Natural Resources, Development and Democracy: The Quest for Mechanisms. Erik Wibbels Department of Political Science University of Washington

Transcription

1 Natural Resources, Development and Democracy: The Quest for Mechanisms Erik Wibbels Department of Political Science University of Washington Ellis Goldberg Department of Political Science University of Washington Abstract The correlation between natural resource abundance, authoritarianism and poor developmental outcomes typically referred to as the resource curse is one of the few findings in comparative politics and political economy on which there is something approaching a scholarly consensus. That the case, there is widespread disagreement as to the causal mechanisms behind the correlation. Indeed, we count 11 distinct causal stories linking resource abundance to political-economic outcomes. This dissensus persists in large part due to the empirical difficulty of testing alternative hypotheses cross-nationally and limits our understanding of the causal processes at work. We address this difficulty with two innovations: first, we integrate insights from the traditional enclave economy argument with work on economic geography to develop a new hypothesis to account for the negative relationship between resource dependence and poor developmental outcomes; second, we test our hypothesis against the 11 hypotheses culled from the literature in a new, better empirical context. We find support for our hypothesis and several others while discarding most of the conventional wisdom. In doing so, we serve to sharpen the debate on the factors underpinning the resource curse.

2 "A place in the throes of an energy boom isn't so different from a person in the throes of addiction: there's the denial that things are out of control; there's the sleeplessness and the moral carelessness, and the fact that you're doing something that you know isn't good for you but you just can't stop." 1 Over the course of three decades, researchers have gathered mounting evidence that natural resource abundance contributes to all manner of dysfunctional outcomes, ranging from poor and uneven economic development (Sachs and Warner 1995) to authoritarianism (Ross 2001) to civil war (Collier and Hoeffler 2001). Collectively, these findings have become known as the resource curse. Some common problems in the literature aside (Ross 1999), these findings represent some of the most consensual in comparative politics and political economy. They also represent some of the most important. The empirical effect of natural resources on some of the most central dependent variables in the social sciences aside, the findings have considerable theoretical importance. The association between endowments in natural resources and poor growth, for instance, represents a serious challenge to the most fundamental theory of international trade, namely comparative advantage. Given the consistency and importance of the findings, it comes as something of a surprise that there is little agreement as to the precise reason or reasons that resource abundance contributes to poor outcomes. Indeed, below we identify 11 separate hypotheses linking resource wealth to authoritarianism and poor developmental outcomes. Political scientists, for instance, have suggested that mineral abundance contributes to authoritarianism via its detrimental impact on everything from the development of a robust civil society to public institutions to corruption to repression. Meanwhile, researchers have chalked up the negative effects on economic development to exchange rate effects, poor institutions, economic volatility, poor fiscal management, and the problems associated with state ownership of economic assets. These hypotheses are not necessarily mutually exclusive, and it might be that many (or even all) of them contribute to the resource curse. That the case, most of the hypotheses in the resource curse literature suffer from one common 1 Alexandra Fuller Boomtown Blues: How Natural Gas Changed the Way of Life in Sublette County The New Yorker Feb. 5, 2007: Page 44.

3 flaw they imply that mineral wealth inevitable leads to poor outcomes. Economic theory and economic history suggest differently. Trade theory provides no basis for suspecting a consistent resource curse. The impact of oil or mineral booms should be positive or negative depending on the sectoral makeup of an economy. 2 Indeed, an entire class of open economy models suggests that a booming sector can generate a level of domestic demand sufficient to generate spillovers to other sectors and ultimately increasing returns to a wide range of economic activity (Murphy, Shleifer, and Vishny 1989; Krugman 1991; Corden 1984). Empirically, there are simply too many cases for which natural resource wealth provided the foundations for protracted economic booms and/or were consistent for democracy. Indeed, mineral wealth seems to have done nothing to preclude the emergence and sustainability of democracy in Norway or Botswana, and Chile s recent decades of economic growth and democratic deepening have occurred at the same time as a boom in natural resource based exports. Likewise, economic historians emphasize the developmental foundations of easily extracted coal in early 19th century Great Britain (Pomeranz 2000), the comparative per capita resource wealth of the U.S. and Australia in the latter decades of the 19th century (Wright 1990), and the positive long-term implications of the California gold rush (McLean 2005). We address this shortcoming by developing a new hypothesis linking natural resources to developmental outcomes via economic geography. Rethinking the traditional emphasis in the literature on the propensity for natural resources to be produced in enclaves (Sachs and Warner 1995; Hirschman 1958) from the point of view of research on economic geography (Krugman 1991), we argue that: first, geography (broadly understood to include local endowments) conditions the costs of researching and developing natural resources with consequences for the initial level of local demand at the time of the resource boom; and second, geographically-determined access to external markets conditions the 2 A resource boom has two key effects the reallocation of resources and increased incomes. Import-competing sectors benefit from the income effect as demand increases but are hurt by the higher wages associated with the reallocation of resources. For these sectors, the net impact is ambiguous. Non-booming traditional exports will be hurt by the rise in wages brought about by the boom sector and benefit little from the income effect. Nontradables, on the other hand, can adjust to higher wages with increased prices and will benefit from increased demand via the income effect of the boom. In this sectoral approach, the overall impact of a resource boom is positive when the economy is weighted to non-tradables and negative to the degree that it relies on non-boom exports.

4 likelihood that natural resource wealth will generate externalities and increasing returns to other kinds of economy activity (and thereby development ). Where local demand is low and transportation is costly, natural resource production is likely to remain isolated in an enclave. As local demand increases and transportation costs fall, the prospects for resource wealth to spillover into other forms of production and foster development improve. Such mechanisms, we suspect, offer the potential to account for apparent outliers the Norways and Californias of the world. Our own hypothesis added to the mix, the proliferation of hypotheses represents a serious shortcoming in our understanding of resource curse and certainly precludes developing coherent policy advice for addressing natural resources supposed detrimental impact on societies. There are several reasons for this divergence of opinion. The quantitative research has been plagued by a dearth of data that might allow for detailed hypothesis testing on the mechanisms linking resource wealth to political and economic outcomes. There are, for instance, no (or very scarce) reliable measures of the richness of civil society, levels of corruption, the quality of institutions, and even real exchange rates across countries all channels that researchers have identified as causing the resource curse. Qualitative research on the subject has produced myriad studies in diverse empirical contexts from diverse theoretical perspectives. While some such research may simply reflect a disregard for hypothesis testing, most has suffered from the common small-n challenge of establishing external validity (Ross 1999). The result of these shortcomings in the comparative literature is that hypotheses have proliferated in the absence of any means to falsify them. We address this problem by turning to a different empirical setting that is considerably more data rich: the U.S. states. The U.S. states offer a number of advantages over the cross-national research on the resource curse including a far richer data environment, a relatively uniform set of rules governing data collection across the states, cultural differences between governments substantially less than in cross-national settings, and a longer time-series dating back as far as 1929 (and therefore containing several international business cycles). These advantages allow us to do two things: first, establish whether or not the resource curse exists in a sample of cases different from that which generated the

5 hypothesis; and second, test the mechanisms underpinning the findings in a much more systematic fashion than is possible in the cross-national setting. We proceed in four parts. In the next section, we overview the resource curse literature with specific attention to the numerous competing hypotheses linking resource dependence to economic and political outcomes. Thereafter, we outline our chief theoretical innovation, namely a hypothesis linking economic geography to the impact of natural resources on development. In the fourth section, we provide the first systematic test of the 11 competing hypotheses that we extract from the resource curse literature. Finally, in the concluding section, we consider the theoretical implications of our findings for the literatures on the resource curse and trade more generally and the implications for future, crossnational work aimed at uncovering the causal relationship between resource abundance, development and democracy. II: The Resource Curse and the Proliferation of Hypotheses Initially articulated in the economic realm as part of the structuralist rethinking of classical economics (Prebisch 1950) and later extended to the political realm in the context of the Middle East (cites), the resource curse literature argues that reliance on natural resources contributes to a number of dysfunctional outcomes. The conventional wisdom is composed of two general propositions that resource abundant nations grow more slowly than resource poor ones and that politics in resource rich societies are authoritarian. Over the last decade, the amount of research lending credence to various aspects of the resource curse is impressive (Ross 2001; Jensen and Wantchekon 2004; Chaudry 1997; Isham et al 2003; Leite and Weidmann 1999; Karl 1997; Smith 2004; Sala-I-Martin & Subramanian 2003; Sachs and Warner 1999; Mehlum, Moene, and Torvik 2006; Morrison 2006). Widespread agreement on the finding aside, there is very little agreement as to the mechanisms or causal processes behind the findings. Indeed, little has changed since Michael Ross wrote nearly 10 years ago that There is now strong evidence that states with abundant resource wealth perform less well

6 than their resource-poor counterparts, but there is little agreement on why this occurs. 3 This is unfortunate, because figuring out why there seems to be a resource curse is of crucial importance for a number of broad literatures. Indeed, the importance of the resource course finding transcends a narrow concern with mineral wealth itself. For instance, the dominant theory of trade is comparative advantage the notion that countries benefit by producing goods employing locally abundant factors of production and engaging in trade. That the production of some locally abundant factors of production might detract from growth and development is an important claim, particularly in light of the emphasis on the advantages of trade openness in much of the research on development in recent decades. 4 As Sachs and Warner note one of the surprising features of modern economic growth is that economies abundant in natural resources have tended to grow slower than economies without substantial natural resources. 5 Likewise, the conventional wisdom holds that higher per capita incomes are the single most important determinant of the sustainability of democracy. That oil-rich countries consistently appear as outliers in such studies underscores the importance of understanding the resource curse for broader theories of regime type. In both literatures, the apparent exceptionalism of resource abundant countries has produced sharp debates around the causes (and existence) of the resource curse. Despite the importance of the finding, increased research on the political economy of natural resource abundance over the last decade has only served to proliferate the number of causal hypotheses. Detailed, though important, concerns with methodology aside (see Stijns 2005), the empirical challenges associated with testing the competing hypotheses in cross-national settings are manifold. First, crossnational economic data is available for a relatively short time period (approximately 30 years) that provides limited traction for studying the impact of resource abundance on long-run dynamics associated with democratization and development. Second, the period includes in many ways the most anomalous era in the history of natural resource markets one in which an international cartel was able 3 Ross 1999: See, for instance, Williamson 1993, Edwards 1998, and Frankel and Romer See Rodríguez and Rodrik 2001 for a skeptical view. 5 Sachs and Warner 1995: 1.

7 to limit the supply of oil and radically increase prices. Third, even where cross-national data is available, the more detailed data necessary to test competing explanations is not. Fourth and finally, there are considerable cross-national differences in everything from data collection standards to legal practices that can be very difficult to measure and thereby might limit the generalizability of the findings. For all of these good reasons, the resource curse literature is stuck in neutral with an apparently strong finding and limited capacity to explain it. In short, while it might be the case that countries become what they produce (Rodrik and Hausman 2006: 2), we do not know much about why that would be the case. While hypotheses mushroom and various kinds of empirical support mount, the challenge remains of actually testing competing hypotheses against each other. In this section, we briefly overview the various hypotheses as a precursor to the first systematic test of the contending arguments in the resource curse literature. 6 II.2: Politics We begin with the impact of natural resources on political competition. One widespread view as to the mechanisms governing the relationship between resource wealth and political competition emphasizes the rentier effect (Crystal 1990). In the rentier state literature, easy revenues accrue to the state from natural resource production. Because production is concentrated at a small number of sites, governments have an easy time controlling and extracting rents from such production. Thanks to resource rents, public officials have little need to develop taxing capacity. Indeed, these cheap revenues allow officials to buy public support and build patronage networks. Lightly taxed, the citizenry only weakly demands representation in government, thereby stunting the development of democratic institutions and attenuating political accountability. There are two steps in the causal chain: first, from natural resources to easy fiscal expansion and patronage networks; and second, from patronage to the capacity to survive under otherwise adverse circumstance to insulate incumbents from the political business cycle. 6 For other good recent overviews of the diverse causal accounts that emerge from the resource curse literature, see Ross (1999), Isham et al (2003), and Sarraf and Jiwanji (2001)

8 Building on the classic work of Moore (1966), a related argument links natural resource dependence to delayed processes of modernization and the failure to develop the social constituencies and cultural practices that provide the foundation for robust, competitive politics. In some of these accounts, states reliant on natural resources actively resist the process of industrialization as a means to head off the creation of alternative sources of social and economic power, such as urban labor, middle classes and industrialists. In the absence of such actors, the government faces limited demands for increased literacy, union organization, education spending, political pluralism, and the like. The net result is that many resource rich societies lack the social capital that is a precursor to democracy (Bellin 1994). Yet a third hypothesis links natural resource abundance to political contestation via its impact on asset inequality and the mobility of capital. In several recent accounts (Boix 2003; Acemoglu and Robinson 2006), democratization occurs when the redistributive demands of labor are muted and elites can, therefore, foresee relatively few threats to their economic power under a democratic order. According to Boix, two factors impact the intensity of redistributive demands: first, the level of economic inequality; and second, the capacity of capital to exit. Consistent with the long-standing median voter tradition (Meltzer and Richard 1981), economic inequality serves to exacerbate redistributive demands, thereby encouraging elites to maintain constraints on political pluralism. Capital mobility has the opposite effect as it reduces the redistributive demands of labor which recognizes that excessive redistributive demands will cause capital to flee. The result is a political elite more predisposed to democratize. According to such an account, natural resource wealth reduces the likelihood of democratic politics by both increasing the level of interpersonal inequality (Leamer et al. 1999) and eliminating capital s exit threat since natural resources are the quintessential fixed asset (Boix 2003). 7 One final hypothesis bears attention, namely that linking resource wealth to poor institutions and 7 In Leamer et al s (1999) account, natural resources serve to exacerbate inequality by pulling physical capital into sectors that require little human capital. The result is a workforce heavily oriented toward low skill, low wage activity and education systems poorly adapted to the accumulation of equality-inducing human capital.

9 thereby to undemocratic politics. The typical argument suggests that natural resource wealth encourages society to engage in rent seeking which militates against the development of institutions and discourages the state from developing the institutional capacity to gather systematic information about society. Both state leaders and large portions of civil society come to share a preference for discretion in the production and distribution of private goods by the state. The result is a tendency for incumbents to view the state as a tool for rewarding supporters and heading off opponents, while government opponents see little opportunity for altering state behavior through institutionalized politics. In such environments, incumbents accrue tremendous benefits from weakly institutionalized constraints and are able to marginalize any potential opposition. Different researchers emphasize the negative impact of natural resources on different institutions ranging from fiscal (Jones-Luoung and Weinthal 2006; Karl 1997) to regulatory institutions (Ross 2001). In these accounts, poor institutions increase the likelihood of rentseeking, corruption and repression and thereby delay the emergence of robust democratic competition. II.3: Growth and Development Previous research has discussed a plethora of plausible mechanisms linking natural resource dependence to poor economic growth and developmental outcomes. One of the most prominent of these arguments links natural resource booms to exchange rate effects commonly referred to as Dutch Disease. The most common version of the argument suggests that natural resource booms lead to appreciation of the real exchange rate and a resulting decline in the competitiveness of traditional exports and import-competing sectors (Corden 1984; Auty 1998; Usui 1997; Mikesell 1997; Sachs and Warner 1995). Related accounts emphasize the role of booming mineral sectors in drawing labor and capital away from traditional sectors of the economy. Particularly where traditional sectors are associated with manufacturing, the resulting process of deindustrialization is likely to contribute to poor developmental outcomes over the long term, and politically inspired attempts to shield negatively affected producers via trade protection and subsidies often serve to exacerbate the problem (Auty 1994; Sachs and Warner 1995; Matsuyama 1992). Although there is some evidence of real exchange rate appreciations in the presence of resource windfalls, opinion is divided as to the long-term developmental

10 costs of reliance on natural resources. While proponents of Dutch disease style arguments typically assume the superior developmental characteristics of reliance on manufacturing and the externalities related to it (Sachs and Warner 1995; Matsuyama 1992), critics cite the lack of evidence that reliance on any particular type of endowment is superior to any other (Stijns 2001; de Feranti et al. 2001; Wright 2001) and emphasize the ease with which governments can sterilize the impact of mineral-inspired inflows on the real exchange rate. The overall cross-national evidence on this count is mixed. 8 A second set of arguments suggest that poor developmental outcomes are a result of the terms of trade associated with commodity exports. The oldest version of this argument suggests that the terms of trade for natural resources will decline through time, resulting in relatively poorer economic performance in commodity-reliant states (Prebisch 1950; Singer 1959). Recent research on commodity markets sheds considerable doubt on these claims (Cashin and McDermott 2002), but a related argument persists, namely that it is the volatility of international commodity markets that contributes to unpredictable export earnings, weak growth and poor development (Nurske 1958; Levin 1960; Isham et al. 2003; Auty 1998; Mikesell 1997). Under conditions where commodity exports crowd out nonresource tradables, these effects are particularly pronounced (Hausmann and Rigobon 2003). Volatility, it is argued, produces an unstable environment for investment, with the result being poor long term economic performance. These arguments go hand in hand with evidence that more volatile economies grow more slowly (Ramey and Ramey 1995) and that the terms of trade are an important determinant of economic growth (Mendoza 1997). That the case, the international evidence on trends in the terms of trade for commodities is mixed, with some arguing there has been a long-term decline (Powell 1991; Bleaney and Greenway 1993) while others paint a more benign picture (Cuddington 1992; Cashin and McDermott 2002). 8 Fardmanash (1990) finds evidence that while resource booms hurt agriculture, they do not hurt manufacturing. Sala-i-Martin and Subramanian (2003), using Easterly and Levine s overvaluation variable, find no evidence of a dutch disease effect. Likewise, in an overview of several papers, McMahon (1997) finds no support for an exchange rate effect. Auty and Evia (2001), on the other hand, suggest that the Bolivian economy showed signs of Dutch disease as evidenced by limited diversification, an overly small agriculture and a protected, uncompetitive manufacturing sector.

11 A third hypothesis suggests that resource abundant nations grow slowly because they systematically overspend and invest in inefficient ways that non-mineral rich nations do not (McMahon 1997). Three interrelated factors might contribute to this dynamic. First, government spending accelerates rapidly during boom periods, but because of the political difficulty of retrenching spending in subsequent bust periods, politicians respond by borrowing. Though this disjuncture between fiscal policy during up- and downturns in the business cycle is a common finding in the public finance literature, McMahon (1997) suggests the scale of the effect is distinctive in mineral states and refers to it as the irreversibility of government expenditure. Second, the negative long-term effects of overindebtedness are likely to be exacerbated by the inefficiency of investments that increase very rapidly in boom time, particularly when pursued via government spending (Sachs and Rodriguez 1999). Such spending, for instance, often goes to inefficient, but politically attractive, capital projects and the expansion of the public bureaucracy. Third and finally, because natural resources represent obvious and easily measureable collateral, banks are willing to lend more to such countries than to others (Cuddington 1989). 9 When easy access to credit is combined with boom-driven spending cycles and inefficient government spending, the result is excessive indebtedness and the misallocation of capital which jointly lower economic performance. Yet if exchange rates, economic volatility and overspending are prominent lines of argumentation, probably the most prevalent hypothesis in contemporary research linking resource dependence to poor growth is that which emphasizes the quality of institutions (Robinson, Torvik and Verdier 2006; Sala-i-Martin and Subramanian 2003). Consistent with a long line of work suggesting that good institutions represent a key first-order ingredient of long-term economic development (North 1981; Olson 1982; Acemoglu and Robinson 2006), such accounts emphasize the role of natural resources in limiting the incentives of states to develop the institutional capacity to gather information from society (through tax collection or regulatory oversight) that might contribute to better public policy. Nor do such 9 Auty and Evia (2001), for instance, provide evidence from Bolivia which used the government revenues accrued as result of a commodity boom in the 1970s as collateral for unsustainable foreign borrowing.

12 states develop institutional mechanisms for members of society to provide systematic feedback into the policymaking process. A recent variant of this argument suggests that state ownership of natural resources militates against the development of institutions since the state can rely on resource rents rather than developing extractive capacity (Weinthal and Jones-Luong 2002). In these accounts, poor institutions are endogenous to resource wealth. In yet other accounts, the quality of institutions are considered exogenous, and it is institutional quality that mediates the relationship between mineral wealth and growth (Mehlum, Moene, and Torvik 2006). When institutions are good, resource booms contribute to income growth. When combined with large rents from natural resource production, however, poor institutions for gathering, receiving and processing information foster rent-seeking (Leite and Weidmann 1999; Tornell and Lane 1999), corruption (Leite and Weidman 1999), weak rule of law, low state investments in bureaucratic capacity (Sala-i-Martin and Subramanian 2003), and even civil war (Collier and Hoeffler 2000) all of which are associated with poor economic growth. In short, in the absence of institutional constraints, interests in society compete to extract from a fundamentally distributive state rather than focusing on market competition (Bates 2005). Finally, a longstanding argument suggests that natural resource exploitation results in a sector poorly linked to the rest of the economy resulting in an enclave economy (Hirschman 1958; Seers 1964; Baldwin 1966). Because such sectors are highly capital intensive, employ relatively few people, rely on imported inputs, and are oriented overwhelmingly toward export markets they fail to foster upstream and downstream investments that might promote a broad-based economic boom. The result is an economy lacking in diversification and profoundly dependent on the whims of the natural resource sector. All of these hypotheses suffer from one key problem. As formulated, they suggest that natural resource abundance always and inevitably leads to bad outcomes. This is problematic because there are many cases both historic and contemporary that do not fit easily within the resource curse paradigm. Historically, natural resource endowments play a central role in the developmental successes of the U.S., Canada, Australia and England. In none of these cases, moreover, did natural resources block or delay in

13 any obvious way the development of competitive, democratic politics. Likewise, the contemporary experiences of countries ranging from Norway to Botswana to Chile show little evidence of suffering from a resource curse. The challenge, therefore, is to account both for cases in which natural resources detract from growth prospects and others in which resource wealth serves as a catalyst for broad-based development. III: Geography, Natural Resources and Development In response to this challenge, we develop a new hypothesis to account for the apparently divergent impact of natural resources on growth across the world. Our hypothesis builds on traditional arguments linking natural resource extraction with undiversified enclave economies. Traditional discussions of enclave economies have emphasize that natural resource extraction is capital intensive, involves a small, specialized workforce, and often relies on imported inputs (Hirschman 1958). The result is weak linkages with the rest of the economy. As with most of the hypotheses outlined above, however, the enclave economy provides no foundation for explaining the many cases in which natural resource wealth served as a catalyst for broader economic development. Thus, while traditional enclave accounts bear strongly on the West Virginian and Bolivian economies, they fail to explain the role of natural resource wealth in fostering broad-based development in places like California and Australia. With specific reference to our empirical cases, we must ask: Why does the coal boom of West Virginia look so different from the gold boom of California from the vantage point of 2007? Though all of the hypotheses outlined above provide logical accounts that might contribute to the negative relationship between natural resources and development, we believe they do not account for one of the most powerful factors mediating the impact of resource abundance, namely geography. Rethinking the traditional emphasis in the literature on the propensity for natural resources to be produced in enclaves (Sachs and Warner 1995; Hirschman 1958) from the point of view of research on economic geography (Krugman 1991), we argue that topography has two potential implications: first, geography (broadly understood to include local endowments) conditions the costs of researching and developing natural resources with consequences for the initial level of local demand at the time of the

14 resource boom; and second, geographically-determined access to external markets conditions the likelihood that natural resource wealth will generate externalities and increasing returns to other kinds of economy activity (and thereby development ). Where local demand is low and transportation is costly, natural resource production is likely to remain isolated in an enclave. As local demand increases and transportation costs fall, the prospects for resource wealth to spillover into other forms of production and foster development improve. Such mechanisms, we suspect, offer the potential to account for apparent outliers the Norways and Californias of the world. Consistent with the insights of Krugman (1991), we expect that natural resource extraction can, under certain conditions, generate positive externalities for other forms of production, thereby generating increasing returns to a rich array of economic activity, urbanization, and development. In explaining when this is likely to happen, Krugman emphasizes two key factors: the size of the local market and transportation costs. As local market size increases, incentives mount for other producers to locate nearby. Transportation costs work in a parallel manner. As economies of scale mount, the incentives to produce in any given location increase as its transportation networks improve only thus will large initial investments be recouped by serving a broader market through trade. How does this discussion bear on the relationship between resource wealth and development? Natural resource deposits are randomly located around the world. This means both that the sites and the sizes are normally distributed a lot of the resource will be in a large number of very small sites and a lot will be in a small number of very large sites. Search and development, however, is determined by cost, so the cheapest (easiest) deposits will be located first. These will usually be deposits near existing use, i.e. where local demand is quite high. Only over time (as more geological and technological information is gained and transportation technology also improves) will more distant sites be located and developed. These are in "harsher environments" on average because they are not where people preferred to live at the earlier stages of search and development. Because populations will be smaller and less dense, local demand will be lower. The returns to natural resources will ensure extractive investments, but the paucity of local demand will militate against increasing returns to other economic activity. Such

15 an economy will be linked via trade to external markets and become an enclave. 10 The geography of transportation costs functions similarly. Broad-based investments and economic concentration (another way of talking about development) will only occur if the costs of serving a wider market are relatively low. Thus, while a large resource boom may generate some initial level of demand, whether that demand is served by investments in local production and external economies resulting from other firms decisions will depend in large part on the costs of transporting other products to wider markets. Again, where the costs of transporting ancillary products to broader markets are high, investments will likely be concentrated only in the high return resource sector the value to weight ratio of other products will simply be too low to warrant producing them for outside markets. In contrast, if transportation costs are low, the local demand and capital generated by a resource boom are more likely to generate spillovers for other sectors of the economy. This is precisely the story told by McLean (2005) of California, where the gold rush generated population inflow, high wages and considerable local demand, but also where propitious geography combined with mineral capital to finance capital intensive, mechanized, large-scale investments in wheat and other agricultural production for outside markets shipped via an emerging San Francisco. The growth of San Francisco strengthened local demand and stimulated early industries, generating increasing returns to a broad swath of economic activity. Thus, the gold boom rippled through the agricultural, service, nascent manufacturing, construction, and transport sectors. Los Angeles emerges as a manufacturing center as part of this process. IV: Data, Methods and Results A decade ago, Chaudhry noted that theories of the rentier state far outstrip detailed empirical 10 If we think of development in terms of GDP/population, harsher environments are thus likely to reduce both the numerator and the denominator. This might help explain the apparently strange finding in the comparative literature, whereby resource endowments are associated with weak growth but high per capita incomes. In the presence of a large resource boom, the depressing effect of bad geography on the denominator of GDP/population will generate high per capita incomes. At the same time, the absence of strong local demand and transportation networks (themselves also a function of geography) is likely to facilitate weak growth.

16 analysis of actual cases. 11 In this section we address that problem by bringing a new dataset to bear on the resource curse literature. In the analysis, we focus on the relationship between natural resource dependence, economic development, and the competitiveness of electoral politics in the U.S states from 1929 through 2002 (see Appendix for data sources). A focus on the U.S. states has a number of advantages over the traditional approaches taken in the literature. First and most important, it allows us to analyze a much longer time-series of data than any previous study. As noted above, existing crossnational research provides little leverage on many national cases that were authoritarian and/or poor before and after the oil-induced swelling of state coffers. Second, the U.S. states evince considerable diversity in their resource abundance, levels of development, and experiences with electoral democracy. 12 Third, the U.S. states provide considerable variation with regards to a number of alternative explanations for political and developmental outcomes. For example, the colonizing nation, factor endowments, and transportation networks vary considerably across the states. Finally, inherent to varying degrees in statistical comparative politics work is a considerable amount of unmeasured crossnational variation that is consumed by either the error term or country dummies this is what researchers either do not know, do not understand, or cannot measure but that has a bearing on explaining outcomes across nations. By analyzing states within a federation, we control for legal practices, institutions of government, party systems, cultural differences, and data collection standards that might impact findings but that are often un- or poorly-measured in cross-national work. In the following pages, we pursue two lines of analysis: first, we provide evidence of both political and economic resource curses in the U.S. states; second, we use detailed state-wise data to examine the four hypotheses linking resource wealth to uncompetitive politics and similar data to test the causal mechanisms commonly posited to impact economic growth as well as our own geographic hypothesis. In the hypothesis testing sections, we first assess whether or not there is any relationship between resource abundance and the mechanism under investigation (institutional quality, for instance) : Mean resource dependence across the U.S. states is very similar to that across countries around the world and shows a higher standard deviation.

17 and then utilize either simultaneous equations or two stage least squares to assess if natural resources work through the hypothesized mechanisms to impact political and economic outcomes. IV2: The Resource Curse in the U.S. SEMINAR PARTICIPANTS CAN SKIP THE NEXT THREE PAGES IF THEY VE READ PAPER 1 Here we present initial evidence that there is a resource curse in the U.S. states. The mere existence of such a curse in an empirical setting entirely different from previous research provides considerable support to the core finding in the comparative politics and political economy literatures. We estimate two sets of models, one designed to assess the prevalence of an economic resource curse and the other focused on the political aspect of the resource curse literature. Where time-series data is available and appropriate, we estimate random effects, cross-sectional time-series models with state fixed effects, a lagged dependent variable and define the errors as clustering on the cross-sections. In the event only a single data point is appropriate for hypothesis testing or available for each state, the report results for simple OLS models with robust standard errors. The results are robust to alternative specifications. Turning first to growth and development, Table 1 reports the results for three different dependent variables: logged per capita income in 2002 (Model 1), the ten year average of log annual differences of state per capita income (Model 2), and annual percent change in state per capita income (Model 3). 13 Per capita income in 2002 is a more direct assessment of long-term development while the latter two assess short- and medium-term economic growth. The key independent variable, natural resource dependence, is measured as annual oil and coal production as a share of state income as our measure of resource dependence. 14 This approach mirrors the typical approach in the resource curse literature and facilitates comparisons with existing findings. 13 There is some debate as to the appropriate measure of growth. See, for instance, Chatterjee and Shukayev s (2006) critique of Ramey and Ramey s (1995) use of average log differences, the standard in the growth literature. They recommend using annual percent change. 14 Focusing on either oil or coal separately has little impact on the results which are not very sensitive to the operationalization of the dependent variable.

18 Nevertheless, some have criticized the use of GDP in the denominator since a high resource/gdp ratio could be the result of a high numerator or a low denominator. An alternative measure, resource production per capita, suffers from its own problems. 15 In any case, our results are robust when using this alternative measure of resource dependence. The measure of resource dependence in Model 1 is the average for the entire period, while in Models 2 and 3, resource dependence is measured as the ten year average and annual lag, respectively. We control for factors commonly emphasized in each body of research. All models control for lagged wealth following Barro s (1989) evidence and theorization of a return to the mean in growth rates. 16 Model 1 also controls for factor endowments, access to external markets, and colonial heritage. Factor endowments have important implications for development. In the context of the U.S. states, Engerman and Sokoloff (2000) argue that the key factoral determinant of long term growth trajectories is the degree to which geographic and climatic conditions created the foundations for either plantation or smallholder agriculture. Plantation agriculture led to slavery, extractive property rights institutions, exclusionary political institutions, and weak human capital development all of which contributed to poor long-term development. Smallholder agriculture in the northeast did the opposite. Thus, we introduce a measure of the percentage of the state population that was enslaved in A second factor oft-associated with development is access to external markets (Hausmann 2001; Acemoglu, Johnson and Robinson 2005). To control for market access, we include a dummy variable for states that have access to rivers, lakes, or an ocean upon which to transport goods to and from foreign markets. 18 Finally, a prominent line of work suggests that colonial origins have implications for long-run 15 Most importantly, if resource abundance affects growth and growth has implications for birth rates (Przeworski et al 2000), the numerator and the denominator of a resource per capita measure are not independent. 16 In Model 1 this is measured as the initial per capita income in 1929, though for Hawaii and Alaska it is per capita income in the year they achieve statehood. In model 2, income is measured as the initial level at the beginning of the 10 year period. In model 3, it is the one year lag of per capita income. 17 Data from Mitchener and Mclean (2003). Note that this measure significantly improves on the atheoretical standard practice of including a dummy for southern states in statistical work on the U.S. 18 Data from Mitchener and Mclean (2003).

19 developmental trajectories (North 1979). The 50 U.S. states had eight different colonial experiences. 19 The most common theme running through the literature is the negative implications of Spanish colonialism. As such, we create a dummy variable taking on a value of 1 for any state in which the Spanish were not involved. In the growth equations, Models 2 and 3 control for each state s capital stock, human capital endowment and government spending consistent with standard models in the crossnational growth literature (see Barro 1997). 20 The findings are consistent natural resource dependence has a negative impact on growth and development. Models 2 and 3 increase our confidence in the findings by spotlighting growth. Focusing on the easier to interpret coefficient in Model 3, the results suggest that a 10 percent increase in natural resource dependence reduces annual growth by 1.4 percent relative to a state with no natural resources. Lest the reader think these cuts in growth rates trivial, they imply that relying on natural resources to the tune of 30 percent of the state economy (think Louisiana or West Virginia) would reduce the average state per capita income by $5,000 over the next decade when compared to a similar state without such resources. Unreported results suggest that these findings are robust to additional controls for which we have less data, the time period under analysis, the use of a per capita resource endowment measure, and the estimation procedure. In short, we find substantial support for an economic resource curse in the context of the American states a fact that lends credence to the cross-national resource curse literature. Table 1 Here Table 2 turns to politics. We measure the competitiveness of the electoral environment with two indicators: the margin of victory in gubernatorial elections and the vote share of the incumbent governor. We are not claiming that resource dependence transform polities into one-party dictatorships. The logic 19 The eight different experiences were: colonized by the English alone, the French alone, the Spanish alone, the English and Dutch, the French and Spanish, the English and Spanish, the English, French and Spanish, and those that were not colonized (or rather were colonized by the U.S.). 20 Capital stock is measured on a per capita basis and is from Garofalo and Yamarik (2002). Human capital is measured as the share of the population with a high school diploma and comes from the U.S. Census. State spending is measured as a share of state income and comes from the Book of the States. The educational attainment data is reported by decade beginning in We interpolate the data for the intervening years. The state spending data is reported at mostly two year intervals from 1940 to 1982 and annually thereafter. We interpolate the missing data.

20 of the resource curse argument does not, however, imply such a claim. Recall that the argument is that leaders of resource-rich states are able to maintain lower direct taxes on their citizens while mineral rights provide them with rents to lavish on key constituencies. As a result, they are expected to survive in office for long periods. It is generally accepted in the business-cycle literature on democracies that incumbents lose votes during periods when the economy either ceases to grow or experiences decline. The presence of state income generated as an external rent should allow political elites to remain in power without regard to the business cycle, and thus a rentier effect should appear in a democratic context and be evidenced by a party or governing elite winning by large electoral margins and winning higher vote shares across the business cycle than for politicians without access to mineral rents. Controls include the same slave state, colonial heritage, and wealth indicators noted above. Given the importance of growth in retrospective election models, we also introduce a control for state-level economic growth the year prior to the election. In Model 2 we also control for the incumbents vote share in the prior election. Table 2 Here Given the poor economic performance of resource dependent states noted above, the U.S. voting behavior and comparative literature on elections would suggest that such states should see significant political turnover. From research on elections in the U.S. states to those across established democracies to newer democracies in poorer regions of the world, weak economic growth is associated with declining electoral fortunes of incumbent governments. The results in Table 2 provide further support for the resource curse hypothesis, though the scale of the impact is relatively small. Turning first to electoral margins, each percent increase in natural resource dependence increases the margin of gubernatorial victory by about.15 percent. Put differently, a U.S. rentier state (defined as one in which resources constitute 20 of state product) is predicted to have gubernatorial victories 3 percent larger than a state without natural resources. Likewise, Model 2 shows a statistically significant, postitive impact of resource abundance on incumbent vote shares. The scale of the effect is relatively small but that it appears at all after controlling for the lagged vote of the incumbent, the business cycle and state

21 dummies suggests a consistent effect that would make a real difference in elections that are relatively close. IV3: The Impact of Natural Resources on the Hypothesized Mechanisms But what exactly is the means through which resources are having this impact? In this section, we present the results of 11 different models designed to measure the impact of resource wealth on the many mechanisms we discuss above. As formulated in the literature, each mechanism is hypothesized to work in two stages: first, mineral wealth has implications for the mechanism, and second, thereby impacts economic and political outcomes. The intuition behind the empirics here is that if resource abundance does not have an impact on the hypothesized mechanism, it cannot thereby impact outcomes. In each model, the hypothesized mechanism serves as the dependent variable and natural resource wealth as the key independent variable. The other independent variables vary from model to model in a manner consistent with the separate literatures on each dependent variable. Our dependent variables assess each of the mechanisms discussed above the rentier/tax, modernization, inequality and asset specificity arguments linking resource abundance to uncompetitive politics; the real price, economic volatility, and fiscal performance arguments linking resource wealth to poor growth; and the hypotheses linking mineral wealth to poor economic and political outcomes via their impact on institutional quality. We measure the rentier effect as the difference between actual taxation and taxing capacity. As argued in the rentier state literature, resource wealth should be associated with limited tax effort. We use Berry and Fording s (1997) annual data from 1960 to We proxy modernization with two measures of social development the first is Putnam s state-wise index of social capital and the second is the richness of the interest group environment as collected by Lowery and Gray (1993). Both measures are increasing in social capital. 21 Inequality is measured by each state s annual gini coefficient from Lowery and Gray measure the interest group environment as the ratio of state GSP divided by the absolute number of interest groups, which is decreasing in social capital. We invert the scale to make interpretation more intuitive. Lowery and Gray s measure of interest groups is available for 1980, 1990 and Putnam s measure is available for a single year in See Putnam (2000: ) for details on the variable. There is no social capital