olivier and Enrico Moretti (2009). “Extreme Weather Events, Mortality, and Migration”. In: The Review of Economics and Statistics 91(4), pp. 659–681. ISSN: 00346535, 15309142. URL: http://www.jstor.org/ stable/25651369. Duranton, Gilles (2016). “Agglomeration effects in Colombia”. In: Journal of Regional Science 56(2), pp. 210–238. Eberle, Ulrich J., Dominic Rohner, and Mathias Thoenig (Dec. 2020). Heat and Hate: Climate Security and Farmer- Herder Conflicts in Africa. CEPR Discussion Papers 15542. C.E.P.R. Discussion Papers. URL: https://ideas. repec.org/p/cpr/ceprdp/15542.html. 146 References Technical Paper 2. Climate Change, Rural Livelihoods and Urbanization: Evidence from the Permanent Shrinking of Lake Chad FAO (2009). “Adaptive water management in the Lake Chad Basin. Addressing current challenges and adapting to future needs”. In: FAO Water Seminar Proceedings of the World Water Week. Fisher, Anthony C et al. (2012). “The economic impacts of climate change: evidence from agricultural output and random fluctuations in weather: comment”. In: American Economic Review 102(7), pp. 3749–60. Gallagher, Justin and Daniel Hartley (Aug. 2017). “Household Finance after a Natural Disaster: The Case of Hurricane Katrina”. In: American Economic Journal: Economic Policy 9(3), pp. 199–228. DOI: 10.1257/pol.20140273. Geels, Jolijin (2006). Niger: The Bradt Travel Guide. Bradt Travel Guides. Gollin, Douglas, Martina Kirchberger, and David Lagakos (2017). “In search of a spatial equilibrium in the developing world”. In: NBER Working Paper. Gray, Clark and Valerie Mueller (2012). “Drought and population mobility in rural Ethiopia”. In: World development 40(1), pp. 134–145. Gröger, André and Yanos Zylberberg (2016). “Internal labor migration as a shock coping strategy: Evidence from a typhoon”. In: American Economic Journal: Applied Economics 8(2), pp. 123–53. Harari, Mariaflavia and Eliana La Ferrara (Oct. 2018). “Conflict, Climate, and Cells: A Disaggregated Analysis”. In: The Review of Economics and Statistics 100(4), pp. 594–608. ISSN: 0034-6535. DOI: 10.1162/rest a 00730. Henderson, J Vernon, Adam Storeygard, and Uwe Deichmann (2017). “Has climate change driven urbanization in Africa?” In: Journal of Development Economics 124, pp. 60–82. DOI: https://doi.org/10.1016/j. jdeveco.2016.09.001. Hsiang, Solomon and Robert E. Kopp (Nov. 2018). “An Economist’s Guide to Climate Change Science”. In: Journal of Economic Perspectives 32(4), pp. 3–32. DOI: 10.1257/jep.32.4.3. Hughes, R., J. Hughes, and G. Bernacsek (1992). A directory of African wetlands. IUCN. Hutchinson, Charles F et al. (1992). “Development in arid lands: lessons from Lake Chad”. In: Environment: Science and Policy for Sustainable Development 34(6), pp. 16–43. Ighobor, Kingsley (2019). Développer le bassin du lac Tchad. URL: https : / / www . un . org / africarenewal/fr/ magazine/d percent 5C percent C3 percent 5C percent A9cembre- 2019 - mars- 2020 /d percent 5C percent C3 percent 5C percentA9velopper-le-bassin-du-lac-tchad. (accessed: 11.26.2020). Jedwab, Remi, Luc Christiaensen, and Marina Gindelsky (2017). “Demography, urbanization and development: Rural push, urban pull and... urban push?” In: Journal of Urban Economics 98. Urbanization in Developing Countries: Past and Present, pp. 6–16. ISSN: 0094-1190. DOI: https://doi.org/10.1016/j.jue.2015.09.002. URL: https://www.sciencedirect.com/science/article/pii/S0094119015000601. Jedwab, Remi and Adam Storeygard (2019). “Economic and Political Factors in Infrastructure Investment: Evidence from Railroads and Roads in Africa 1960–2015”. In: Economic History of Developing Regions 34(2), pp. 156–208. DOI: 10.1080/20780389.2019.1627190. eprint: https://doi.org/10.1080/20780389.2019.1627 190. URL: https://doi.org/10.1080/20780389.2019. 1627190. Jedwab, Remi and Adam Storeygard (2020). “The average and heterogeneous effects of transportation investments: Evidence from Sub-Saharan Africa 1960-2010”. In: NBER Working Paper. Jedwab, Remi and Dietrich Vollrath (2015). “Urbanization without growth in historical perspective”. In: Explorations in Economic History 58(C), pp. 1–21. URL: https://EconPapers. repec.org/ RePEc:eee:exehis:v:58:y:2015:i:c:p:1-21. Jedwab, Remi and Dietrich Vollrath (2019). “The urban mortality transition and poor-country urbanization”. In: American Economic Journal: Macroeconomics 11(1), pp. 223–75. References 147 Lake Chad Regional Economic Memorandum  |  Development for Peace Jessoe, Katrina, Dale T. Manning, and J. Edward Taylor (June 2017). “Climate Change and Labour Allocation in Rural Mexico: Evidence from Annual Fluctuations in Weather”. In: The Economic Journal 128(608), pp. 230–261. ISSN: 0013-0133. DOI: 10.1111/ecoj.12448. Jones, Benjamin F and Benjamin A Olken (2010). “Climate shocks and exports”. In: American Economic Review 100(2), pp. 454–59. Kalemli-Özcan, Sebnem, Alex Nikolsko–Rzhevskyy, and Jun Hee Kwak (2020). “Does trade cause capital to flow? Evidence from historical rainfall”. In: Journal of Development Economics 147, p. 102537. ISSN: 0304-3878. DOI: https://doi.org/10.1016/j.jdeveco.2020.102537. Kirchberger, Martina (2017). “Natural disasters and labor markets”. In: Journal of Development Economics 125, pp. 40–58. ISSN: 0304-3878. DOI: https://doi.org/10.1016/j.jdeveco.2016.11. 002. Kleemans, Marieke and Jeremy Magruder (Nov. 2017). “Labour Market Responses to Immigra- tion: Evidence from Internal Migration Driven by Weather Shocks”. In: The Economic Journal 128(613), pp. 2032–2065. DOI: 10.1111/ecoj.12510. Kocornik-Mina, Adriana et al. (2020). “Flooded cities”. In: American Economic Journal: Applied Economics 12(2), pp. 35–66. Luxereau, Anne, Pierre Genthon, and Jean-Marie Ambouta Karimou (2012). “Fluctuations in the size of Lake Chad: consequences on the livelihoods of the riverain peoples in eastern Niger”. In: Regional Environmental Change 12(3), pp. 507–521. Maccini, Sharon and Dean Yang (2009). “Under the weather: Health, schooling, and economic consequences of early- life rainfall”. In: American Economic Review 99(3), pp. 1006–26. Mahajan, Parag and Dean Yang (2020). “Taken by storm: Hurricanes, migrant networks, and us immigration”. In: American Economic Journal: Applied Economics 12(2), pp. 250–77. Miller, Steve et al. (Feb. 2021). “Heat Waves, Climate Change, and Economic Output”. In: Journal of the European Economic Association. DOI: 10.1093/jeea/jvab009. Murray, Senan (2007). Lake Chad fishermen pack up their nets. URL: http://news.bbc.co.uk/2/ hi/africa/6261447.stm. (accessed: 11.27.2020). Okpara, Uche T, Lindsay C Stringer, and Andrew J Dougill (2016). “Lake drying and livelihood dynamics in Lake Chad: Unravelling the mechanisms, contexts and responses”. In: Ambio 45(7), pp. 781–795. Olivry, Jean-Claude et al. (1996). Hydrologie du lac Tchad. Vol. 12. Orstom. Onyia, Chukwuma (2015). “Climate Change and Conflict in Nigeria: The Boko Haram Challenge”. In: American International Journal of Social Science 4(2), pp. 181–190. Park, Chang-Eui et al. (2018). “Keeping global warming within 1.5 C constrains emergence of aridification”. In: Nature Climate Change 8(1), p. 70. Partridge, Mark D., Bo Feng, and Mark Rembert (May 2017). “Improving Climate-Change Modeling of US Migration”. In: American Economic Review 107(5), pp. 451–55. DOI: 10.1257/ aer.p20171054. Peri, Giovanni and Akira Sasahara (2019). “The impact of global warming on rural-urban migrations: evidence from global big data”. In: NBER Working Paper. Ploeg, Frederick van der (June 2011). “Natural Resources: Curse or Blessing?” In: Journal of Economic Literature 49(2), pp. 366–420. DOI: 10.1257/jel.49.2.366. Rosenzweig, Mark R. and Christopher Udry (May 2014). “Rainfall Forecasts, Weather, and Wages over the Agricultural Production Cycle”. In: American Economic Review 104(5), pp. 278–83. DOI: 10.1257/aer.104.5.278. Sarch, Marie-thérèse and Charon Birkett (2000). “Fishing and farming at Lake Chad: Responses to lake-level fluctuations”. In: Geographical Journal 166(2), pp. 156–172. 148 References Technical Paper 2. Climate Change, Rural Livelihoods and Urbanization: Evidence from the Permanent Shrinking of Lake Chad Schlenker, Wolfram, W. Michael Hanemann, and Anthony C. Fisher (Mar. 2005). “Will U.S. Agriculture Really Benefit from Global Warming? Accounting for Irrigation in the Hedonic Approach”. In: American Economic Review 95(1), pp. 395–406. DOI: 10 . 1257 / 0002828053828455. Schlenker, Wolfram, W. Michael Hanemann, and Anthony C. Fisher (2006). “The Impact of Global Warming on U.S. Agriculture: An Econometric Analysis of Optimal Growing Conditions”. In: The Review of Economics and Statistics 88(1), pp. 113–125. URL: http://www. jstor.org/stable/40042963. Sédick, Ahmed (n.d.). “Le Lac Tchad et ses tributaires”. In: (). Shah, Manisha and Bryce Millett Steinberg (2017). “Drought of Opportunities: Contemporane- ous and Long-Term Impacts of Rainfall Shocks on Human Capital”. In: Journal of Political Economy 125(2), pp. 527–561. DOI: 10.1086/690828. Spitzer, Yannay, Gaspare Tortorici, and Ariell Zimran (2020). “International Migration Responses to Natural Disasters: Evidence from Modern Europe’s Deadliest Earthquake”. In: NBER Working Paper. Tol, Richard S. J. (June 2009). “The Economic Effects of Climate Change”. In: Journal of Economic Perspectives 23(2), pp. 29–51. DOI: 10.1257/jep.23.2.29. Topel, Robert H (1986). “Local labor markets”. In: Journal of Political economy 94(3, Part 2), S111–S143. Torvik, Ragnar (2002). “Natural resources, rent seeking and welfare”. In: Journal of Development Economics 67(2), pp. 455–470. ISSN: 0304-3878. DOI: https://doi.org/10.1016/S0304-3878(01) 00195-X. Venables, Anthony J. (Feb. 2016). “Using Natural Resources for Development: Why Has It Proven So Difficult?” In: Journal of Economic Perspectives 30(1), pp. 161–84. DOI: 10.1257/jep.30.1. 161. References 149