Ethnopolitical demography and democracy in sub-Saharan Africa

Ethnic fragmentation is largely presumed to be bad for democracy. However, many African countries belie this claim, as democracy has recently sprouted in several of its multiethnic states. We argue that African countries that have demographic patterns where the largest ethnopolitical group is at least a near-majority and is simultaneously divided into nested subgroups produce Africa's most democratic multiethnic societies. This large-divided-group pattern, which has gone largely unnoticed by previous scholars, facilitates transitions to democracy from authoritarian rule. The large group's size foments the broad-based multiethnic social agitation needed to pose a genuine threat to a ruling autocrat, while its internal divisions reassure minorities that they will not suffer permanent exclusion via ethnic dominance under an eventual democracy. We support our claim with cross-national quantitative evidence on ethnic fragmentation and regime type.


Introduction
The argument that ethnic fragmentation inhibits the emergence and sustainability of democracy has a long pedigree in political science. 1 The modified argument that certain institutions can significantly mitigate complications caused by ethnic cleavages is only slightly younger. 2 These two visions have dominated subsequent research on the relationship between ethnic fractionalization and regime type 3 and continue to characterize most scholarly thinking on the topic. 4 At first glance, the sub-Saharan African experience with democracy seems to provide ample support for these conventional wisdoms. In support of the claim that ethnic fragmentation inhibits democracy, scholars have attributed Africa's status as a latecomer to democratization and its many remaining authoritarian regimes to the fact that it has the highest rates of ethnic fractionalization in the world. 5 In support of the institutionalist claim, African authoritarianism may be caused by the continent's relative lack of the formal political arrangements -proportional representation, federalism, cross-ethnic vote pooling -that allegedly encourage power sharing and mitigate the deleterious effects of ethnic fractionalization on democracy. African countries feature ill-advised winner-take-all systems: almost all are presidential and unitary, and most employ plurality-or majority-rule electoral systems. Overall, African nations seem to converge on a reality -high ethnic fragmentation, wrong institutions, relatively high rates of autocracythat supports both conventional wisdoms.
However, a more thorough look at Africa's recent political history, and in particular the emergence of democracy in many parts of the continent since the third wave of democracy began there in the early 1990s, belies both hypotheses. Threefourths of sub-Saharan Africa's 48 countries have been, on average over the period 2000 to 2012, either "free" (N ¼ 8) or "partly free" (N ¼ 28) according to Freedom House. This is a far higher share of democracies and semi-democracies than the Middle East and North Africa, a region whose societies are less ethnically fragmented. Moreover, several of the states in which democracy has taken root have moderate or high levels (Benin, South Africa) of ethnopolitical diversity. Contrary to institutionalists' claims, some of the continent's most robust democracies -Botswana, Ghana, Mauritius -employ winner-take-all electoral rules with unitary systems. These conventional wisdoms thus leave a glaring unanswered puzzle for sub-Saharan Africa: Where and why has democracy taken root in African countries that are ethnically fragmented?
In this paper, we provide an answer to this puzzle by arguing and demonstrating that multiethnic societies in which the largest ethnopolitical group is at least a nearmajority of the population and is simultaneously divided into nested subgroups are more democratic than other types of multiethnic societies in Africa. We argue that these large-divided-group societies are more likely to experience transitions from autocracy to democracy than other types of multiethnic societies. The presence of a large ethnopolitical group facilitates the creation of a movement that poses a genuine threat to a sitting autocrat, yet its division into salient subgroups reassures minorities that ethnic dominance by the largest group is unlikely in an eventual democracy. This theoretical argument allows us to explain why democracy has not emerged in places with very different ethnic configurations, like highly fragmented Uganda and dominant-majority Djibouti, but has emerged in countries such as Ghana and South Africa. We provide support for this argument by analysing a quantitative dataset that classifies 385 ethnopolitical groups in 48 countries of Sub-Saharan Africa by their number and size and also by their internal structurethat is, whether there are nested subgroups within larger groups.

Ethnic demography and democracy
Scholars have uncovered numerous factors (for example, international diffusion, colonial legacy, prosperity levels) that influence a country's prospects for democratizing, but a particularly fruitful line of research has looked at ethnic diversity. The primary scholarly viewpoints that see ethnic demography, meaning the number and sizes of a society's ethnic groups, as having a major impact on a country's regime type fall into two camps. 6 The diversity-inhibits-democracy school of thought has long been sceptical about the prospects for democracy in ethnically diverse societies, dating back to John Stuart Mill's writings in the nineteenth century 7 and then reappearing in some of the classics of modern comparative politics. 8 According to this school of thought, a culturally diverse society that lacks a unified national identity is likely to experience intractable, zero-sum disagreements among groups over policy priorities and institutional configurations, such as where the limits of state power lie. 9 Rather than suppressing ethnic demands in the national interest, office-seeking politicians would play to them, risking the escalation of disagreement into (at best) political gridlock or (at worst) political violence. Ultimately, the mutual distaste and lack of cooperation among groups would lead to persistent authoritarianism or, if it emerged at all, the breakdown of peaceful democratic competition. In sum, the diversity-inhibits-democracy school sees the low likelihood of democracy in plural societies as resulting from a coordination dilemma: an excessive number of groups makes it difficult to reach societal consensus on public goods provision, the limits of state repression, and the geographical extent of state authority. The large number of faultlines makes it likely that at least one group will defect from peaceful competition. Subsequent empirical testing has lent some support to this conventional wisdom. 10 In contrast, some scholars see the degree of ethnic diversity to be operating in a nonlinear fashion, whereby societies with just two major groups, one being a majority and the other a sizeable minority, are at greatest risk of sustained authoritarian rule or even civil war. 11 In this scenario, it is ethnic dominance, meaning the repression and permanent exclusion of a minority by the majority (an "ethnocracy" in Horowitz's terms), and not diversity per se that makes democracy unlikely to emerge or be sustained. 12 Even if the minority gains power autocratically, it will seek to stay there via undemocratic means, since it stands little chance of regaining office under democratic rules. Thought of differently, this argument holds that highly diverse societies are more likely to be democratic than countries with a two-group, majority-minority configuration, because the former have lots of small groups with no single one able to impose ethnic dominance. Indeed, from this view, the large number of small groups in highly fractionalized countries grants a certain flexibility to create a wide variety of governing coalitions, thus enabling groups to move in and out of power under peaceful alternations. 13 In sum, this dominance-inhibits-democracy view poses diversity's threat to democracy as more of a credible commitment problem: any commitments from the majority to not exclude the minority will lack credibility because of the majority's numerical dominance. 14 Despite the valuable insights of both sides in this debate, we see three problems with them collectively. First, no a priori reason or overwhelming evidence exists to suggest that one kind of problem -coordination or credible commitment -is more hazardous than the other for democratic emergence and sustainability. In actuality, both are surely causes of authoritarian persistence and democratic breakdowns, so a theory that recognizes the threat of both problems and the need for multiethnic societies to successfully balance the two is necessary.
Second, the literature has largely ignored the potentially different effects that ethnic demography may have on democratic emergence and democratic sustainability. The publication of Przeworski and Limongi's "Modernization: Theories in Facts" in 1997 15 heightened many scholars' attention to the conceptual difference between democratic emergence and democratic survival. However, those studying ethnic demography and democracy have largely overlooked the differing frequencies and natures of these two processes and the role that ethnopolitical demography may play in them.
Third, both sides in the debate use a unidimensional conceptualization of identity that views ethnic demography as overly fixed and that assumes an intrinsic unity to ethnic groups that often does not exist. Recent innovations within the constructivist school of thought have soundly debunked these assumptions. 16 One such innovation is the realization that individuals have multiple identities from which they can choose and that, in many contexts, some identities are nested within others. For example, in sub-Saharan Africa, many ethnic groups have nested subgroups within them, so individuals avail themselves of different identities in different contexts. 17 The recognition and measurement of this nestedness reveals the large-divided-group ethnic demography pattern, which prevails in about a dozen of the region's countries. We argue below that this pattern has induced elites in large-divided-group countries to successfully hit the sweet spot by balancing the challenges of coordination and credible commitment to craft many of Africa's emergent democracies.

Ethnopolitical demography and democracy in sub-Saharan Africa
Ethnicity, like all sources of social cleavages, can be activated strategically by social actors, often to advance political goals. Objective ethnic markers -ascriptive attributes, such as common descent, "tribal" membership, culture, language, race, religion and/or common territory -are often politicized by political entrepreneurs to define group interests and organize collective action in pursuit of political goals. 18 When activated for such purposes, ethnic groups become "ethnopolitical groups" (politicized ethnic groups), a more specific term that we use in this paper, because many small African groups are politicized as parts of larger but similar ethnopolitical groups. 19 Sub-Saharan Africa's 48 countries vary dramatically in their ethnopolitical demographies. To provide some order to this seeming cacophony, we group African societies into four demographic types: (1) Homogeneous societies, (2) Only Small Groups societies, (3) Large Undivided Group societies, and (4) Large Divided Group societies. The first type is a rarity in Africa, the homogeneous society, where all residents hail from the same ethnopolitical group. At the opposite extreme is the only-small-groups society, which we take as any multiethnic society in which the largest group comprises less than a near-majority of the population. We use 40% of the population as the cutoff. 20 Large-undivided-group countries are those in which there is one group that constitutes a majority or near-majority of the population and that is not subdivided into nested subgroups. Finally, large-divided-group (LDG) societies are those in which the largest ethnopolitical group exceeds 40% of the population and is divided into nested subgroups.

Theory
We argue that LDG societies are the non-homogenous societies with the greatest chance of experiencing a transition to democracy. Our argument is thus about democratization more than about democratic survival, and in particular the third wave democratizations that began in Africa in 1990. 21 The problems posed by certain patterns of ethnic demography are typically ones, we argue below, that cause authoritarianism to persist, either under the dominance of one longstanding ruler or via illegal changes in leadership from one autocrat to another.
For an autocracy to become a democracy, there must be agitation among the citizenry against either an autocratic incumbent's policies or regime legitimacy, and this social unrest must be of a kind that provides a genuine threat to the incumbency of the sitting autocrat. 22 For example, the early 1990s in sub-Saharan Africa, the era in which the third wave began, was met by strikes, mass demonstrations, and other forms of protest. 23 Moreover, for democratization to occur, the incumbent must consider the costs of repressing citizens who are fomenting the unrest to be excessively high. Among other reasons, a leader could conclude that the costs of repression are too high, because a large number of important and powerful societal groups are involved in the agitation. If the autocrat concludes as such, he is likely to negotiate with his citizen opponents and concede to them various political rights, including the convening of elections that could lead to his regime's demise. Democratization also requires a credible commitment by all major societal players to engage in peaceful democratic competition and to accept spells out of power, should they occur, as well as a minimal degree of political centralization -that is, the ability to broadcast state authority throughout the entire territory. 24 The ethnopolitical configurations of only-small-groups societies and largeundivided-group societies suppress the emergence of these causes of transition to democracy in different ways. The diversity-inhibits-democracy claim is present in considering the challenges of democratization in only-small-groups countries. The citizenries of only-small-groups countries face significant coordination or collective action problems in creating anti-regime unrest that is sufficiently sweeping to provide a genuine threat to the incumbent. Since engaging in such unrest carries high costs (such as the risk of repression and lost productivity), some ethnic groups will simply prefer to free ride on the protest efforts of other groups. As a result, anti-regime activity is likely to be sparse and the incumbent largely unthreatened. If any does emerge, it may be isolated to one or two small groups and, thus, relatively costless for the incumbent to repress. Moreover, even if unrest is such that the regime is under genuine threat, the many faultlines and mutual distaste among competing ethnopolitical groups increase the probability that at least one has an armed wing, a fact that could signal to other groups that not everyone would accept peaceful political competition under a potential democracy. Finally, the presence of so many groups retards the provision of public goods, including the establishment of the bureaucratic competency necessary to create a relatively uniform degree of "stateness," or legitimate political authority, throughout the territory. 25 By contrast, the dominance-inhibits-democracy problem is evident in largeundivided-group societies. Here, social agitation that poses a genuine threat to a ruling autocrat may fail to emerge not because of the coordination failure among multiple ethnic groups. Rather, if the ruling autocrat is a member of the largest group, the dominant group has no incentive to challenge the policies or legitimacy of the incumbent, since the country's largest group is heavily favoured in the distribution of government spoils (such as Afwerki's Eritrea). 26 If minority ethnopolitical groups foment unrest or rebel, their diminutive size lowers the chances that they will cause a genuine threat to the incumbent. In contrast, if a member of a minority group is the autocrat (such as Kagame's Rwanda or Eyadema's Togo), he and his coethnics will be loathe to relinquish power and support democratization, since they are unlikely to hold office in a democracy. In other words, the majority struggles to credibly commit to minority inclusion under democracy in large-undivided-group societies.
In contrast to both of these scenarios, LDG societies create an atmosphere that is conducive to the emergence of democracy, because political actors can successfully balance between the coordination failures of only-small-groups societies and the ethnic dominance of large-undivided-group ones. The shared identity within the large group helps to solve the collective action problem by fostering the creation of a united front in any uprising against the autocrat, and its size means the mobilization poses a genuine threat to the incumbent that would be costly to repress. Its relatively large size can also facilitate the creation of political order and state legitimacy throughout much of the territory. At the same time, of great importance is backwards induction, meaning actors' (and particularly minority groups outside the LDG) anticipation of future behaviour. The presence of nested subgroups within the largest group lessens the fears among other ethnopolitical groups that the largest group will be internally cohesive enough to impose domination and permanently exclude outside groups from power under an eventual democracy. After all, any of the minority outsider groups can court subgroups within the large group to defect and eventually form, or threaten to form, an alternative winning coalition within an existing or a newly formed political party. The LDG cleavage structure provides signals to elites from all groups that there are opportunities for a wide range of strategic negotiations and prior or post-election multi-ethnic coalition formations among small groups outside the largest group and subgroups from within it. 27

An African illustration
Most descriptions and measures of ethnic diversity hold to the longstanding assumption that ethnopolitical groups are internally unified and that a single index of fractionalization adequately captures cross-national differences in ethnic diversity. 28 However, many African societies are characterized by a complex multiethnicity in which "higher-level" groups, defined as those identity groupings that are the largest ethnopolitical groups just below the level of nationality, have smaller and less inclusive nested subgroups within them. 29 Higher level groups and nested subgroups have often been politicized at different times and in different ways. In countries with ethnopolitical subgroups nested inside higherlevel groups, fragmentation scores are higher (in some countries, substantially higher) if subgroups are tallied as separate groups.
Consider Benin, an example of the LDG type (according to the Scarritt and Mozaffar data set described below). As illustrated in Table 1, Benin has an ethnopolitical demography pattern that cannot be captured in a unidimensional fragmentation index. A large higher-level group (Fon) comprises more than 50% of the population, yet this large group is further subdivided into nested subgroups of varying sizes (as is the smaller Bargu higher-level group). This gives Benin two different fragmentation indices, depending on whether one tallies only higher-level groups (.620) or each subgroup separately (.847). We refer to the former as Higher-level fragmentation and the latter as Total fragmentation.
How did the large-divided-group pattern facilitate Benin's transitions to democracy? Benin became the first mainland African country to make a third wave transition in 1991 by toppling the 19-year Marxist military dictatorship of Mathieu Kérékou (1972Kérékou ( -1991, himself a northerner of the minority Somba/ Bargu ethnicity. Kérékou's ouster began with social agitation from below in 1989, as various urban sectors in the south launched strikes and protests against austerity and economic decline. 30 Facilitated by the size of the majority Fon group and alliances between it and other Southern ethnopolitical groups, the protests forced Kérékou to call a national conference. The conference had a pan-ethnic constituency that enabled it to assume executive power at the expense of Kérékou, although he was granted immunity. In 1990 a new democratic constitution that was approved by 96% of the population in a nationwide referendum called for parliamentary elections and two-round presidential elections that guaranteed majority support for the second round winner. In 1991 Kérékou lost in the second round of largely peaceful elections. The new government, led by Nicephore Soglo (1991 -1996) of the majority Fon, was supported by a post-election three-party multi-ethnic coalition in the legislature. This pattern of post-election coalitions among relatively small parties representing specific ethnopolitical groups has persisted since then and has enabled democracy to survive. Defections from the ruling coalition have been frequent, but other parties have quickly replaced the defectors. 31

Rethinking the measurement of ethnopolitical demography in Africa
To test our proposition about the advantages of LDG countries, we use the Scarritt and Mozaffar dataset (henceforth "SM dataset") on ethnopolitical groups and cleavages in 48 sub-Saharan African countries. 32 The SM dataset features two methodological improvements on other operationalizations of ethnic fragmentation. First, the SM dataset includes only ethnopolitical groups and cleavages relevant to the analysis of political contestation in these countries. 33 Most studies of the effects of fragmentation assume all ethnic groups are politically relevant. In contrast, the SM dataset tallies only politicized ethnic groups. Many African ethnic groups, most of them very small ones, are not politicized, so fragmentation of ethnopolitical rather than ethnic groups should be analysed. The existence of ethnopolitical groups was determined through an extensive search of comparative and country-specific interdisciplinary datasets (namely the Black Africa Handbook 34 and Minorities at Risk 35 ) and literature on the processes of ethnopolitical group identity construction in the colonial and pre-1990 post-independence periods. For more information on coding, see the Online Appendix.
Second, the SM dataset records ethnopolitical groups as multi-layered and, thus, avoids the inaccurate and oversimplifying assumptions of unidimensional fractionalization indices. As illustrated with the Benin example, the dataset codes higher-level groups and, if they exist, any subgroups nested within them. Numerous scholars have acknowledged the importance of nested levels of identity and inclusiveness, 36 but few ever operationalize it in an explicit and meaningful way. 37 Based on Scarritt and Mozaffar's coding, we classified 48 sub-Saharan African countries into the four country types discussed above. Table 2 reports the category classifications for the 48 countries in the dataset. The two shaded columns represent the country types that we expect to be the most democratic. As a preliminary test of our hypotheses, we report average Freedom House (FH) scores (from 2000 to 2012) for each country and category in Table 2. FH scores run from 1 to 7 with higher values meaning more political freedom. Scores of 6 and 7 denote "free" societies, 3 through 5 signify "partly free" ones, and 1 and 2 denote "not free" countries. Note that, as is customary, we flip the original scores. According to the simple FH means reported in Table 2, the most democratic societies in Africa have indeed been homogeneous and LDG ones.
Table 2 also reports the higher-level and total fragmentation indices, averaged by each country type. Among the three multiethnic society types (that is, ignoring the homogenous countries), these indicators are seemingly useless in predicting average FH scores. Only-small-groups and large-undivided-groups countries are at opposite observed extremes on both fragmentation indices, despite having similar FH averages. Nor is it clear which indicator (higher-level or total) one should choose. To address this, we develop a new indicator, one that simultaneously measures the size of the largest group and the degree to which it is subdivided, that we expect will capture what is more relevant to regime type in Africa. This indicator, which we call Weighted fragmentation within the largest group, is the fragmentation index for the subgroups within the largest higher-level group weighted by the size of this higher level group: where p s is subgroup s's proportion in the popultion, N is the number of subgroups within the largest higher-level group H, and P H is the largest higher level group's proportion of the population. The indicator (which has a theoretical range of 0 to 1) is relatively large for LDG countries, since they have fragmentation within their largest group and a large highest level group. For example, Benin's value is .35. In contrast, in large-undivided-group countries the value is zero, since the fragmentation score within the largest group (that is, the first term in the equation) is zero. 38 In onlysmall-groups countries, the value can be non-zero, but it is generally low, since the size of the largest higher level group (that is, the second term in the equation) is small.  Note: Numbers in "FH" columns are country-level mean Freedom House scores (flipped from the original scores so that 7 is the most free and 1 is the least free) from 2000 to 2012. Years in parentheses are those in which a post-1989 transition to democracy occurred. Boix et al., "A Complete Data Set of Political Regimes". Ethnic group names in Large Divided Groups column are these countries' large divided group with its share of the population after the comma.
We expect this variable, once controlling for homogenous societies, to be positively correlated with democracy in Africa. Table 2 provides preliminary evidence that this is the case, as the average weighted fragmentation score is at its highest for the multiethnic category (LDG) with the highest average FH score. The difference in mean FH scores is a full two points on a scale with just a 6-point range, a huge substantive effect. Figure 1 depicts the relationship graphically at the country level of analysis by plotting the simple bivariate relationship between a country's weighted fragmentation indicator and its average FH score in the twenty-first century. The figure facilitates identification of the 12 LDG societies by circling them in grey. The bivariate correlation is +.42 and is strong despite the existence of influential outlying cases, Namibia and Zimbabwe.

Methods and measures
While informative, use of these averaged Freedom House scores prohibits us from distinguishing effects on democratic emergence from those on democratic sustainability, since higher scores reflect a combination of both. To remedy this, we run a series of dynamic probit models that estimate the impact of covariates on democratic emergence and on democratic survival separately. 39 The models are pooled time series with the unit of analysis being the country year: each model contains 40-plus units (that is, countries) and a T around 20, as we set each series to begin in 1990. The dependent variable is 1 for country years in which the prevailing regime type is democratic and 0 in years in which the country's regime is non-democratic. We include the one-year lag of the dependent variable and interact it with each of the covariates. Each model, then, produces two sets of coefficients: One set of coefficients is the effects of the independent variables on democratic emergence and is estimated using the country years in which the lag of democracy is zero. The other set is the effects of independent variables on democratic survival and is estimated using the country years in which the lag of democracy is one. 40 In terms of measuring our dependent and key independent variables, we take an inclusive approach. We measure each concept in multiple ways and run multiple models to discern the extent to which our main argument and findings are robust to different measurement assumptions. For our dependent variable, we employ four different measures of democracy. Two are the datasets created by Boix, Miller and Rosato and Cheibub, Gandhi, and Vreeland, both of which code democracy as a binary variable. 41 The other two are the Polity IV and Freedom House indices, 42 which we converted from their original ordinal scales to binary ones using the standard thresholds. 43 For independent variables, we take three approaches to measuring our key causal concept. First, we include a dummy variable for the Large divided group countries and another for Homogeneous countries. We treat the omitted baseline category as Other multiethnic countries -that is, those with large undivided groups and those with only small groups. 44 Second, we construct a slightly different categorical variable called Large divided groups countries (50% cutoff) that treats 50%, not 40%, as the cutoff for a large group, since 50% is the value at which a group becomes a majority. Third, to avoid some of the arbitrariness of these binary classifications, we use the quantitative indicator we introduced above: weighted fragmentation within the largest group. In models with this indicator, we control for ethnic homogeneity and test the standard diversity-inhibitsdemocracy linear claim by including the fragmentation indices of Ethnolinguistic Fractionalization (ELF) or Politically Relevant Ethnic Groups (PREG). 45 In checking for the robustness of these various combinations of independent and dependent variables, we ran and report 16 different statistical models.
We also control for other potential causal factors that have emerged from the literature on democratization in Africa. We include measures of each country's electoral system to test the common claim that proportional representation is more representative of minorities and, thus, more likely to facilitate democratization than plurality rule. 46 While not all African countries have free and fair elections, virtually all of them do have elections and, thus, electoral systems. We code assembly electoral systems with a Plurality electoral system dummy variable, treating systems with proportional representation as the omitted base category. 47 We also control for politico-institutional history. Bratton and van de Walle claim that, even in Africa's post-independence authoritarian era (1960 -1990), meaningful political rules and routines were established that ultimately shaped whether a country democratized in the 1990s. 48 For example, they argue that countries that had some semblance of democracy within a long-ruling party were more likely to experience post-1990s transitions. For this, we include Years as a competitive one-party regime between independence and 1989. 49 We also control for who colonized each country. A small body of literature on Africa suggests that British colonization, because rule was more indirect and placed more emphasis on competitive elections, led to better post-independence outcomes. 50 We include two dummy variables -Former British colony and Former French colony -and treat colonies of other European powers as the omitted category. We control for average wellbeing with GNI per capita (Atlas Method) to capture standard modernization theory arguments. 51 Finally, we also control for Years since independence, since temporal distance from colonialism may allow for more national unity or unity within divided higher-level groups, and Population size, since smaller countries tend to be more ethnopolitically homogenous.
A word on our choice of time period is also in order. We seek to explain which African countries have been participants in the third wave of democratization. This means that we limit our analysis to 1990 and onwards. 52 Of course, this is where most of the action is, since the region has had, depending on the dataset, roughly twice as many transitions in nearly thrice as many countries since 1989 as it had in the longer post-independence period before 1989. More importantly, looking only at post-1989 country years protects us from charges of endogeneity -that is, the assertion that ethnopolitical configurations are caused by regime types, instead of the reverse. 53 It is certainly not the case that post-1990 transitions are causing LDG ethnopolitical configurations to emerge. Aside from the fact that our ethnopolitical data is based on sources from the late 1980s or early 1990s, most of Africa's contemporary ethnopolitical identities have roots in linguistic and territorial differences that date back generations, and the politicization of groups based on these markers occurred primarily during the late colonial and early post-independence periods. 54 That said, an endogeneity problem could lie in the possibility that the presence of democracy before 1990 caused both the emergence of LDG ethnopolitical configurations and the eventual post-1990 transitions to democracy that have tended to occur in LDG countries. While theoretically plausible, it is not empirically supported. For example, the few countries that had democratic spells before 1990 include ongoing autocracies and non-LDG countries like Republic of Congo, Nigeria, Somalia, Sudan, and Uganda. 55 Regardless, we include Number of democratic years before 1990 as a control variable to account for this possible alternative explanation. 56 Table 3 reports the most important results of 16 statistical models. 57 To reiterate, each model estimates two sets of coefficients. One set, reported in Table 3, is effects on democratic emergence, our main consequence of interest. The other set is the list of effects on democratic survival, which we report and discuss only in the Online Appendix, since our primary focus is on democratization. 58 The coefficients that test our hypotheses regarding LDG societies are shaded in grey. The results in Table 3 lend strong support to the argument that LDG countries have been more likely to experience transitions to democracy than other types of multiethnic societies in sub-Saharan Africa. Regardless of how we measured democracy or ethnopolitical demography, coefficients on our measures of LDG are statistically significant in explaining democratic emergence: all 16 hypothesis tests yield positive results. Although the precise estimates of substantive effects vary for each model, the predicted probabilities of a democratic transition occurring in a given year in a non-LDG, multiethnic autocracy (one with average measured traits) vary between .003 and .015. In contrast, the predicted probabilities in an LDG autocracy with otherwise equivalent measured traits vary between .06 and .15. In general, the models predicted a transition to democracy to be 5 -12 times more likely in LDG autocracies than in other multiethnic ones. The results also show that it is not ethnic fractionalization per se, as measured by ELF or PREG, that causes transitions to democracy, but the extent to which smaller groups are nested within the largest group. While the coefficients on ELF and PREG are almost always in the diversity-inhibits-democracy claim's expected (negative) direction, they are never statistically significant. Moreover, using our dichotomous measure of homogeneous societies, ethnic homogeneity does seem to contribute to democratic emergence, yet even this finding is less robust than the LDG one.

Model results
Relative to the control variables, our measures of the LDG pattern have a far more robust effect on democratic emergence. A few control variables, such as Number of years as a democracy before 1990 and British colonial possession, make appearances as statistically significant predictors of transition, but these positive findings are highly dependent on which measures of democracy one uses. In the end, we are confident not only that ethnopolitical demography matters for democratic emergence, but that it is more important than the rival explanations.

Conclusion
Many observers still attribute Africa's ills to multiethnicity. However, in this paper, we pinpointed -amidst the continent's diverse multiethnic palette -the particular arrangement that foments democracy. Countries with a large ethnopolitical group that is divided into identifiable ethnopolitical subgroups have sprouted third wave democracies and sustained some of them at higher rates than both other types of multiethnic countries. Countries with a large group can overcome the strategic coordination problems that often plague attempts to establish peaceful democratic competition, but when this group is also divided into nested subgroups, those outside the large group are less fearful of being victimized by Africa's all-too-common temptation of permanent exclusion.
The strength of our findings for the African case leads us to suspect that the advantages of the LDG pattern for democracy may hold elsewhere. With the emergence of democracy in some of Asia's most ethnically diverse states, some scholars have cast doubt on the notion that ethnic demography holds relevance for regime type. 59 However, some of Asia's democratic success stories seem to feature an LDG pattern. For example, the Hindu majority in democratic India is subdivided by caste, and political alliances between lower-caste Hindus and members of the Muslim majority (as our theory would predict) are not uncommon. 60 In democratic Indonesia, the Javanese near-majority ( 40%) is similarly subdivided by religion, and highly diverse Papua New Guinea has a large higher-level group of Melanesians. 61 In the end, the extent to which LDG societies support democracy elsewhere will only be revealed when scholars analyse ethnicity at different levels of inclusiveness beyond Africa. of ethnic groups. However, Posner codes groups relevant for economic policy and thus fails to recognize the nestedness of identities. 34. Morrison et al.,Black Africa. 35. Gurr,Minorities at Risk. 36. Habyarimana et al.,Coethnicity;McLaughlin,"Beyond Racial Census";Posner,ing Ethnic Fractionalization," 852; Fearon, "Ethnic Structure and Cultural Diversity." 37. An exception is Ferree, who exploits the multiple levels in the SM data in two published papers. Ferree, "How Fluid is Fluid?"; Ferree, "The Social Origins of Electoral Volatility." 38. Homogeneous societies also have values of zero for this reason. 39. This has become the standard statistical approach in the literature. See Przeworski and Limongi, "Modernization"; Boix and Stokes, "Endogenous Democratization"; Houle, "Inequality and Democracy." That said, to exploit the added information in the ordinal measures of democracy created by Freedom House and Polity, earlier versions of this paper reported the results of ordinary least squares regressions in which the dependent variable was each country's average democracy score (as used in Table 2 and Figure  1). These always returned a strong and statistically significant relationship between our measures of large-divided-group countries and democracy. 40. More specifically, we parsed each independent variable (except the lagged dependent variable) into two variables. One is the original variable recoded to zero for all years in which the lagged dependent variable is zero; the coefficient on this variable estimates its effect on democratic emergence. The other is the original variable recoded to zero for all years in which the lagged dependent variable is one; its coefficient estimates the covariate's effect on survival. This allows for a much more straightforward reading of effect estimates and their standard errors than the (mathematically equivalent) "slope dummy" approach occasionally used in this literature (see Boix  57. All models were estimated using multiple imputation techniques for missing data. King et al., "Analyzing Incomplete Political Science Data"; Royston, "Multiple Imputation of Missing Values." 58. Although split across two tables, readers should keep in mind that coefficients falling under the same column number in the two different tables were estimated jointly in the same model. 59. Hardgrave, "India: Dilemmas of Diversity"; Reilly, "Democracy, Ethnic Fragmentation, and Internal Conflict." 60. Osella and Osella, Social Mobility in Kerala. 61. Geertz, The Religion of Java.