In this article, we examine empirically both the direct and indirect links between ethnic fragmentation and economic growth. We find that both ethnic fractionalization and polarization are negatively associated with growth if considered in isolation; an effect that is though primarily attributed to their link to other growth-related activities (i.e., investment, conflict, control of corruption, fertility). We study the corresponding transmission channels and calculate their relative importance in explaining a development curse based on ethnic diversity. For both measures of ethnic fragmentation, we find the corruption channel to be the most important one. (JEL C21, Oil, Z13)

I. INTRODUCTION

It is a common assumption in the development economics literature that ethnic fragmentation (fractionalization, polarization) tends to frustrate economic growth. Several empirical and theoretical studies have recently linked ethnic fragmentation to poor economic performance and its determinants (e.g., see Alesina, Baqir, and Easterly 1999; Alesina et al. 2003; Alesina and La Ferrara 2005; Baggio and Papyrakis 2010; Easterly and Levine 1997; Esteban and Ray 2011; Habyarimana et al. 2007; Hodler 2006; Montalvo and Reynal-Querol 2005a, 2005b). A common argument in these studies is that ethnic fragmentation reduces cooperative behavior in the society and raises transaction costs with individual ethnic groups often favoring short-term opportunism over long-term planning (Collier 2001).

In the literature, several indirect transmission channels have been identified and investigated through which ethnic fragmentation leads to lower economic growth. Ethnic fragmentation often leads to reduced infrastructure investment

as a result of diverse preferences over the range of capital goods to be delivered (see Alesina, Baqir, and Easterly 1999 in particular) as well as a riskier environment for investors (often due to uncertainty over property rights protection, see Busse and Hefeker 2007; Svensson 1998). Ethnically heterogeneous societies tend to suffer disproportionately more from corruption as a result of ethnic favoritism, with obvious repercussions for the efficient allocation of talent and public revenues (see LaPorta et al. 1999; Mauro 1995; Pellegrini and Gerlagh 2008). Another stream of the literature emphasizes that ethnic fragmentation (primarily ethnic polarization) leads to conflictual behavior among ethnic groups and consequent loss of human life, which naturally obstructs the productive capacity of affected economies (Esteban and Ray 1999; Montalvo and Reynal-Querol 2005b, 2010). Last, ethnically heterogeneous societies are often characterized by higher fertility rates (with the latter often linked to lower capital intensity and hence economic growth, see Becker and Tomes 1976; Brander and Dowrick 1994; Bloom et al. 2009; Lee, Mason, and Miller 2001); (strategic) political competition at the group level encourages pronatalism, where relative group size is associated with political leverage and division of benefits (see Janus 2010, as well as Anson and Meir 1996; Basu 1997; Camps and Engerman 2008).

In this article, we contribute to this strand of the literature by jointly investigating the transmission mechanisms identified in the literature through which ethnic heterogeneity may hamper economic growth and quantifying their relative importance. To our knowledge, this is the first empirical paper that attempts to quantify the relative contribution of these indirect mechanisms to the overall negative association between ethnic diversity and growth. Our analysis follows the methodology set out by Mo (2000, 2001) and Papyrakis and Gerlagh (2004, 2007) who explore the intermediate channels through which inequality, corruption, and resource dependence link to growth. Through cross-country growth regressions we investigate the empirical relationships between ethnic diversity and investment, control of corruption, conflict and fertility, and estimate the share of each transmission channel in accounting for the overall negative association between ethnic heterogeneity and growth. We find that, on average, ethnic heterogeneity is negatively linked to these growth-promoting variables, which explains the largest part of a "development curse" based on ethnic diversity.

While many of the aforementioned studies linking ethnicity and economic growth have relied exclusively on ethnic fractionalization as a measure of ethnic fragmentation, many recent papers have placed more emphasis on ethnic polarization as a potential cause of poor economic performance. A number of prominent papers by Montalvo and Reynal-Querol (2005a, 2005b, 2008) for instance claim that ethnic heterogeneity, measured by polarization rather than fractionalization, can be a stronger determinant of conflictual behavior in the economy. For this reason, we make use of both ethnic fractionalization as well as polarization indices to estimate the effects of ethnic heterogeneity on socioeconomic performance.

It is also worth noting that some recent studies point to nonlinear relationships between fragmentation and several economic development outcomes. For example, Ashraf and Galor (2011) find a hump-shaped relationship between their measure of genetic diversity (proxied by migratory distance to Eastern Africa) and current income per capita levels (that might suggest that moderate ethnic fragmentation can allow for an expansion of an economy's production possibilities frontier through diversity-driven knowledge accumulation). Cerqueti, Coppier, and Piga (2012) also find an empirical inverse-U relationship between growth and fractionalization (although for the very short run). They also discuss a theoretical model, where corruption can be high for homogenous and highly fragmented societies, but low for moderate levels of ethnic diversity (i.e., moderate fragmentation can play a positive role in controlling corruption by increasing the level of control across different ethnic groups). Dincer (2008), finds that corruption is at its highest level for moderate levels of fragmentation when he examines panel data for U.S. states. Tangeras and Langerlof (2009) also present a game theoretical model, where the risk of civil war is comparatively high for intermediate levels of ethnic diversity (given the larger amount of resources needed to be invested in conflict per group in the presence of a large number of ethnic groups). (1)

The next section is devoted to the empirical analysis on ethnic fractionalization/polarization and economic growth. We find that ethnic fractionalization, in particular, is negatively associated with economic growth, although the correlation substantially decreases in magnitude once the intermediate transmission variables are accounted for. Section III studies empirically these transmission channels and compares their relative weight in the overall negative link between ethnic diversity and economic growth. There are significant policy implications. In the case of ethnic fractionalization, for instance, we find that the negative association with investment and control of corruption explains more than half of the total negative correlation between ethnic heterogeneity and economic growth. Section IV provides some reflection on the exogeneity of ethnic diversity measures (and hence moves from discussing partial correlations toward discussing causal mechanisms). Our analysis is, hence, in the spirit of King and Fevine (1993) who focus on the links between economic growth and another growth factor (i.e., financial development), examine the association of the latter with other growth determinants (capital accumulation, efficiency), and finally comment on the ability of financial depth to act as a predictor of growth. Section V summarizes our main results and offers concluding remarks.

II. ETHNIC FRAGMENTATION AND ECONOMIC GROWTH

To identify the links between growth and ethnic fragmentation (fractionalization/polarization) we estimate cross-country growth regressions in the tradition of Kormendi and Meguire (1985), Grier and Tullock (1989), Barro (1991), Sachs and Warner (2001), and Akqomak and ter Weel (2009). We include initial income per capita in our empirical analysis to check for the conditional convergence hypothesis that predicts higher growth in response to lower starting income per capita, keeping the other explanatory variables constant. Thus, per capita income growth from period [t.sub.0] = 1967 to [t.sub.T] = 2007, denoted by [G.sup.i] = (1/T) In([Y.sup.i.sub.T]/[T.sup.i.sub.0]), depends on initial per capita income [Y.sup.i.sub.0]), on a measure of ethnic heterogeneity [E.sup.i]--i.e., ethnic fractionalization or ethnic polarization--and a vector of other explanatory variables [Z.sup.i]:

(1) [G.sup.i] = [[alpha].sub.0] + [[alpha].sub.1]ln([Y.sup.i.sub.0]) + [[alpha].sub.2][E.sup.i] + [[alpha].sub.3][E.sup.i] + [[epsilon].sup.i],

where i corresponds to each country in the sample. The coefficients [[alpha].sub.1], [[alpha].sub.2], and [[alpha].sub.3] measure the expected change in G (i.e., in growth) if the corresponding explanatory variable (i.e., In([Y.sub.0]), E, or Z) changes by one unit, assuming that all other regressors remain constant (i.e., the so-called ceteris paribus condition holds). For example, [[alpha].sub.2] measures the expected differences in growth between a country that is ethnically homogenous and one that is either highly fractionalized or polarized (i.e., with either the ethnic fractionalization or polarization index taking values close to 1), assuming that values for all other explanatory variables are held constant. The coefficient do corresponds to the constant term (i.e., the expected rate of growth if all explanatory variables are equal to 0). The error term e is the unobserved residual component of growth (i.e., the difference between the observed growth values and the approximated values given by the estimated a coefficients and the actual values of the corresponding explanatory variables), which is assumed to be normally distributed with zero mean and constant variance and independently distributed of the other regressors in the model.

A. Data

Ethnic Heterogeneity (E). Much of the literature on ethnic fragmentation uses fractionalization and polarization indices of the following form: fractionalization = 1 - [[summation].sup.N.sub.i=1][[pi].sup.2.sub.i] and polarization = 1 - [[summation].sup.N.sub.i=1] [((0-5 - [[pi].sub.i])/0.5).sup.2][[pi].sub.i];, respectively, where n, stands for the proportion of the total population belonging to the i-th ethnic group and N stands for the number of groups. The ethnic fractionalization index captures the probability of two randomly chosen individuals from the general population belonging to different ethnic groups, while the polarization index places emphasis on the relative size of these competing groups in the society. (2) For both indices, a value of zero corresponds to a perfectly homogenous society. The fractionalization index approaches unity as the number of different ethnic groups in the economy increases, while the polarization index takes the value of one in the case of a bipolar distribution of two ethnic groups of equal size. In other words, the difference between the fractionalization and polarization index is that the latter weighs the probabilities of two individuals belonging to different ethnic groups by the relative size of the groups (see Montalvo and Reynal-Querol 2005b).

Data on ethnic fractionalization and polarization are provided by Montalvo and Reynal-Querol (2005a, 2005b). The ethnic fractionalization/polarization indices are constructed using data (for a large cross-section of countries in 1970, 1975, and 1980) from the World Christian Encyclopedia, which combines information on race, skin color, ethnolinguistic families/dialects, and culture (for more detailed information, see the World Christian Encyclopedia 1982, pp. 107-15)--Montalvo and ReynalQuerol (2005a, 2005b, 2008) construct their ethnic fragmentation indices from these data taking into account the most important ethnic divisions (i.e., 71 ethnolinguistic families). The indices by Montalvo and Reynal-Querol (2005a, 2005b) are some of the most commonly used proxies of ethnic fragmentation in the economic literature (e.g., see Bhavnani and Miodownik 2009; Krieger and Meierrieks 2010; van der Ploeg and Poelhekke 2009) and largely correlate with other indices that use slightly different levels of disaggregation among ethnic groups (e.g., the ones by Alesina et al. 2003).

Other Growth Variables (Z). In growth specification (1), we include a set of variables commonly used in the literature as economic growth regressors. We always control for initial income per capita levels in 1967 to check for the conditional convergence hypothesis. Data on purchasing power parity (PPP) adjusted income levels are provided by the Penn World Tables (Heston, Summers, and Aten 2010) and have been extensively used in the empirical growth literature (e.g., see Alesina et al. 2003; Bjqrnskov 2012; Papyrakis and Gerlagh 2004). We further include in our specification a number of other growth-related variables that are found in the empirical literature to be both significantly correlated to growth, as well as associated with ethnic fragmentation. These include the share of (public and private) investment in GDP and the fertility rate at the beginning of the period (data provided by Heston, Summers, and Aten 2010 and World Bank 2010, respectively). The investment variable, capturing changes in physical capital (production plants, equipment, nonresidential construction), is one of the most commonly used and robust variables in empirical growth analysis (Aisen and Veiga 2013; Bjornskov 2012; Mankiw, Romer, and Weil 1992). The fertility variable relies on data collected by the United Nations Population Division based on input from civil registration systems and census/survey estimates (see Bloom et al. 2009 and Klasen and Lamanna 2009 on the use of the variable in growth econometrics). Furthermore, we incorporate into the analysis two additional variables that appear prominently in the literature linking ethnicity with economic growth--i.e., a proxy of conflict and a control of corruption index (data provided by the World Bank 2010 and Kaufmann, Kraay, and Mastruzzi 2009, respectively). The conflict variable is proxied by the number of deaths attributed to conflict as a share of total population based on data collated by the Uppsala Conflict Data Program (one of the most reliable data collection projects on organized violence; see Gates et al. 2012 and Esteban, Mayoral, and Ray 2012 for recent empirical studies that link the variable with economic growth and ethnicity). The control of corruption index is based on data summarizing the views by a large number of (enterprise, citizen, and expert survey) respondents regarding the extent of bribery in the economy and the level of protection offered by anti-corruption and accounting institutions (and is part of the worldwide governance indicators (WGI) project funded by the World Bank). The index is known to have one of the largest country coverages among other similar corruption indices available (see Svensson 2005) and several studies have linked it to economic growth and ethnic fragmentation (e.g., see Aidt, Dutta, and Sena 2008 and Shen and Williamson 2005). A matrix reporting all pairwise correlations between the key variables in our analysis is presented in Table Al of the Appendix. (3)

B. Growth Analysis

We now estimate growth Equation (1) using ordinary least squares, gradually increasing the set of explanatory variables. As a starting point we regress income growth only on initial income per capita in 1967 (In [Y.sub.67]). As a second step, we include the extent of ethnic fractionalization as a regressor in our empirical specification. The results are presented in columns (1) and (2) of Table 1 and indicate a highly significant and negative relationship between economic growth and ethnic fractionalization. The correlation between ethnic fractionalization and economic growth points to a difference in annual income growth of approximately 1.91% between an ethnically homogenous country (such as Denmark) and a highly fractionalized nation (such as Tanzania, index value close to 1).

We now turn to the possible transmission channels linking ethnic fractionalization with economic growth. We expect that the largest part of a negative correlation between ethnic fractionalization and economic growth is likely to be attributed to a negative statistical relationship observed between ethnic diversity and other growth-promoting factors captured by the vector Z1 (rather than a direct impact on growth). In other words, a negative statistical relationship between [E.sup.i] and [Z.sup.i] is likely to account for the largest part of the negative correlation between [E.sup.i] and [G.sup.i] in the second regression of Table 1. When the vector [Z.sup.i] is sufficiently rich to capture most of the indirect links between fractionalization and growth, we expect that its inclusion in the empirical analysis would substantially decrease the coefficient of fractionalization on growth. As our next step, we thus extend the vector [Z.sup.i], by progressively adding variables commonly used to explain growth, such as investment, fertility, conflict, and control of corruption, and we correspondingly examine the magnitude and significance of the fractionalization coefficient [[alpha].sub.2] (Papyrakis and Gerlagh 2004, 2007 use a similar argument in their empirical analysis on the direct and indirect links of resource abundance).

In column (3) of Table 1, we include the share of investment (public and private) in GDP as an additional regressor. The variable refers to the beginning of the period in order to avoid endogeneity problems. The coefficient of the investment variable accords with intuition (e.g., see Adams 2009; Crafts 2009; Levine and Renelt 1992), suggesting a significant positive link between capital accumulation and economic growth. A 10% increase in the share of investment in GDP corresponds to a higher rate of economic growth by 0.65%. We now also observe a negative sign for the coefficient [[alpha].sub.1] (that is now statistically significant), providing hence support to the conditional convergence hypothesis. More importantly, we notice a substantial reduction in the magnitude and statistical significance of the fractionalization coefficient (suggesting that part of the negative association between fractionalization and growth can be attributed to the investment channel).

In column entry (4) we include the fertility rate in 1967 as an additional explanatory variable, expecting that higher fertility rates correspond to lower growth rates as a result of reduced capital deepening (Becker and Tomes 1976; Bloom et al. 2009; Lee, Mason, and Miller 2001; Madsen, Ang, and Banerjee 2010). In subsequent column entry (5), we include our measure of conflict, proxied by the conflict-related deaths per population in 1967. Next, we include the proxy for the control of corruption, capturing "perceptions of the extent to which public power is exercised for private gain." The index corresponds to the control of corruption in 1996, the first year for which data are available (Mo 2001 argues that endogeneity is less likely to be an issue of concern for the corruption variable since institutions tend to evolve slowly). (4) We expect conflict to be a significant deterrent of economic growth (and control of corruption to be a strong stimulant)--conflict obstructs the productive capacity of the economy and destroys all types of productive capital (Esteban and Ray 1999; Montalvo and ReynalQuerol 2005b, 2010; Olsson 2007), while corruption leads to an inefficient allocation of talent and public revenues (see Fisman and Svensson 2007; LaPorta et al. 1999; Mauro 1995; Pellegrini and Gerlagh 2008). We indeed observe that fertility, conflict, and control of corruption have the expected signs, as suggested by the literature. The sequence of regressions in Table 1 suggests that the gradual inclusion of these explanatory variables steadily reduces both the magnitude and statistical significance of the coefficient of fractionalization (while the opposite holds for the coefficient of initial income, providing hence strong support for the conditional convergence hypothesis). This is an important finding; in column (6) the coefficient of fractionalization has been reduced approximately by a factor of 3.5 and has become totally insignificant. This suggests that the largest part of a development curse based on fractionalization is likely to be explained by the indirect association between ethnic diversity and other growth-related activities (i.e., investment, fertility, conflict and control of corruption). Once one controls for these indirect links (i.e., the transmission channels), fractionalization is only weakly correlated with growth. (5)

It is of interest to explore whether there is a differentiated link between different types of ethnic fragmentation (fractionalization vs. polarization) and economic growth, given the recent increased interest in ethnic polarization as a determinant of adverse economic outcomes (Baggio and Papyrakis 2010; Esteban and Ray 2008; Montalvo and Reynal-Querol 2005a, 2005b, 2008). For this reason, we reestimate all regressions of Table 1, where we now include the extent of ethnic polarization in all specifications in place of the ethnic fractionalization measure. Data on ethnic polarization are provided by Montalvo and Reynal-Querol (2005a, 2005b). (6) We find a much weaker relationship between economic growth and ethnic polarization (column (7)), although it is still of interest to explore whether the observed correlation may be largely (and indirectly) attributed to any links of polarization to the growth-related activities captured by the vector Z (which is the focus of our next section). Interestingly enough, the sequence of regressions in Table 2 reveals a similar pattern to the one identified in our previous results. The gradual inclusion of explanatory variables steadily reduces the magnitude of the coefficient of polarization, suggesting hence that any (even weak) negative correlation between polarization and economic growth is likely to be attributed to the intermediate channels, i.e., the association between polarization and other growth-related activities. As a matter of fact, in column entries (10) and (11), the coefficient of polarization turns positive (although still of low statistical significance). Although this may look surprising at first sight, it simply suggests that polarization is not harmful to growth per se, once one accounts for any indirect links between the former and the growth-relevant variables of vector [Z.sup.i] (i.e., investment, fertility, conflict, and control of corruption). (7) The signs of all other growth-related variables (investment, fertility, conflict, control of corruption as well as ln [Y.sub.67]) accord with intuition.

III. TRANSMISSION CHANNELS

In this section, we turn our attention to the transmission channels and estimate the links between ethnic fractionalization (and polarization) and investment, fertility, conflict, and control of corruption (and the indirect links, thereof, to economic growth). We then calculate the relative importance of each of these channels in accounting for the poor economic performance of ethnically fractionalized (or polarized) countries.

Before proceeding with the empirical investigation, we discuss in brief the variables that entered our growth regressions and their plausibility to act as a transmission channel. The first transmission channel we explore points to a negative link between ethnic fragmentation and investment. Political economy theories and empirical evidence suggest that ethnic heterogeneity often leads to reduced infrastructure investment as a result of divergent preferences over the range of capital goods to be delivered (Alesina, Baqir, and Easterly 1999; Miguel and Gugerty 2005). Governments in ethnically fragmented societies often tend to favor the interests of particular ethnic groups; given that there is a high risk of being overturned by an opposition favoring alternative ethnic groups, governments often become short-sighted devoting a larger share of their budget to immediate consumption rather than investment (Azzimonti 2011). Another stream of the literature posits that ethnic diversity often creates a riskier environment for (both domestic and foreign) investors. Ethnic fragmentation is often associated with political instability, incoherent public policies, and uncertainty over property rights protection (Busse and Hefeker 2007; Levine 2005; Svensson 1998).

As a second transmission channel we consider the role of ethnic fragmentation in explaining observed cross-country variation in fertility rates. The literature suggests that fertility rates are often the outcome of a (strategic) political competition at the group level. Such a strategic interaction among ethnic groups often encourages pronatalism, since relative group size is associated with political leverage and division of benefits (see Janus 2010, as well as Anson and Meir 1996; Basu 1997; Camps and Engerman 2008). Ethnic group leaders have hence an incentive to discourage fertility-reducing policies (e.g., family planning); every household with a high fertility rate in effect contributes to a public good for the community (ethnic group) by sustaining or increasing its relative size. It is no surprise that during periods of ethnic tensions and heightened nationalism ethnic groups and their leaders deliberately invest in fertility promotion (Horowitz 2000).

The third transmission channel we consider focuses on the link between ethnic heterogeneity and conflict. Several papers note that ethnic fragmentation (both ethnic fractionalization and polarization, but primarily the latter) leads to ethnic conflict and consequent loss of human life (e.g., see Esteban and Ray 1999; Montalvo and Reynal-Querol 2005b, 2010). Ethnic conflict among groups naturally obstructs the productive capacity of affected economies (e.g., by destroying productive capital, creating mistrust among different ethnic groups and restricting trade to individuals of the same community). Collier and Hoeffler (1998) claim that such conflictual behavior intensifies once a large ethnic group identifies with the government, while another one (of similar size) with the interests of a rebel group. (8)

The last transmission channel we explore considers the relationship between ethnic fragmentation and corruption. Ethnically heterogeneous societies tend to suffer disproportionately more from corruption as a result of ethnic favoritism (see LaPorta et al. 1999; Mauro 1995; Pellegrini and Gerlagh 2008). Bureaucrats tend to favor members of their own group with obvious repercussions for the efficient allocation of talent and public revenues. The positive impact of ethnic favoritism on corruption may be further amplified by the fact that ethnically fragmented societies are characterized by an under-provision of public goods (i.e., ethnic diversity tends to reduce collective action that is often necessary for investment in public goods). As a result, dependency on special bonds and ethnic ties becomes increasingly important in terms of securing access to essential services (Pellegrini and Gerlagh 2008). (9)

Now we turn to the data and estimate the statistical association of each of these variables with ethnic fractionalization [E.sup.i] and initial income ln([Y.sup.i.sub.0]):

(2) [Z.sup.i] = [[beta].sub.0] + [[beta].sub.1] ln([Y.sup.i.sub.0]) + [[beta].sub.2][E.sup.i] + [[mu].sup.i],

where [E.sup.i], ln([Y.sup.i.sub.0]), [[beta].sub.0], [[beta].sub.1], [[beta].sub.2], and [[mu].sup.i]) are specified for investment, fertility, conflict, and control of corruption. The coefficients [[beta].sub.1] and [[beta].sub.2] measure the expected change in Z (i.e., in investment, fertility, conflict, and control of corruption) if the corresponding explanatory variable (i.e., ln ([Y.sub.0]) or E) changes by one unit. The coefficient [[beta].sub.0] corresponds to the constant term. The error term [mu] is the unobserved residual component of each transmission variable Z (i.e., the difference between the observed values of Z and the approximated values given by the estimated [beta] coefficients and the actual values of initial income and ethnic fractionalization), which is assumed to be normally distributed with zero mean and constant variance and independently distributed of the other regressors in the model. To avoid having different sample sizes due to data availability, we confine the analysis on the transmission channels to the 102 countries used in the last regression of Table 1 (for which data are available for all variables). (10) A negative statistical relationship between ethnic diversity and the growth-promoting variables captured by the vector Z (in combination with a corresponding positive statistical relationship between the growth-promoting variables and subsequent economic growth, as found in Tables 1 and 2) is likely to account for a large part of the negative correlation between ethnic diversity and growth. Results of these transmission channels (i.e., the statistical associations between ethnic fractionalization and the growth-related variables captured by Z) are presented in Table 3 and indicate that ethnic fractionalization relates to lower investment and control of corruption and higher fertility and conflict, as suggested by the literature. All coefficients of fractionalization (apart from the conflict channel (11)) are significant at the 1% level and their sign provides support to a development curse based on ethnic diversity. The correlation between ethnic fractionalization and investment (column (12)) points to a difference in investment rates of approximately 10% between an ethnically homogenous country (such as Denmark) and a highly fractionalized nation (such as Tanzania, index value close to 1). Similarly, the estimated correlations between ethnic fractionalization and the rest of the growth-related variables, captured by the vector Z, suggest that in highly fractionalized nations (compared to ethnically homogenous ones): (a) a woman bears, on average, one additional child during her lifetime (column (13)) and (b) corruption is significantly higher (column (15))--the "control of corruption" index is lower by 0.8 units, which points to a correlation of substantial magnitude given that the variable has a mean of 0.10 and a standard deviation of 1.12. These partial correlations between ethnic fractionalization and the growth-related variables captured by Z are derived after controlling for the level of economic development (with higher levels of income per capita pointing to increased investment, lower corruption, and lower fertility rates).

Since ethnic fragmentation explains part of the variation in investment and the other growth-related variables of vector Z, by substitution of Equation (2) into (1) we can calculate the magnitude of the overall (direct and indirect) statistical association between fractionalization and growth:

(3) [G.sup.i] = ([[alpha].sub.0] + [[alpha].sub.3][[beta].sub.0]) + ([[alpha].sub.1] + [[alpha].sub.3][[beta].sub.1]) ln([Y.sup.i.sub.0])

+ ([[alpha].sub.2] + [[alpha].sub.e][[beta].sub.2])[E.sup.i] + [[alpha].sub.3][[mu].sup.i] + [[epsilon].sup.i],

where [[alpha].sub.2][E.sup.i] and [[alpha].sub.3][[beta].sub.2][E.sup.i] capture the direct and indirect links between fractionalization and growth, respectively. In other words, Equation (3) now only includes the component of the vector of growth-related variables Z that is not associated with ethnic fractionalization or initial income (i.e., the [mu] residuals). Table 4 presents the estimated values for all coefficients of Equation (3) (for the case of ethnic fractionalization). The coefficient of fractionalization now includes both direct and indirect links (and points to a 2.04% difference in growth rates between an ethnically homogenous economy and a fully fractionalized one).

Our next step is to quantify the relative importance of each transmission channel in accounting for the overall negative link between ethnic fractionalization and economic growth. The direct association is given by [[alpha].sub.2], while [[alpha].sub.3][[beta].sub.2] captures the channel-specific link for each of the intermediate transmissions mechanisms considered (see Equation (3)). Results are presented in Table 5. We see that the largest part (close to three quarters) of a fractionalization-based development curse can be attributed to the indirect transmission channels (i.e., investment, fertility, conflict, control of corruption). Control of corruption appears to be the most important transmission mechanism, accounting for 30% of the overall negative association between fractionalization and growth (which jointly with the investment channel, the second most important one in terms of relative contribution, explain more than 50% of the overall correlation).

We turn our attention once more to ethnic polarization, in search of differentiated links between different types of ethnic diversity and growth-relevant variables and transmission channels. We reestimate Equation (2) having ethnic polarization as our measure of ethnic fragmentation in place of ethnic fractionalization; results are presented in Table 6. Results accord with our earlier findings of Table 3; that is, ethnically polarized societies tend to suffer from reduced investment, inadequate control of corruption, and higher fertility. (12) The correlation between ethnic polarization and investment (column (17)) points to a difference in investment rates of approximately 7% between an ethnically homogenous country (such as Denmark) and a highly polarized nation (such as Jordan, index value close to 1). Similarly, the estimated correlations between ethnic fractionalization and the rest of the growth-related variables, captured by the vector Z, suggest that in highly polarized nations (compared to ethnically homogenous ones): (a) women bear, on average, a larger number of children during their lifetime (a difference in the fertility rate close to 1.4; see column (18)) and (b) corruption is significantly higher (column (20))--the "control of corruption" index is lower by 0.7 units. (13) These partial correlations between ethnic polarization and the growth-related variables captured by Z are derived after controlling for the level of economic development. Comparing Tables 3 and 6, one can observe that a highly fractionalized nation is likely to experience lower levels of investment and higher corruption rates compared to a highly polarized nation, other things equal (while the opposite holds for fertility rates).

We then substitute Equation (2) into (1) (where now ethnic polarization is the independent variable) and calculate the magnitude of the overall (direct and indirect) association between ethnic polarization and growth. We reestimate hence Equation (3) for the case of ethnic polarization; results are presented in Table 7. The coefficient of polarization (that now captures both direct and indirect links with growth) points to a 0.94% difference in growth rates between an ethnically homogenous economy and a fully polarized one--the overall correlation is half the magnitude of the corresponding one in the case of fractionalization in Table 4.

Last, we quantify the relative importance of each transmission channel in accounting for the overall negative association between ethnic polarization and economic growth. Results are presented in Table 8. Consistent to what we observed earlier (Table 2), polarization is likely to have a positive (albeit weak) direct link to growth, once we control for the indirect associations with other growth-related variables (i.e., the transmission channels); the size of the positive direct link is approximately equal to 51% of the overall negative correlation between polarization and growth. Control of corruption still appears to be the dominant transmission mechanism, through which polarization is linked to economic growth (the size of the correlation is approximately equal to 64% of the overall negative correlation with growth). Fertility is now the second most important channel, followed by investment. When we exclude the direct link between polarization and growth (i.e., [[alpha].sub.2]) and focus only on the indirect links (i.e., [[alpha].sub.3][[beta].sub.2]), we can see that the corruption and fertility channels account for 42% (0.60/1.42) and 37% (0.53/1.42) of the indirect negative association between polarization and growth.

IV. SOME REFLECTION ON THE EXOGENEITY OF ETHNIC FRAGMENTATION

Cross-country empirical studies of growth typically capture partial correlations between growth and other variables, often with ambiguity about the direction of causality (and for that reason empirical results often need to be interpreted with some caution). In the case of ethnic fragmentation, it is more likely to be the case that ethnic diversity mainly affects economic growth, rather than the other way round (although to a certain extent causality can run both ways). Ethnic fragmentation remains rather stable over time--at least for the majority of the countries (see also Montalvo and Reynal-Querol 2005a, 2005b for a discussion). In almost all empirical analyses, the indices of fractionalization and polarization enter the regressions as time invariant variables (it is customary in the economics literature to treat ethnic diversity as a non-time-varying variable due to limited data availability; e.g., see Arezki and Bruckner 2012; Montalvo and Reynal-Querol 2005a, 2005b, 2010). This is also the reason why indices of ethnic fragmentation have often been used as exogenous instruments in regression analysis (e.g., for the level of corruption or political stability; see Cole 2007, Mauro 1995, and Reynal-Querol 2005)--although, as Brock and Durlauf (2001) correctly point out, the validity of any time-invariant instrument depends on it being uncorrelated with other omitted growth factors (captured by the residuals).

Ideally one would wish to investigate whether predetermined ethnic fragmentation is strongly linked to subsequent economic growth (in the vein of King and Levine 1993, who establish that predetermined financial depth is a good predictor of subsequent growth). To the best of our knowledge, the only dataset that provides time-varying data on ethnic fractionalization (and, hence, allows for some form of similar experimentation similar to the one used by King and Levine 1993) is the one by Campos and Kuzeyev (2007) and Campos, Saleh, and Kuzeyev (2011). The dataset, though, constructs an ethnic fractionalization index based on census data only for 26 former Communist economies and for four periods (1989, 1993, 1999, 2003). This provides an interesting natural experiment, since the ethnic composition of the population is likely to have changed over time in these former Communist economies, given that with the collapse of Communism international labor mobility restrictions (to and from the countries) were relaxed, while many members of their Russian ethnic minorities migrated to Russia (Campos, Saleh, and Kuzeyev 2011). In Table 9, we see that even for this small sample of ex-Communist economies, predetermined ethnic fractionalization (for the four periods for which data are available; i.e., 1989, 1993, 1999, 2003) is strongly linked with subsequent economic growth, control of corruption and fertility rates (averaged over the three subsequent years). While results are not comparable to the ones from the earlier analysis (both because of the different sample, as well as the larger volatility in growth and growth-related factors that is more commonly observed in the shorter term), our findings provide some support to the hypothesis that ethnic diversity acts as a predictor of subsequent economic growth and other growth-related factors (fertility, control of corruption), at least for the 26 former Communist economies for which time series data on ethnic composition are available. For the same variables, we also run several Granger-causality tests (using either one or two lags for the respective dependent variable and ethnic fractionalization)--the hypothesis that the coefficients of lagged ethnic fractionalization are equal to zero (and hence that ethnic fragmentation does not Granger-cause the dependent variable) is rejected at the 1% level of significance in the case of economic growth, fertility, and control of corruption (as dependent variables).

V. CONCLUSIONS

There has been an increasing interest in the links between ethnic fractionalization (and polarization more recently) and economic growth (and several of its intermediate determinants; e.g., investment, conflict, control of corruption, fertility). The literature suggests that ethnically fragmented societies suffer from conflict, high fertility rates, and reduced investment and control of corruption (i.e., channels through which long-term growth becomes frustrated). In this article, we contribute to this strand of the literature by jointly investigating the transmission mechanisms identified in the literature through which ethnic fragmentation hampers economic growth and quantifying their relative importance. To our knowledge, this is the first empirical attempt to quantify the relative contribution of each of these mechanisms to the overall negative association between ethnic fragmentation and growth. For both measures of ethnic fragmentation, we find corruption to be the most important mechanism. Investment plays the second most important role in explaining a development-curse based on ethnic fractionalization (and fertility in the case of ethnic polarization).

We find that both ethnic fractionalization and polarization are negatively associated with growth if considered in isolation; an effect that is though primarily attributed to their links with other growth-related activities. The findings have significant policy implications. Although ethnic fractionalization (and polarization to a certain extent) may impact negatively on economic growth, this is by no means an iron law. A better understanding of the indirect mechanisms is essential for adopting policy measures that can prevent the negative association between ethnic heterogeneity and economic growth. Fractionalized economies that seriously tackle corruption (as well as adopt measures that promote investment and lower fertility) are more likely to remain unharmed by the "fractionalization curse." In the case of ethnic polarization, the curse might even turn into a blessing if the potential indirect damaging effects of fragmentation (on other growth-related factors) are controlled.

We have various extensions in mind of our analysis. Further analysis could focus on religious rather than ethnic fragmentation and test whether the key results of our analysis hold for religious fractionalization/polarization. Furthermore, we need to acknowledge that the current polarization and fractionalization indices are still far from perfect, as they focus exclusively on population structures, ignoring hence the relative financial and military power of groups, or potential long-term alliances among them. Developing fractionalization and polarization indices with such qualitative attributes will certainly improve the accuracy of our estimations.

ABBREVIATIONS

GDP: Gross Domestic Product

PPP: Purchasing Power Parity

WGI: Worldwide Governance Indicators

doi: 10.1111/ecin.12070

APPENDIX

TABLE A1
Correlation Matrix

                      [G.sub.1967-                        Ethnic
                         2007]       Ln [Y.sub.67]   Fractionalization

[G.sub.1967-2007]        1.000
Ln [Y.sub.67]            0.164           1.000
Ethnic                   -0.348         -0.441             1.000
  fractionalization
Ethnic polarization      -0.147         -0.111             0.574
Investment               0.434           0.624            -0.471
Fertility                -0.433         -0.780             0.473
Conflict                 -0.042         -0.141             0.092
Control of               0.472           0.786            -0.508
  corruption

                      Polarization   Investment   Fertility   Conflict

[G.sub.1967-2007]
Ln [Y.sub.67]
Ethnic
  fractionalization
Ethnic polarization      1.000
Investment               -0.204        1.000
Fertility                0.269         -0.311       1.000
Conflict                 -0.073        0.101        0.032      1.000
Control of               -0.241        0.355       -0.243      -0.055
  corruption

                      Corruption

[G.sub.1967-2007]
Ln [Y.sub.67]
Ethnic
  fractionalization
Ethnic polarization
Investment
Fertility
Conflict
Control of              1.000
  corruption

TABLE A2
List of Variables Used in the Regressions

[G.sub.1967-2007]          Average annual growth in real GDP per
                           person from 1967 to 2007, calculated as G =
                           (ln([Y.sub.2007] /[Y.sub.1967])/40) x 100%
                           (Penn World Tables, Heston, Summers, and
                           Aten 2010).

Ln [Y.sub.67]              Log of real GDP per capita in 1967 at 2005
                           international prices (Penn World Tables,
                           Heston, Summers, and Aten 2010).

Investment                 Share of investment (private and public) in
                           GDP (for 1967) (Penn World Tables, Heston,
                           Summers, and Aten 2010).

Fertility                  Fertility rate in 1967 (the number of
                           children that would be born to a woman if
                           she were to live to the end of her
                           childbearing years and bear children in
                           accordance with current age-specific
                           fertility rates; World Bank 2010).

Control of corruption      Control of corruption index in 1996,
                           measuring the extent of bribery and
                           protection given by anticorruption and
                           accounting institutions. Measure ranging
                           from -3 (most corrupt) to 3 (most
                           transparent) (Kaufmann, Kraay, and
                           Mastruzzi 2009).

Conflict                   Deaths attributed to conflict per
                           population (x 100) (1967) (World
                           Development Indicators, World Bank 2010).

Ethnic fractionalization   Ethnic fractionalization index (0-1
                           continuous scale) (Montalvo and
                           Reynal-Querol 2005a, 2005b). Time-variant
                           data for the 26 former Communist economies
                           provided by Campos, Saleh, and Kuzeyev
                           (2011).

Ethnic polarization        Ethnic polarization index (0-1 continuous
                           scale) (Montalvo and Reynal-Querol 2005a,
                           2005b). Time-variant data for the 26 former
                           Communist economies provided by Campos,
                           Saleh, and Kuzeyev (2011).

TABLE A3
Descriptive Statistics

                                   Standard
Variable                   Mean    Deviation   Minimum   Maximum

[G.sub.1967-2007]          1.74      1.68      -3.60      6.96
Ln [Y.sub.67]              8.11      1.01       6.11      9.98
Ethnic fractionalization   0.47      0.28       0.01      0.96
Ethnic polarization        0.52      0.24       0.02      0.98
Investment                 0.19      0.12       0.02      0.49
Fertility                  5.40      1.81       2.02      8.20
Conflict                   0.005     0.03       0         0.24
Control of corruption      0.10      1.12      -1.70      2.20

REFERENCES

Adams, S. "Foreign Direct Investment, Domestic Investment, and Economic Growth in Sub-Saharan Africa." Journal of Policy Modelling, 31, 2009, 939-49.

Aidt, T., J. Dutta, and V. Sena. "Governance Regimes, Corruption and Growth: Theory and Evidence." Journal of Comparative Economics, 36, 2008, 195-220.

Aisen, A., and F. J. Veiga. "How Does Political Instability Affect Economic Growth?" European Journal of Political Economy, 29, 2013, 151-67.

Akcomak, I. S., and B. ter Weel. "Social Capital, Innovation and Growth: Evidence from Europe." European Economic Review, 53, 2009, 544-67.

Alesina, A., and E. La Ferrara. "Ethnic Diversity and Economic Performance." Journal of Economic Literature, 43, 2005, 762-800.

Alesina, A., R. Baqir, and W. Easterly. "Public Goods and Ethnic Divisions." Quarterly Journal of Economics, 114, 1999, 1243-84.

Alesina, A., A. Devleeschauwer, W. Easterly, S. Kurlat, and R. Wacziarg. "Fractionalization." Journal of Economic Growth, 8, 2003, 155-94.

Anson, J., and A. Meir. "Religiosity, Nationalism and Fertility in Israel." European Journal of Population, 12, 1996, 1-25.

Arezki R., and M. Bruckner. "Commodity Windfalls, Polarization, and Net Foreign Assets: Panel Data Evidence on the Voracity Effect." Journal of International Economics 86(2), 2012, 318-26.

Ashraf, Q, and O. Galor. "The 'Out of Africa' Hypothesis, Human Genetic Diversity, and Comparative Economic Development." NBER Working Paper No. 17216, 2011.

Azzimonti, M. "Barriers to Investment in Polarized Societies." American Economic Review, 11, 2011, 2182-204.

Baggio, J., and E. Papyrakis. "Ethnic Diversity, Property Rights and Natural Resources." Developing Economies, 48, 2010, 473-95.

Barro, R. J. "Economic Growth in a Cross Section of Countries." Quarterly Journal of Economics, 106, 1991, 407-41.

Basu, A. M. "The 'Politicization' of Fertility to Achieve Non-Demographic Objectives." Population Studies, 51, 1997, 5-18.

Becker, G. S., and N. Tomes. "Child Endowments and the Quantity and Quality of Children." Journal of Political Economy, 84, 1976, 143-62.

Bhavnani, R., and D. Miodownik. "Ethnic Polarization, Ethnic Salience, and Civil War." Journal of Conflict Resolution, 53, 2009, 30-49.

Bjornskov, C. "How Does Social Trust Affect Economic Growth?" Southern Economic Journal, 78, 2012, 1346-68.

Bloom, D. E., D. Canning, G. Fink, and J. E. Finlay. "Fertility, Female Labor Participation, and the Demographic Dividend." Journal of Economic Growth, 14. 2009, 79-101.

Brander, J. A., and S. Dowrick. "The Role of Fertility and Population in Economic Growth." Journal of Population Economics, 7, 1994, 1-25.

Brock, W. A., and S. N. Durlauf. "What Have We Learned from a Decade of Empirical Research on Growth? Growth Empirics and Reality." World Bank Economic Review, 15. 2001, 229-72.

Busse, M., and C. Hefeker. "Political Risks, Institutions and Foreign Direct Investment." European Journal of Political Economy, 23, 2007, 397-415.

Campos, N. F., and V. Kuzeyev. "On the Dynamics of Ethnic Fractionalization." American Journal of Political Science, 51, 2011, 620-39.

Campos, N. F., A. Saleh, and V. Kuzeyev. "Dynamic Ethnic Fractionalization and Economic Growth." Journal of International Trade and Economic Development, 20, 2011, 129-52.

Camps, E., and S. Engerman. "What Do Women Do? World Population Growth and Fertility Patterns, 1960-2000." Working Paper 363, Barcelona Economics Working Paper Series, 2008.

Cerqueti, R., R. Coppier, and G. Piga. "Corruption, Growth and Ethnic Fractionalization: A Theoretical Model." Journal of Economics, 106, 2012, 153-81.

Cole, A. M. "Corruption, Income and the Environment: An Empirical Analysis." Ecological Economics, 62, 2007, 637-47.

Collier, P. "Implications of Ethnic Diversity." Economic Policy, 16, 2001, 127-66.

Collier, P., and A. Hoeffler. "On Economic Causes of Civil Wars." Oxford Economic Papers, 50, 1998, 563-73.

Crafts, N. "Transport Infrastructure Investment: Implications for Growth and Productivity." Oxford Review of Economic Policy, 25, 2009, 327-43.

Dincer, O. C. "Ethnic and Religious Diversity and Corruption." Economics Letters, 99, 2008, 98-102.

Easterly, W., and R. Levine. "Africa's Growth Tragedy: Policies and Ethnic Divisions." Quarterly Journal of Economics, 112, 1997, 1203-50.

Easterly, W., J. Ritzen, and M. Woolcock. "Social Cohesion, Institutions, and Growth." Economics and Politics, 18, 2006, 103-20.

Esteban, J., and D. Ray. "Conflict and Distribution." Journal of Economic Theory, 87, 1999, 379-415.

--. "On the Salience of Ethnic Conflict." American Economic Review, 98, 2008, 2185-202.

--. "Linking Conflict to Inequality and Polarization." American Economic Review, 101, 2011, 1345-74.

Esteban, J., L., Mayoral, and D. Ray. "Ethnicity and Conflict: An Empirical Study." American Economic Review, 102, 2012, 1310-42.

Fafchamps, M., and C. Moser. "Crime, Isolation and Law Enforcement." Journal of African Economies, 12, 2003, 625-71.

Fearon, J. D., and D. D. Laitin. "Ethnicity, Insurgency, and Civil War." American Political Science Review, 97, 2003, 75-90.

Fisman, R., and J. Svensson. "Are Corruption and Taxation Really Harmful to Growth? Firm Level Evidence." Journal of Development Economics, 83, 2007, 63-75.

Gates, S., H. Hegre, H. M. Nygard, and H. Strand. "Development Consequences of Armed Conflict." World Development, 40, 2012, 1713-22.

Grier, K. B., and G. Tullock. "An Empirical Analysis of Cross-National Economic Growth, 1951-80." Journal of Monetary Economics, 24, 1989, 259-76.

Habyarimana, J., M. Humphreys, D. N. Posner, and J. M. Weinstein. "Why Does Ethnic Diversity Undermine Public Goods Provision?" American Political Science Review, 101, 2007, 709-25.

Hegre, H., and N. Sambanis. "Sensitivity Analysis of Empirical Results on Civil War Onset." Journal of Conflict Resolution, 50, 2006, 508-35.

Heston, A, R. Summers, and B. Aten. Penn World Table Version 6.3. Philadelphia, PA: Center for International Comparisons of Production, Income and Prices, University of Pennsylvania, 2010. http://pwt.econ.upenn.edu/php_site/pwt_index.php.

Hodler, R. "The Curse of Natural Resources in Fractionalized Countries." European Economic Review, 50, 2006, 1367-86.

Horowitz, D. Ethnic Groups in Conflict. Berkeley: University of California Press, 2000.

Janus, T. "Fertility and Social Divisions." Working Paper, Department of Economics and Finance, University of Wyoming, 2010.

Kaufmann, D., A. Kraay, and M. Mastruzzi. "Governance Matters VIII: Governance Indicators for 1996-2008." Policy Research Working Paper No. 4978, World Bank, 2009.

King, R. G., and R. Levine. "Finance and Growth: Schumpeter Might Be Right." Quarterly Journal of Economics, 108, 1993, 717-37.

Klasen, S., and F. Lamanna. "The Impact of Gender Inequality in Education and Employment on Economic Growth: New Evidence for a Panel of Countries." Feminist Economics, 15, 2009, 91-132.

Kormendi, R. C., and P. G. Meguire. "Macroeconomic Determinants of Growth: Cross-Country Evidence." Journal of Monetary Economics, 16, 1985, 141-63.

Krieger, T., and D. Meierrieks. "Terrorism in the Worlds of Welfare Capitalism." Journal of Conflict Resolution, 54, 2010, 902-39.

LaPorta, R., F. Lopez-de-Silanes, A. Shleifer, and R. Vishny. "The Quality of Government." Journal of Law, Economics and Organization, 15, 1999, 222-79.

Lee, R. D., A. Mason, and T. Miller. "Saving, Wealth, and Population," in Population Matters: Demographic Change, Economic Growth, and Poverty in the Developing World, edited by N. Birdsall, A. C. Kelley, and S. W. Sinding. Oxford: Oxford University Press, 2001, 137-64.

Levine, R. "Law, Endowments, and Property Rights." NBER Working Paper No. 11502, 2005.

Levine, R., and D. Renelt. "A Sensitivity Analysis of Cross-Country Growth Regressions." American Economic Review, 84, 1992, 942-63.

Madsen, J. B., J. B. Ang, and R. Banerjee. "Four Centuries of British Economic Growth: The Roles of Technology and Population." Journal of Economic Growth, 15, 2010, 263-90.

Mankiw, G. N., D. Romer, and D. N. Weil. "A Contribution to the Empirics of Economic Growth." Quarterly Journal of Economics, 107, 1992, 407-37.

Mauro, P. "Corruption and Growth." Quarterly Journal of Economics, 110, 1995, 681-712.

Miguel, E., and M. K. Gugerty. "Ethnic Diversity, Social Sanctions, and Public Goods in Kenya." Journal of Public Economics, 11-12, 2005, 2325-68.

Mo, P. H. "Income Inequality and Economic Growth." Kyklos, 53, 2000, 293-316.

--. "Corruption and Economic Growth." Journal of Comparative Economics, 29, 2001, 66-79.

Montalvo, J. G., and M. Reynal-Querol. "Ethnic Diversity and Economic Development." Journal of Development Economic, 76, 2005a, 293-323.

--. "Ethnic Polarization, Potential Conflict, and Civil Wars." American Economic Review, 95, 2005b, 796-816.

--. "Discrete Polarization with an Application to the Determinants of Genocides." Economic Journal, 118, 2008, 1835-65.

--. "Ethnic Polarization and the Duration of Civil Wars." Economics of Governance, 11, 2010, 123-43.

Olsson, O. "Conflict Diamonds." Journal of Development Economics, 82. 2007, 267-86.

Ostby, G. "Polarization, Horizontal Inequalities and Violent Civil Conflict." Journal of Peace Research, 45, 2008, 143-62.

Papyrakis, E., and R. Gerlagh. "The Resource Curse Hypothesis and Its Transmission Channels." Journal of Comparative Economics, 32, 2004, 181-93.

--. "Resource Abundance and Economic Growth in the United States." European Economic Review, 51, 2007, 1011-39.

Pellegrini, L., and R. Gerlagh. "Causes of Corruption: A Survey of Cross-Country Analyses and Extended Results." Economics of Governance. 9, 2008, 245-63.

Reynal-Querol, M. "Does Democracy Preempt Civil Wars?" European Journal of Political Economy, 21, 2005, 445-65.

Richard, O. C. "Racial Diversity, Business Strategy, and Firm Performance: A Resource-Based View." Academy of Management Journal, 43, 2000, 164-77.

Richard, O. C., B. P. S. Murthi, and K. Ismail. "The Impact of Racial Diversity on Intermediate and Long-Term Performance: The Moderating Role of Environmental Context." Strategic Management Journal, 28, 2007, 1213-33.

Sachs, J. D., and A. M. Warner. "Natural Resources and Economic Development: The Curse of Natural Resources." European Economic Review, 45, 2001, 827-38.

Shen, C., and J. B. Williamson. "Corruption, Democracy, Economic Freedom and State Strength: A Cross-National Analysis." International Journal of Comparative Sociology, 46, 2005, 327-45.

Svensson, J. "Investment, Property Rights and Political Instability: Theory and Evidence." European Economic Review, 42, 1998, 1317-41.

--. "Eight Questions about Corruption." Journal of Economic Perspectives, 19, 2005, 19-42.

Tangeras, T. P., and N.-P. Langerlof. "Ethnic Diversity, Civil War and Redistribution." Scandinavian Journal of Economics, 111, 2009, 1-27.

van der Ploeg, F., and S. Poelhekke. "Volatility and the Natural Resource Curse." Oxford Economic Papers, 61, 2009, 727-60.

World Bank. World Development Indicators 2010. Washington, DC: World Bank, 2010.

(1.) We experimented with the quadratic forms of our indices of fractionalization and polarization in the empirical analysis that follows--in all cases there was very weak support of a nonlinear relationship.

(2.) There is a nonlinear relationship between these two indices of social fragmentation, and an in-depth discussion on their construction can be found in Montalvo and Reynal-Querol (2005a).

(3.) Table A2 lists all variable descriptions and data sources. Descriptive statistics are presented in Table A3 (for the sample of 102 countries, for which data are available for all key variables).

(4.) Due to the high correlation of the index across years, we find almost identical results once we include the average of the measure for the 1996-2007 period instead.

(5.) The inclusion of additional regressors slightly decreases our sample size from 111 (column (1)) to 102 observations (columns (5) and (6)). Replicating Table 1 for the 102 countries for which data are available for all variables provides very similar results.

(6.) Similar to Hodler (2006) and Montalvo and Reynal-Querol (2005a), we avoid introducing the fractionalization and polarization proxies jointly into the estimated specifications to avoid multicollinearity.

(7.) There is some tentative evidence of such a positive relationship at a more micro scale in the business and management literature (e.g., see Richard 2000 and Richard, Murthi, and Ismail 2007 who discuss how such diversity can enhance creativity and innovation in the workplace).

(8.) There is also evidence suggesting a positive link between ethnic heterogeneity and (nonfatal) crime more broadly (the extent of theft, rape, number of injuries, etc.). See Fafchamps and Moser (2003) for a discussion.

(9.) For a broader discussion and empirical evidence on the links between ethnic fragmentation and institutions, see Easterly, Ritzen, and Woolcock (2006).

(10.) Running the same regressions for the largest possible sample for each transmission channel yields almost identical results.

(11.) This does not come as a surprise as there is much empirical evidence in the literature suggesting that the conflict transmission channel is the weakest of all (e.g., see Fearon and Laitin 2003, Hegre and Sambanis 2006, and Ostby 2008).

(12.) The conflict transmission channel again appears to be the weakest one (and the coefficient does not suggest a conflict-enhancing impact of ethnic polarization).

(13.) The correlation between conflict and ethnic fractionalization is again of both low statistical significance as well as small magnitude.

ELISSAIOS PAPYRAKIS and PAK HUNG MO *

* The authors thank the editor and an anonymous referee for their many helpful comments on the paper.

Papyrakis: Senior Lecturer (Associate Professor) in Economics, School of International Development, University of East Anglia, Norwich NR4 7TJ, UK; Institute for Environmental Studies, Vrije Universiteit Amsterdam, De Boelelaan 1087, 1081 HV Amsterdam, The Netherlands. Phone 44-1603592338, Fax 44-1603451999, E-mail e.papyrakis@uea.ac.uk

Mo: Associate Professor in Economics, Department of Economics, Hong Kong Baptist University, Kowloon, Hong Kong. Phone 852-34117546, Fax 852-34115580, E-mail phmo@hkbu.edu.hk

TABLE 1
Growth Regressions as in Equation (1) with
Fractionalization as an Independent Variable

Dependent Variable:
[G.sub.1967-2007]           (1)        (2)         (3)

Constant                   -0.16      2.20        3.92
Ln [Y.sub.67]               0.24      0.05       -0.34 *
                           (0.15)    (0.19)      (0.21)
Ethnic fractionalization            -1.91 ***   -1.35 **
                                     (0.62)      (0.60)
Investment                                      6.50 ***
                                                 (1.72)
Fertility

Conflict

Control of conniption

[R.sup.2] adjusted          .02        .10         .21
N                           111        105         105

Dependent Variable:
[G.sub.1967-2007]             (4)         (5)         (6)

Constant                     12.33       13.26       13.73
Ln [Y.sub.67]              -1.00 ***   -1.12 ***   -1.34 ***
                            (0.27)      (0.26)      (0.27)
Ethnic fractionalization    -0.95 *      -0.80       -0.55
                            (0.55)      (0.54)      (0.52)
Investment                 5.28 ***    6.10 ***    5.05 ***
                            (1.62)      (1.58)      (1.34)
Fertility                  -0.56 ***   -0.59 ***   -0.34 **
                            (0.13)      (0.13)      (0.15)
Conflict                               -8.93 ***   -8.66 ***
                                        (2.77)      (2.40)
Control of conniption                              0.77 ***
                                                    (0.26)
[R.sup.2] adjusted            .35         .37         .43
N                             103         102         102

Note: Robust standard errors for coefficients in parentheses.
*, **, and *** correspond to a 10%, 5%, and 1% level of significance.

TABLE 2
Growth Regressions as in Equation (1) with
Polarization as an Independent Variable

Dependent Variable:
[G.sub.1967-2007]          (7)         (8)         (9)

Constant                  0.05        2.57        11.91
Ln [Y.sub.67]             0.26        -0.25     -1.01 ***
                         (0.19)      (0.21)      (0.29)
Ethnic polarization       -0.85       -0.42       -0.42
                         (0.64)      (0.61)      (0.57)
Investment                          7.26 ***    6.03 ***
                                     (1.77)      (1.62)
Fertility                                       -0.61 ***
                                                 (0.13)
Conflict

Control of corruption

[R.sup.2] adjusted         .03         .17         .33
N                          105         105         103

Dependent Variable:
[G.sub.1967-2007]         (10)        (11)

Constant                  12.97       13.64
Ln [Y.sub.67]           -1.13 ***   -1.38 ***
                         (0.26)      (0.27)
Ethnic polarization       0.32        0.48
                         (0.57)      (0.57)
Investment              6.77 ***    5.49 ***
                         (1.56)      (1.30)
Fertility               -0.64 ***   -0.38 **
                         (0.13)      (0.15)
Conflict                -9 73 ***   -9.10 ***
                         (3.05)      (2.62)
Control of corruption               0.82 ***
                                     (0.26)
[R.sup.2] adjusted         .36         .42
N                          102         102

Note: Robust standard errors for coefficients in parentheses.
*, **, and *** correspond to a 10%, 5%, and 1% level of significance.

TABLE 3
Links Between Fractionalization and Other Growth
Determinants (Transmission Channels--Equation (2))

                           Investment   Fertility
                              (12)        (13)

Constant                     -0.24        15.24
Ln [Y.sub.67]               0.06 ***    -1.27 ***
                             (0.01)      (0.12)
Ethnic fractionalization   -0.10 ***    1.03 ***
                             (0.03)      (0.40)
[R.sup.2] adjusted            .43          .62
N                             102          102

                           Conflict     Control of
                             (14)     Corruption (15)

Constant                     0.03          -5.80
Ln [Y.sub.67]               -0.003       0.77 ***
                           (0.004)        (0.08)
Ethnic fractionalization    0.004        -0.80 ***
                           (0.007)        (0.25)
[R.sup.2] adjusted           .03            .64
N                            102            102

Note: Robust standard errors for coefficients in parentheses.
*, **, and *** correspond to a 10%, 5%, and 1% level of significance.

TABLE 4
Growth Regression (Equation (3)) Taking
Account of the Indirect Links Between
Fractionalization and Other Growth
Determinants

Dependent Variable: [G.sub.1957-2007]        (16)

Constant                                     2.53
Ln [Y.sub.67]                               0.02
                                            (0.14)
Ethnic fractionalization                  -2.04 ***
                                            (0.51)
Investment ([[micro].sub.1])               5.05 ***
                                            (1.34)
Fertility ([[micro].sub.2])                -0.34 **
                                            (0.15)
Conflict ([[micro].sub.3])                -8.66 ***
                                            (2.40)
Control of corruption ([[micro].sub.4])    0.77 ***
                                            (0.26)
[R.sup.2] adjusted                           .43
N                                            102

Note: Robust standard errors for coefficients in parentheses.
*, **, and *** correspond to a 10%. 5%, and 1% level of significance.

TABLE 5
Relative Importance of Transmission Channels Through
Which Fractionalization Affects Growth

Transmission Channels      [[alpha].sub.3]   [[alpha].sub.2]
                              (Table 1)         (Table 3)

Ethnic fractionalization
Investment                      5.05              -0.10
Fertility                       -0.34             1.03
Conflict                        -8.66             0.004
Control of corruption           0.77              -0.80
Total

Transmission Channels       Contribution to      Relative
                           [[alpha].sub.2] +   Contribution
                            [[alpha].sub.3]
                            [[beta].sub.2]

Ethnic fractionalization         -0.55             27%
Investment                       -0.50             25%
Fertility                        -0.35             17%
Conflict                         -0.03              1%
Control of corruption            -0.61             30%
Total                            -2.04             100%

TABLE 6
Links Between Polarization and Other Growth Determinants
(Transmission Channels--Eauation (2))

                                                          Control of
                      Investment   Fertility   Conflict   Corruption
                         (17)        (18)        (19)        (20)

Constant                -0.34        15.72       0.04       -6.43
Ln [Y.sub.67]          0.07 ***    -1.36 ***    -0.004     0.85 ***
                        (0.01)      (0.10)     (0.005)      (0.07)
Ethnic polarization    -0.07 *     1.41 ***     -0.01     -0.74 ***
                        (0.04)      (0.45)      (0.01)      (0.24)
[R.sup.2] adjusted       .40          .64        .02         .64
N                        102          102        102         102

Note: Robust standard errors for coefficients in parentheses.
*, **, and *** correspond to a 10%, 5%, and 1% level of significance.

TABLE 7
Growth Regression (Equation (3)) Taking Account of the Indirect
Links Between Polarization and Other Growth Determinants

Dependent Variable:
[G.sub.1967-2007]                          (21)

Constant                                   0.22
Ln [Y.sub.67]                         0.25 ** (0.12)
Ethnic polarization                   -0.94 * (0.55)
Investment ([[micro]1])              5.49 *** (1.30)
Fertility ([[micro]2])               -0.38 ** (0.15)
Conflict ([[micro]3])                -9.10 *** (2.62)
Control of corruption ([[micro]4])   0.82 *** (0.26)
[R.sub.2] adjusted                         .42
N                                          102

Note: Robust standard errors for coefficients in parentheses.

*, **, and *** correspond to a 10%, 5%, and 1% level of
significance.

TABLE 8
Relative Importance of Transmission Channels
Through Which Polarization Affects Growth

Transmission Channels   [[alpha].sub.3]   [[beta].sub.2]
                           (Table 2)        (Table 6)

Ethnic polarization
Investment                   5.49             -0.07
Fertility                    -0.38             1.41
Conflict                     -9.10            -0.01
Control of corruption        0.82             -0.74
Total

Transmission Channels    Contribution to      Relative
                        [[alpha].sub.2] +   Contribution
                         [[alpha].sub.3]
                         [[beta].sub.2]

Ethnic polarization           0.48              -51%
Investment                    -0.38             41%
Fertility                     -0.53             56%
Conflict                      0.09              -10%
Control of corruption         -0.60             64%
Total                         -0.94             100%

TABLE 9
Links Between Initial Ethnic Fractionalization and Subsequent
Growth and Determinants (26 Former Communist Economies)

                                     Growth     Investment   Fertility
                                      (22)         (23)        (24)

Constant                              17.95       23.35        1.11
Ethnic fractionalization (lagged)   -40.31 **     -4.46      1.94 ***
                                     (20.53)     (18.07)      (0.61)
[R.sup.2] adjusted (within)            .11         .03          .15
N                                      78           78          78

                                    Conflict     Control of
                                      (25)     Corruption (26)

Constant                             -0.01          0.07
Ethnic fractionalization (lagged)     0.01        -1.34 **
                                     (0.01)        (0.66)
[R.sup.2] adjusted (within)           .04            .08
N                                      78            75

Note: Standard errors are robust, clustered by country, and appear in
parentheses.

*, **, and *** correspond to a 10%, 5%, and 1% level of significance.