1 Introduction Developing countries have urbanized dramatically over the last 50 years. This episode of massive urbanization shares many similarities with the historical urbanization process of today's developed countries. As in the historical experience, urbanization in today's developing countries is closely linked to increasing income per capita. Indeed, the correlation is so strong that urbanization is often used as a proxy for income per capita in comparisons across time and space (De Long and Shleifer 1993; Acemoglu et al. 2002). There is also a virtuous circle between economic development and urbanization (Henderson 2010; Duranton 2013; Glaeser 2013; Jedwab and Vollrath 2015b). As countries develop, people move out of rural areas and agricultural activities into urban centers, where they engage in manufacturing and service activities (Gollin et al. 2002; Michaels et al. 2012). These non-agricultural sectors are generally thought to have high rates of productivity growth, and agglomeration effects in urban areas are also believed to promote further economic growth (Glaeser et al. 1992; Duranton 2008; Glaeser and Gottlieb 2009). The historical pattern in the European and Neo-European countries that have undergone significant urbanization (beyond 50 %) has been that urbanization has accompanied industrialization; with the factories came the cities. In many developing countries, however, urbanization has deviated from this pattern. Many of today's developing countries have high rates of urbanization with little significant industry. Kuwait, Gabon, Saudi Arabia, Libya, Algeria, Angola and Nigeria are as urbanized as Uruguay, Taiwan, South Korea, Mexico, Malaysia, South Africa and China respectively, and yet the former countries have not industrialized to the same extent as the latter. This raises several questions. What has driven the urbanization process in Angola, Nigeria, and the others, in the absence of industry? Why have so many cities in today's developing world never been factory cities, in stark contrast to the historical experience of today's developed countries? If these cities have a different origin, does it matter for economic development? In this paper, we document that the expected relationship between urbanization and industrialization is absent in large parts of the developing world. The breakdown in this relationship occurs because of a large number of natural resource exporting countries that urbanized without significantly increasing the share of output from manufacturing and/or tradable services in GDP. Using a sample of 116 developing countries observed each decade from 1960-2010, we show that under a variety of specifications there is a statistically significant and economically meaningful association of resource exports with the urbanization rate. This is estimated holding constant the share of manufacturing and services in GDP. In our preferred specification, a one standard deviation increase in resource exports is associated with a 0.51 standard deviation increase in the urbanization rate, which equates to a roughly 13 % point increase. The effect of resource exports on urbanization holds not only in the cross-section but also in a panel that uses within-country variation for identification. These specifications are robust to adding region-decade fixed effects (Western Africa in the 1960s, etc.). Our results hold when incorporating a variety of controls for confounding effects. Different means of measuring the role of natural resource exportsas a percent of GDP, value per capita, or including only fuel and mining exportsdo not qualitatively alter the results. Following the existing literature (Sachs and Warner 2001; Brckner 2012; Henderson et al. 2013b), we also use resource discoveries and international commodity price shocks as instruments for resource exports to provide further evidence of causality. One can think of our approach as being similar to a "difference-in-difference" estimator, where countries that find a new resource (e.g. Botswana discovering diamonds in 1968) are the treatment 123 J Econ Growth (2016) 21:35-70 37 group. The control groups are countries that either never discovered a resource in this period or were producing a natural resource the entire time. The IV results are consistent with our cross-section and panel findings. Having established this relationship between natural resource dependence and urbanization, we then consider whether urbanization in resource-exporting countries differs substantially from urbanization in industrializing countries. In the cross-section, the strong positive relationship between income and urbanization is similar regardless of the source of urbanization. In fact, conditional on income per capita, urbanization rates are unrelated to the share of resource exports in GDP or the share of manufacturing and services in GDP. Urbanization is a function of income per capita across all countries. However, the composition of urban employment differs starkly between resourceexporters and non-exporters, holding constant income levels and urbanization rates. We use IPUMS census micro-data, labor force surveys, and household survey data to recreate the sectoral composition of urban areas for a sub-sample of 88 countries. Using this novel data set, we find that cities in resource-exporting countries are what we term "consumption cities", with a larger fraction of workers in non-tradable services such as commerce and transportation or personal and government services.1 Cities in countries that do not export significant resources, on the other hand, appear to be "production cities", with more workers in industrial sectors such as manufacturing or in tradable services like finance. The differences between consumption cities and production cities are not limited to employment patterns. We show that, in addition, cities in resource-exporters tend to have higher poverty rates and shares of population in slums than non-exporters, holding constant income levels and urbanization rates. Unconditionally, however, resource exporters are better off than non-exporters on these various measures of urban welfare. Our interpretation is that the income boost from resource exports makes cities richer, but it does not appear to translate into improved quality of life to the same degree that an income boost through industrialization would provide. Our findings could be seen as an example of "premature urbanization" (rather than "premature deindustrialization" as in Rodrik 2013a). Whether this constitutes a resource curse is not clear. As we show, the high urbanization rates in resource exporters are driven by high incomes; so in that sense, there is no evidence for a resource curse. Yet the "consumption cities" of resource exporters may not offer all the welfare advantages of "production cities".2 To understand how resources are related to urbanization, we develop a model of labor allocations between rural and urban areas that allows for two types of urban employment: tradables and non-tradables.3 An exogenous increase in resource export earnings raises incomes, which in turn increases demand for all types of output. The export earnings from resources are used to purchase additional food and other tradable goods on the world market; meanwhile, the increased demand for urban non-tradables is met by an increase in labor in that sector. This result echoes the standard Balassa-Samuelson framework and also corresponds in some respects to a standard model of "Dutch Disease" (Corden and Neary 1982; Matsuyama 1992; Harding and Venables 2013), where resource exports lead to a contraction of employment in other tradable sectors but expand non-tradable employment. In our model, resource rents cause an increase in the urbanization rate as the rural sector contracts, and there is a shift in the composition of urban employment towards non-tradable workers. Urbanization generated by resource rents differs from urbanization generated by industrialization, which occurs when industrial productivity increases. In the latter, substitution effects pull labor into the urban industrial sector, in addition to the income effects which draw labor into non-tradable production. Our results illustrate that urbanization and industrialization are not synonymous and that significant amounts of urbanization can and will take place in response to income shocks from resource exports. But resource-exporting countries may urbanize without acquiring the industrial sectors that we typically associate with development. The "consumption cities" of resource-exporters may nevertheless exert an influence on growth.4 These cities have higher shares of non-tradable service workers and lower shares of industrial sector workers. Much research in the growth literature (Lucas 2009; Duarte and Restuccia 2010; Rodrik 2013b) suggests that convergence is faster in industry than in services. This suggests that the source of urbanization (resource exports or industrialization), while inconsequential with regard to the level of urbanization, may be consequential for long-run development. To pick up the example of the countries mentioned above, Kuwait, Libreville, Riyadh, Tripoli, Algiers, Luanda and Lagos have relatively fewer workers in manufacturing and tradable services than Montevideo, Taipei, Seoul, Mexico City, Kuala Lumpur, Johannesburg and Shanghai respectively, in spite of similar aggregate levels of income and urbanization. It is not clear that resource-led consumption cities will generate the same kind of productivity growth associated with cities in industrializing countries. The remainder of this paper is organized as follows. Section 2 presents a model of structural change that frames the empirical investigation to follow. Section 3 describes the data and presents stylized facts regarding resources and urbanization.

Read the text above carefully and answer the questions below.

what specific question are the authors trying to address? Why should we care about this question from an economic perspective? Is it an unproven theoretical result? What are its key findings? Is this paper the first to address an important research question? If not, how does this paper compare with previous similar efforts?