Towards An Alternative Explanation for the Resource Curse: Natural Resources, Immigration, and Democratization by David H. Bearce Associate Professor of Political Science University of Pittsburgh and University of Colorado, Boulder and Jennifer Laks Ph.D. Candidate in Political Science University of Pittsburgh 2 nd draft (March 2010) Abstract: Why do many resource-rich countries maintain autocratic political regimes? The answer proposed in this paper focuses on the causal effect of labor imports, or immigration. Using the logic offered by Acemoglu and Robinson s democratization model, we posit that immigration makes democratization less likely because it facilitates redistributive concessions to appease the population within an autocratic regime and also makes repression less costly for the autocratic elites. This immigration argument applies directly to the political resource curse since many resource-rich countries tend also to be labor-scarce, leading them to import foreign laborers. Consistent with our primary hypothesis, we find a statistically significant negative relationship between net immigration per capita and democratization in future periods. Our statistical results also show that when controlling for this immigration effect, the standard resource curse variables lose statistical significance in a democratization model. This latter result suggests that much of the so-called resource curse stems not from resource endowments per se, but rather from the labor imports that are often required to exploit them. Acknowledgements: We thank Sarah Brooks, Bill Clark, Jerry Cohen, Daniela Donno, Michael Goodhart, Julia Gray, Nate Jensen, Quan Li, Irfan Nooruddin, Maggie Peters, Dennis Quinn, Ron Rogowski, Peter Rosendorff, Nita Rudra, Ken Scheve, and Jim Vreeland for helpful comments and suggestions on the earlier draft of this paper.
1 Why do many, but certainly not all, resource-rich countries maintain autocratic political regimes? One standard explanation for this political resource curse is known as rentier state theory. Resource-rich governments earn so much revenue (i.e. rents) from their natural resources that they can pay for a set of basic goods and services without taxing their citizens. Consequently, their citizens do not demand a more democratic political regime. But even if this basic rentier state logic is correct (and we will not argue otherwise in this paper), it is clearly limited in what it can explain with regards to the political resource curse. For example, there are many resource-rich countries, notably those in Latin American (Dunning 2008), that have democratized despite substantial resource rents, a development that rentier state theory alone has trouble explaining. Consistent with this understanding, Ross (2001) added government spending and taxation variables next to the standard resource curse variables (oil and mineral rents) in a democracy model and found that the inclusion of these rentier state controls only reduced the causal effect of the resource variables by 6-17 percent. This set of results suggests that while the basic rentier state logic may be correct, it nonetheless leaves most of the political resource curse unexplained. Another explanation for this political resource curse is repression theory. Resource-rich governments can use their natural resources to pay for a larger military and police force that can be used to repress their citizens who are demanding more political rights and freedoms. While this basic logic seems reasonable, it simply cannot explain much of the variation in terms of the political resource curse since there are many repressive governments that are not particularly well-endowed in terms of natural resources. Indeed, when Ross (2001) added military spending and personnel controls next to the resource rent variables in his democracy model, he found that while these military controls were negatively signed and statistically significant (indicating their inverse
2 relationship with democracy), they also had little effect on the negative coefficients for resource rent variables. In other words, while the simple logics underlying rentier state and repression theory may well be correct, these arguments by themselves offer surprisingly little traction in explaining the observed political resource curse. This understanding us back the research question stated above: why do many, but not all, resource-rich countries maintain autocratic political regimes? This paper seeks to advance a new explanation for political resource curse, one focusing on the causal effect of labor imports, or immigration. Using the logic offered by Acemoglu and Robinson s (2006) democratization model, we argue that more immigration makes democratization less likely because it facilitates redistributive concessions to appease the population within an autocratic regime and also makes repression a less costly strategy option for the autocratic elites. This immigration argument applies directly to the political resource curse since under-developed land- or resource-rich countries tend also to be labor-scarce (Rogowski 1989, chapter 1), requiring them to import foreign laborers in order to exploit their natural resource endowments. For example, the Middle Eastern oil governments (e.g. Bahrain, Kuwait, Oman, Qatar, Saudi Arabia and the United Arab Emirates) maintain very repressive political regimes and import large amount of foreign labor from South Asia to support their energy production. Indeed, the argument presented in this paper would make rentier state and repression theory endogenous to immigration in this set of resource-rich countries. Stated differently, these governments would have trouble making redistributive concessions to their citizens or paying for their repressive capabilities without the foreign labor imports necessary to exploit their energy resources. In fact, the argument presented here can also help explain why another set of energy-rich countries - notably the oilproducers in Latin America such as Ecuador and Venezuela - have been able to democratize even with substantial natural resource endowments. They are relatively labor-abundant and do not need
3 to import foreign workers, thus making democratization more likely in this set of resource-rich countries. While we offer our immigration argument as an explanation for the observed political resource curse, its causal logic should also apply more broadly, or should cover more than just resource-rich countries. Indeed, we would expect more immigration to be negatively associated with future democratization in any autocratic country regardless of its natural resource endowments. Thus, after describing our immigration and democratization argument in the next section of the paper, we first test for our hypothesized negative relationship between immigration and democratization. Then after establishing this negative relationship, we examine how much of the political resource curse can be explained away by this negative immigration effect, following a test like those first offered by Ross (2001). To preview our empirical results, we find a statistically significant negative relationship between net immigration per capita and democratization over the following five years. This result is robust to using several different operational measures for democracy, including Polity2, Freedom House and a dichotomous regime indicator. This negative relationship gets even stronger with longer time lags and when using various instruments for our immigration variable, consistent with the understanding that reverse causality tends to work against our hypothesized negative relationship given that immigrants likely prefer more (not less) democratic destinations. Finally, we also show that when controlling for this immigration effect, the standard resource curse variables lose statistical significance, even taking on a positive coefficient, in a democratization model. This result suggests that much of the so-called resource curse stems not from resource endowments per se, but rather from the labor imports that are often required to exploit them. The remainder of the paper is structured into three sections, plus an appendix. The first section presents our theory explaining why immigration should hinder democratization. The second
4 section tests our hypothesis about a negative relationship running from immigration to democratization. The third section then shows how the political resource curse can be explained away by immigration. Finally, we take this opportunity to draw the reader s attention to the appendix where we explore the determinants of country/year net immigration per capita. These immigration models have two primary purposes. First, they show how resource-rich, and especially energy-rich, countries have been large importers of foreign labor. Second, they help establish the strength of our immigration instruments used in the second section of the paper. 1. The Argument Our argument about how immigration tends to hinder democratization uses the democratization model offered by Acemoglu and Robinson (2006) [henceforth A&R]. Thus, we first describe (in non-formal terms) some key features associated with their model, focusing on two key theoretical concepts: 1) the credibility of elite concessions to appease the poor as an alternative to democratization, and 2) the costs of engaging in greater repression, another alternative to democratization. We will then discuss (in separate sub-sections) how immigration affects both the credibility of concessions and the costs of repression, leading to our primary hypothesis that immigration should be negatively associated with democratization. The A&R Democratization Model Acemoglu and Robinson s democratization model begins in an autocracy where a small group of rich elites hold de jure political power, or control over the institutions. Given that the elites obviously benefit from such a system, they prefer to maintain their autocratic regime. The elite are opposed in this preference by a larger group of poor citizens, who would like to democratize
5 because they stand to gain a larger slice of the economic pie in a more democratic regime following the logic of the median voter theorem. While this poor population lacks de jure political power, it may have some de facto political power based on its size advantage over the smaller group of rich elites. Given some de facto power and a favorable set of external circumstances (e.g. political and/or economic crises), the population could revolt to obtain de jure political power, but such a revolution would be costly because it destroys a portion of the economic assets within the national economy, thus reducing the size of the economic pie that could be redistributed in a more democratic post-revolutionary regime. While somewhat costly for the population, a revolution would be extremely costly for the rich elites because they stand to lose all of their economic assets via expropriation. To avoid a revolution, their worst possible outcome, the elite have three strategy options: 1) make redistributive concessions to appease the poor within an autocratic political regime, 2) engage in even greater repression to prevent a popular revolution, or 3) acquiesce to the population s demand for a more democratic regime. Given these three strategy options, democratization (the third option) becomes more likely as concessions (the first option) become less credible and repression (the second option) becomes more costly. As Acemoglu and Robinson (2006, 30) neatly summarized on this point: democracy arises when concessions are not credible and repression is not attractive because it is too costly. The latter consideration - that democracy is more likely when repression is more costly - is relatively straightforward. But the former - that democracy is less likely when concessions are more credible - is not obvious, so it may be helpful to explain the underlying logic from the A&R model. As just described, the elite can offer the citizens some redistributive concessions to reduce the threat of a popular revolution. But the citizens will not accept these concessions unless the elite s offer is credible. So when will the elite s offer be credible? It will be more credible when the
6 population s de facto political power is larger and more permanent. When its de facto political power is only transitory, the elite s offer becomes less credible because the citizens know that promised redistributive concessions could be later retracted. Knowing that these concessions are only temporary due to their transitory political power, the citizens have a greater incentive to revolt in the present (or whenever the external circumstances for this event become favorable), thus leading to a more democratic regime via costly social revolution. This conclusion that democratization becomes less likely when the population has more permanent de facto power may appear counter-intuitive (Acemoglu and Robinson 2006, 161) since it is more natural to expect that democracy would emerge when the citizens who want a more democratic regime are more powerful relative to the autocratic elites. But when they have permanent de facto political power, the citizens do not necessarily need de jure power (or control over the political institutions) because the elite will regularly make them economic concessions and because a revolution would be economically costly even for the citizens who would gain de jure political power as a result. It is worth noting that this logic about regular economic concessions within an autocratic regime is similar to that of the rentier state theory, a point that will be developed in more detail below. Although we have not yet offered our argument linking immigration to democracy, its basic causal logic is previewed in Figure 1. The negative relationship between the credibility of elite concessions and democratization, as well as the positive relationship between the costs of repression and democratization, come from the A&R model. Our contribution will thus come from linking immigration positively to the credibility of elite concessions and negatively to the costs of repression. Together these arguments explain why immigration should be negatively associated with democratization.
7 Figure 1: How Immigration Affects Democratization Immigration + _ Credibility of Concessions _ Democratization _ Costs of Repression + Immigration and the Credibility of Concessions Our argument about why immigration tends to make the elite s promise for redistributive concessions to the poor more credible concerns immigration s effect on the de facto power of the population. Stated simply, immigration tends to increase the permanent de facto power of the population. When migrants can enter the country and become citizens, they form part of an even larger non-elite population, permanently increasing this group s revolutionary strength. But immigrants may also increase the permanent de facto power of the non-elite bloc even when there is no possibility of the immigrants ever becoming citizens, especially if a new set of guest workers regularly enters the national economy because another set has just been deported. This latter proposition is based on the understanding that immigrant preferences tend to be more closely aligned with those of the citizen bloc in an autocratic regime and counter to those of the wealthy elite even when these immigrants can only stay in the country for a limited period of time. This certainly makes sense in the context of a factoral model of the national economy where the owners of same factor of production (e.g. capital, land, or labor) have convergent political and
8 economic preferences, while owners of different factors of production are presumed to have divergent preferences. Using this three factor framework in the context of an autocratic polity, capital and land tend to be concentrated in the hands of the elite, while the population has little more than its labor as an economic asset (Milner and Kubota 2005, 115). The latter would also be true of most immigrants who enter the autocracy looking for work with only their labor to offer to potential elite employers. Hence, the domestic population and immigrants together form the labor class in the national economy, aligned against the capital- and land-owing elites. Indeed, the fact that certain autocratic governments, including those in the Gulf Cooperation Council (GCC), refuse to extend citizenship to foreigners working in their domestic economy, treating them instead as guest workers without any political rights or civil liberties, is consistent with the understanding that this immigrant labor potentially threatens their de jure political power. As Castles (1995, 512-3) wrote about the kafala system in the GCC countries, migrants were not allowed to settle nor bring in dependents, and lacked civil or political rights. They were generally segregated in barracks. They could be deported for misconduct, and were often forced to work very long hours. The literature on the treatment of guest workers in the GCC countries often mentions that their lack of civil rights stems from the heightened risk of political unrest associated with a large number of foreign laborers (Abella 1995, 420) and that this risk varies with the ratio of foreign workers relative to the native population (Longva 1999, 22). 1 1 One can think of this lack of freedom as a way for the autocratic elites to manage the additional revolutionary risk posed by foreign laborers. As Smith (2008, 781) wrote on this point, political rights and civil liberties, particularly freedoms of association, enhance the ability of potential revolutionaries to organize against the government. This point is echoed in Bueno de Mesquita and Smith (2009, 172): freedom of assembly, free speech, free press, and transparent government improve the ability of citizens to form views about what the governments is doing and to organize and coordinate in the event that they are displeased with their government.
9 But if immigration - whether as a regular stream of temporary guest workers or a one-time increase in the size of the citizen population permanently increases the de facto power of the poor citizen bloc, thus threatening the de jure power of the autocratic elites, then why would the latter permit foreigners to enter into their national economy? Presumably they would not in a laborabundant national economy where there is no need to import additional workers. Indeed, in a laborabundant economy, one might even expect the autocratic elite to export excess, or unemployed, workers in an effort to weaken the de facto power of poor citizen population and reduce their revolutionary threat. 2 But in a labor-scarce national economy, the land-owning elites may need to import foreign labor in order to exploit their natural resources given the shortage of domestic labor willing to do this often dirty and dangerous work. This logic seems to fit the petro-governments in the Middle East (Abella 1995, 418) because their de jure political power depends in large part of the revenue that they can generate from oil exports (Bellin 2004). But this oil cannot be exported until it has been extracted and refined, which are labor-intensive production activities. Thus given the laborscarcity of these so-called frontier societies (Rogowski 1989, 31), the elites who own the land and natural resources must import workers from other (presumably labor-abundant) national economies to earn their resource rents despite the attendant political risks, especially when some of these political risks can be managed with redistributive concessions offered to the citizen population. 2 As Boone (1986, 868) argued on this point, a large population of unemployed and unmarried young men ultimately creates a highly competitive, volatile situation at the societal level. Rulers must choose between dispersing these individuals or facing disorder and overthrow on the home front. Likewise, Hudson and Den Boer (2002, 26) wrote: There is only one short term situation for dealing with a bare-branch problem [an excess supply of unemployed unmarried males]: Reduce their numbers. There are several traditional ways to do so: Fight them, encourage their selfdestruction, or export them. Thus, it is not surprising to see the governments of labor abundant national economies actively marketing their workforce abroad and signing agreements with foreign governments to supply labor. For example, Abella (1995, 419) notes that the South Korean and Philippine governments have been involved in such activities since the early 1970s.
10 In fact, the argument presented here about labor imports, or immigration, increasing the permanent de facto power of the non-elite population, thus leading the elite to offer redistributive concessions, directly relates to the logic of rentier state theory (Karl 1997). In its basic form, rentier state theory posits that the governments in energy-producing countries, especially those in the Middle East, earn so much revenue (i.e. rents) from their energy exports that they can provide a basic set of goods and services to their population without taxing them to pay for these goods and services. Thus, their citizens do not demand a more democratic political system because they are already receiving a reasonable slice of the economic pie. But in terms of this rentier state logic, the argument advanced here about immigration suggests that the elite s ability to provide these redistributive concessions may be endogenous to their labor imports at least in labor-scarce national economies. Stated differently, the elite would not be able to provide these redistributive concessions without importing foreign labor to extract and refine their oil, for example. Thus, without the ability to offer redistributive concessions made possible through immigration, the elite would have only two options available to them per the A&R model repression or democratization - thus making democratization more likely. We will now also discuss how immigration may make repression less costly, or an easier option for the elite to select instead of democratization. Immigration and the Costs of Repression A democratic regime is also less likely to emerge when the costs of repression are relatively low. Given the three options available to the elite (appease, repress, and democratize) in the A&R model, the elite may be able to forestall democracy by further repressing its population even when it cannot credibly promise redistributive concessions to appease the poor citizens within its autocratic regime. Thus, factors which make repression less costly or reduce the opportunity costs associated
11 with using the government s revenues to fund police and military forces should also make democratization less likely to occur. At first glance, more immigration might seem to increase the cost of repression since the autocratic elites now have an even larger group of poor people to repress. But this logic ignores the economies-of-scale in terms of repression infrastructure. The same police and the same military used to repress the domestic population can also be used to repress the imported laborers, especially when the guest workers are segregated into very limited living and work spaces as done in the GCC countries, for example. Another way to think about the costs of repression concerns the opportunity costs associated with paying for the repression infrastructure (i.e. the police and the military). When the autocratic government s revenues are smaller and/or it faces a tighter budget constraint, then the opportunity costs of paying police and military salaries and equipment get larger because the same scarce revenue is needed elsewhere to provide other goods and services. Consequently, factors which raise the revenues of autocratic governments and make their budget constraint less tight should ultimately serve to make repression less costly. As already discussed, the wealthy elite tend to own the land and natural resources, which can be exploited and sold to finance all the goods and services (both public and private) that the autocracy wants and needs to provide, including the repression infrastructure. For example, Middle Eastern governments use their oil revenues to pay for a well-equipped police force and relatively large military. But given the labor-scarcity of these national economies, the autocratic elites need to import foreign labor to exploit their natural resources and thus obtain their resource revenues. Consequently without immigration, their resource revenues would be reduced and their budget constraints would be tighter, thus making police and military goods and services more costly to provide. Conversely with immigration, the budget constraint in these labor-scarce national
12 economies gets looser, thereby reducing the opportunity costs associated with a large military and police force that can be used to repress the domestic population and the imported laborers alike. Indeed, this understanding leads directly to the second explanation (the first being rentier state theory) often offered to explain the political resource curse: repression theory (Ross 2001, 335-6). Repression theory simply posits that resource-rich governments use their resource revenues to fund a large military which can be used to repress their citizens. The argument advanced here not only builds from the basic logic of repression theory (that resource revenues decrease the opportunity costs associated with paying for repression infrastructure), but also makes the logic endogenous to immigration in labor-scarce national economies, which is a description fitting most (e.g. the GCC countries) but certainly not all (e.g. Iran) Middle Eastern oil-producing countries. With their labor imports used to exploit their energy resource endowments, the wealthy elites in these countries have sufficient revenues either to appease their population (per rentier state theory) or to pay for the infrastructure necessary to repress them (per repression theory). But it is also important to state clearly that the immigration/democratization argument advanced here, while it applies to resource-rich countries because they tend to be large importers of foreign labor as shown in the appendix, also applies to any resource-scarce country that imports foreign labor. Thus, if immigrants are entering the national economy to work on urban construction projects or to provide childcare and housecleaning, then they would still increase the permanent de facto power and the revolutionary threat of the population, thus making the elites promise to provide redistributive concessions more credible. Likewise, immigrants into resource-scarce countries would also reduce the opportunity costs associated with repression provided that their labor contribution to the national economy increases the revenue available to the autocratic government. And given that autocrats tend to control who enters (and exits) their national economy,
13 it seems unlikely that immigrants or guest workers would be allowed to enter if the condition above were not met. Testable Hypotheses Given its broad applicability, our primary hypothesis simply posits that immigration should be negatively associated with democratization. Following Mainwaring (1992, 298 emphasis in original), we define democratization as movement toward democracy which may or may not include a democratic transition, whereby a formerly autocratic regime becomes a democratic one. Stated in terms of our statistical unit of analysis, we hypothesize that country-years experiencing greater net immigration should also experience less democratization in future periods and thus should also be less likely to undergo a democratic transition. In this paper, we make no hypothesis concerning immigration and democratic consolidation (i.e. the probability of an already democratic regime remaining democratic in future periods). It is important to note that our hypothesized negative relationship between immigration and democratization runs contrary to the more obvious positive relationship between these two variables based on the understanding that immigrants likely prefer more democratic destinations in an effort to obtain greater political rights and civil liberties (Leblang, Fitzgerald, and Teets 2008). Indeed, this understanding about a positive relationship running from democracy to immigration suggests that endogeneity in the form of reverse causality should make it harder, not easier, to find statistical support for our hypothesis about a negative relationship running from immigration to democratization. Our argument also advances a secondary hypothesis concerning immigration as an alternative explanation for the resource curse. Inasmuch as the standard resource curse explanations (i.e. rentier state and repression theory) depend in large part on immigration for their operation in labor-scarce countries, we hypothesis that when holding constant the effect of net immigration (our
14 primary independent variable), the standard resource curse independent variables (e.g. oil/mineral rents per capita or energy production per capita) should lose power in a model of democratization. At the limit, these resource curse variables may even become statistically insignificant in a model of democratization when controlling for immigration. 2. The Evidence Statistical Model The statistical model used to test our first hypothesis is described in following equation: Democracy it = Democracy it-5 + Net Immigration Per Capita it-5 + Controls it-5 (1). There are several operational indicators for Democracy, our dependent variable, but we begin with the 21-point Polity2 scale, which has become the standard measure used in the resource curse literature (e.g. Ross 2001, Dunning 2008). We will also use the Freedom House scale and an updated version of Przeworski et al s (2000) dichotomous regime measure (Cheibub and Ghandi 2004) as robustness checks. Our first independent variable is the dependent variable lagged five years. This model specification means that we are effectively measuring the change in Democracy (i.e. democratization) over a five-year period, while controlling for the previous level of the dependent variable. 3 We also lag all other independent variables by five years as a first cut to help assure that our coefficients are measuring X s causal effect on Y, and not the reverse relationship (we will later present estimates using instrumental variable regression). Our primary independent variable is Net Immigration Per Capita (lagged five years) using data from the United Nations. Measured every five years (i.e. 1960, 1965, 1970, etc.), these data are 3 We obtain an identical set of results if we make our dependent variable the change in Democracy (i.e. Democracy) over the past five years and then control for its previous level. The initial five year lag is chosen because our primary independent variable Net Immigration Per Capita is measured only in five year intervals. We will also vary this time lag in models using interpolated values of Net Immigration Per Capita
15 country/year estimates of the number of immigrants minus the number of emigrants (i.e. net immigration). We make these data comparable across countries of different size by dividing net immigration by population and, thus, label this variable Net Immigration Per Capita. Our primary hypothesis predicts that this independent variable should take on a negative coefficient and be statistically distinguishable from zero. Equation (1), which will be estimated via Ordinary Least Squares with robust standard errors clustered on the country unit, also includes a host of control variables. Per modernization theory (Lipset 1959), we include a lagged value of the natural log of GDP Per Capita to capture the country/year s level of economic development. We also include the growth rate of GDP per capita (Growth) based on the understanding that governments presiding over more rapidly growing national economies may be able to delay democratization because their citizens would be more satisfied with the status quo (Przeworski et al 2000). As there is currently much debate about how income inequality may influence a country s political regime type in either a positive or negative direction, we use the capital share measure from Ortega and Rodriguez (2006) including interpolated observations from Houle (2009) to measure inter-class income inequality (Inequality). 4 Consistent with the nonlinear relationship between inequality and democracy advanced in Acemoglu and Robinson (2006), we also include a squared value of income inequality (Inequality 2 ). Our control variables also a measure of ethnolinguistic fractionalization using the data provided by Taylor and Hudson (1972). In the next section of the paper, we will explore how our argument fares statistically as an alternative explanation for the oil curse. Consequently, we need to control for another alternative explanation for why the Middle Eastern oil producers, in particular, may engage in less political liberalization. Many (e.g. Lipset 1994, Midlarsky 1998) have argued that 4 We make use of these interpolated observations from Houle (2009) because the capital shares measure, like most other measures of income inequality such as the Gini coefficient, is missing for the Middle Eastern oil countries, which are central to our theoretical argument.
16 the Islamic faith is hostile to democratic values and institutions so we add Islam, a variable which captures the percentage of Muslim individuals in the population using data from (Cheibub and Ghandi 2004). Finally, to pick up on systemic pressures associated with democratic diffusion and contagion, we include a full set of year fixed effects, following the specification offered by Ross (2001). It should be noted that our specification in this section of the paper does not include a control variable that directly measures the effect of natural resources on democratization. The resource curse variables namely, oil rents per capita and energy production per capita, entered separately - will be added to our model in the next section of the paper in order to assess how much of the standard resource curse result is washed away in the presence of our immigration variable. But in this section of the paper, our first goal is to demonstrate a strong negative relationship between immigration and democratization. Indeed, as shown in our appendix, these resource variables are strong positive predictors of Net Immigration Per Capita, a result that is consistent with the stylized fact introduced earlier that resource-rich countries also tend to be labor-scarce, which is why the autocratic elites in these national economies need to import foreign workers to exploit their resource endowments. Hence, there is much collinearity between resource endowments and labor imports, serving to weaken the coefficients of both independent variables when included together. 5 Which variable weakens less than the other and does a better job explaining democratization in the presence of the other will be explored in the last section of the paper. 5 The bivariate correlation between Net Immigration Per Capita and Oil Rents Per Capita, which will be introduced below, is 0.6. The bivariate correlation between Net Immigration Per Capita and Energy Production Per Capita (which will also be introduced below) is even stronger: 0.7.
17 Statistical Results The estimates of equation (1) are presented together in Table 1. Despite the small sample size (N=537) owing to the fact that our primary independent variable was measured only every five years, Net Immigration Per Capita nonetheless returns a negative coefficient (-9.5) that is statistically distinguishable from zero with more than 99 percent confidence. In terms of the substantive significance of this coefficient, its negative size means that a one standard deviation increase (0.05) in Net Immigration Per Capita would be associated with a 0.5 decrease in the Polity2 score over the next five years. There will be more discussion about the substantive significance of this independent variable below, suggesting that the effect reported in Table 1 may be substantially understated. Given that we have data on Net Immigration Per Capita measured every five years, it becomes relatively easy to interpolate linearly the internal values of this independent variable (e.g. 1961, 1962, 1963, and 1964 using the values from 1960 and 1065). So in model 1.2 of Table 1, we re-estimate equation (1) using a larger sample that includes these interpolated observations (N=2545). This additional estimate allows us to take fuller advantage of the information in the control variables that were coded every year, and it will also allow us to vary the time lag on Net Immigration Per Capita in other than five-year increments (as will be done later in the paper). The results show that Net Immigration Per Capita remains statistically significant with a negative coefficient (-11.9) of much the same size as the first estimate, suggesting that sample size has not biased the coefficient on our primary independent variable. We thus proceed using this larger statistical sample. In model 1.3, we perform the first robustness check on our dependent variable using a 13- point Freedom House scale (civil liberties plus political rights), which has been inverted so that larger values indicate more Democracy. These results also show a negative coefficient for Net Immigration Per Capita (-4.2), which is statistically different from zero with more than 95 percent confidence.
18 For a second alternative measure of Democracy, we use the dichotomous indicator originally created by Przeworski et al. (2000) and then updated by Cheibub and Ghandi (2004), which we inverted so that 1 indicates a democracy and 0 indicates an autocracy. Using this dichotomous measure, we can estimate a dynamic probit model, which allows us to separate the effects of our independent variables with regards to a democratic transition (i.e. an autocracy becoming a democracy) versus a democratic consolidation (i.e. a democracy remaining a democracy). 6 In this regard, our theoretical argument concerns immigration s effect on a democratic transition, and the dynamic probit results show a statistically significant negative effect for Net Immigration Per Capita (- 5.2) in terms of this particular democratization phenomenon. The results also show no statistically significant immigration effect with regards to democratic consolidation. Endogeneity? Despite the five-year lag on our independent variables, readers may still be concerned that the Net Immigration Per Capita coefficient is biased by endogeneity in the form of reverse causality. Indeed, it seems reasonable to think that democratization may have some causal effect on immigration; presumably international migrants would prefer ceteris paribus more democratic destination countries. In fact, if there is some positive reverse causality between immigration and democracy, then endogeneity tends to work against the negative immigration coefficients reported in Table 1, suggesting that our earlier set of negative estimates may actually be understated, or biased in a positive direction. This positive bias could also help explain why the Net Immigration Per Capita 6 For those who are unfamiliar with the dynamic probit specification, it interacts all of the substantive independent variables with the lagged dependent variable, which is coded as 0 for autocracies and 1 for democracies. Thus, the probability of an autocracy becoming a democracy, or engaging in a democratic transition, is given by the noninteracted version of the independent variable, or the constitutive term (e.g. Net Immigration Per Capita). The probability of democratic consolidation, or a democracy remaining a democracy, is given by the linear combination of the constitutive and interaction terms (e.g. Net Immigration Per Capita + Net Immigration Per Capita*Lagged Democracy).
19 coefficient, while statistically significant, did not exhibit a substantively larger negative effect with regards to democratization in Table 1. As a quick test to determine how reverse causality may be affecting the Net Immigration Per Capita coefficient, we can simply re-estimate our model while varying the time lag on our independent variables. To the extent that reverse causality runs in a positive direction and thus tends to weaken our immigration coefficient even when the variable is lagged five years, we should be able to observe that longer time lags tend to make its coefficient even larger in a negative direction. We should also be able to observe that shorter time lags tend to push the immigration coefficient in a positive direction. Figure 2: The Net Immigration Per Capita Coefficient with Various Time Lags Coefficient 0-5 -10-15 -20-25 -30-35 Years Lagged 1 2 3 4 5 6 7 8 9 10 Plotting the Net Immigration Per Capita coefficient versus its time lag in years, the results in Figure 2 demonstrate exactly this relationship using the Polity2 dependent variable. While Net Immigration Per Capita has a negative sign and is statistically distinguishable from zero with 95 percent confidence at any time lag between 1 and 10 years, its coefficient is only -2.3 with a one year lag.
20 The coefficient grows to -11.9 with a five year lag (as reported earlier in model 1.2) and almost doubles to -20.8 with a ten year lag. Using this last coefficient, the substantive effect of a one standard deviation increase in Net Immigration Per Capita would be a full one point decrease in the Polity2 score over the next ten years. Indeed, given that the lower bound of the 95 percent confidence interval reaches -32.6, the substantive effect of this independent variable could be as large as a 1.6 point decrease in the Polity2 score over the following ten years. We can put these larger negative coefficients for Net Immigration Per Capita on a more solid empirical footing by using an instrument for this primary independent variable. As shown in our appendix, it is not necessarily hard to find a strong instrument for Net Immigration Per Capita. But it is challenging to select an instrument that is plausibly excludable in the Democracy equation since one can always create a story for why some variable might be correlated with domestic regime type through some channel other than those associated with immigration. Thus, we present three different instrumental variable regressions; each uses a single different instrument for Net Immigration Per Capita, thus exactly identifying each model. We then defend each instrument s excludability, primarily using control variables, which should pick up on other causal channels through which the instrument might influence Democracy, the second stage dependent variable measured using the Polity2 indicator. These instrumental variable regressions are presented together in Table 2. Our first instrument is the Labor Force Participation Rate based on the understanding that national economies with a more fully utilized domestic labor stock will need to import foreign laborers to expand their production given the limits to simply increasing domestic labor productivity. This single instrument provides a strong F statistic (>10, following Stock, Wright and Yogo 2002) and passes the Kleibergen-Paap under-identification test. One might argue that this instrument poses excludability problems because democracies allow greater freedom for their citizens to work; thus these variables should be positively correlated even without considering immigration. But we
21 strongly discount this possibility since a reduced form Democracy equation shows Labor Force Participation Rate to have a statistically significant negative coefficient, which is hard to explain except through our immigration channel. And when we use this first instrument for Net Immigration Per Capita, it returns a statistically significant negative coefficient as shown in model 2.1. Our second instrument was chosen to pick up on foreign labor imports based on the demand for new construction and additional childcare. Domestic population growth creates the demand for both new housing and for additional household help, potentially leading to increased male and female immigration. To make sure that our population growth variable is not picking up on migration flows either in or out of the national economy (and is thus caused by immigration rather than being a cause for immigration), we subtract net immigration from the domestic population and then calculate the annual percentage change to create a new instrumental variable labeled Population Growth (Net of Migration). 7 One might object that domestic population growth also affects political regime type through the economic development channel per modernization theory: population growth is a rough measure of economic development with lesser developed countries experiencing greater population growth due to the lack of birth control availability. But our model directly controls for this other causal channel with GDP Per Capita (logged), the standard measure for economic development that has been used to test modernization theory. 8 This second instrument, which offers an even stronger F statistic than the first instrument, also yields a statistically significant negative coefficient for Net Immigration Per Capita in model 2.2. Our third instrument is Industrial Value Added Per Capita, based on the understanding that where national production is highly labor intensive (e.g., concentrated in industrial sectors), a 7 We find much the same results even if we do not subtract net immigration from population, so we are confident that our subtraction is not creating an artificial correlation between Population Growth (Net of Migration) and Net Immigration Per Capita. 8 The results presented in model 2.2 change very little if we add a squared measure of GDP Per Capita (logged) to pick up any non-linear effects associated with economic development.
22 country may need to import additional labor in order to increase its aggregate output. One potential excludability problem is that some of the resources associated with the political resource curse, especially oil and minerals, may be counted as industrial production (rather than as mining/ agricultural production). This means that this instrument may be negatively associated with Democracy through some channel other than immigration, especially since we have not yet demonstrated how much of the so-called resource curse can be explained away by immigration. To deal with this possibility, we add two more control variables, Oil Rents Per Capita and Mineral Rents Per Capita, to pick up on alternative causal channels associated with the resource curse. Even when controlling for oil and minerals rents, our primary independent variable returns a statistically significant negative coefficient using this third instrumental variable. In model 2.3, it is also interesting to note that neither of the resource rent variables has a statistically significant negative coefficient in the second stage Democracy equation when controlling for Net Immigration Per Capita. To the extent that either rent variable is statistically significant, Oil Rents Per Capita actually takes on a positive coefficient, which would suggest a resource blessing rather than a resource curse when holding constant the effect of immigration. We now turn to exploring this result in greater detail. 3. A Resource Curse? Our second hypothesis posited that resource variables should lose their negative explanatory power with regards to democracy when controlling for immigration based on the understanding that many resource-rich countries are also big importers of foreign labor. [For some statistical evidence on this point, we urge the reader to examine the models in Table 4 where Net Immigration Per Capita becomes the dependent variable.] Model 2.3 was consistent with this secondary hypothesis, but it is important to put this last result on a more solid empirical foundation. Would the resource curse
23 variables (Oil Rents Per Capita and Mineral Rents Per Capita) have even been statistically significant with a negative sign (indicating a resource curse, rather than a resource blessing) had we not controlled for immigration? As shown in model 3.1 in Table 3, the answer to this question is yes at least for Oil Rents Per Capital, which is negatively signed and statistically significant with greater than 99 percent confidence. The sample in this model was deliberately restricted to include only those country/year observations for which we have Net Immigration Per Capita data, including interpolated observations (N=2473). This sample restriction means that when we add the immigration variable back into the equation, any changes to the resource rent coefficients will be due to the inclusion of this variable and not due to changes in the statistical sample. In model 3.2, Net Immigration Per Capita is added back into the equation, which has a large effect on Oil Rents Per Capital. The coefficient on this resource curse variable is more than cut in half (from -0.041 in model 3.1 to -0.019 in model 3.2) and it loses statistical significance at any conventional level in the latter model. This result would suggest that a large part of the observed oil curse, in particular, can be explained by labor imports, consistent with the results in the appendix showing that oil-rich countries tend to be large imports of foreign labor. The results in the appendix also show that mineral-rich countries tend not to be such large importers of foreign labor, likely because many of the mineral-rich countries, especially those in sub-saharan Africa, are also relatively labor-abundant. This understanding could help explain why Mineral Rents Per Capita did not exhibit a resource curse result even in model 3.1: mineral rents are not strongly associated with labor inflows: therefore there is not much of a negative democracy effect to explain away by controlling for immigration (as appears to be the case for oil rents). In models 3.3 and 3.4, we perform a final robustness check using another resource curse variable: Energy Production Per Capita. In model 3.3, we first show that this variable - when not