WPS6481 Policy Research Working Paper 6481 Is Urbanization in Sub-Saharan Africa Different? J. Vernon Henderson Mark Roberts Adam Storeygard The World Bank Development Research Group Environment and Energy Team June 2013 Policy Research Working Paper 6481 Abstract In the past dozen years, a literature has developed differences, however, at the sector level. Agricultural arguing that urbanization has unfolded differently in trade effects that improve farm prices deter African post-independence Sub-Saharan Africa than in the rest urbanization, while they promote urbanization elsewhere. of the developing world, with implications for African Potential reasons include differences in land ownership economic growth overall. While African countries are institutions and the likelihood of agricultural surpluses more urbanized than other countries at comparable being invested in urban production. Positive shocks to levels of income, it is well-recognized that total and modern manufacturing spur urbanization in the rest of sector gross domestic product data are of very low the developing world, but effects are dependent on the quality, especially in Africa. When instead viewed from level of development. Thus many countries in Africa, the perspective of effective technology, as suggested with their lower level of development, do not respond to in endogenous growth frameworks (and as proxied these shocks. Finally, historical indicators of the potential by educational attainment), the African urbanization for good institutions promote urbanization both inside experience overall matches global patterns. There are and outside Africa. This paper is a product of the Environment and Energy Team, Development Research Group. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://econ.worldbank.org. The authors may be contacted at j_henderson@brown.edu, mroberts1@worldbank.org and adam.storeygard@tufts.edu. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team Is Urbanization in Sub-Saharan Africa Different? J. Vernon Henderson, Mark Roberts, and Adam Storeygard JEL Codes: O10, O55, R12 Key words: Africa, Urbanization, Economic Growth Sector: Urban Is Urbanization in Sub-Saharan Africa Different? 1 J. Vernon Henderson, Mark Roberts, and Adam Storeygard A growing literature argues that, in the period since independence, Sub-Saharan Africa’s urbanization process has differed fundamentally from that in the rest of the developing world. 2 In particular, this literature argues that two stylized facts that hold for the rest of the world do not hold for Africa. First, while rapid urbanization in other developing regions has been accompanied by fast macro-economic growth, Africa has seen urbanization without growth (e.g., World Bank 1999, 2009; Fay and Opal 2000; Barrios, Bertinelli, and Strobl 2010; Gollin, Jedwab, and Vollrath 2012). Second, while, in the rest of the world, urbanization has generally been accompanied by a sector transformation from agriculture to manufacturing, Africa has suffered from urbanization without industrialization (e.g., Fay and Opal 2000; Collier and Venables 2007; Gollin et al. 2012). 3 Linking these two stylized facts, Jedwab (2012) argues that urbanization in Africa is driven by spending of agricultural resource income on consumption in cities, rather than investment in manufacturing (see also Gollin et al., 2012). We find rather different results in this paper. African urbanization follows fundamental growth drivers in the same way as the rest of the world. Africa does differ in sector responses, but not in the way characterized in the recent literature. On the urbanization process and the first stylized fact, while we confirm that urbanization bears a stronger relationship to income in the rest of the world than it does in Africa, we note that this result may be driven by substantial measurement error in GDP data (Henderson, Storeygard and Weil 2012; Young 2012). More fundamentally, although, in general, we expect to find a positive correlation between urbanization and, when accurately measured, income, this relationship should not be seen as a causal one. Rather, both a country’s level of urbanization and its 1 J. Vernon Henderson is with Brown University and the National Bureau of Economic Research. Mark Roberts is with the World Bank. Adam Storeygard is with Tufts University. The authors thank Nat Tharnpanich and Marine Gassier for their excellent research assistance. They also thank participants at the 2012 North American Regional Science Association-Urban Economics Meetings and a World Bank workshop on African Urbanization for their helpful feedback on earlier versions of this paper. Financial support from the World Bank’s Knowledge for Change Program is gratefully acknowledged. Responsibility for the contents of the paper is the authors’ alone and should not be attributed to the World Bank, its Executive Directors, or member countries. 2 For brevity, in the remainder of this paper we frequently refer to Sub-Saharan Africa as either SSA or simply Africa. Similarly, we often refer to the rest of the developing world as just the “rest of the world� or NSSA (non- Sub-Saharan Africa). 3 Explanations for this lack of industrialization include small country sizes restricting specialization, crowd-out by primary exports, exchange rate overvaluation, high labor costs relative to quality, poor infrastructure, and inadequate institutions (e.g., Johnson et al., 2007, Collier and Venables 2007, Barrios et al., 2010). 2 level of income should be viewed as the equilibrium outcome of allocations between sectors which are influenced by capital accumulation and technological sophistication, as well as trade relationships and resource availability. Thus, models of urbanization set in endogenous growth frameworks feature human capital accumulation as the driver of technological progress, leading to productivity growth in both the urban and rural sectors, as well as a shift of labor from the rural (or traditional) sector to the urban (or modern) sector (Galor and Zeira 1993; Lucas 2004; Henderson and Wang 2005; Galor, Moav and Vollrath 2009). Consistent with this, we find that urbanization responds to human capital accumulation in Africa as it does in the rest of the world. With respect to the second stylized fact, recent literature argues that, unlike the rest of the world, African urbanization exhibits an unusual positive response to agricultural shocks which is inconsistent with simple two-sector models of urbanization (Fay and Opal 2000; Barrios, Bertinelli, and Strobl 2006; Collier and Venables 2007; Jedwab 2012). Most notably, Jedwab (2012) and Gollin et al. (2012) argue that African cities are not classic “producer cities� whose growth is fueled by manufacturing scale economies. Rather, they are “consumer cities� that arise to fulfill increased demand for non-traded, non-agricultural goods, where increased demand is based on increased agricultural export rents. Thus, if the agricultural sector experiences a positive shock, urbanization in Africa increases. In contrast, we find that Africa’s urbanization response to a positive agricultural shock is negative. Instead it is in the rest of the world where we observe the “unusual� positive response. Africa is different, but not in the way characterized in the literature. In an attempt to explain our key finding of a negative urbanization response to agricultural trade shocks for Africa and a positive response for the rest of the world, we will explore some conceptual issues concerning the definition of agriculture, differences between agriculture and natural resources, and the definition of what are rural versus urban activities. Then we will turn to two inter-related questions: namely, who are the main beneficiaries from agricultural windfalls? And do these beneficiaries differ between Africa and the rest of the world? In this context, we will analyze three broad sets of institutional factors that help determine the beneficiaries of shocks: (1) monopsony purchasing of agricultural products (Jayne and Jones 1997) and distribution of agricultural surpluses to cities, as highlighted in the urban bias literature (Renaud 1980; Henderson 1988; Ades and Glaeser 1995; Davis and Henderson 2003; McCormick and Wahba 2003); (2) specific institutions related to landownership and how agricultural surpluses may be invested; and (3) other political and economic institutions that affect the distribution of agricultural surpluses. 3 Next we will turn to the industrial sector, where results are more nuanced. In particular, for what we term the “modern manufacturing� sub-sector 4, urbanization in developing countries outside Africa responds positively to trade shocks. The presence of such a response, and its absence in other sub- sectors, is consistent with the existence of strong urban scale effects in this sub-sector (Henderson, Lee and Lee 2001). For Africa, however, we observe no response of urbanization to trade shocks for all industrial sub-sectors, including the modern manufacturing sub-sector. The evidence suggests that the response to modern manufacturing shocks is increasing in a country’s level of development, or human capital; and, on average, African countries have lower levels of human capital than do other developing countries. In analyzing sector responses, the econometric problem is that sector composition and urbanization are simultaneous outcomes influenced by many factors we do not observe. To achieve pseudo- randomization, we estimate the urbanization response of developing countries to external trade shocks, given historical trade patterns. For agriculture, we have detailed historical export data differentiated by crop type. We use changes in current world commodity prices to shock the agricultural sector of each developing country, using the historical export data to weight crop prices for each exporter. For the industrial sector, we model shocks to the demand for a developing country’s exports using OECD country income variation. For each developing country, we know historical bilateral trade-patterns to OECD countries, indicating historical pathways or networks of trade flow destinations by commodity and country. We use these to weight the exogenous shocks (changes in OECD country incomes), much like a Card (2001)-type instrument. Using import data from OECD and other more developed countries has the added advantage that it is more likely to be reliable than developing country sector composition data. Finally, we relate aspects of the African urbanization debate to concerns about political institutions that have been raised in the larger related literature on Africa’s poor growth performance. One strand of this literature emphasizes the long run effects of pre-colonial institutions on their modern counterparts. Acemoglu and Robinson (2010), for example, argue that governance in many African countries is “neopaternalistic� with the ruler standing above the law in his ability to expropriate and confer rights. Weak institutions perpetuate appropriation risk and bureaucratic rent seeking, thereby lowering private citizens’ incentives for saving and investment, as well as governments’ incentives to supply public goods, which are important for both agricultural and industrial development (Johnson, 4 This sub-sector includes both electrical and non-electrical machinery, telecommunications and transport equipment, instruments, and various other miscellaneous items (arms and ammunition, toys, musical equipment). 4 Ostry, and Subramanian 2007; Binswanger and Townsend 2000; Block and Bates 2011). 5 However, while weak institutions may inhibit development, we seek to address a different question: do they have any further effect on urbanization, conditional on the degree of development? One hypothesis is that countries with weak underlying determinants of institutions may find it harder to upgrade from the informal institutions that tend to prevail in rural societies to the more formal, structured institutions that are characteristic of urban settings. As such, these countries may be less likely to urbanize. In addressing the relationship between political institutions and urbanization we are confronted with the well-known problem of the endogeneity of contemporaneous institutions. To deal with this problem we focus on time invariant and pre-determined country conditions thought to independently influence the development of institutions, such as a long history of statehood or less ethno-linguistic fractionalization, to see, first of all, if these have any effect and, if so, whether this effect differs between Africa and the rest of the world. The remainder of the paper is organized as follows. The next section, Section 1, presents basic correlations which help to set the scene and motivate the subsequent analysis. Section 2 then provides a model of structural change and urbanization. Section 3 presents our econometric specification which derives from this model, as well as our basic results which follow from the estimation of this specification. Section 4 considers the impact of pseudo-random trade shocks on urbanization, while Section 5 examines the relationship between institutions and urbanization. Finally, Section 6 concludes. 1. Patterns in the data For this paper, the sample is all countries that, in 1970, had a population in excess of 300,000 and a level of GDP per capita that was less than the world sample mean for all countries meeting the population criterion. 6 With an average level of GDP per capita of $1,877 in 2009, Sub-Saharan African (SSA) 5 Binswanger and Townsend (2000) argue that monopolistic dominance in the purchase of output and inputs (seeds and fertilizer) by government or private concerns, low functioning and biased capital markets, heavy taxation of farm output, and lack of political voice of farmers (who are often disproportionately women) reduce incentives in, and, therefore, also the performance of, agriculture. Block and Bates (2011) argue that TFP in African agriculture is low, with a decline in capital investment and use of fertilizer between 1985 and 2007 due to poor incentives. Using World Bank enterprise surveys for 10 African countries, Kalemli-Ozcan and Sorensen (2012) argue that access to capital markets is restricted and capital is misallocated compared to other regions of the world. In their limited sample, African countries with better institutions, property rights and legal systems have better allocations in capital markets. 6 The world sample mean level of GDP per capita in 1970 for all countries with a population greater than 300,000 was $5,317 (constant 2005 international dollars). In 1970, Gabon was the only SSA country (for which data is available) with a level of GDP per capita greater than this. 5 countries in the sample are much poorer on average than countries in the rest of the developing world (non-SSA, or NSSA), where the average is $6,235 (Table 1). As an alternative definition, we could restrict to countries with incomes less than half of the world mean. Only two of 40 SSA countries in the original sample would be removed, but 19 of 46 non-SSA countries would be. This would leave us with a small non-SSA sub-sample that is more similar to SSA in terms of both GDP per capita and industrialization, but with the African countries still poorer and less industrialized on average. 7 Our results are generally similar for both samples. We prefer the larger sample because it provides more power. We try to address this limited overlap between the two samples in other ways. Data sources and definitions, as well as a complete list of sample countries, are provided in the Appendix. As the literature suggests, compared to the rest of the developing world, Africa’s urbanization is both less related to income levels and proceeding more quickly relative to growth in real GDP per capita (World Bank 1999/2000, 2009; Fay and Opal 2000). Table 1 shows growth rates from 1970 to 2010 (2009 for GDP per capita). The fraction urban grew at an average annual rate of 2.09% in Africa compared to just 1.29% in the rest of the developing world. In contrast, over this same time period, real GDP per capita in Africa grew at a much slower rate: 0.53% per year versus 2.40% per year in the rest of the world. The differential in the average growth rate of the absolute urban population was even larger: 4.78% per year in Africa compared to 3.15% elsewhere, reflecting higher overall fertility rates in Africa (McGranahan, Mitlin, Satterthwaite, Tacoli, and Turok 2009). Consistent with these numbers, Figure 1a and Table 2 (column 1) show that the association between GDP per capita and urbanization levels that is present in much of the world is absent in Africa in 2010, as suggested in the literature (Fay and Opal 2000, Barrios et al. 2006, Collier and Venables 2007). However, two additional facts make us wary of drawing any conclusions from correlations between income and urbanization. First, for the period 1970-2010, there is no evidence of a significant long difference relationship between GDP per capita and urbanization, and this is true for both Africa and the rest of the world (Figure 2a and Table 3, cols 1 and 2). Second as noted earlier, GDP data are particularly suspect in Africa (Young 2012, Henderson et al. 2012). As such, these associations are measured with a high degree of error that may differ substantially across regions. 8 7 The two SSA countries lost are Namibia and South Africa. Restricting attention to countries with less than 50% of 1970 mean income, the average level of GDP per capita for SSA in 2009 is $1,651. 8 Concerns also exist about the quality of urbanization data (Potts 2012, Satterthwaite 2010), especially when censuses are not regularly conducted, as has been the case in many African countries. A 2013 study commissioned by the World Bank (but not yet released) suggests that the UN World Urbanization Prospects data, which we use, are better than other sources of data, reporting past urbanization fairly reliably, even if projections are questionable. We return to this issue later. 6 As argued in the introduction, we learn more about the path of African urbanization by considering it in relation to evolving accumulation processes rather than another outcome variable, such as income. Thus, when, consistent with endogenous growth models of urbanization (Lucas 2004, Henderson and Wang 2005), we instead view urbanization as a function of educational attainment, Africa looks like the rest of the world. More specifically, in Figure 1b and Table 2 (col. 3), we see that the relationship between urbanization and the average years of high school and college per adult for Africa is virtually identical to that for the rest of the world. Similarly, in Figure 2b (see also Table 3, col. 3), we see an association between changes in urbanization and growth in effective technology, as reflected by the growth of educational attainment, that is both significantly positive overall, and not significantly different for Africa. Later panel regressions will substantiate these associations. Africa’s faster rate of growth of effective technology, as determined by its pace of human capital accumulation, matches with its higher rate of change in urbanization. 9 Next we turn to sector differences. Africa is much less industrialized than the rest of the developing world. In 2010, manufacturing value added as a share of GDP was, on average, 10.6% in Africa, much lower than the 16% average in the rest of the world. This difference is more than accounted for by the higher average share of agriculture in Africa – 23% of GDP compared to 14% elsewhere (Table 1). The puzzle, which has been highlighted in the literature (e.g., Fay and Opal 2000; Collier and Venables 2007; Gollin et al. 2012), is that the distinct cross-section relationship between higher urbanization and lower agricultural share (higher manufacturing share) that exists elsewhere in the world is not present in Africa (Figure 3a and Table 4, cols 1 and 2). Figure 3b shows a negative long difference relationship between agricultural growth and urbanization in the rest of the developing world, but no relationship for Africa. However, the differenced relationships are all insignificant (Table 4, cols 3 and 4), although sample sizes are small. While we are concerned about the quality of African GDP data, we are even more concerned about the accuracy of the sector shares. Hence our econometric analysis will rely on trade data collected by more developed countries. 2. Conceptualizing urbanization In thinking about urbanization in developing countries, researchers have traditionally been guided by classic two-sector models of rural-urban migration (Harris and Todaro 1969; Williamson 1986; Mills and Becker 1986). Implicit in these models is the presumption that economic development involves a 9 We also obtain very similar results when we use average years of schooling at levels other than high school and college as our measure of educational attainment. 7 transition of employment from a backward rural sector to a modern urban manufacturing sector. 10 More recently, growth models of urbanization have endogenized the employment transition, modeling the switch from a rural or backward sector to an urban or modern sector as human capital accumulates. In Henderson and Wang (2005) this involves a shift from rural production to the production of urban goods. In other models, however, there is no shift per se in the good produced. Rather, the transition is from a sector in which production relies on land and unskilled labor to a sector which produces the same good, but instead using skilled labor (e.g., Lucas 2004, Caselli and Coleman 2001, Galor and Zeira 1993) and, in some cases, also physical capital (Galor et al. 2009). We do not repeat these models here. Rather, we simply adopt their presumption that the urbanization process is driven by human capital accumulation. Given this presumption, we explicitly consider the impact on this process of trade conditions and shocks, which affect the price of agricultural and natural resource products, relative to manufacturing goods (i.e. the terms of trade). We thus start with a typical two sector model of rural-urban migration, assuming a small open economy with both rural agriculture and urban manufacturing production. As is typical, missing from the specification are non-traded locally produced goods within both the urban and rural sectors. In many developing countries, it is typical for significant portions of rural employment to be engaged in non- agricultural, mostly non-export production, including locally manufactured and consumed food products, plates, cups, cooking utensils, furniture, apparel, and trunks, as well as housing and retail services. For example from the IPUMS in India (2004), Indonesia (2010) and Vietnam (2009), the percentages of non-agricultural employment in the respective sectors were 32, 40 and 35. For Sub- Saharan Africa, a) the only lower income countries with IPUMS data from the last 10 years are Malawi (2008), Sierra Leone (2004), and Sudan (2008), with non-agricultural employment percentages of respectively 34, 12, and 44. Urban sectors have huge local non-traded good sectors. Adding a non- traded component to each of the urban and rural sectors won’t change our general results. 11 But the point is that, in the data, “rural� is not just agriculture and “urban� is not just modern manufacturing. To facilitate a focus on rural-urban transitions in the classic two sector model, we embed this in an endogenous growth model of urbanization based on Henderson and Wang (2005).12 Within this 10 This is consistent with the common portrayal of cities as “engines of growth� (see, for e.g., World Bank 1999/2000, 2009). 11 However if all such non-agricultural goods are defined to be produced only in the urban sector (and exported domestically to the agricultural sector) as in Gollin et al (2012), then one can get fundamentally different results. 12 Henderson and Wang’s model is itself adapted from Black and Henderson (1999), with a sophisticated version in Rossi-Hansberg and Wright (2008) that adds in a stochastic process, multiple rather than just two sectors, and physical as well as human capital accumulation. 8 framework, human capital accumulation fuels urbanization under a fairly straightforward set of conditions, which we note later. We focus on how urbanization is influenced by both the terms of trade and national population growth. In the rural sector per person production and income is given by θa =I a pa ha g ( K / N a ), g ' > 0 (1) where I a is per person rural income. The RHS of (1) contains the per person production technology. pa is the price of agricultural output. ha is per person human capital (0 <θa< 1) and combines both skill level and sector level of effective technology: the ability of rural workers to absorb and adapt global agricultural technologies. K / N a is the land to labor ratio in the sector. Farmers are assumed to each have an equal claim on returns from land as under, for example, communal based land ownership in Africa (Bruce 1998). As such, they receive their average product as specified in (1). The total value of θa agricultural output is pa ha g ( K / N a ) N a . In the Appendix, we refer to a Cobb-Douglas version of ψ 1−ψ θ technology where the total value of agricultural output is instead given by pa ha a K N a . Total population, N , is divided between the urban and rural sectors, or N= a N−N (2) where N is the urban population. In the urban sector, there is a single city 13 where residents commute to work in the center, which is a point. In urban models of a city, there are external scale economies in production that generate initial benefits to having a larger city, but there are also diseconomies in terms of potential work time that is lost in commuting, which increase as city population and spatial area rise. In such models there is a residential sector spatial equilibrium with a rent gradient, where rents for people who live further from the center decline monotonically, as a compensating differential for increased commuting times. In most models, the city is closed by having total land rents collected distributed as an equal dividend to all residents. We give a summary expression for per person urban real income: θu =I u hu f ( N ), f N ( N ; N > N *) < 0 (3) Urban income, I u , which, in equilibrium, is the same for all workers in the city, is defined to be income available to spend on market goods apart from residential land, where all workers consume a lot of fixed size. The price of the urban good is the numeraire and hu is human capital per worker in the urban sector. We assume that returns to human capital are higher in the urban than in the rural sector, 13 Total urban population is, therefore, given by the population of this city. 9 i.e. θu > θ a where 0 <θu≤ 1. It is this differential in returns that drives urbanization in this open economy model. f ( N ) is assumed to be an inverted-U shaped function of city population, which achieves its maximum at N * . In models with multiple cities, in a stable equilibrium, all city sizes must be equal to or greater than N * , in which case urban real incomes decline as city size expands. Here, as we will see later, as the urban population expands relative to rural, for stability we need to assume that the rate of growth of urban incomes is less than that of rural incomes. A specific example from Duranton and Puga = (2004) gives the form θu I u hu N δ (1 − τ N )1+δ to city real income where δ is the degree of scale externalities and τ the cost of commuting a unit distance in the city. We analyze the allocation of labor between the urban and rural sectors in a small open economy, where agricultural goods are exported internationally, as are urban goods. We note all of the developing countries in our sample exported some modern manufactured products even in the 1960s. With both products traded internationally, we close the model for an instantaneous equilibrium by assuming an absence of rural-urban migration costs. Labor thus moves between the urban and rural sectors to equalize real incomes. Substituting in N= a N − N and equating (1) with (3) we have 14 θa θu Ia =pa ha g ( K / ( N − N )) =Iu =hu f (N ) (4) Our estimating equation will be based on a differenced version of (4). We define: θa θu M ≡ d [ pa ha g ( K / ( N − N )) − hu f ( N )] / dN > 0 (5) M is positive in a stable allocation between the urban and rural sectors. That is, as the urban sector expands, the rise in productivity in the rural sector from an increased capital-labor ratio, K / ( N − N ) , exceeds that in the urban sector, which, under the assumption that N ≥ N * , is less than or equal to zero. In our estimating equation we look at the change in the urban share over time or d ( N / N ) . Differencing (4) and rearranging into the estimating form for a LHS of d ( N / N ) we have N  θa KN dN θa K dK θa dp dh dh  =d M −1 N −1 [ pa ha g '(⋅) − MN ] − pa ha g '(⋅) + pa ha g (⋅){− a + [θu u − θ a a ]} (6) N  (N − N ) 2 N (N − N ) K pa hu ha  While dN / dN > 0, the effect on the urban share, N / N , of an increase in national population is ambiguous. Everything else is straightforward. An increase in the relative price of the agricultural good decreases urbanization, because the enhanced returns to agriculture draw people out of the urban 14 θu θa Or, using the functional forms mentioned above, I u = hu N δ (1 − τ N )1+δ = Ia = pa ha K ψ ( N − N ) −ψ 10 sector. An increase in land reduces urban population and share as agriculture becomes more productive. For human capital, as in growth models of the urbanization process, we generally assume dhu / hu ≥ dha / ha . Then, given the assumption θ u > θ a , with fixed terms of trade, human capital growth leads to increased urbanization, given the relative gain in labor productivity in the urban versus rural sector. Given its non-linearities, this model does not, in the closed economy case, lend itself to simple closed form solutions (see Henderson and Wang, 2005). As explained in the Model Appendix, for the specific functional forms we use here for a small open economy, we simply assume that there is eventual convergence to an approximate steady-state and that developing countries are still accumulating human capital. In general, urban human capital levels exceed rural, because the return to human capital is higher in the urban sector. Growth rates of human capital in the two sectors may, however, be the same. Our empirical results on Africa will conform to the results in equation (6). Most notably, apart from the effects of changes in human capital, we find that positive shocks to African agriculture associated with international trade tend to reduce the urban share, in contrast to some of the literature. However, for non-African countries we will find that increased prices in agriculture spur urbanization. Given the predictions of our simple model, what might explain this result in non-African countries? We consider two additional features that might lead to this result: (1) monopsony purchasing by state boards and (2) a class of landowners that invests agricultural profits in cities. 2.1 Monopsony purchasing in agriculture A government purchasing board that exercises monopsony rights to agricultural products fixes prices paid to farmers at below market prices, pa , selling at an international price in excess of pa and distributing surplus rents. In this context, the impact on urbanization of a positive shock to agricultural prices will depend on who receives the additional surplus. In particular, if the additional surplus is distributed to the representative urban resident, then the urbanization response will be positive. However, if the additional surplus is instead distributed only to elites, there will be no urbanization response because, in these circumstances, the shock will not change the margin of choice for the representative rural-urban migrant. 15 To see this more clearly, consider an example. The income to a rural resident is now given by θa pa ha g ( K / ( N − N )) . Since rural and urban incomes are equal, national demand for domestically θa consumed agricultural output can be written as D( pa ha g (⋅), pa ) N , where for simplicity we assume all 15 This is the case irrespective of whether the surplus goes to urban or to rural elites. 11 θ domestic consumers pay pa . The volume of agricultural exports is then ha g (⋅)( N − N ) − D(⋅) N and the a θ surplus ( pa − pa )[ha g (⋅)( N − N ) − D(⋅) N )] . If this surplus is divided equally among urban residents we a now have a new urban-rural migration equilibrium condition: θa θa θu pa ha g ( K / ( N − N )) = ( pa − pa )[ha g (⋅)( N − N ) − D (⋅) N )]N −1 + hu f (N ) (7) Given the stability requirement that  pa ha M1 ≡ d  θa ( θa g ( K / ( N − N )) − ( pa − pa )[ha θu g (.)( N − N ) − D(.) N )]N −1 − hu )  / dN > 0 , it is f (N )  straightforward to show that θa = dN / dpa M 1−1{[ha g (⋅)( N − N ) − D(⋅) N )]N −1} > 0 . (8) The problem with this explanation for the positive response of non-African urbanization to agricultural price shocks is that monopsony purchasing boards have historically been more prevalent in Africa than elsewhere (Jayne and Jones 1997, Binswanger and Townsend 2000, Block and Bates 2011). Given this, we turn to a second potential explanation based on another important difference between Africa and the rest of the world. 2.2 Private ownership of land Our model treats rural land as communally owned, and this does fit Africa better than it does the rest of the developing world. How would private ownership change the comparative statics? In particular, could increases in agricultural land rents be used in a way that attracts rural residents to cities? To consider this question, we introduce a class of landowners, distinct from workers, as featured in overlapping generations growth models such as Galor and Zeira (1993) and Galor et al. (2009). Such models have a “joy of giving� bequest formulation, in which the older generation works, consumes, and invests their bequest in some form of capital for their children to receive in the next period. These are one sector models with a backward (rural) and a modern (urban) technology. But if multi-sector versions of these models were developed, they could offer insights about how responses to agricultural price shocks vary between communal and private land ownership regimes. Consider a small open economy with sectors producing an agricultural good, a modern manufactured good, and a “traditional� manufactured good (e.g., furniture, food utensils, and apparel) which is produced in the rural, as well as potentially the urban, sector. In a set-up like this, increased export demand for either agriculture or modern manufacturing can promote urbanization. 16 To see this, suppose agriculture is produced with land and unskilled labor, modern manufacturing with physical and human capital, and traditional manufacturing with unskilled labor. Human capital augments unskilled 16 This result will not be possible in a two-sector model with one relative price. 12 labor and requires investments by parents (as part of their bequest). Land is owned by a class of landowner dynasties, in which parents have the knowledge and skills to invest their bequests in physical capital in the urban sector. Urban residents could also invest bequests in physical capital. Rural unskilled labors have neither the mechanisms nor information to undertake such investments. In some versions of such a model, there would be no capital market per se, so all investments are part of bequests. In this context, an increase in the international price of agriculture has two effects. First, it raises the marginal product of unskilled labor in farming, deterring urbanization. But any institution that limits the responsiveness of farm wages, such as tenancy, can reduce this effect. The second effect of the increase in agricultural prices is to raise the return to land and bequests of landowners. This raises their investment in physical capital, which raises the marginal product of efficiency units of labor in the urban sector. Depending on the magnitude of the first effect relative to the second, the wages in agriculture relative to manufacturing may rise or fall and urbanization may be deterred or spurred. 17 In many developing countries outside of Africa, agricultural land has been privately owned for at least the last 150 years. Landowners have invested agricultural surpluses in physical capital, starting with small village businesses serving agriculture before moving into bigger businesses such as large scale food processing and textiles and eventually more modern manufacturing. Sugar cane plantation owners in Brazil, for example, progressed from food processing to modern light manufacturing (Baer 1979). In the vast majority of African countries, by contrast, rural land is communally owned (Bruce 1998). Such land may be allocated by a village head, with use and, perhaps, control rights. However, if such allocations do not include transfer rights they cannot be used as collateral. And, in general, direct collection of land rental income is rare. In essence, there therefore exists some large class of unskilled farmers with small holdings who earn their average product. This difference in landownership may explain some the difference in the effect of agricultural price shocks on urbanization in Africa versus the rest of the developing world. 17 Galor and Zeira (1993) also consider both inequality in land holdings and capital market imperfections. If there is no capital market, having a class of landowners means those with larger holdings have larger bequests to invest in physical capital, which is even more important if there are fixed costs to investment which must be overcome. With imperfect capital markets, small human and physical capital loans may be largely unsecured, raising the costs of borrowing. Human capital collateral is subject to the no-slavery constraint and the secondhand market for immobile, rapidly depreciating special-order machinery used in manufacturing may be very limited. If there is a capital market, families with larger landholding, incomes and bequests can avoid the high costs of unsecured borrowing and are more likely to invest in their own businesses (with some degree of fixed costs). And they can also use their land as collateral to reduce their borrowing costs. In a related political economy story, landowners make urbanization unattractive by under-investing in public education to maintain cheap farm labor (Galor et al. 2009). A related empirical literature focuses on tenancy reform and its impact on poverty and agricultural productivity in India (e.g., Besley and Burgess 2000). 13 3. Econometric specification and base results The econometric specification is based on equation (6). The base specification examines changes in the urban share as a function of two variables: 1) the national population growth rate and 2) the growth rate in the average years of post-primary (i.e. high school and college) education in the adult population (over age 25), which represents both changes in effective labor units and changes in effective technology. Time invariant country conditions, including geography, culture and underlying institutional determinants, are differenced out. Such features could also affect growth per se and we consider this in later variants of our specification. For now, we note only that national land area ( K ) may interact with other covariates. For example, in a small country, the urbanization response to rapid population growth may be stronger than in a large country. The basic specification for country i is N ∆10   =  N it ( ) b0 ∆10 hit + b1∆10 ln N it + b2 ln K + b3 ln K ∆10 ln N it + β∆10 X i ;t − s + Tt + ∆10 eit . (9) In this specification, the unit of analysis is first-differences at 10 year intervals, as denoted by ∆10 . We N pool 4 time periods: 1970-80, 1980-90, 1990-2000, and 2000-10. 18 ∆10   , ∆ 10 h , and ∆ 10 ln N are N changes in the urban share, human capital (and, hence, effective technology), and national population, respectively. We include time dummies (Tt) to account for global changes in, for example, the world technology level to which all developing countries are adapting. The X term in different specifications includes factors that shock income such as commodity price indices, trade partner income indices, rainfall, and institutional influences. When these are time varying, we generally lag the differences by 2 or 5 years to allow the shocks time to have an impact. We also sometimes smooth measures such as prices and rainfall to reduce noise and give more weight to more persistent shocks. Finally, we cluster errors by country to account for possible country-specific serial correlation (Stock and Watson, 2008). 18 Since censuses in this sample are carried out at least 10 years apart, using shorter intervals where data are interpolated is unlikely to add any new information to the dependent variable. 14 3.1 Identification Our chief estimation concern is identification. Everything is potentially endogenous and while we can estimate correlations, claiming causality is a challenge. We address this problem in two ways. First, we consider correlations among country conditions and spell out the assumptions required for effects to be causal in our base case. Second, in later sections, we construct plausibly exogenous shocks to gain pseudo-randomization. What do we need for identification of the base specification? Let’s start with national population growth. Consider a time-varying unobservable such as the urban-rural differential in quality of health care. If such an amenity enters the utility function separably, the effect of differences in relative quality levels is differenced out in the growth specification. But relative differences are likely to change over time. Suppose that over the decade t − 10 to t , urban health care improves both absolutely and relative to rural health care. This improvement could draw more rural migrants to cities thus having an unobserved direct effect on urbanization. But it could also reduce national population growth, if better access to health facilities increases the provision and social acceptance of birth control in urban areas, reducing birth rates aged 5-30 over the decade in cities and therefore also nationally. If both links exist, our estimated effect of population growth on urbanization will be understated. Of course better relative improvements in health care in cities could also reduce national death rates with an opposite bias. Education presents a related endogeneity concern. Relative improvements in urban education facilities in a decade may cause both a move to cities and directly enhance national educational achievement. Here, however, timing may alleviate the problem, making it less salient. We consider educational attainment of the population age 25 or older. The change in this variable between t and t − 10 includes information only about people who were aged at least 15 in t − 10 . It may be reasonable to assume that decisions about their educational progress were effectively made before contemporaneous changes in relative urban-rural education quality. However, if changes in rural-urban educational quality differentials are correlated across decades, then the problem remains. We first present results using this base specification. Then we turn to trade shocks to the modern sector, rain and price shocks to the agricultural sector and the role of institutions. 3.2 Base results Table 5 presents results from the estimation of our base specification. In column 1, urbanization is positively and significantly affected by both the growth of national population and the growth of effective technology. In column 2, we estimate the same relationship as in column 1 for a restricted 15 sample in order to reduce concerns about extrapolation in the UN projections (Potts 2012). Specifically, we limit to country-years no more than 5 years after the most recent census. Results are very similar. In column 3, using the original sample, we add interactions with an SSA dummy for all variables (including time dummies) and report its effect on slope coefficients. 19 The SSA differential is not significant for growth in national population or in effective technology, individually or jointly. The point estimates are not small, but small sample sizes limit precision. If anything, changes in effective technology have stronger effects in SSA. 20 If we use alternative measures of education, such as the fraction of the population over 25 with at least some secondary schooling, we get similar results. In column 4, we add land area and land area interacted with national population growth to account for changes in density. As might be expected, the effect of population growth on urbanization is more modest in countries with larger areas and thus more farm land on average. We keep these additional variables in our base specification for the rest of the paper. 21 Finally, in column 5, we find no compelling SSA differentials in the column 4 specification. How big are these effects? In column 4, for a country of average ln(land area), a one standard deviation change in the national population growth rate leads to a 0.57 percentage point increase in the percentage of the population classified as urban. For a country with one standard deviation more land, the impact of a one standard deviation change in population growth on the urban share drops to 0.043. A one standard increase in the growth of education (0.30 years) leads to a 1.05 percentage point increase in the growth of urban share (equivalent to 0.28 standard deviations of growth in urban share). The average change in urban share per decade in the sample is 4.67, so our results suggest that a standard deviation increase in education growth would raise this to about 5.7. As we discussed above, both population growth and education are potentially endogenous. However, an instrumental variables strategy poses two major problems. First, we are unable to independently predict both population growth and education growth. Possible instruments include 19 Whenever we introduce SSA differences in a specification in the paper we always interact all covariate coefficients including time dummies. We only, however, report the interactions for the current variables of interest. Although interacting all covariates reduces precision, it also reduces the risk that we are misinterpreting other interaction effects as our effect of interest. 20 If we measure change in effective technology using growth in GDP per capita, we get similar, albeit slightly weaker results. Explanatory power falls and the ∆10lnGDPpc coefficient is insignificant at the 5% level. The ∆10lnGDPpc coefficient (s.e.) and the coefficient on ∆10lnGDPpc interacted with SSA are respectively 2.858 (1.455) and -0.705 (1.960). Note here the SSA differential is negative. In our data, while the relationship between ∆10lnGDPpc and ∆10education is positive in the rest of the world it is negative in SSA. We believe the GDP per capita relationships are weak and distorted because of the measurement problems with GDP discussed above. 21 We also interacted education with land area. The interaction term is positive but insignificant, and the education effect at mean size is the same as in column 4. The coefficient (s.e.) is 0.453 (0.688). 16 lagged (t − 20) population shares of both youths (aged 0-15) and the elderly (over 65), and sex-specific lagged education measures. These generally predict growth in both population and education, and the joint Kleibergen-Papp F-statistic is, at best, between 3 and 4. The second problem is that these variables are unlikely to meet the exclusion restriction, since all may affect contemporaneous migration decisions. 4. Trade shocks to agriculture and manufacturing In this section we examine the effect of exogenous trade shocks on urbanization, with an emphasis on how Africa might respond differently to these economic forces. We start with trade shocks to agriculture and also natural resources, sectors where we will argue shocks are better defined, before turning to manufacturing. 4.1 Shocks to agriculture and natural resources We start this section looking at shocks to agriculture overall, and that is our main focus. Then in the last part we distinguish food and non-food agriculture, as well as consider natural resource price shocks. 4.1.1 Agriculture We examine two kinds of agricultural shocks: rainfall and international agricultural prices. Our rainfall measure is based on an annual average aggregated for each country from interpolated gauge data. For price shocks we use an index suggested in Bruckner and Ciccone (2010) and Collier and Goderis (2009). Each country is a long term exporter of a specific set of agricultural commodities such as coffee and cocoa, with their individual export bundles depending primarily on physical geographic factors. As world prices fluctuate for these primary products, so do potential incomes of exporting countries. Our model in Section 2 (equation 6 in particular) suggests that increased agricultural product prices will expand agricultural employment and slow national urbanization. However, as equation (8) suggests, if the surplus generated in agriculture does not go to farmers, but is spent instead in cities, then higher prices will instead fuel urbanization. The expression for the export price environment in country j in time t is PE jt = EXSH j ,1962 −9 ln PI jt , (10) n PI = −1 [� pktkj ,1962−9 ]CPIUSAt , ∑ akj ,1962 −9 = n a =jt k 1 1 k =1 Following Collier and Goderis (2009), we use normalized international prices pkt (in current US $) in year t for commodity k, and weight by the historical (1962-69) share of agricultural commodity k in country 17 j ’s total commodity exports, akj ,1962 −9 from Feenstra et al (2005). By construction, weights sum to unity for each exporter. We deflate by the USA CPI because the prices are reported in nominal terms. We chose 24 agricultural products for which we have consistent price series, and for which trade data on a relatively unprocessed form of the good exists. 22 For the price environment, PE jt , we take the log of the index PI jt to ensure that final effects will be invariant to units of commodities used to define unit prices and then multiply by EXSH j1962 −9 , the (average) share of agricultural exports in country j’s GDP in the 1960s. This latter adjustment allows for changes in the price of an export commodity to have a bigger impact on the price environment for countries that export more commodities overall. The 10- year change in the price environment, ∆PE jt , enters the regressions below. 23 We adjust both the price and rainfall data in two further ways. First, we lag the shock by 5 years to allow time for an urbanization response. 24 Hence, the shock variable is ∆PE jt −5,t −15 and we model urbanization during, for example, the 1970s, as a function of price changes from 1965 to 1975. Second we smooth by using three-year centered averages around t − 5 and t − 15 (for rainfall and for each commodity price deflated by the CPI before applying weights). This reduces noise and places greater weight on more persistent shocks. 25 Results are reported in Table 6. We show the price and rainfall shocks together. Coefficients on each are robust to the inclusion of the other. If African weather conditions are affecting world prices for any product then by including both we are separating out the effects. In column 1, rainfall has a negative effect, consistent with the story that rainfall improves agricultural productivity (effective land, in terms of our theoretical model), raising incomes in agriculture and deterring urbanization. Consistent with Barrios et al (2006), the estimated coefficient on the SSA interaction with rainfall in column 2 is negative, implying that the negative effect of rainfall on urbanization is larger in Africa than in the rest of the world. However, unlike Barrios et al, we do not find this differential to be significant. We were surprised by the weakness of our results in this area. 22 These goods are listed in the data appendix. 23 In an attempt to expand the sample we also experimented with the use of weights based on 1970-72 data in the definition of the price environment. GDP data are available for a greater number of countries in this period. The results we obtained were similar to those reported. However, since 1970-72 falls within our sample-period, our preferred results are those based on the 1962-69 weights. 24 We also report results with a 2 year lag. 25 If fewer than 3 years of data are available for a particular year-variable we smooth over the number of years that are available. 18 Our main results concern agricultural price shocks. In column 1 for the overall sample, there is a positive but insignificant price shock effect. Column 2 indicates a strong and highly significant positive effect of 9 on urbanization for non-SSA countries; and a strong net negative significant effect of about - 7.6 for SSA countries. 26 For the rest of the world, a one standard deviation increase in the price shock variable (0.046) leads to 0.41 increase in the urban share, where the average increase is 4.74. In SSA, such a price increase leads to a 0.35 decline in the urban share. To consider the possibility that our results are driven by outliers, we iteratively dropped individual observations whose absence had a noticeable impact on the coefficients of interest (the price shock variable and the SSA differential for that variable). Doing so only strengthened the effects. 27 In column 3, we show weaker results with a 2- year lag structure. Why do results differ for African versus non-African countries? No data that is consistently available across countries quantifies the phenomenon of capturing agricultural rents and spending them in cities or not. We consider government consumption as a share of GDP as a crude proxy. A regression of changes in government consumption on our base covariates, agricultural price shocks, and all SSA interactions yields a coefficient (s.e) on the agricultural price shock variable of 0.125 (.0657) and the SSA interaction of -0.234 (0.219). This is modest support for the idea that surpluses fund increased government procurement of actual goods and services delivered in cities outside of Africa, with no such reaction in Africa. In section 2.2, we also speculated that land ownership systems might drive differentials between SSA and non-SSA. We characterized Africa as having communal ownership where there is no separate land rent income and land cannot be used as collateral, while the rest of the world is more likely to have a class of landowners who might invest surpluses in urban businesses. In the absence of a data set on land ownership in non-SSA countries, we cannot test this hypothesis directly. To investigate whether landownership might relate to differentials within Africa, we consider measures of land ownership systems from Bruce (1998). Nine out of 27 African countries in our sample have significant (not necessarily dominant) private ownership. When this private ownership dummy is interacted with the price shock variable in the SSA sample, the prick shock coefficient is -17.8 and its interaction with the dummy is 8.2. While the sign of that interaction is consistent with the land ownership hypothesis, it is 26 When estimated on the SSA sample, the price index coefficient (s.e.) is -7.59 (3.07), significant at the 5% level. 27 We ran the “dfbeta� command in Stata to find these outliers. First, there are no major outliers of the kind researchers worry about in a small sample: the dfbeta numbers on the two coefficients are all under 0.45 in absolute value. Dropping outliers, defined either as dfbeta values over 0.35 in absolute value or the one country with repeated outliers, strengthens the results. 19 insignificant. And it says nothing about the comparison to those NSSA countries where private ownership dominates. Another potential reason for a differential is that effects are somehow conditional on the “stage of development�, an issue we will explore for modern manufacturing development. We tried the 1965 level of our education variable as a measure of overall “stage of development�. The base price shock coefficient is 9.92 and its interaction with education is -14.2. Neither is significant and the interaction coefficient is opposite in sign to expectations given that SSA countries have lower education levels. Differences in effects might alternatively be related to geography. We considered geographic variables affecting agriculture (Collier and Venables 2007), including distance from the equator, rainfall, fraction of the country in tropical zones, and agricultural potential as modeled by Ramankutty et al (2002). None of these variables has noticeable, let alone significant, interactions with agricultural price shocks. Below, we look at interactions with political/institutional variables. They also have no effect. 4.1.2 Non-food agriculture and natural resources Our results differ from Jedwab (2012). For two African countries, he finds growth is spurred in towns local to areas experiencing positive agricultural rent shocks. His work does not deal with overall national urbanization per se, and some local towns may or may not be defined to be in the urban sector, so his specific results are not necessarily inconsistent with ours. However the general claims differ. In this paper positive shocks to agriculture deter urbanization in Africa, while Jedwab (2012) and Gollin et al. (2012) argue the opposite. To be fair, Jedwab (2012) focuses on “resource rents�, in his case rents from the production of non-food agriculture, whereas we draw on data for all agricultural exports. For comparison, we formed a category of “non-food� products (palm oil, coffee, cocoa, linseed oil, wool, tobacco, cattle hides, copra, sisal, rubber, tea and cotton). Column 4 shows results for this class. Both the SSA and non-SSA price coefficients increase in magnitude, with the same signs as in the specification including food crops, but lose significance. The results are again consistent with consumer cities in the rest of the world but not in SSA. Finally we turned to natural resource rents from both minerals and oil and gas. Collier and Venables (2007) presume increases in rents make rural populations less mobile and reduce urbanization in Africa. A more typical story is that rents and aid are captured by urban residents and spent in cities, increasing urbanization (Ades and Glaeser 1996; Davis and Henderson 2003). Analogously to our strategy for agriculture, we look at exogenous price shocks to the natural resource sector. Natural resources are listed in the Appendix; note the absence in the data of gold and diamonds. We get exactly the same SSA and non-SSA divide we had for agriculture for natural resource shocks, but in this case results are 20 statistically insignificant. Moreover, as soon as we consider outliers, all effects go away. 28 We therefore conclude there is no strong evidence that the limited set of natural resource price shocks in our data affect urbanization either inside or outside of Africa. 4.2 Trade shocks to manufacturing Manufactured products are highly heterogeneous within the categories reported in the trade data, so, unlike agriculture, there are not price data. Instead, we consider shocks to demand, as proxied by income shocks in high-income importer countries. The underlying assumptions are that: (1) demand for certain manufactured products is income elastic, so increased incomes generate demand shocks for manufactured exports from less developed countries, and (2) such products tend to be produced in urban rather than rural areas. While these shocks are probably less salient than price shocks, we focus on a class of goods we call modern manufactured products that: (1) may have relatively high income elasticity, and (2) are relatively more affected by urban scale economies in developing countries (Henderson et al. 2001). In order to focus on the income effect of these shocks, we use data on trade flows from a fixed period (1962-69) before our period of interest. We limit our sample of importers to Organization for Economic Co-operation and Development (OECD) members as of 1974, excluding Turkey, which is a developing country based on our definition in this paper. These 23 countries purchase the vast majority of African exports. 29 The data, as recorded by the importer, are from the NBER trade database (Feenstra et al. 2005). Our demand shocks are income growth of OECD countries applied to historical gross exports (“historical pathways�) from developing country j to OECD country i . Exports of commodity c from country j to OECD country i are denoted b jic ,62 − 69 and national income by GDP. The “62-69� refers to b jic ,62 − 69 1969 b jic , s the fact that GDPj ,62 − 69 =[ ∑ s =1962 GDPj , s ] / 8 , so the weights are an average for the 1960s, and define the historical importance of exports of commodity c from a given developing country to a given OECD 28 The overall coefficient (standard error) on the natural resource price shock and the SSA differential are respectively 12.8 (10.8) and -18.6 (12.1). With removal of just one outlier (|dfbeta| > 0.5 ) these become 0.93 (14.3) and -6.8 (15.2) respectively, with further deterioration if 3 outliers (|dfbeta| > 0.3) are removed. 29 The list is Australia, Austria, Belgium-Luxembourg, Canada, Denmark, Finland, France-Monaco, Germany, Greece, Iceland, Ireland, Italy, Japan, Netherlands, New Zealand, Norway, Portugal, Spain, Sweden, Switzerland- Liechtenstein, the United Kingdom and the United States. Countries linked by a hyphen in this list are treated as one in the trade data. Similarly, one region these countries import from in some years is the “South African Customs Union“, which includes South Africa, Botswana, Namibia, Swaziland and Lesotho. We apply those base numbers just to South Africa as it is by far the largest economy in the group. 21 country.30 The shock weighted by the size of this pathway is ∆10GDPi ,t −l , the change in OECD country i ’s GDP from t to t − 10 , lagged, in general, by l years to allow time for developing country urbanization to respond to trade shocks. We use a 5-year lag. Hence, we consider, for example, 1970-80 urbanization in developing country j as a function of weighted OECD income changes from 1965 to 1975. For income demand shocks, results for a 2-year lag are very similar. More specifically, the demand shock index for country j that we use in a first differenced equation is:  b jic ,62 − 69  =DI jct ∑  GDP  ∆ GDPi ,t −l  10 (10) i  j ,62 − 69  The income growth shocks use unlogged changes in real GDP measured in PPP units, thereby implying bigger importing countries have a bigger impact. We define these shocks for several commodity classes, c : (1) heavy manufactures (ferrous and non- ferrous metals and metal products in SITC 67-69), (2) apparel and accessories (SITC 83-86), (3) chemicals (SITC 51-59), (4) traditional products made in rural villages, in addition to other places (leather, rubber & wood products, non-metallic minerals, textiles, wood & paper products, furniture and fixtures in SITC 61-66, 81, 82), and (5) modern products (SITC 71-79, 87-89). We are most interested in modern, which covers non-electrical machinery, electrical machinery and telecommunications equipment, transport equipment, instruments, and miscellaneous (arms and ammunition, toys, musical equipment). This is where we think urban scale economies in developing countries are likely to be strongest, in contrast to what we expect to see for traditional industries and resource-based manufacturing carried out in smaller towns and rural areas. The SITC system is not well-suited for some of these categories, especially modern, so we tried some variants, with little effect on results. Total exports (not just to OECD countries) of manufactures from our sample of developing countries increased from 2.2% to 6.8% of their aggregate GDP between 1962-69 and 1998-2000. Interestingly, the SSA countries start more manufacturing-intensive and end less so than non-SSA countries. 31 The big growth is in the modern sector, which starts at an average of 0.22% and increases to 2.7% by 1998-2000. 30 As an alternative to applying fixed weights from the 1960s, we considered time-varying weights for the period 14-16 years before each of our sample years. This might increase the explanatory power of the proxy as it would incorporate the effects of import destinations that have become prominent since the 1960s. However, we worry about endogeneity, since other factors affecting urbanization may also affect exports, and so any results could be driven just by this rise in exports which is correlated with rising urbanization (as opposed to the OECD income shocks). Thus we rely on the results using the 1960s weights for all years. 31 Eleven countries in our main sample lack 1960s trade data: Albania, Bangladesh, Botswana, Bulgaria, Laos, Lesotho, Malta, Mongolia, Namibia, Swaziland and Vietnam. 22 For modern, SSA countries start higher than non-SSA, at 0.32% versus 0.15%, but end lower at 2.5% versus 2.9%. If one removes Liberia, which reported substantial exports but a GDP reduced massively by war in 1998-2000, the ending SSA value is reduced to 0.38%, raising the regional differential considerably. All countries in our sample report some modern exports in the 1960s, so these changes are not on the extensive margin at the country level. Of the five separate manufacturing commodity classes, modern products is the only one for which OECD income shocks have a significant positive effect on urbanization, either overall or for non-SSA or SSA countries, in all specifications of lag structures and bases. Traditional manufactures has a significant negative effect overall, which is consistent with village-based production in the rural sector. As our focus is on urbanization, the remaining results emphasize modern products, though some formulations also include all manufacturing commodity classes with the exception of modern products grouped together as a control. We report our results in Table 7. In column 1, the OECD income shock variable for the modern sector is positive but not significant. Column 2 shows that this result masks two sharply different regional effects. For the rest of world (NSSA), the income shock effect is positive and significant. A one standard deviation ($4.1 billion) increase in the modern shock leads to a 1 percent point increase in the change in the urban share which has a mean of 4.58 in the overall sample (4.81 in NSSA). The effect is both sharp and large. Column 2 also shows that the net effect (25.38 – 23.12) for SSA countries is zero. To consider the possibility that our results are driven by outliers, we dropped individual observations whose absence noticeably changed the coefficients of interest (the modern manufacturing income shock or the SSA differential for that variable). There are only two such observations, and our results are robust to their removal.32 As mentioned above, our definition of modern manufacturing is not ideal. In particular, we counted SITC code 89, “Miscellaneous manufactured articles� as modern, and it is a substantial category for some countries. As a robustness check, we removed some sub-categories that are obviously not modern such as works of art, collectors’ pieces and antiques (896), basketwork (8997) and small wares and toilet articles (8998) with essentially no effect on results. In column 3, we control for shocks to all other (i.e. non-modern) manufacturing sectors. It has a negative coefficient with an enhanced negative effect for SSA countries, but those effects are not significant. There is no impact of the added controls on the modern sector results. 32 We used the dfbeta command in Stata with the same 0.35 criterion used with the price shock data. Two observations are dropped, and the coefficient on the basic modern manufacturing shock variable increases in absolute magnitude to 4.37, with SSA countries still having a net zero effect. 23 Why do modern sector trade shocks not spur urbanization in Africa, while they do in the rest of the world as expected? To answer this, we considered whether many African countries with low levels of human capital lack the capacity (in terms of labor force quality and institutions) to respond to trade shocks to the modern sector. We use mean years of high school and college in the adult population in 1965 as a pre-determined control for capacity. As noted earlier (see Table 1), African countries have much lower education levels on average, but there is an overlap in the distributions. The top quartile of SSA countries is above the bottom quartile of NSSA countries. In column 4, the base coefficient on the modern sector shock is insignificant and negative, but the interaction with 1965 education is positive and highly significant. The point estimate of the base coefficient implies that income shocks have a positive effect starting at a low level of 0.16 average years of education. For the mean education in the overall sample of 0.34, a one standard deviation increase in the trade shock increases urban growth by 0.25 points ((-5.13 + 0.34*32.7) * 0.041 hundred billion), a large effect that increases with development. Interacting all covariates with the education variable yields almost identical coefficients on the shock variables. We tried a horse race for column 4 to see if we interacted everything with SSA whether the education variable worked the same in both regions. This is too much for the data: none of the relevant coefficients are significant at the 5% level and only one at the 10% level (SSA interacted with education X the shock variable). The point estimates of the shock effects are generally positive for most non-SSA country education values and for a number of SSA county values as well. 33 In column 5, we allow 2 years for shocks to take effect rather than 5, but the results are essentially the same. 5. Institutions In this section, we examine the role of general political and institutional development on urbanization. While weak institutions are widely believed to inhibit development, we ask whether they have any further effect on urbanization, conditional on the degree of development. One hypothesis in this regard is that countries with weak underlying determinants of institutions may find it harder to upgrade from the informal institutions that tend to prevail in rural societies to the more formal, and structured, institutions that are characteristic of urban settings. As such, these countries may be less likely to urbanize. This question is difficult to study empirically, because urbanization may also affect the quality of institutions. Key governmental institutions are located in cities and employ urban residents. Governments may look increasingly to urban residents for political support as urbanization increases. 33 The coefficients on the base shock, that shock interacted with SSA, the base shock interacted with education, and that interaction interacted with SSA are respectively 0.00021, -0.00024, -0.000187, and 0.00033. 24 We address this question initially by using the well-known polity2 and related measures that attempt to capture the degree of democratization and authoritarianism and quality of institutions as they evolve over time. In an effort to mitigate the endogeneity, we use level measures from 15 years ago relative to the present, and also 10-year changes with a 10-year lag (i.e. change in the decade prior to the current decade’s urbanization change). However the issue of endogeneity remains critical. Thus, we also focus on two measures of long pre-determined historical circumstances that are likely to shape current institutions. First, ethno-linguistic fractionalization (ELF) from the early 1960s, used by Easterly and Levine (1997), Alesina et al. (2011), and others, represents the probability that two independently drawn random residents of a country are from the same ethno-linguistic group. It captures potential for within-country conflict that may slow democratization, as well as a force that may reduce the strength of a central government. Second, the length of ancient statehood (Putterman 2007; see also Bockstette et al. 2002, Alesina et al. 2011) is an index of the geographic and (discounted) temporal extent of domestic governance beyond the tribal level from 1 to 1950 C.E. This is intended to capture the potential for a stronger central government and development of solid institutions. We enter each of these measures in separate urbanization regressions, but report all four coefficients in column 1 of table 8. The polity measure is completely insignificant, both on its own and interacted with SSA (not shown). 34 ELF is stronger but also insignificant. Years of ancient statehood increases urbanization, and the effect is significant at almost the 5% level. A one standard deviation increase in years of ancient statehood (0.86) results in a 0.40 increase in the growth of the urban share (mean 4.74). The explanation is that, while informal institutions may suffice in certain rural settings, formal institutions in cities with their complexity and anonymity are needed for better functioning. Weaker institutions and capability inhibit functioning of urban land markets, of formal sector development, and urban and transport planning. Thus overall faster urbanization may occur in places with stronger institutions. None of the four measures show significant interactions with SSA, meaning slope coefficients for SSA are no different from those for the rest of the world. For example, for years of ancient statehood, in the usual specification where all variables are interacted with SSA, the base years of ancient statehood coefficient (s.e.) is 0.631 (0.460) and the SSA differential is 0.072 (0.583). There is also the notion that institutions may mediate the other effects we have found for agricultural price shocks, and, in particular, that they might explain the SSA dummy. We interacted all of our measures of institutions with agricultural price shocks. No interaction terms are ever significant at a reasonable level. In columns 2 and 3 of Table 8 we present a sample of these results, for years of 34 Examining just the “constraints on the executive� dimension of polity yields similar results. 25 ancient statehood and polity2 levels. Although the signs of the estimated coefficients are consistent with the hypothesis that better institutions act to make the effect of agricultural price shocks on urbanization positive, the effects are completely insignificant. 6. Conclusions Does urbanization in Africa differ from the rest of the world? When, instead of income, urbanization is matched to effective technology, as proxied by educational attainment, the African experience matches global patterns. At a sector level, while positive trade shocks to modern manufacturing promote urbanization in the rest of the developing world, this effect is absent in Africa. We suggest that this may be explained by differences in level of development, where the capacity at low levels of development to respond to modern manufacturing sector shocks may be limited. Our starkest results relate to agricultural price shocks. Thus, whilst recent literature has argued that Africa exhibits an unanticipated positive impact of agricultural shocks on urbanization, we find the opposite: urbanization in Africa responds negatively to trade price shocks, just as we would expect from our theoretical model. On the contrary, it is for the rest of the developing world where we find a perverse response of urbanization to agricultural trade shocks. We explore two possible explanations for the differential responses of Africa and the rest of the world to such shocks – the first based on monopsony purchasing in agriculture and the second, which we consider more plausible, on the different ownership structures of land. Hence, while urbanization does differ in Africa, it does not differ in the ways that have been previously thought. 26 Appendix 1: Data Definition of samples Our sample consists of all countries that satisfied two criteria in 1970: (i) a population in excess of 300,000; and (ii) a level of GDP per capita that was less than the world sample mean for all countries that satisfied (i). Only countries that have data on both urban share and GDP per capita in 1970 and 2009 or 2010 are included in the sample. Other variables have some missing values within this sample. As discussed in the main text, we also experimented with a narrower sample restricted to countries with a 1970 GDP per capita level that was less than 50 % of the world sample mean (again with the mean defined for all countries with a population in excess of 300,000). Table A1 provides a full list of all countries in our sample. Countries in bold are not included in the more restrictive sample. Definition of variables and data sources Population and urban share Estimates of population and urban share are taken from the CD-ROM edition of World Urbanization Prospects: the 2011 Revision (WUP-2011). Although, where possible, WUP-2011 adjusts data to ensure that urban areas within individual countries are consistently defined over time, definitions of urban vary across countries. Estimates are generally based on population and housing censuses, although, in some cases, other sources are incorporated. 35 GDP per capita and openness GDP per capita is measured in constant 2005 international dollars at purchasing power parity exchange rates (chain series), whilst openness is measured as the share of exports plus imports in GDP at constant 2005 prices. Data on both variables is from the Penn World Table v. 7.0 (http://pwt.econ.upenn.edu/php_site/pwt_index.php). Effective technology Average years of high school (secondary) and college-level (tertiary) education in the adult population aged 25 and over is used as a proxy for effective technology. Data are from the Barro-Lee Educational Attainment Dataset, 2011 version (http://www.barrolee.com/). We also experimented with an alternative measure of effective technology – the fraction of the population aged 25 and over with at least some high schooling, from the same source. The share of the female population aged 15 and over with at least completed primary school, used as an instrument in section 3.2 of the main text, is also from the same source. Democratization 35 For a more detailed description of the methodology employed in WUP-2011 to arrive at urban share and population estimates see: http://esa.un.org/unpd/wup/Documentation/faq.htm. 27 Democratization is measured using the revised combined polity score (POLITY2) taken from the Polity IV Dataset (Marshall, Jaggers and Gurr 2010; http://www.systemicpeace.org/polity/polity4.htm). POLITY2 is measured on a -10 to +10 scale where -10 represents a strongly autocratic political regime and +10 a strongly autocratic political regime. 36 Ethnolinguistic fractionalization index The ethnolinguistic fractionalization (ELF) index was taken from the replication data for Alesina et al (2011), who describe it as follows: “The literature of ethnolinguistic fractionalization has normally focused on one index of fractionalization, the Herfindhal index, which captures the probability that two randomly drawn individuals from the population of the country belong to different groups. The original index was based on a linguistic classification of groups from a Soviet source (the Atlas Narodov Mira, Bruk and Apenchenko 1964).� State Antiquity Index The State Antiquity Index (“Statehist�) Version 3.1 was taken from the replication data for Alesina et al (2011). It is described further below by Louis Putterman as follows (http://www.econ.brown.edu/fac/louis_putterman/State_Antiquity_Index_V3%201.doc ; accessed 12 October 2012) 37: "The index used by Bockstette et al. was constructed as follows. They began by dividing the period from 1 to 1950 C.E. into 39 half centuries. Years before 1 C.E. were ignored on grounds that the experience of more than 2000 years ago would be unlikely to have much effect today, and in order to avoid low-return research effort using low quality information. For each period of fifty years, they asked three questions (and allocated points) as follows: 1. Is there a government above the tribal level? (1 point if yes, 0 points if no); 2. Is this government foreign or locally based? (1 point if locally based, 0.5 points if foreign [i.e., the country is a colony], 0.75 if in between [a local government with substantial foreign oversight]; 3. How much of the territory of the modern country was ruled by this government? (1 point if over 50%, 0.75 points if between 25% and 50%, 0.5 points if between 10% and 25%, 0.3 points if less than 10%). Answers were extracted from the historical accounts on each of 119 countries in the Encyclopedia Britannica. The scores on the three questions were multiplied by one another and by 50, so that for a given fifty year period, what is today a country has a score of 50 if it was an autonomous nation, 0 if it had no government above the tribal level, 25 if the entire territory was ruled by another country, and so on. To combine the data of the 39 periods, Bockstette et al. tried alternative rates for discounting the influence of the past, ranging from 0 to a discount of 50% for each half century. At a 50% discount rate, for example, the contribution to our index of having had an autonomous state over the whole territory from 1850 to 1900 is 50x(1.5)-1 = 33.33. The bulk of the analysis in the paper used statehist05, which has 36 Full details of the methodology underlying the construction of the POLITY2 measure can be found in the Polity IV Users’ Manual (http://www.systemicpeace.org/inscr/p4manualv2010.pdf). 37 Note that while this paragraph refers to an earlier version of the dataset, the same document goes on characterize all subsequent edits as additions of new countries using the same methods. 28 a discount rate of 5% (i.e., 0.05). Finally in order to make the series easier to interpret, the sum of the discounted series was divided by the maximum possible value the series could take given the same discount rate. Thus the value that the index could take for any given country lay between zero and one." Rainfall The rainfall data are in millimeters per year and are aggregated from Matsuura and Wilmott’s (2012) 0.5 X 0.5 degree grid (approximately 3000 square km at the equator) to countries using the country grid of Mitchell et al (2002). Price and income shock data The price shock indices draw data from several sources. Exports by commodity for 1962-1969 and 1970- 1972 are from Feenstra et al (2005). Price data for the 37 commodities listed in table A2 are from World Bank (2012) and UNCTAD (2012). We define all products that are grown, except for logs and sawnwood, as agricultural, and the remainder as (non-agricultural) natural resources. These commodities are a subset of those with world prices listed in these two sources. In order to be included in the present study they had to 1) have uninterrupted price series from either source between 1960 and 2010, 2) correspond to an export category in the Feenstra database in a form embodying little or no processing. CPI is from the US Bureau of Labor Statistics. GDP in current dollars (the units of the exports data) at PPP for 1962-1972 is from Heston, Summers and Aten (2011), except for 6 countries 38 with missing 1960s data. For these, log GDP was predicted based on annual cross-sectional linear regressions of log GDP at PPP from Heston, Summers and Aten (2011) on log GDP in current dollars from the World Development Indicators, log of total population and the urban population share. The income shocks use the Feenstra et al. (2005), with product categories defined in the text. As represented in Feenstra (2005), the OECD countries are Australia, Austria, Belgium-Luxembourg, Canada, Denmark, Finland, France-Monaco, Germany, Greece, Iceland, Ireland, Italy, Japan, Netherlands, New Zealand, Norway, Portugal, Spain, Sweden, Switzerland-Liechtenstein, the United Kingdom, and the United States. Sector shares and land area Data on the shares of agriculture, manufacturing and services value added in GDP, as well as land area, are from the World Development Indicators 2011 (http://data.worldbank.org/data-catalog/world- development-indicators/wdi-2011). 38 Afghanistan, Cambodia, Liberia, Somalia, Sudan, and Swaziland. 29 Table A1.1: Full list of countries included in sample Sub-Saharan Africa (SSA) Non-SSA Developing (NSSA) WBID Country GDP pc Urban share WBID Country GDP pc Urban share 19701 1970 2010 1970 1970 2010 ZAF South Africa 0.99 47.81 61.55 PER Peru 0.97 57.41 76.91 NAM Namibia 0.84 22.29 37.82 URY Uruguay 0.91 82.37 92.45 AGO Angola 0.46 14.96 58.38 CHL Chile 0.86 75.23 88.94 ZMB Zambia 0.43 30.35 38.73 BRA Brazil 0.84 55.91 84.34 MUS Mauritius 0.37 42.03 41.78 TUR Turkey 0.83 38.23 70.49 LBR Liberia 0.35 26.03 47.80 MLT Malta 0.83 89.70 94.67 SDN Sudan 0.27 16.52 33.08 SLV El Salvador 0.78 39.40 64.28 CIV Cote dIvoire 0.27 28.16 50.56 GTM Guatemala 0.76 35.55 49.33 NGA Nigeria 0.26 22.71 49.00 DZA Algeria 0.76 39.50 72.02 CMR Cameroon 0.25 20.30 51.51 NIC Nicaragua 0.69 47.03 57.25 MRT Mauritania 0.24 14.56 41.23 PAN Panama 0.65 47.64 74.61 SEN Senegal 0.24 30.00 42.25 ROM Romania 0.60 40.32 52.79 TGO Togo 0.23 21.28 37.53 ECU Ecuador 0.59 39.28 66.86 COG Congo, Rep. 0.23 39.13 63.22 COL Colombia 0.58 54.82 75.02 SWZ Swaziland 0.23 9.71 21.32 KOR Korea, Rep. 0.57 40.70 82.93 BWA Botswana 0.21 7.83 60.98 DOM Dominican Rep 0.55 40.21 69.07 ZAR Congo Dem. Rep. 0.20 30.30 33.73 BOL Bolivia 0.52 39.78 66.40 CAF Central Afr. Rep. 0.19 27.33 38.85 JOR Jordan 0.52 55.97 82.47 KEN Kenya 0.19 10.30 23.57 BGR Bulgaria 0.51 52.30 72.52 SLE Sierra Leone 0.18 23.40 38.88 FJI Fiji 0.47 34.76 51.84 MDG Madagascar 0.18 14.10 31.93 ALB Albania 0.46 31.74 52.32 GHA Ghana 0.18 28.97 51.22 HND Honduras 0.44 28.90 51.58 SOM Somalia 0.17 22.68 37.29 MYS Malaysia 0.39 33.45 72.01 TCD Chad 0.17 11.57 21.74 PRY Paraguay 0.39 37.07 61.37 BEN Benin 0.17 16.69 44.26 TUN Tunisia 0.39 43.48 66.10 GMB Gambia, The 0.17 19.50 56.66 GUY Guyana 0.38 29.43 28.31 UGA Uganda 0.15 6.66 15.16 SYR Syrian Arab Rep 0.34 43.35 55.67 GIN Guinea 0.15 15.98 34.97 HTI Haiti 0.31 19.76 51.99 NER Niger 0.15 8.79 17.62 THA Thailand 0.30 20.89 33.73 RWA Rwanda 0.15 3.19 18.81 PHL Philippines 0.30 32.98 48.65 MWI Malawi 0.13 6.05 15.54 MAR Morocco 0.28 34.48 56.68 TZA Tanzania 0.12 7.85 26.28 MNG Mongolia 0.27 45.05 67.57 BFA Burkina Faso 0.11 5.75 25.67 PNG Pap New Guinea 0.25 9.80 12.43 LSO Lesotho 0.10 8.61 26.85 EGY Egypt, Arab Rep 0.25 42.21 43.38 MLI Mali 0.09 14.33 34.28 PAK Pakistan 0.22 24.82 35.88 ETH Ethiopia 0.09 8.59 16.76 LKA Sri Lanka 0.21 19.51 15.04 MOZ Mozambique 0.08 5.78 30.96 KHM Cambodia 0.21 15.97 19.81 BDI Burundi 0.07 2.38 10.64 IND India 0.17 19.76 30.93 ZWE Zimbabwe 0.06 17.36 38.13 IDN Indonesia 0.16 17.07 49.92 GNB Guinea-Bissau 0.06 15.13 43.22 AFG Afghanistan 0.16 11.03 23.24 BTN Bhutan 0.16 6.09 34.79 BGD Bangladesh 0.15 7.59 27.89 NPL Nepal 0.13 3.96 16.66 LAO Lao PDR 0.12 9.63 33.12 VNM Vietnam 0.11 18.30 30.39 CHN China 0.07 17.40 49.23 1 Expressed as a proportion of the world sample mean level of GDP per capita in 1970 for all countries with a 1970 population in excess of 300,000. 30 Table A1.2: Commodities used in price indices commodity code agricultural Source Aluminum high grade, London Metal Exchange, cash ALUM no UNCTAD Bananas, Central America and Ecuador, U.S. importer's price, FOB BANA yes UNCTAD U.S. ports (¢/lb.) Beef, Australia & New Zealand, frozen boneless, U.S. import price BEEF yes UNCTAD FOB port of entry (¢/lb.) Cattle hides, U.S. Chicago packer's heavy native steers, FOB CATH yes UNCTAD shipping point (¢/lb.) Coconut oil, in bulk, Philippines, CIF Rotterdam CCNO yes UNCTAD Cocoa beans, average daily prices New York/London (¢/lb.) COCO yes UNCTAD Coffee, average of 5 prices COFF yes UNCTAD Copper, wire bars, US producer, FOB refinery (¢/lb.) COPP no UNCTAD Copra, in bulk, Philippines/Indonesia, CIF N.W. European ports COPR yes UNCTAD Cotton, U.S. average of Memphis/Eastern and COTT yes UNCTAD Memphis/Orleans/Texas, Midd.1-3/32, CFR Far Eastern quotations (¢/lb.) Crude petroleum, average of UK Brent (light)/Dubai CRUD no UNCTAD (medium)/Texas (heavy) equally weighted ($/barrel) Cottonseed oil, in bulk, United States, PBSY, FOB Gulf CTSO yes UNCTAD Groundnut oil, in bulk, any origin, CIF Rotterdam GNDO yes UNCTAD Iron ore, Brazilian to Europe, Vale Itabira SSF, 64.5% Fe content, IRON no UNCTAD FOB (¢/dmt Fe unit) Jute, Bangladesh, BWD, FOB Mongla JUTE yes UNCTAD Lead, average of London Metal Exchange, cash settlement and LEAD no UNCTAD North American producer price Linseed oil, in bulk, any origin, ex-tank, Rotterdam LNSO yes UNCTAD Maize ($/mt) MAIZ yes World Bank Manganese ore, 48/50% Mn Max 0.1% P, FOB metallurgical MANG no UNCTAD Natural gas, average of Europe and US ($/mmbtu) NGAS no World Bank Pepper, white Muntok, faq spot PEPP yes UNCTAD average of Palm kernel oil, in bulk, Malaysia, CIF Rotterdam and PLMO yes UNCTAD Palm oil, in bulk, Malaysia/Indonesia, 5% ffa, CIF N.W. European ports Rice, Thailand, white milled, 5% broken, nominal price quotes, FOB RICE yes UNCTAD Bangkok Rubber, SGP/MYS (cents/kg) RUBB yes World Bank Silver (cents/toz) SILV no World Bank Sisal, Tanzania/Kenya, average of n° 2 & 3 long and n° 3 & UG, FOB SISA yes UNCTAD Sunflower oil, in bulk, European Union, FOB N.W. European ports SNFO yes UNCTAD Soybeans, in bulk, United States, n° 2 yellow, CIF Rotterdam SOYB yes UNCTAD Sugar, in bulk, average of I.S.A. daily prices, FOB Caribbean ports SUGA yes UNCTAD (¢/lb.) Soybean oil, in bulk, The Netherlands, FOB ex-mill SYBI yes UNCTAD Tea, average of Mombasa, Colombo and Kolkata auctions, TEA yes World Bank (cents/kg) Tin, Kuala Lumpur Tin Market, ex-smelter, and London Metal TIN no UNCTAD exchange, cash) 31 Tobacco, US import u.v. ($/mt) TOBA yes World Bank Logs, average of Cameroon and Malaysia TRPL no World Bank Sawnwood, Malaysian TRPS no World Bank Wheat, United States, n° 2 Hard Red Winter (ordinary), FOB Gulf WHEA yes UNCTAD Zinc, average of Prime Western, delivered, North America and ZINC no UNCTAD special high grade, London Metal Exchange, cash settlement Appendix 2: Model Growth properties of the model To see some properties of a growth process we consider two simple growth formulations with exogenous savings, avoiding the even more complicated endogenous growth details as developed in Henderson and Wang (2005). We use the specific functional forms footnoted in the text. We assume all goods are traded internationally at fixed prices, that human capital is converted at the numeraire, and  /N =g. that population grows at an exogenous rate N For the simplest case, suppose s fraction of income is saved by all people and that h= a h= u h, θ so total savings are pa ha  , the total change in human capital. K ψ ( N − N ) −ψ N , which then equals H a Given equalized urban and rural incomes it is easy to show that the rate of change in per person human /h H where= h  / H − g is /h= h spa ( N − N ) −ψ K ψ hθa −1 − g . (a) While it might look like there can be steady state growth if θ a = 1 and convergence to a steady state level if θ a < 1 , it isn’t that simple. As h rises for the same population, we know dN / dha > 0 as long as θu > θ a , Second, as N rises, dN / dN > 0 . Thus, N or h rising both cause N to increase. The term ( N − N ) −ψ in (a) is changing. Totally differentiating the expression and using dN / N = g and dh / h as defined in (a) we couldn’t definitively sign the change in ( N − N ) −ψ . Moreover since population on its own leads ( N − N ) −ψ to decline we can’t define a steady state. What we would have to assume for this simple  / h > 0 and θ is sufficiently below 1 that we do not have context is that we are at point in time where h a explosive growth. Once we allow human capital to differ across sectors these issues do not go away. However it is helpful to examine the case briefly to establish relative urban-rural differences in human capital allocations. Families each have the same human capital h which they rent in capital markets. The return to capital r must be equalized in uses so that 32 θu −1 δ θ a −1 ψ =r θu p y hu N (1 − τ N = )1+δ θ a pa ha K ( N − N ) −ψ . Incomes net of capital rental costs must be equalized across sectors so that θu θa ψ p y hu δ N δ (1 − τ N )1+= − rhu pa ha K ( N − N ) −ψ − rha . Using this expression we can then show that hu / ha > 1 iff θu > θ a . 33 References Acemoglu, Daron, and James A. Robinson. 2010. "Why is Africa Poor?" Economic History of Developing Regions 25(1): 21–50. Ades, Alberto, and Edward Glaeser. 1995. "Trade and Circuses: Explaining Urban Giants." The Quarterly Journal of Economics 110(1): 195–227. Alesina, Alberto, William Easterly and Janina Matuszeski. 2011. “Artificial States.� Journal of the European Economic Association 9(2):246–277. Baer, Werner. 1979. The Brazilian Economy: Growth and Development. Boulder, CO: Lynne Rienner Publishers. Barrios, Salvador, Luisito Bertinelli, and Eric Strobl.2006. "Climatic change and rural-urban migration: The case of sub-Saharan Africa." Journal of Urban Economics 60(3): 357–371. Barrios, Salvador, Luisito Bertinelli, and Eric Strobl. 2010. "Trends in Rainfall and Economic Growth in Africa: A Neglected Cause of the African Growth Tragedy." Review of Economics and Statistics 92(2):350–366. Besley, Timothy, and Robin Burgess. 2000. “Land Reform, Poverty Reduction, and Growth: Evidence from India.� Quarterly Journal of Economics 115(2): 389–430. Binswanger, Hans, and Robert Townsend. 2000. "The Growth Performance of Agriculture in Subsaharan Africa." American Journal of Agricultural Economics 82(5): 1075–1086. Black, Duncan, and Vernon Henderson. 1999. "A Theory of Urban Growth." Journal of Political Economy 107(2): 252–284. Block, Steven A., and Robert H. Bates. 2011. "Revisiting African Agriculture: Institutional Change and Productivity Growth." Unpublished. Bockstette, Valerie, Areendam Chanda, and Louis Putterman . 2002. “States and Markets: The Advantage of an Early Start.� Journal of Economic Growth 7: 347–369. Bruce, John W. 1998. “Country Profiles of Land Tenure: Africa, 1996.� University of Wisconsin Land Tenure Center Research Paper 130. Brückner, Markus, and Antonio Ciccone. 2010. "International Commodity Prices, Growth and the Outbreak of Civil War in Sub-Saharan Africa." Economic Journal 120: 519–534. Bruk, Solomon, and V. S. Apenchenko, eds. 1964. Atlas Narodov Mira. Moscow: Glavnoe Upravlenie Geodezii i Kartografii. Card, David. 2001. "Immigrant Inflows, Native Outflows and the Local Labor Market Impacts of Higher Immigration." Journal of Labor Economics 19(1): 22–64. Caselli, Francesco, and Wilbur John Coleman. 2001. “The US Structural Transformation and Regional Convergence: A Reinterpretation.� Journal of Political Economy 109(3): 584–616. 34 Collier, Paul, and Benedikt Goderis. 2009. "Structural Policies for Shock-prone Developing Countries." Oxford Economic Papers 61(4): 703–726. Collier, Paul, and Anthony Venables. 2007. "Rethinking Trade Preferences: How Africa Can Diversify Its Exports." World Economy 30(8): 1326–1345. Davis, James, and J. Vernon Henderson. 2003. "Evidence on the political economy of the urbanization process." Journal of Urban Economics 53(1): 98–125. Duranton, Gilles, and Diego Puga. 2004. "Micro-foundations of Urban Agglomeration Economies." In Handbook of Regional and Urban Economics, Volume 4, edited by J-F Thisse and J.V. Henderson, 2063– 2117. Easterly, William, and Ross Levine. 1997. "Africa's Growth Tragedy: Policies and Ethnic Divisions." Quarterly Journal of Economics 112(4): 1203–1250. Fay, Marianne, and Charlotte Opal. 2000. "Urbanization without growth : a not-so-uncommon phenomenon." World Bank Policy Research Working Paper 2412. Feenstra, Robert, Robert Lipsey, Haiyan Deng, Alyson Ma, and Hengyong Mo. 2005. “World Trade Flows: 1962-2000.� National Bureau of Economic Research Working Paper 11040. Galor, Oded, Omer Moav, and Dietrich Vollrath. 2009. “Inequality in Land Ownership, the Emergence of Human Capital Promoting Institutions, and the Great Divergence.� Review of Economic Studies 76(1): 143-179. Galor, Oded, and Joseph Zeira. 1993. “Income Distribution and Macroeconomics.� Review of Economic Studies 60: 35–52. Gollin, Douglas, Remi Jedwab, and Dietrich Vollrath. 2012. “Urbanization with Structural Transformation: Evidence from Sub-Saharan Africa.� Unpublished. Harris, John and Michael P. Todaro. 1970. "Migration, Unemployment and Development: A Two-Sector Analysis." American Economic Review 60(1): 126–142. Henderson, J. Vernon. 1988. Urban Development: Theory, Fact, and Illusion. New York: Oxford University Press. Henderson, J. Vernon. 2003. "The Urbanization Process and Economic Growth: The So-What Question." Journal of Economic Growth 8(1): 47–71. Henderson, J. Vernon, Todd Lee and Y-J Lee. 2001. "Scale Externalities in a Developing Country.� Journal of Urban Economics 49: 479–504. Henderson, J. Vernon, Adam Storeygard, and David N. Weil. 2012. "Measuring Economic Growth from Outer Space." American Economic Review 102(2): 994–1028. Henderson, J. Vernon, and Hyoung G. Wang. 2005. "Aspects of the Rural-urban Transformation of Countries." Journal of Economic Geography 5(1): 23–42. Heston, Alan, Robert Summers and Bettina Aten, Penn World Table Version 7.0, Center for International Comparisons of Production, Income and Prices at the University of Pennsylvania, (June 2011). 35 Jayne, T.S. and Stephen Jones. 1997. “Food Marketing and Pricing Policy in Eastern and Southern Africa: A Survey.� World Development, 25: 1505–1527. Jedwab, Rémi. 2012. "Why Is African Urbanization Different? Evidence from Resource Exports in Ghana and Ivory Coast." Unpublished. Johnson, Simon, Jonathan Ostry, and Arvind Subramanian. 2007. "Prospects for Sustained Growth in Africa: Benchmarking the Constraints." National Bureau of Economic Research Working Paper 13120. Kalemli-Ozcan, Sebnem, and Bent E. Sorensen. 2012. "Misallocation, Property Rights, and Access to Finance: Evidence from Within and Across Africa," National Bureau of Economic Research Working Paper 18030. Lucas, Robert E. 2004. "Life Earnings and Rural�Urban Migration." Journal of Political Economy 112(1): 29–59. Marshall, Monty G., Keith Jaggers, and Ted Robert Gurr. 2010. Polity IV Project: Political Regime Characteristics and Transitions, 1800-2010. Vienna, VA: Center for Systemic Peace. http://www.systemicpeace.org/polity/polity4.htm Matsuura, Kenji, and Cort J. Willmott. 2012. “Terrestrial Precipitation: 1900-2010 Gridded Monthly Time Series, Version 3.01.� http://climate.geog.udel.edu/~climate/html_pages/download.html#T2011 Accessed 4 September 2012. McGranahan, Gordon, Diana Mitlin, David Satterthwaite, Cecilia Tacoli, and Ivan Turok. 2009. "Africa’s Urban Transition and the Role of Regional Collaboration." Unpublished. Mills, Edwin S., and Charles M. Becker. 1986. Studies in Indian Urban Development. New York: Oxford University Press. Mitchell, T.D., M. Hulme, and M. New. 2002. “Climate data for political areas.� Area 34: 109–112. Potts, Deborah. 2012. “What Do We Know About Urbanization in Sub-Saharan Africa and Does It Matter?� International Development Planning Review 34(1): v–xxii. Ramankutty, Navin, Jonathan A. Foley, John Norman and Kevin McSweeney. 2002. “The global distribution of cultivable lands: current patterns and sensitivity to possible climate change.� Global Ecology and Biogeography 11(5): 377–392. Renaud, Bertrand. 1981. National Urbanization Policy in Developing Countries. New York: Oxford University Press. Rossi-Hansberg, Esteban and Mark Wright. 2007. "Urban Structure and Growth.� Review of Economic Studies, 74(2): 597–624. Satterthwaite, David. 2010. “Urban myths and the mis-use of data that underpin them.� World Institute for Development Economics Research Working paper 2010/28. Stock, James H. and Mark W. Watson. 2008. “Heteroskedasticity-Robust Standard Errors for Fixed Effect Panel Data Regression.� Econometrica 76(1): 155–174. 36 United Nations Conference on Trade and Development (UNCTAD). 2012. UNCTADStat, http://unctadstat.unctad.org/ReportFolders/reportFolders.aspx Accessed 12 June 2012. Wahba, Jackline and Barry McCormick. 2003. "Did Public Wage Premiums Fuel Agglomeration in LDCs?" Journal of Development Economics. 70(2): 349–379. Williamson, Jeffrey G. 1986. Migration and Urbanization in the Third World. Cambridge, MA: Harvard Institute for Economic Research. World Bank. 1999. World Development Report 1999/2000: Entering the 21st Century. Washington, DC: World Bank. World Bank. 2009. World Development Report 2009: Reshaping Economic Geography. Washington, DC: World Bank. World Bank. 2012 World Bank Commodity Price Data (Pink Sheet) Accessed 23 August 2012 http://econ.worldbank.org/WBSITE/EXTERNAL/EXTDEC/EXTDECPROSPECTS/0,,contentMDK:21574907~ menuPK:7859231~pagePK:64165401~piPK:64165026~theSitePK:476883,00.html (update, 5 July 2012) Young, Alwyn. 2012. “The African Growth Miracle.� Journal of Political Economy 120(4): 696–739. 37 Table 1. Summary statistics for key variables Sub-Saharan Africa Non-SSA developing Std. Std. Variable n Mean Min. Max. n Mean Min. Max. dev dev Urban share (%) 1970 40 18.13 10.8 2.4 47.8 46 35.43 19.13 4 89.7 2010 40 36.74 14 10.6 63.2 46 54.65 21.9 12.4 94.7 GDP p.c. (2005 internat. $) 1970 40 1221 986 300 5281 46 2359 1346 390 5139 2009 40 1877 2208 143 9484 46 6235 4800 1170 25029 Aver. yrs schooling 1970 31 0.241 0.29 0.03 1.12 45 0.636 0.41 0.08 1.569 2010 31 1.319 0.81 0.168 3.11 45 2.66 1.2 0.593 5.9 % GDP agriculture 1970 30 37.69 16.7 7.16 70.6 28 27.5 13.04 6.91 67.29 2010 30 23.17 16.1 2.86 61.3 45 14.14 8.6 1.83 35.79 % GDP manufacturing 1970 24 10.18 5.3 3.32 22.8 27 16.52 6.62 3.71 33.75 2010 28 10.6 8.22 3.07 45.2 45 16.2 6.6 5.9 34.15 Unweighted growth rates, 1970-2009/10 (% pa): Urban pop. share 40 2.09 1.15 -0.01 5.27 46 1.29 0.988 -0.65 4.45 Urban pop. 40 4.78 1.31 1.12 8.1 46 3.15 1.4 0.018 6.77 GDP p.c. 40 0.53 1.77 -3.91 5.44 46 2.4 1.57 -1.31 7.69 Avg. yrs. schooling 31 5.23 1.7 1.61 9.12 45 4 1.46 1.39 8.37 Notes: The means are the unweighted sample means across countries. Growth rates are calculated for the period 1970-2010 with the exception of GDP per capita, for which the period is 1970-2009. Samples consist of all countries with non-missing data for which 1970 population > 300,000 and GDP per capita < global sample mean. GDP per capita is measured in PPP at constant 2005 international prices (chain series). Aver. yrs of schooling is the mean number of years of schooling (high school + college) of the adult (aged 25+) population. Urban share/population data is from World Urbanization Prospects 2011; GDP p.c. data from the Penn World Tables v 7.0; aver. yrs of schooling is from the 2011 version of the Barro-Lee Educational Attainment Dataset. All other data is derived from the 2011 World Development Indicators. 38 Table 2. Urbanization level vs. income per capita and effective technology, 2010 (1) (2) (3) (4) GDP Schooling Schooling GDP levels levels levels levels - 0.2324* Constant -1.3753*** 0.2689*** 0.5720*** ** (0.2254) (0.1342) (0.0643) (0.0324) SSA 1.2498*** -0.0533 (0.2805) (0.0724) ln(GDP per capita) 0.2262*** 0.1319*** (0.0259) (0.0170) SSA*ln(GDP per capita) -0.1568*** (0.0349) 0.1161* ln(years of schooling) 0.1061*** ** (0.0247) (0.0168) SSA*ln(years of postprimary schooling) 0.0119 (0.0330) F(Joint significance of SSA) 10.11*** 0.41 (p = 0.0001) (p = 0.6655) df=(2,82) df=(2,72) 2 Adj R 0.539 0.458 0.463 0.463 N 86 86 76 76 Notes: Dependent variable is urbanization level. *** p<0.01, ** p<0.05, * p<0.1; robust standard errors. 39 Table 3. Urbanization rates and growth in income and effective technology: 1970-2010 long difference (1) (2) (3) (4) GDP GDP Schooling Schooling difference difference difference difference constant 0.0099*** 0.0158*** 0.0003 0.0028 (0.0023) (0.0013) (0.0035) (0.0033) SSA 0.0095*** 0.0126 (0.0027) (0.0081) Growth in GDP per capita 0.1274 0.0545 (0.0936) (0.0688) SSA*Growth in GDP per capita 0.1536 (0.1590) Growth in effective technology 0.2983*** 0.2817*** (0.0989) (0.0841) SSA*Growth in effective technology -0.1563 (0.1791) F(Joint significance of SSA) 8.71*** 3.26** (p = 0.0004) (p = 0.0440) df=(2,82) df=(2,72) Adj R2 0.202 -0.003 0..213 0.171 N 86 86 76 76 Notes: Dependent variable is the urbanization rate, or change in fraction urban. *** p<0.01, ** p<0.05, * p<0.1; robust standard errors. 40 Table 4. Urbanization and economic structure: 2010 levels and 1970-2010 long difference (1) (2) (3) (4) Levels Long difference constant 0.6384*** 0.8324*** 0.4204*** 0.3387*** (0.0414) (0.0314) (0.0431) (0.1232) SSA -0.4067*** 0.0734 (0.0582) (0.1319) agriculture share -0.9070*** -2.0182*** -0.1114 -0.375 (0.2088) (0.1565) (0.0871) (0.2798) SSA*agriculture share 1.7906*** 0.3603 (0.2326) (0.2896) F(Joint significance of SSA) 30.5 1.2 (p < 0.00005) (p = 0.3104) Df=(2,71) Df=(2,45) Adj R2 0.296 0.561 0.007 0.021 N 75 75 49 49 time period 2010 2010 1970-2010 1970-2010 Notes: *** p<0.01, ** p<0.05, * p<0.1; robust standard errors. Agriculture share denotes agricultural value added expressed as a share of GDP. 41 Table 5. Base results on urbanization rate (1) (2) (3) (4) (5) Must have Simple recent SSA Base SSA formulation census differential case differential Δln(national population) 7.166** 8.356** 8.457** 42.86** 36.84 (3.105) (3.585) (4.033) (19.65) (30.26) Δln(postprimary education) 3.491*** 3.746*** 2.892** 3.515*** 2.978** (1.268) (1.375) (1.356) (1.264) (1.368) ln(land area) 0.739** 0.565 (0.294) (0.384) ln(land area)*Δln(national population) -3.006* -2.383 (1.587) (2.438) SSA*Δln(national population) -3.365 33.43 (6.435) (41.01) SSA*Δln(postprimary education) 2.29 2.615 (3.816) (3.685) SSA*ln(land area) 0.981 (0.617) SSA*ln(land area)*Δln(national population) -3.321 (3.303) Observations 304 255 304 304 304 R-squared 0.1 0.116 0.118 0.126 0.153 countries 76 76 76 76 76 sample full recent full full full census Notes: Dependent variable is Δln( urban population share). Robust standard errors, clustered by country, are in parentheses. *** p<0.01, ** p<0.05, * p<0.1. 42 Table 6. Shocks to Agriculture (1) (2) (3) (4) 2 year SSA lag Non- Base interactions structure food Δln(national population) 59.11** 67.82 73.42 65.98 (25.55) (48.14) (49.63) (47.32) ln(land area) 0.978** 0.969 1.049 0.911 (0.453) (0.800) (0.821) (0.774) ln(land area)*Δln(national population) -4.286** -4.645 -5.067 -4.439 (2.021) (3.790) (3.900) (3.713) Δln(postsecondary education) 3.354*** 4.151** 4.195** 4.205*** (1.229) (1.620) (1.602) (1.565) Δln(rainfall) -3.219* -2.448 -0.519 -2.503 (1.703) (1.999) (2.070) (1.914) Agricultural price shock 3.28 9.022** 13.40* 48.97 (3.12) (4.12) (7.23) (32.44) SSA*Δln(rainfall) -2.17 -1.734 -2.386 (3.69) (2.61) (3.581) SSA*Agricultural price shock -16.61*** -12.12 -59.45* (5.11) (10.18) (34.42) Observations 260 260 260 260 R-squared 0.156 0.202 0.178 0.209 countries 65 65 65 65 rain and price lag (years) 5 5 2 Agriculture class all all all non-food Notes: dependent variable is Δln( urban population share). Robust standard errors, clustered by country, are in parentheses. *** p<0.01, ** p<0.05, * p<0.1. Rainfall is a 3-year smoothed average. 43 Table 7. Trade Shocks to Industry (1) (2) (3) (4) (5) SSA Other Stage of 2 year Base interactions sectors development lag Δln(national population) 63.88** 76.58 76.65 63.71** 63.56** (25.05) (47.69) (47.69) (24.91) (25.01) ln(land area) 1.071** 1.187 1.191 1.097** 1.091** (0.446) (0.812) (0.810) (0.431) (0.432) ln(land area)*Δln(national population) -4.654** -5.331 -5.336 -4.702** -4.692** (1.980) (3.758) (3.757) (1.960) (1.967) Δln(postprimary education) 3.337*** 3.722** 3.785** 3.504*** 3.515*** (1.194) (1.598) (1.727) (1.210) (1.209) modern manu income shock 6.423 25.38** 25.39* -5.134 -4.889 (4.484) (12.65) (12.73) (3.987) (3.969) other industry income shock -0.214 (1.896) SSA*modern manu income shock -23.12* -22.92* (12.74) (12.82) SSA*other industry income shock -1.820 (1.966) 1965 education*modern manu shock 32.70*** 27.89** (10.68) (11.03) 1965 education -0.950 -0.926 (0.844) (0.840) Observations 260 260 260 260 260 R-squared 0.140 0.184 0.198 0.150 0.147 countries 65 65 65 65 65 shock lag (years) 5 5 5 5 2 Notes: dependent variable is Δln( urban population share). Robust standard errors, clustered by country, are in parentheses. *** p<0.01, ** p<0.05, * p<0.1. Shocks are measured in hundred billions of dollars. 44 Table 8. Institutions and mediating effects (1) (2) (3) Political and Rents and political historical interactions measures 0.0398 0.0166 polity 2, lagged 15 years (0.0503) (0.0467) -0.0557 Δpolity 2, lagged 10 years (0.0385) Ethnolinguistic -0.0125 fractionalization [ELF60] (0.0124) ln(years of ancient 0.455* 0.408 statehood (Putterman) (0.252) (0.255) -10.02 4.167 Agricultural price shock (21.01) (5.987) ln(years of ancient 2.614 statehood)* Agric. price (4.826) shock 0.405 polity 2* Agric. price shock (0.790) Observations 256 249 R-squared 0.145 0.140 countries 64 64 Notes: dependent variable is Δln( urban population share). Column 1 shows coefficients from four separate regressions. Robust standard errors, clustered by country, are in parentheses. *** p<0.01, ** p<0.05, * p<0.1. Shocks are measured in hundred billions of dollars. See text for further information 45 Figure 1: Urbanization level vs. income per capita and effective technology, 2010: (a) natural logarithm of income per capita; (b) effective technology < Mean(Global GDP pc) - GDP pc = PWT7.0 < Mean(Global GDP pc) - GDP pc 1 1 'MLT' 'MLT' 'URY' 'URY' 'CHL' 'CHL' 'JOR' 'BRA' 'KOR' 'BRA' 'JOR' 'KOR' .8 .8 'PER' 'COL''PAN' 'COL' 'PER' 'BGR' 'BGR' 'DZA' 'PAN' 'DZA' 'TUR' 'MYS' 'TUR' 'DOM' 'MYS' 'MNG' 'DOM' 'MNG' 'BOL' 'ECU''TUN' 'SLV' 'TUN' 'SLV' 'COG' 'ECU' 'BOL' Urban share Urban share 'COG' 'PRY' 'ZAF' 'BWA' 'ZAF' 'PRY' 'BWA' .6 .6 'GMB' 'NIC' 'MAR' 'AGO' 'GMB' 'MAR''NIC' 'SYR' 'SYR' 'GHA''HTI''CMR' 'FJI' 'HND' 'ALB' 'ROM' 'HTI' 'HND' 'CMR' 'ALB' 'ROM' 'FJI' 'LBR' 'CIV' 'NGA' 'PHL' 'IDN' 'GTM' 'CHN' 'GTM' 'CIV' 'IDN' 'LBR' 'CHN' 'GHA' 'PHL' 'GNB''BEN''SEN' 'MRT' 'EGY' 'MUS' 'SEN' 'MRT' 'BEN' 'EGY' 'MUS' .4 .4 'ZWE' 'SOM' 'CAF' 'SLE' 'TGO' 'ZMB' 'NAM' 'SLE' 'CAF' 'ZMB' 'NAM' 'TGO' 'ZWE' 'GIN' 'MLI' 'PAK' 'BTN' 'MLI' 'SDN' 'PAK' 'ZAR' 'MDG' 'LAO' 'SDN' 'THA' 'ZAR' 'LAO''THA' 'MOZ' 'IND' 'VNM' 'MOZ' 'IND' 'VNM' 'BGD' 'LSO' 'GUY' 'LSO' 'BGD' 'GUY' 'BFA''TZA' 'KEN' 'TZA' 'KEN' 'AFG' 'TCD' 'KHM' 'SWZ' 'AFG' 'SWZ' .2 .2 'ETH' 'RWA' 'NER' 'NPL' 'RWA' 'KHM' 'NER' 'NPL' 'MWI' 'UGA' 'LKA' 'MWI' 'UGA' 'LKA' 'BDI' 'PNG' 'PNG' 'BDI' 0 0 5 6 7 8 9 10 0 2 4 6 ln(GDP per capita) Aver yr school (high + college) SSA NSSA SSA NSSA Figure 2: Urbanization rates and growth in income and effective technology: (a) growth in income per capita; (b) growth of effective technology < global mean(GDP pc) - GDP pc = (PPP, PWT7.0) < global mean(GDP pc) - GDP pc = (PPP, PWT7.0) .06 .06 'BWA' 'BWA' Growth rate of urban share 'RWA' Growth rate of urban share 'RWA' 'BTN' 'MOZ' 'MOZ' .04 .04 'BDI' 'BFA' 'NPL' 'BDI' 'NPL' 'AGO' 'BGD' 'BGD' 'TZA' 'LAO' 'TZA' 'LAO' 'LSO' 'LSO' 'IDN' 'IDN' 'GMB' 'MRT' 'GMB' 'GNB' 'CHN' 'CHN' 'MRT' 'MWI''BEN' 'HTI' 'CMR' 'MLI' 'HTI' 'MWI''BEN' 'CMR' 'MLI' .02 .02 'ZWE' 'MDG' 'KEN' 'UGA' 'GIN''AFG' 'SWZ' 'ZWE''SWZ' 'KEN' 'UGA' 'NER' 'NGA' 'SDN' 'ETH' 'MYS' 'KOR' 'KOR' 'SDN' 'MYS' 'AFG' 'NER' 'LBR' 'TGO' 'CIV' 'TCD' 'GHA''DZA' 'HND' 'TUR' 'GHA' 'LBR' 'TUR' 'CIV' 'HND' 'DZA' 'TGO' 'SOM' 'SLE' 'NAM''BOL' 'ECU' 'DOM' 'THA' 'NAM' 'ALB' 'BOL' 'VNM' 'DOM''ECU''SLE' 'PRY' 'SLV' 'SLV' 'PRY' 'COG' 'FJI' 'JOR' 'PHL' 'MAR' 'BRA' 'MNG' 'ALB' 'PAN''IND' 'VNM' 'TUN' 'PHL' 'PAN' 'BRA' 'MNG' 'PAK' 'IND' 'TUN' 'FJI''SEN' 'JOR' 'THA' 'COG' 'MAR' 'CAF' 'SEN' 'GTM' 'PAK' 'COL' 'BGR' 'BGR' 'GTM''COL' 'CAF' 'PER' 'ROM' 'PNG''ZAF' 'CHL' 'ROM' 'PER' 'NIC''ZMB' 'ZAF' 'KHM' 'SYR' 'PNG' 'CHL' 'KHM' 'NIC''SYR' 'ZMB' 'ZAR' 'URY' 'URY' 'MLT' 'ZAR' 'MLT' 'EGY' 'GUY' 'EGY' 0 0 'GUY' 'MUS' 'MUS' 'LKA' 'LKA' -.02 -.02 -.05 0 .05 .1 .02 .04 .06 .08 .1 Growth rate of GDP pc Growth rate of aver yrs school (high + college) SSA NSSA SSA NSSA Figure 3: Urbanization and economic structure: (a) urbanization vs. value added in agriculture (% GDP); (b) urbanization vs. value added in manufacturing (% GDP) 2010 cross-section (< Mean(Global GDP pc)) 1 'MLT' 'URY' 'CHL' 'KOR''BRA' 'JOR' .8 'PER' 'COL' 'PAN' 'BGR' 'MYS' 'DZA' 'TUR' 'DOM' 'ECU' 'TUN' 'BOL' 'MNG' 'SLV' Urban share 'COG' 'ZAF' 'BWA' 'PRY' .6 'AGO' 'MAR' 'NIC''SYR' 'GMB' 'ROM' 'HND' 'FJI' 'ALB' 'GHA' 'CHN' 'PHL' 'IDN' 'GTM' 'CIV' 'LBR' 'MUS' 'EGY' 'SEN''MRT' .4 'ZMB' 'NAM' 'ZWE' 'SLE' 'CAF' 'GIN' 'BTN' 'PAK' 'THA' 'SDN' 'MDG' 'LAO' 'ZAR' 'IND' 'VNM' 'MOZ' 'LSO' 'GUY' 'BGD' 'TZA' 'TCD' 'KEN' 'AFG' 'SWZ' .2 'KHM' 'RWA' 'LKA' 'UGA' 'NPL' 'MWI' 'ETH' 'PNG' 0 0 .2 .4 .6 Value added in agriculture (% GDP) SSA NSSA 46