Federalism,the Geographic Location of Groups,and Conflict

Abstract

In the literature on civil conflicts, federalism is often touted as a useful institution to address regional demands. However, diversity in the groups present in a country is also associated with a higher tendency for conflicts. In this article we examine how the geographic distribution of groups across a country affects the ways in which federalism contributes to conflict resolution. Of tantamount importance in assessing these effects of federalism is whether particular types of distributions of groups across a territory make the adoption of federal institutions more likely. We find federal countries with strong ethno-federal arrangements to be particularly conflict-prone.

Keywords

civil war conflict ethno-federalism federalism minorities

1. Introduction

Federalism and regional autonomy frequently appear among the proposed remedies for societal conflict (e.g. Nordlinger, 1972; Rabushka and Shepsle, 1972; Cohen, 1997; Congleton, 2000; Gurr, 2000; Bermeo, 2002; McGarry and O’Leary, 2005). Such arrangements would grant disgruntled minorities at the national level autonomy to run their affairs at the regional level. As scholars of federalism have noted for some time, however, such arrangements are quite delicate (e.g. Riker, 1964; Lemco, 1991; Watts, 1998; Filippov et al., 2004; Bednar, 2008), and many federal systems have collapsed (see e.g. Watts, 1998; Stepan, 1999). Hence, recent research has focused more heavily on the question what types of federal arrangements are more conflict-prone, and what characteristics may make them more stable (e.g. Bunce, 2003a; Hale, 2004; Bunce and Watts, 2005; Bakke and Wibbels, 2006; Brancati, 2006, 2009). At the center of this debate is the question whether ethnofederalism, that is, a federal system where regional borders follow as closely as possible settlement patterns of minorities, is advisable or not.

Assessing whether particular makeups of federal units lead to more or less conflict is, however, a difficult endeavor. First of all, detailed information on the composition of federal units and the geographic location of ethnic groups is, especially in conflict-prone regions, difficult to come by. Second, the general makeup of the country is likely to influence the presence or absence of federalism and the type of federalism chosen. A nice illustration to this effect appears in Glazer (1983), who shows that in the United States after 1789 new states were admitted to the Union only once WASPs (white Anglo-Saxon protestants) formed a clear majority (see also McGarry and O’Leary, 2005). This aspect has been largely ignored in the literature so far. Hence, we propose to assess in this article whether federal arrangements in which subunits are designed so that nationwide minorities may be in a majority in some lower-level units, are more conflict-prone. More specifically, we argue that conflict-proneness increases with larger shares of minority-controlled subunits, but decreases when this share approximates one. To demonstrate this we rely on detailed information on the ethnic makeup of subnational units in all federal states and control for possible endogeneity biases in our empirical analyses.

In the next section we briefly review the literature on federalism, especially as it pertains to this institution’s potential to mitigate societal conflicts. In section 3 we discuss and present data on the ethnic configuration of federal countries. Preliminary analyses presented in this section suggest that the ethnic makeup of countries affects the presence and type of federal arrangements. In section 4 we assess whether particular types of ethnic makeups make conflict onset more likely in federal countries than in others. Section 5 concludes and charts future research.

2. The Debate on Federalism

Whether federalist arrangements or regional autonomy might prove an appropriate institutional means to mitigate societal conflicts is an increasingly hotly debated question (see the excellent reviews in Lake and Rothchild, 2005; Bakke and Wibbels, 2006). The literature dealing with this question relies more and more often on the general literature on federal arrangements that highlights the latter’s inherent instability (e.g. Riker, 1964; Lemco, 1991; Watts, 1998; Filippov et al., 2004; Bednar, 2008). When the potential benefits of federal arrangements are seen in this larger context, very quickly the question arises under what conditions federal arrangements are adopted. This problem of endogeneity of federal arrangements, which is already well discussed in Riker’s (1964) incisive work (but since then often forgotten), makes the results discussed in the literature on the effects of federalism (e.g. Cohen, 1997; Saideman et al., 2002) rather questionable. While there exist many case-based descriptive studies assessing why states become federal, systematic evidence, especially evidence obtained from theoretically guided research, is rare. Commonly it is assumed (e.g. Panizza, 1999) that large territories and/or populations or diverse populations make the adoption of a federal system more likely (see also Diaz-Cayeros, 2004; Elkins and Sides, 2007).

The idea that federalism is used to “hold together” diverse populations or to let them “come together” (Stepan, 1999) is closely related to the notion of ethnofederalism. The debate over whether such federal systems, where federal borders follow closely settlement patterns of minorities, increase societal conflict or help mitigate it, is a longstanding one in political science. Lipset (1963: 81) argues against federal units corresponding to the settlements of ethnic groups and is supported in this by, among others, Nordlinger (1972) (see also McGarry and O’Leary, 2005). More recently, Snyder (2000: 199ff) very forcefully argued against ethnofederalism, especially in the light of the experiences after the fall of the Iron Curtain.¹ His argument relies on the conception that ethnofederal arrangements were not forced on the communist regimes, but were used for strategic reasons by the communist leadership, and where they were adopted, a breakup was programmed.

More recently scholars have attempted to figure out under what conditions ethnofederal arrangements are more conflict-prone. Hale (2004), for instance, suggests that the existence of a core ethnic region may lead to the collapse of states and/or civil war.² Brancati (2006), on the other hand, argues that regional parties are detrimental to the stability of federal systems. Obviously, ethnofederal systems are likely to create structures fostering the development of regional parties. Bakke and Wibbels (2006), in their study of federal arrangements, look at fiscal decentralization, inequality, and the ethnic makeup of federal units and find interesting interaction effects when trying to explain conflicts. According to their study, ethnic heterogeneity, when combined with high income inequality, leads to more conflict, and the effect of fiscal decentralization depends on the economic development of the ethnic regions.

Bunce (2003b), based on her work on the breakup of ethnofederal states in Central and Eastern Europe, suggests that nationalist mobilization failed to lead to conflict in all cases (see also Bunce and Watts, 2005). Subsequently Bunce (2003a) proposes as explanation, derived from a comparison of a small number of cases, the bargaining style in ethnofederal states. Closely related is Roeder’s (2007) argument that segmental states lead to nationalism and secession. Even though Roeder (2007) distinguishes segmental states from federal units, the logic, namely that giving institutional means to minority groups may foster and harden identities, appears also, for instance, in the work by Horowitz (1985).

Clearly there are many arguments appearing in the literature concerning whether and under what circumstances ethnofederal arrangements might work. In the present article our focus will be on the following two main arguments. First, as for instance Bunce (2007) argues, ethnofederal arrangements strengthen ethnic identities, since at the subnational level ethnic collaboration is not required (see also Breton, 1964; Rothchild and Roeder, 2005). Second, as most prominently put forward by Horowitz (1985), federal arrangements that follow ethnic lines may allow for easier mobilization and thus foster violent conflict (see also Cornell, 2002; Roeder, 2007). Against these two arguments, other authors oppose the claim that ethnofederal arrangements allow appeasement of ethnic groups that are nationally a minority (e.g. Stepan, 1999; Bermeo, 2002). These arguments do not, however, distinguish between different types of ethnofederal arrangements. We propose to assess this by looking closely at the territorial arrangements. More precisely, we agree with the critics of ethnofederal arrangements that they may harden ethnic divisions, but argue that this depends on the share of federal units controlled by national minorities. If this share is small, ethnic divisions are likely to harden, but violent conflict will be less probable as minorities only profit from a small share of the regional institutions. As this share increases, the national minorities start to control a more sizeable share of the subnational institutions, thus yielding more resources for mobilization against the central government. If this share increases even further (i.e. beyond the 0.5 threshold), however, such shares essentially reflect situations where the national majority is either concentrated in a small number of large (and thus powerful) units, or dispersed over a large part of the national territory. In the former case, the large and powerful units controlled by the national majority are likely to deter actions by national minorities. In the latter case, however, in each federal unit interethnic relations increase in importance and, as Bunce (2007) would argue, make strengthening ethnic identities more difficult (see for instance Kymlicka, 1998). Thus we would expect in such situations a reduced conflict risk. Consequently, between the two extreme situations, that is, when the share of minority controlled federal units is about one-half, conflicts are most likely. This is due to the fact that this may be the result of two ethnic groups being separated in two federal units (e.g. Sudan or Czechoslovakia) or the national majority may face a series of minorities with their own federal units, as a consequence multiplying the likelihood of violent conflict.

To assess whether our claim is correct, we propose to consider the federal design in terms of subunits as an endogenous policy choice and assess the effect of this design while taking into account its endogenous character. Before attempting to address these problems, we need to know what we consider as federal arrangements.

3. Federalism and the Makeup of Subunits

When defining countries as federal, scholars often refer to Riker’s (1964: 11) rightly famous definition suggesting that “[a] constitution is federal if (1) two levels of government rule the same land and people, (2) each level has at least one area of action in which it is autonomous, and (3) there is some guarantee (even though merely a statement in the constitution) of the autonomy of each government in its own sphere.” It is clear, however, that this definition is difficult to operationalize, and thus it is not surprising that scholars often differ in their lists of federal countries. Bednar (2008: 21) convincingly argues that definitions relying on both structural features and “federal practices” are not very useful when attempting to assess the effect of federal arrangements. She pleads for a strictly structural definition with three necessary components, namely geopolitical division, independence, and direct governance. With the geopolitical divisions element, she requires that territorial units are “constitutionally recognized and may not unilaterally be abolished” (Bednar, 2008: 21) and thus rules out non-territorial arrangements. With independence it is assured that different levels of governments have “independent bases of authority”, while direct governance assures that “each citizen is [directly] governed by at least two authorities” (Bednar, 2008: 21). This clear-cut and crisp definition, which, despite its similarity with Riker’s (1964) definition, strengthens the importance of constitutional recognition of federal units, allows us to cut through the maze of different types of federal arrangements and competing lists of federal countries.

Bednar (2008) identifies 26 entities as having fulfilled all three of these criteria, at least for some time between 1990 and 2000, and thus classifies them as federal. The list of countries Bednar (2008) proposes comes very close to the list appearing in McGarry (2002). What McGarry (2002; McGarry and O’Leary 2005) provides in addition is a classification of the federal countries as national or multinational federations,³ as well as start and end dates of the federal arrangements. Similarly, Roeder (2009) provides a distinction between federations and ethnofederations that we also include in Table 1, which lists the countries we consider federal.

Table 1.

Federal countries

Country	Bednar (2008) ^a	McGarry (2002)	Roeder (2009) ^b
	National Federations		Federations
Argentina	−	1853–	1853–
Australia	−	1901–	1901–
Austria	−	1920–	1918–1938, 1945–
Brazil	−	1891–	1891–1934, 1946–
Germany	−	1949–	1867–1938, 1949–
Mexico	−	1917–	1917–
Venezuela	−	1960–	1864–
United Arab Emirates	−	1971–	1971–
USA	−	1789–	1789
	Multinational Federations		Ethnofederations
Belgium	1994–	1993–	1980–
Bosnia	1995–	1995–	1995–
Burma		1948–	1948–1962
Cameroon		1961–1972	1961–1972
Canada	−	1867–	1867–
Czechoslovakia	1990–1992	1968–1992	1969–1992
Ethiopia	1995–2000	1992–	1952–1962
India	−	1950–	1953–
Malaya		1957–1963	?
Malaysia	−	1963–	1963–1965
Mali		1960	F: 1960
Nigeria	1998–2000	1960–	1960–
Pakistan	1990–1999	1947–1971	1947–
Russia	1992–2000	1993–	1991–
St Kitts-Nevis		1983–	F 1983–
South Africa	1993–2000	1996–	A: 1972–1994
Soviet Union	1990–1991	1918–1991	1922–1991
Switzerland	−	1848–	F: 1848–
West Indies Federation		1958–1962	?
Yugoslavia (former)	1990–1991	1953–1992	1945–1992
Yugoslavia	1992–2000	1992–	1992–

Countries listed as federal by Bednar (2008) for the whole period between 1990 and 2000. We omit from her list the European Union, which she codes as federal for 1999–2000.

A: Countries listed by Roeder (2009) as having autonomous ethnic regions; F: countries listed as federations by Roeder. In addition, he lists Indonesia (1949–50), Tanzania (1964–) and Spain (1978–) as ethnofederations, which we omit, given that neither Bednar (2008) nor McGarry (2002) lists these cases.

Table 1 contains two additional pieces of information. First, we provide information on the time period for which Bednar (2008: 26) and McGarry (2002) consider these countries as federal. Second, from McGarry (2002) we also employ, as a first indication of the ethnic makeup of federal units, the distinction between national and multinational federations.⁴

As Table 1 illustrates, there are very few contradictions between the two sets of authors. Omitting small differences in timing (e.g. Belgium becoming federal in 1993 or 1994⁵) the only contradictions concern Burma, St Kitts-Nevis, and Micronesia. The first two countries are coded by McGarry (2002) as federal but are considered as quasi-federal and unitary, respectively, by Bednar (2008: 26), while for Micronesia the exact opposite is the case.⁶

Having at hand a list of federal countries, a first question to address is what distinguishes these countries from other countries in terms of the overall ethnic configuration. First, if we just consider the ethnic composition of federal and unitary states, we find on the basis of Fearon’s (2002) dataset that the ethnic fractionalization is practically identical in federal and unitary states, being on average 0.485 compared to 0.476 (F-statistic 0.017, statistically not significant). When we distinguish between the two types of federations proposed by McGarry (2002), we find average ethnic fractionalization indices of 0.381 and 0.601 for the national and multinational federations. Despite the marked difference, and even when compared to the averages for unitary states, we fail to find a statistically significant relationship.

The result is slightly different if we look at the overall number of groups per country as reported in Fearon’s (2002) dataset. We find on average 5.144 groups per country in the 160 countries for which he provides this information. When distinguishing between federal and non-federal countries we find hardly any differences. In federal states there are on average 5.412 groups present, while this number is only slightly lower in unitary states, namely 5.112 (F-statistic 0.111, statistically not significant). However, when we distinguish between the two types of federal systems, namely national and multinational, differences become more marked. In national federations we find 3.556 groups on average, while this number increases to 7.500 in multinational federations (F-statistic 2.811, p = 0.063). These differences mask to some degree the distribution of the number of groups depicted in Figure 1. This figure suggests that unitary states have a very wide spectrum of numbers of groups, while national federations are very much concentrated on the left side of the graph, and, not surprisingly, multinational federations more on the right side of the graph. It is interesting to note that the highest number of groups appears in unitary countries.

Figure 1.

Number of groups in different types of federations

As noted above, some authors claim that the concentration of groups in particular areas leads to higher probabilities of conflict outbreak in general (e.g. Toft, 2003) and more particularly in ethnofederal countries (Saideman et al., 2002; Hale, 2004). For this reason, in Table 2 we depict the information on group concentration for each federal country, as collected by the Minorities At Risk project (Gurr, 1993), as well as the number of groups according to Fearon (2002).⁷ The country tables in the web appendix show the group concentration for all minorities at risk in the federal countries considered here, as well as information on their relative group size obtained from Fearon (2002) and Elkins and Sides (2007).

Table 2.

Group concentration and number of groups in federal countries^a

	Groupcon Groups Spatial Distribution (Fearon, 2002)
Country	0	1	2	3	N	N
	National Fedederations
Austria	−	−	−	−	0	4
Australia	1	0	0	0	1	3
Mexico	1	0	2	0	3	4
Venezuela	0	0	1	1	2	4
Brazil	0	0	0	2	2	3
Argentina	0	1	0	1	2	3
Germany	0	1	0	0	1	3
USA	2	0	1	1	4	4
	Multinational Federations
Belgium	−	−	−	−	0	4
Bosnia	0	1	1	1	3	3
Canada	2	0	0	1	3	6
Czechoslovakia	0	0	0	2	2	4
Ethiopia	0	0	2	5	7	11
India	1	1	3	5	10	12
Malaysia	0	1	1	5	7	9
Mali	1	1	0	2	4	5
Nigeria	0	0	1	4	5	7
Russia	2	2	6	17	27	5
South Africa	2	1	2	3	8	13
Switzerland	1	0	0	1	2	9
Yugoslavia	1	0	4	3	8	7

Austria and Belgium have no minorities at risk.

What is immediately transparent from Table 2 is that in all multinational federations with “minorities at risk”, some of the latter are concentrated in a region or form a majority in the region (codes 3 and 2, respectively, in Table 2). In national federations, however, having regionally concentrated “minorities at risk” is rare, with Venezuela and Brazil being the exceptions to the rule. Hence, following Toft’s (2003) argument we would expect multinational federations and federations with federal units populated mostly by a national minority to be much more conflict-prone.

By looking more closely at the distribution of groups across the various federal units, we may gain a much more detailed assessment of what types of federal arrangement are related to conflict. Table 3 shows this information for all federal states covered in our empirical analyses.⁸ We list in this table first the national and then the multinational federal states before indicating the number of federal units and in how many of these the national majority is a minority. After having assessed the effect of federalism we now turn to our main independent variable. We measure the degree to which a federal arrangement follows the ideas of ethnofederalism by using the proportion of federal units in which a national minority forms the majority group in the population.⁹ In the last three columns we list the national majority, the federal units controlled by a minority, and the minority concerned.

Table 3.

Federations and Dominant Majorities or Minorities Federations

	N Fed. Units	N Maj. / Min. States	Nationwide Dominant Group	Maj. / Min. Fed. Unit	Maj. / Min. Group
National Federations
Argentina	24	0	White (85–97%)	−	−
Sources: http://www.indec.mecon.gov.ar/ Table: 14.1 Total de hogares y hogares donde al menos una persona se reconoce perteneciente o descendiente de un pueblo indígena. Total del país según provincias
Australia	8	0	Caucasian (84%–92%)	−	−
Sources: Census 2001: http://www.abs.gov.au/websitedbs/d3310114.nsf/home/Previous\%20Censuses:\%20Census\%20data
Austria	9	0	Austrians (90–93%)	−	−
Sources: Population Estimations 2006: http://www.statistik.at/fachbereich\_03/bevoelkerung\_tab8.shtml
Brazil	27	17	White (54–55%)	Acre	Mulatto
				Alagoas	Mulatto
				Amap	Mulatto
				Amazonas	Mulatto
				Bahia	Mulatto
				Cear	Mulatto
				Maranho	Mulatto
				Mato Grosso	Mulatto
				Par	Mulatto
				Paraba	Mulatto
				Pernambuco	Mulatto
				Piau	Mulatto
				Rio Grande do Norte	Mulatto
				Rondnia	Mulatto
				Roraima	Mulatto
				Sergipe	Mulatto
				Tocantins	Mulatto
Sources: Census 2000: http://www.sidra.ibge.gov.br/ Table 136
Germany	16	0	Germans (91%–95%)	−	−
Sources: Estimations 2004 http://www.statistik-portal.de/Statistik-Portal/de\_jb01\_jahrtab2.asp
Mexico	32	0	Mestizo (60%)	Oaxaca, Chiapas and Yucatan	Strong Indigenous Minorities -
Sources: Census 2000: Instituto Nacional de Estadistica Geografia e informatica; http://www.inegi.gob.mx/est/contenidos/espanol/rutinas/ept.asp?t=mlen02\&c=3327; Library of Congress
United Arab Emirates	7	0	Ethnic Arabs (87%)	−	−
Sources: Library of Congress: Foreign workers who represent about 70% are not included in the calculation since they are not citizens of UAE
United States	51	3	White (67%–82%)	District of Columbia	Black
				Hawaii	Asian
				New Mexico	Hispanics
Sources: Population Division, U.S. Census Bureau: Annual Estimates of the Population by Race Alone and Hispanic or Latino Origin for the United States and States: July 1, 2004 (SC-EST2004-04). The Census Bureau provides statistics for non Hispanic or latino origins and Hispanic and latino origins. All "white" hispano have been classified as hispanics. Other groups have been classified according to their group. E.g. Hispanics Asian alone have been classified as Asian.
Venezuela	25	2	Mestizo (59%–68%)	Delat Amacuro	White
				Monagas	White
Sources: World Value Survey 2000; http://www.ine.gov.ve/censo/fichascenso/fichacenso.asp
Multinational Federations
Belgium	13	7	Flemming (58%)	Brabant Wallon	Wallon (French)
				Brussels	Wallon (French)
				Hainaut	Wallon (French)
				Liege	Wallon (French)
				Luxembourg	Wallon (French)
				Namur	Wallon (French)
				Wallonia	Wallon (French)
Sources: World Value Survey 1999 and various official internet pages
Bosnia	3	1	Bosniak (Muslims) (44%–48%)	Republika Srpska	Serbs
Sources: Estimations because the last census was held in 1991. Brsko District seems to be equally populated by Bosniaks and Serbs (about 40% for each group). http://en.wikipedia.org/wiki/Republika\_Srpska. http://en.wikipedia.org/wiki/Bosnia\_and\_Herzegovina
Burma (Myanmar)	14	7	Burman (68%)	Chin State	Chin
				Kachin State	Kachin
				Kayah State	Kayah
				Kayin State	Kayin
				Mon State	Mon
				Rakhine	Rakhine
				Shan State	Shan
Sources: Reported results based on the last census (1983). All states are dominated by a minority group.
Cameroon	10	9	Bamilike-Central (21%)	Adamoua	Peulh and Gbaya
				Centre	Beti
				East	Meka and Gbaya
				Extreme North	Mboum, Mafa and Peulh
				North	Mboum, Guidar, Fali and Mafa
				North West	Ring, Ngemba, Wimbum-Yamba and Momo
				South	Boulou-Fang
				South West	Ngoe-Oroko
Sources: DHS Survey 2004; http://www.unesco.org/most/dpzognong.htm and http://www.ethnonet-africa.org/data/camer/genpop.htm
Canada	13	1	English speaker	Quebec	Quebecois (French)
Sources: Ottawa: Statistics Canada – Language Composition of Canada Cat. No. 97F0007XCB2001001
Czechoslovakia	2	1	Czech (63%)	Slovak Republic	Slovaks
Sources: (Alemán & Treisman, 2005)
Ethiopia	11	6	Oromo (35%)	Addis	Amharra
				Amhara	Amharra
				Affar	Affar
				Somali	Somalie
				Tigray	Tigray (Tigraway)
				Southern Nations, Nationalities and Peoples	Guragie, Sidama and many small groups
				States where Oromo is as strong as any other minority:
				Benishangul-Gumaz	Amharra (28%); Oromo (25%) and Beni-Snagul-Gumuz (? %)
				Gambela	Amharra (18%), Oromo (10%) and Gambela (? %)
				Dire Dawa	Oromo (42%); Amharra (39%)
Sources: DHS Survey 2000; http://lcweb2.loc.gov/frd/cs/ethiopia/et02\_04a.pdf; http://commons.wikimedia.org/wiki/Image:Ethiopia\_regions\_english.png; http://www.csa.gov.et/
India	35	21	Religion: Hindus (81%)		Ethnic (language) and Religious majority:
			Language: Hindi (40%)	Jammu and Kashmir	Kashmiri and Dogri (in Jammu) / Muslims
				Lakshadweep	Malayalam / Muslims
				Punjab	Punjabi / Sikhs
				Meghalaya	Khasi / Christian
				Mizoram	Lushai/Mizo / Christian
				Nagaland	Ao, Sema and Konyak / Christian
			Hindi (40%)		Ethnic (language) minority:
				Assam	Assamese
				Dadra and Nagar Haveli	Bhili/Nhilodi
				Daman and Diu	Gujarati
				Goa	Konkani
				Gujarat	Gujarati
				Karnataka	Kannada
				Kerala	Malayalam
				Maharashtra	Marathi
				Manipur	Manipuri
				Orissa	Oriya
				Pondicherry	Tamil
				Sikkim	Nepali
				Tamil Nadu	Tamil
				Tripura	Bengali
				West Bengal	Bengali
Sources: Census 2001 Religious Data http://www.censusindia.net/religiondata/index.html; Census 1991 Three Main Languages in every State 1991, Table 26 http://www.censusindia.net/cendat/datatable26.html and http://www.kashmirstudygroup.net/awayforward/mapsexplan/languages.html. The Kashmir, Tamil, and Sikh minorities are the most militant, along religious cleavages (Kohli, 2004)
Malaya (1957–63)	11	1	Malay	Penang (Pinang)	Chinese
Malaysia (1963–65)	14	4	Malay	Penang	Chinese
				Sabah	Kadazan-Dusun
				Sarawak	Iban
				Singapore	Chinese
Note: August 7, 1965: Singapore is expelled from the Malaysian Federation
Malaysia (1966–)	13	3	Malay	Penang	Chinese
				Sabah	Kadazan-Dusun
				Sarawak	Iban
Sources: http://www.statistics.gov.my/english/frameset\_census.php?file=pressdemo ; Wikipedia (see Table for Malaysia for more details). We do not include the 3 Federal Territories of Malaysia; Kuala Lumpur (Capital, 1972), Labuan (Offshore territory, 1984) and Putrajaya (Administrative centre, 1995). All are dominated by Malays and are governed by the Ministry of Federal Territory.
Mali	8	5	Bambara (34%)	Kayes	Sarakole, Soninke, Marka and Malinke
				Mopti	Dogon and Peulh
				Tombouctou	Sonrai
				Gao	Sonrai
				Kidal	Tamacheck
Sources: DHS Survey 1999 and http://region.segou.net/Donn\%E9es\%20sur\%20le\%20Mali.htm
Nigeria	37	24	Hausa-Fulani (23%)	Kogi	Yoruba, Igbala and Ibira
				Ekiti	Yoruba
				Kwara	Yoruba
				Lagos	Yoruba
				Ogun	Yoruba
				Ondo	Yoruba
				Osun	Yoruba
				Oyo	Yoruba
				Taraba	Wurkum, Mumuje and many small groups
				Plateau	Tera, Hausa, Birom and many small groups
				Niger	Nupe
				Adamawa	Many groupes (81%)
				Bayelsa	Ijaw
				Delta	Igbo, Urhobo and Ijaw
				Abia	Igbo
				Anambra	Igbo
				Ebonyi	Igbo
				Enugu	Igbo
				Imo	Igbo
				Benue	Idoma and Tiv
				Akwa Ibom	Ibidio
				Cross River	Ekoi and many small groups
				Edo	Edo
				Borno	Bura, Kanuri and many other groups
Sources: DHS Survey 1999
Pakistan	8	3	Punjabi (44%)	Sindh	Sindhis and Urdu
				North-West Frontier Province (NWFP)	Pakhtuns (Pathan)
				Baluchistan	Brohi, Pakthuns and Baluchi
Sources: Alemán & Treisman (2005), World Value Survey 1999 and DHS Survey 1990/1991
Russia	88	12	Russian (80%)	Chechen Republic	Chechens
				Chuvashi Republic	Chuvash
				Republic of Dagestan	Avars
				Republic of Ingushetia	Ingushs
				Kabardino-Balkarian Republic	Kabardians
				Republic of Kalmykia	Kalmyks
				Karachaevo-Chercessian Republic	Karachais
				Republic of North Ossetia Alania	Osetians
				Republic of Sakha (Yakutia)	Yakuty
				Republic of Tatarstan	Tatars
				Republic of Tuva	Tuvinians
				Aginsky Buryat autonomous district	Buriats
Sources: Census 2002; National Composition for Regions of the Russian Federation: http://www.perepis2002.ru/ct/doc/English/4-2.xls
South Africa	9	2	Black African (79%)	Northern Cape	Coloured
				Western Cape	Coloured
Sources: South African Census 2001 by Province and Population group: http://www.statssa.gov.za/extract.htm ; DHS Survey 1998
Soviet Union	15	14	Russian (51%)	Armenian Republic	Armenians
				Azerbaydazhan Republic	Azeris
				Belorussian Republic	Belarusans
				Estonian Republic	Estonians
				Georgian Republic	Georgians
				Kazakh Republic	Kazakhs
				Kirgiz Republic	Kyrkyz
				Latvian Republic	Latvians
				Lithuanian Republic	Lithuanians
				Moldavian Republic	Moldavians
				Tadzhik Republic	Tajiks
				Turkmen Republic	Turkmen
				Ukrainian Republic	Ukrainians
				Uzbek Republic	Uzbeks
Sources: Library of Congress
Switzerland	26	7	German (62%–65%)	Fribourg	Romand (French)
				Geneva	Romand (French)
				Jura	Romand (French)
				Neuchatel	Romand (French)
				Tessin	Ticinese (Italian)
				Valais	Romand (French)
				Vaud	Romand (French)
Sources: Census 2000; http://www.bfs.admin.ch/bfs/portal/fr/index/themen/bevoelkerung/sprachen\_\_religionen/blank/kennzahlen0/sprachen.html
Former Yugoslavia (1991)	6	5	Serbs (36%, 1991)	Bosnia-Herzegovina	Muslims (Bosniaks)
				Croatia	Croats
				Macedonia	Macedonians
				Montenegro	Montenegrins
				Slovenia	Slovenes
Sources: Woodward (1995: 32–35, Table 2)
Yugoslavia (Serbia and Montenegro) (1992–2006)	2	1	Serbs	Montenegro	Montenegrins

We might again relate this information to Toft’s (2003) argument that concentrated groups are most conflict-prone. Taking such cursory evidence as an indication that multinational federations (or ethnofederalism) is something to avoid, is, however, a step too far. What many defenders of federal arrangements in general and multinational federations in particular argue is that even though multinational federations may lead to conflict or state collapse, the necessary counterfactual would be to look at what would have happened if the country had not been federal (e.g. McGarry and O’Leary, 2005). This obviously suggests (again) that federal arrangements are chosen in particular situations to address a set of problems. Hence, empirical analyses trying to assess the effect of federalism and its specific form need to take into account the endogenous nature of these institutional choices. Given the important role the numbers of ethnic groups and “minorities at risk” appear to play when comparing federal and non-federal systems, we will employ these variables in our strategy to address the endogenous nature of federalism. In addition we also include the colonial tradition—more precisely the British, Dutch, French, and US traditions—as an instrumental variable.¹⁰ The data for this variable stem from the Quality of Government data (Teorell et al., 2008) and are based on Hadenius and Teorell’s (2007) work.

Clearly the colonial heritage of a country is exogenous to current conflict processes, but as Brancati (2007: 661) convincingly argues, it has clearly affected the territorial institutions. The number of ethnic groups and minorities may not be equally exogenous, but since we have more instruments than instrumented variables (i.e. a situation of overidentification) this is less of a problem.

4. The Geographic Location of Groups and its Effect on Conflict

To assess the effect of the ethnic makeup of federal countries, we employ as a starting point Fearon and Laitin’s (2003) analysis of the conditions leading to civil war, but focus only on the onset of ethnic civil wars as coded by these authors. We first analyze the effect of federal arrangements.¹¹ In Table 4 we first report a simple analysis which replicates Fearon and Laitin’s (2003) analysis but we estimate it as a probit model and add a federalism indicator.¹² The coefficient for federalism suggests that such arrangements marginally decrease the probability of ethnic conflict onset. To assess whether this indicator suffers from endogeneity bias different strategies are possible (e.g. Wilde, 2000; Monfardini and Radice, 2008; Freedman and Sekhon, 2010). For simplicity’s sake we estimate a bivariate probit model in which the federalism indicator appears in one equation as a dependent and in another as an independent variable. A test of the hypotheses that the correlation between the two error terms of the bivariate probit model is different from zero is equivalent to a test of endogeneity (see Wilde, 2000). To predict the presence of federalism we use as instruments the number of ethnic groups, minorities at risk, and the colonial tradition. In this approach it is assumed that these instrumental variables are exogenous and can be used to predict the presence of federal arrangements. Our analyses presented in the previous section seem to suggest that the ethnic makeup of a society is clearly linked to these presumably exogenous variables. Following Acemoglu (2005) we also include all other exogenous variables in this auxiliary regression. As a consequence, other variables associated with the presence of federalism are used in this auxiliary regression, such as those employed by Panizza (1999), Elkins and Sides (2007), and Diaz-Cayeros (2004).¹³

Table 4.

The Effect of Federalism and its Form on Ethnic Civil War Onset (with Robust Standard Errors)

	1	2		3	4	5
	probit	bivariate probit		probit	probit	probit
	onset	onset	federalism	onset	onset	onset
	b	b	b	b	b	b
	(r.s.e.)	(r.s.e.)	(r.s.e.)	(r.s.e.)	(r.s.e.)	(r.s.e.)
Federalism	−0.299	−0.133		−0.236	−1.064	−0.854
	(0.165)	(0.258)		(0.288)	(0.572)	(0.318)
Proportion of minority				−0.361	4.537	3.295
federal units / (pred.)				(1.078)	(2.881)	(1.536)
Proportion of minority				0.340	−3.400	−2.289
federal units / (pred.)				(0.995)	(2.482)	(1.379)
Proportion of minority					−3.280
federal units (res.)					(1.646)
Proportion of minority					1.809
federal units (res.)					(1.829)
Prior war	−0.389	−0.382	−0.988	−0.389	−0.337	−0.350
	(0.149)	(0.151)	(0.467)	(0.151)	(0.153)	(0.151)
Per capita income	−0.123	−0.130	0.107	−0.126	−0.118	−0.124
	(0.030)	(0.032)	(0.034)	(0.032)	(0.037)	(0.036)
log(population)	0.177	0.159	0.394	0.178	0.154	0.159
	(0.035)	(0.040)	(0.123)	(0.037)	(0.041)	(0.040)
log(mountainous terrain)	0.064	0.059	0.156	0.063	0.072	0.068
	(0.044)	(0.044)	(0.132)	(0.045)	(0.042)	(0.041)
Noncontiguous state	0.179	0.207	0.044	0.180	0.119	0.136
	(0.115)	(0.120)	(0.417)	(0.116)	(0.153)	(0.149)
Oil exporter	0.420	0.423	0.098	0.421	0.383	0.388
	(0.125)	(0.128)	(0.409)	(0.125)	(0.146)	(0.146)
New state	0.779	0.781	0.521	0.777	0.800	0.797
	(0.180)	(0.181)	(0.202)	(0.178)	(0.181)	(0.181)
Instability	0.167	0.183	0.218	0.168	0.158	0.163
	(0.106)	(0.107)	(0.167)	(0.106)	(0.125)	(0.125)
Democracy (Polity)	0.011	0.010	0.011	0.011	0.011	0.012
	(0.008)	(0.025)	(0.005)	(0.008)	(0.009)	(0.009)
Ethnic fractionalization	0.358	0.322	−0.656	0.361	0.279	0.288
	(0.220)	(0.230)	(0.589)	(0.222)	(0.195)	(0.193)
Religious fractionalization	0.653	0.599	2.025	0.654	0.556	0.564
	(0.300)	(0.309)	(0.830)	(0.300)	(0.260)	(0.259)
Number of ethnic groups			0.242(0.109)
Number of minorities at risk			0.088
		(0.057)
Dutch colony			0.673
		(0.244)
US colony		0.062
		(0.117)
British colony		0.075
		(0.101)
French colony		0.267
		(0.080)
constant	−4.304	−4.106	−7.223	−4.309	−4.067	−4.104
	(0.399)	(0.476)	(1.374)	(0.426)	(0.420)	(0.413)
ρ		−0.177
		−0.154
Log-likelihood	−337.917		−1415.445	−337.890	−335.343	−335.538
N	6327		6327	6327	6327	6327

The results of the bivariate probit model suggest a rather reduced effect for the presence of federalism on the onset of ethnic civil wars. The estimated coefficient remains negative but becomes statistically insignificant. At the same time the correlation between the two error terms (r) is small and cannot be distinguished from zero on statistical grounds, suggesting that with the instrumental variables employed we find no evidence for endogeneity. We also enter this variable squared to allow for the hypothesized curvilinear effect. The estimated coefficients reported for Model 3 suggest that federal states with either a small or a large share of minority controlled units are the most conflict-prone, contrary to our hypothesized effect. The two estimated coefficients are, however, far from reaching statistical significance. Again suspecting possible endogeneity bias, we instrument this variable with the number of ethnic groups and minorities at risk, as well as the number of federal units. Clearly as the number of federal units increases, the likelihood that one is dominated by a national minority increases, hence this variable is a useful instrument. We also believe that as such it is largely exogenous to conflict. Governments may react to or anticipate conflicts and adjust the number of federal units, but this will depend on the number of relevant groups. The latter variables again, as noted above, are more likely not to be completely exogenous, but given that our model is overidentified regarding instrumental variables, this should be less problematic.

In Table 5 (Columns 1 and 2) we report the results of the auxiliary regressions, both for the proportion of minority controlled federal units and its square.¹⁴

Table 5.

Explaining the Proportion of Minority-Majority Federal Units (Auxiliary Regressions)

	OLS		Heckman (twostep)
			outcome	selection	outcome	selection
	X	X	X	X	X	X
	b	b	b	b	b	b
	(s.e.)	(s.e.)	(s.e.)	(s.e.)	(s.e.)	(s.e.)
Proportion of minority		1.945			1.159
federal units (pred)		(0.070)			(0.077)
Prior war	−0.029	−0.012	−0.250	−0.987	0.160	−0.987
	(0.004)	(0.003)	(0.036)	(0.107)	(0.033)	(0.107)
Per capita income	−0.005	−0.003	−0.014	0.107	−0.006	0.107
	(0.000)	(0.000)	(0.002)	(0.006)	(0.001)	(0.006)
log(population)	0.009	0.005	0.045	0.394	−0.007	0.394
	(0.001)	(0.001)	(0.009)	(0.028)	(0.007)	(0.028)
log(mountainous terrain)	−0.008	−0.005	−0.055	0.156	0.000	0.156
	(0.001)	(0.001)	(0.007)	(0.028)	(0.006)	(0.028)
Noncontiguous state	0.005	−0.002	−0.096	0.045	−0.039	0.045
	(0.004)	(0.003)	(0.020)	(0.092)	(0.015)	(0.092)
Oil exporter	0.013	0.002	0.008	0.100	−0.096	0.100
	(0.004)	(0.003)	(0.022)	(0.102)	(0.017)	(0.102)
New state	−0.002	0.003	0.057	0.516	−0.063	0.516
	(0.008)	(0.006)	(0.056)	(0.206)	(0.039)	(0.206)
Instability	0.005	0.001	0.009	0.216	−0.042	0.216
	(0.004)	(0.003)	(0.021)	(0.092)	(0.015)	(0.092)
Democracy Polity	0.000	−0.001	−0.008	0.011	0.000	0.011
	(0.000)	(0.000)	(0.001)	(0.005)	(0.001)	(0.005)
Ethnic fractionalization	0.025	0.011	0.336	−0.658	0.069	−0.658
	(0.005)	(0.004)	(0.041)	(0.148)	(0.032)	(0.148)
Religious fractionalization	0.029	0.012	0.177	2.027	0.029	2.027
	(0.006)	(0.004)	(0.047)	(0.187)	(0.033)	(0.187)
Number of ethnic groups	−0.007	0.001	0.010	0.243	−0.027	0.243
	(0.001)	(0.000)	(0.006)	(0.017)	(0.004)	(0.017)
Number of minorities at risk	0.026	0.017	0.017	0.087	−0.002	0.087
	(0.001)	(0.007)	(0.002)	(0.019)	(0.002)	(0.019)
Number of federal units	−0.003	0.017	−0.001		0.000
	(0.000)	(0.001)	(0.001)		(0.000)
Federalism	0.243	−0.029
	(0.006)	(0.002)
Dutch colony				0.674		0.674
				(0.054)		(0.054)
US colony				0.062		0.062
				(0.019)		(0.019)
Britsh				0.075		0.075
				(0.024)		(0.024)
French				0.266		0.266
				(0.027)		(0.027)
Constant	−0.066	−0.036	−0.227	−7.226	0.280	−7.226
	(0.010)	(0.007)	(0.120)	(0.329)	(0.089)	(0.329)
Adj R-squared	0.597
Root mean squared error	0.099
ρ				−0.003		−0.570
n	6327	6327	959	6327	959	6327

Adding in Model 4 the residuals based on the results of this auxiliary regression suggests a moderate problem of endogeneity (a likelihood ratio test generates a p-value close to 0.1). Given this, in Model 5 we replace the proportion of federal units dominated by a national minority (as well as its square) by its predicted values from the auxiliary regression.¹⁵ The results suggest that increasing this proportion has a positive (though statistically not significant) effect on the onset of civil wars, but this effect decreases when this share exceeds approximately three-quarters.

While this first analysis already suggests that if ethnofederal arrangements have an effect it is more likely to increase the likelihood of conflict onset, the analysis also has an important shortcoming—in considering the effect of the proportion of minority controlled federal units we also include all unitary states, for which this main independent variable is 0. Hence, even when controlling for the presence of a federal arrangement, the coefficient for our main independent variable conflates both the effect of federal arrangements per se and the effect of the ethnic makeup of the federal units. A quick way to resolve this problem is presented in Alemán and Treisman (2005) and Bakke and Wibbels (2006): consider only federal states. The problem with this solution is obviously that federal institutions are often chosen with particular goals in mind. Hence, instead of having an endogeneity problem, focusing only on federal states creates a potential selection bias problem.

To address these two problems, we use in what follows a Heckman (1976) selection model in a probit setup (see Dubin and Rivers, 1990), where the selection corresponds to the choice (and maintenance) of a federal system. The results for this second approach to assess the effect of ethnofederal arrangements appear in Table 6.

Table 6.

Explaining Ethnic Civil War Onset in Federal Countries: Heckman Probit Selection Model (with Robust Standard Errors)

	outcome	selection	outcome	selection	outcome	selection
	b	b	b	b	b	b
	(r.s.e.)	(r.s.e.)	(r.s.e.)	(r.s.e.)	(r.s.e.)	(r.s.e.)
Proportion of minority	−0.259		3.314		11.151
federal units / pred.	(4.223)		(2.916)		(2.483)
Proportion of minority	−0.108		−2.839		−7.978
federal units / pred.	(3.858)		(2.683)		(1.969)
Proportion of minority			−10.737
federal units res.			(2.623)
Proportion of minority			10.794
federal units res.			(2.574)
Prior war	−1.010	−0.987	−1.280	−0.986	−0.460	−0.986
	(0.417)	(0.468)	(0.365)	(0.394)	(0.364)	(0.468)
Per capita income	−0.297	0.107	−0.524	0.107	−0.331	0.107
	(0.189)	(0.034)	(0.237)	(0.022)	(0.233)	(0.034)
log(population)	0.405	0.393	0.811	0.394	0.273	0.394
	(0.413)	(0.123)	(0.199)	(0.088)	(0.253)	(0.123)
log(mountainous terrain)	0.414	0.155	0.027	0.156	0.255	0.156
	(0.179)	(0.132)	(0.211)	(0.100)	(0.168)	(0.132)
Noncontiguous state	0.274	0.042	0.573	0.045	0.557	0.046
	(0.403)	(0.416)	(0.428)	(0.359)	(0.589)	(0.417)
Oil exporter	0.590	0.099	−0.211	0.097	−0.126	0.097
	(0.315)	(0.410)	(0.359)	(0.375)	(0.456)	(0.410)
New state	0.858	0.509	0.434	0.520	0.406	0.522
	(0.403)	(0.202)	(0.471)	(0.177)	(0.453)	(0.201)
Instability	0.309	0.221	0.124	0.210	0.043	0.210
	(0.276)	(0.167)	(0.428)	(0.159)	(0.379)	(0.166)
Democracy Polity	0.010	0.010	−0.052	0.011	−0.014	0.011
	(0.023)	(0.025)	(0.051)	(0.017)	(0.044)	(0.025)
Ethnic fractionalization	−0.911	−0.657	−1.435	−0.662	−2.030	−0.660
	(0.952)	(0.588)	(1.160)	(0.558)	(1.073)	(0.587)
Religious fractionalization	1.724	2.023	2.990	2.033	−0.099	2.031
	(1.602)	(0.828)	(1.672)	(0.648)	(1.887)	(0.831)
Number of ethnic groups		0.243		0.243		0.243
		(0.109)		(0.044)		(0.108)
Number of “minorities at risk”		0.088		0.087		0.087
		(0.059)		(0.053)		(0.057)
Dutch colony		0.669)		(0.676)		(0.676)
		(0.245)		(0.217)		(0.244)
US colony		0.060		0.063		0.063
		(0.117)		(0.078)		(0.117)
British colony		0.074		0.075		0.075
		(0.101)		(0.083)		(0.102)
French colony		0.263		0.267		0.267
		(0.083)		(0.069)		(0.080)
Constant	−6.845	−7.203	−11.159	−7.226	−6.743	−7.225
	(4.203)	(1.371)	(2.578)	(1.096)	(3.221)	(1.373)
ρ		−0.611		0.787		0.680
		(0.231)		(0.201)		(0.406)
log-likelihood		−1112.010		−1106.585		−1107.491
n	959	6327	959	6327	959	6327

In the first column of Table 6 we report the results of the first Heckman model that estimates simultaneously a selection equation, namely whether a country is federal, and an outcome equation assessing whether these federal countries experience in particular years the onset of an ethnic civil war. In terms of explanatory variables we proceed as above and use the number of ethnic groups and minorities at risk, as well as the colonial tradition as main variables explaining the presence of a federal arrangement, but also include the remaining explanatory variables of Fearon and Laitin (2003). For the outcome equation we use again this latter set of independent variables but add our main explanatory variable, namely the proportion of federal units controlled by a national minority and its square. Not surprisingly the estimation results suggest that an increasing number of ethnic groups increases the probability that a country has a federal arrangement. Similarly, federalism is more prevalent in former Dutch, French, and British colonies. Controlling for this selection mechanism, the estimation results for the outcome equation suggest that increasing the proportion of minority controlled federal units increases the probability of an ethnic conflict onset, again in a curvilinear fashion. This effect fails to reach statistical significance, however. Given that with the more simple approach presented above we also initially found such a statistically non-significant effect among all country-years used by Fearon and Laitin (2003), we need again to worry about possible endogeneity bias. For this reason we again estimate an auxiliary regression model to predict the proportion of minority controlled federal units. Given that this regression might again be subject to a selection bias if we focus only on federal countries, we estimate this auxiliary regression with a classic Heckman (1976) model (see Columns 3–4 and 5–6 in Table 5 ). The residuals from the outcome equation and their square appear as additional explantory variables in the second Heckman probit model in Table 6. While most of the estimated coefficients remain stable compared to the previous model, the effect of the proportion of minority controlled federal units decreases dramatically and the coefficient for the variables related to the residuals are large and statistically significant. Hence, with our instruments we can be sure that there is a problem of endogeneity. Consequently, in the last two columns of Table 6 we provide the results of a model where we use the predicted values of the auxiliary regression to replace the proportion of minority controlled federal units and its square. Doing so results in coefficients which again support a curvilinear effect of the share of minority controlled federal units. The strongest effect can be found for federal states with approximately 60% of the federal units being controlled by minorities. States with larger or smaller shares are less conflict-prone. This is illustrated by Figure 2, which depicts predicted probabilities of ethnic conflict onset (with their 95% confidence interval) for the full range of shares of federal units controlled by national minorities.¹⁶ These predicted probabilities become larger as this share increases up to a level of approximately 0.6, at which these probabilities start to decrease. Countries that have values close to this threshold value are Brazil (0.63) and Nigeria (0.65). Figure 2 also shows that the uncertainty related to these predicted probabilities is quite important when the share of minority controlled units is large. Nevertheless, the differences in the predicted probabilities from the lowest share to the share that results in the highest likelihood of conflict onset are statistically significant.

Figure 2.

Predicted Probabilities of Ethnic Conflict Onset and Proporition of Minority Controlled Federal Units

Hence, we can conclude from this analysis that if there is an effect among federal states of the proportion of minority controlled federal units, then it is much more likely to be a positive one, since only a few states have a larger share of minority controlled units than 60%. As the results of the first analysis reported in Table 4 go in the same direction (without reaching statistical significance), scholars like Horowitz (1985) seem closer to the mark in their assessment of the effect of ethnofederal arrangements.

5. Conclusion

If federal arrangements are chosen to allow diverse populations to “come together” or to be “held together” (Stepan, 1999), two things follow quite directly. First, the stability of federal arrangements and the latter’s conflit-proneness are likely to be related to the “ethnic map” of the country. Second, if trying to assess the effect of different federal arrangements such as ethnofederalism, we also need to consider what brought about these particular arrangements.

In trying to offer elements pertaining to the first point, we have also offered ways to address the second point. Using very detailed information on the ethnic makeup of federal countries, we assessed through a rather simple indicator the effect of different types of ethnic configurations of federal units on civil war onset. Once we control for the endogenous nature of federal arrangements, we find that larger shares of federal units dominated by a national minority considerably increase the likelihood of ethnic civil war onset in federal countries.

This result has to be interpreted with some caution, and future research has to demonstrate its robustness. First of all, it would be useful to build into our empirical model more detailed information on the geographic location of groups. Even with more and better data, however, the most daunting endeavor still seems to be to get a theoretical handle on the question of why and possibly how and what type of federal arrangements are introduced to mitigate societal conflicts.

Footnotes

1

The experiences of the Soviet Union and the countries in its sphere of influence are discussed by a series of authors (e.g. Roeder, 1991; Cornell, 2002; Bunce, 2003a,b; Hale, 2005).

2

This point relates to Toft’s (2003) claim that a group’s geographic concentration and desire to control territory is closely linked to the outbreak of violence. However, Fearon and Laitin (2002), who also find an effect for the newly constructed concentration measure for groups, argue that this has more to do with strategic issues. Interestingly, find that group concentration of minorities at risk positively affects both protest and rebellion. When separating low from high concentration groups, the authors find that federalism fails to have a statistically significant effect for low concentration groups, but that the negative effect on rebellion is statistically significant for high concentration groups (however, this may in part be due to the different sample sizes rather than any substantive differences).

3

Unfortunately, these notions are nowhere clearly defined, but they come very close to the notion of ethnofederalism. Bunce (2003a), however, refers to Azerbaijan, Georgia, Russia, and Serbia-Montenegro as ethnofederalist states. Of these only the last two are considered by McGarry (2002) as multinational federations. Similarly, the POLITY III indicator classifies both Azerbaijan and Georgia as mixed regimes (between unitary and federal), while Henisz (2002) codes Georgia as not federal. For Georgia, also suggest that the subnational units fail to have sufficient authority.

4

The attractiveness of McGarry’s (2002) distinction between national and multinational federations is that it is obviously closely related to the ethnic makeup of the countries concerned. Related notions are “ethnically mined” federations in which one group is a majority within one of the constituent units and a minority within the federation as a whole. These authors classify nine countries in this category, namely Belgium, Canada, Czechoslovakia, India, Nigeria, Pakistan, Soviet Union, Switzerland, and former Yugoslavia.

5

We also put in this category the rather large differences in timing when it comes to Nigeria and Pakistan. Bednar (2008) considers Pakistan as federal between 1990 and 1999, while McGarry (2002) assumes that the federal experience was limited to 1947–71. We follow the latter coding decision, despite the fact that coding suggests that between 1971 and 1990 Pakistan was still federal.

6

Griffiths and Nerenberg (2005) consider all these cases as federal, with the exception of Burma/Myanmar, while Watts (1996) excludes in addition Micronesia. Given that none of our data on ethnic groups and minorities at risk nor the data on conflicts have Micronesia, St Kitts-Nevis, and the West Indies Federation as observations, we will no longer consider these cases in the remainder of the article. For Burma, however, we follow and consider it as a multinational federation.

7

propose an alternative measure for group concentration. Since this dataset covers only 101 groups so far, we refrain from using it here.

8

In a web appendix we report, for each country, details of the type of data we collected to determine the makeup of federal units according to groups. The federal units were determined on the basis of information from the CIA World Factbook (https://www.cia.gov/cia/publications/factbook/geos/), accessed 21 July 2006, and the Administrative Divisions of Countries (Statoids) (http://www.statoids.com/), accessed 21 July 2006. To identify groups we relied on Fearon (2002) and the Minorities at Risk data (Gurr, 1993) as reported in Elkins and Sides (2007). For the group sizes we relied on data from EVS2006 as well as other surveys (as noted in ), which we also used to complement the list of groups if appropriate. When possible we use census data at the level of the federal units. The main information used in this article is the number of federal units (column 2) and the number of units in which the national majority is a minority (column 3). The share of these units among all federal units is the proportion of units controlled by a national minority.

9

As we rely mostly on survey measures and official statistics to determine the group composition of federal units, this information is static and covers countries roughly 1995–2005. As Nigeria had undergone considerable changes in federal units before this period, we checked for the robustness of our results when dropping the years before 1991 for this country. The main results remained largely the same.

10

We thank Warren Camber for suggesting these additional instruments.

11

We fully concur with the view articulated in Achen (2005) and that this is probably not the best strategy. The alternative, however, would be a fully specified theoretical model on the outbreak of civil wars. Such a model is, however, quite clearly still absent in this literature.

12

We employ the replication data made available by and thus refrain from a discussion of the data.

13

We also considered an endogeneity test based on Hausman’s (1978) work in the context of nonlinear models (see Rivers and Vuong, 1988; Bollen et al., 1995) employing either a linear probability or a probit model in the first stage (see Alvarez and Glasgow, 2000). The substantive implications of these other models are identical to those reported in the main text.

14

We follow here : 236f) suggestion to instrument the square of the proportion of federal units controlled by a national minority by the same variables plus the square of the predicted value of the auxiliary regression calculated for the non-squared values of this variable.

15

To correct for the biased standard errors in this model as discussed by Rivers and Vuong (1988), we follow Alvarez and Glasgow (2000) and report bootstrapped standard errors (see Efron, 1981).

16

To simulate the predicted probabilities, all other variables were held at their mean values, except for dichotomous variables which were held constant at their modal category.

References

Acemoglu

Daron

. 2005. Constitutions, politics and economics: A review essay on Persson and Tabellini’s “The Economic Effect of Constitutions”. Journal of Economic Literature 43(4): 1025–1048.

Achen

Christopher H.

2005. Let’s put garbage-can regressions and garbage-can probits where they belong. Conflict Management and Peace Science 22(4): 327–339.

Alemán

Eduardo

Treisman

Daniel

. 2005. Fiscal Politics in “ethnically mined,” developing, federal states: Central strategies and secessionist violence. In Sustainable Peace: Power and Democracy After Civil Wars, ed. Roeder

Philip G.

Rothchild

Donald

, pp. 173–216. Ithaca, NY: Cornell University Press.

Alvarez

Michael

Glasgow

Garrett

. 2000. Two-stage estimation of non-recursive choice models. Political Analyis 8(2): 147–166.

Bakke

Kristin M.

Wibbels

Erik

. 2006. Diversity, disparity, and civil conflict in federal states. World Politics 59(1): 1–50.

Beck

Thorsten

Clarke

George

Groff

Alberto

Keefer

Philip

Walsh

Patrick

. 2001. New tools in comparative political economy: The database of political institutions. World Bank Economic Review 15(1): 165–176.

Bednar

Jenna

. 2008. The Robust Federation. New York: Cambridge University Press.

Bermeo

Nancy

. 2002. The import of institutions. Journal of Democracy 13(2): 96–110.

Bollen

Kenneth A.

Guilkey

David K.

Mroz

Thomas A.

1995. Binary outcomes and endogenous explanatory variables: Tests and solutions with an application to the demand for contraceptive use in Tunisia. Demography 32(1): 111–131.

10.

Brancati

Dawn

. 2006. Decentralization: Fueling the fire or dampening the flames of ethnic conflict and secessionism. International Organization 60(3): 651–685.

11.

Brancati

Dawn

. 2007. The origins of regional parties. British Journal of Political Science 38(1): 135–159.

12.

Brancati

Dawn

. 2009. Design over Conflict: Managing Ethnic Conflict and Secessionism Through Decentralization. Oxford: Oxford University Press.

13.

Breton

Albert

. 1964. The economics of nationalism. Journal of Political Economy 72(4): 376–386.

14.

Bunce

Valerie

. 2003a. Cooperation, compromise, or conflict: Three models of center–regional bargaining in ethnofederal states. Paper presented at the annual meeting of the American Political Science Association, Philadelphia, PA, 27 August–5 September.

15.

Bunce

Valerie

. 2003b. Rethinking recent democratization: Lessons from the postcommunist experience. World Politics 55(2): 167–192.

16.

Bunce

Valerie

. 2007. Minority politics in ethnofederal states: Cooperation, autonomy or secession? Einaudi Center for International Studies, Cornell University, Working Paper Series 08-07.

17.

Bunce

Valerie

Watts

Stephen

. 2005. Managing diversity and sustaining democracy: Ethnofederal versus unitary states in the postcommunist world. In Sustainable Peace: Power and Democracy After Civil Wars, eds Roeder

Philip G.

Rothchild

Donald

, pp. 133–158. Ithaca, NY: Cornell University Press.

18.

Clarke

Kevin

. 2005. The phantom menace: Omitted variable bias in econometric research. Conflict Management and Peace Science 22(4): 341–352.

19.

Cohen

Frank S.

1997. Proportional versus majoritarian ethnic conflict management in democracies. Comparative Political Studies 30(5): 607–630.

20.

Congleton

Roger D.

2000. A political efficiency case for federalism in multinational states: Controlling ethnic rent-seeking. In Competition and Structure: The Political Economy of Collective Decisions: Essays in Honor of Albert Breton, eds Galeotti

Slamon

Wintrobe

, pp. 365–397. New York: Cambridge University Press.

21.

Cornell

Svante E.

2002. Autonomy as a source of conflict: Caucasian conflicts in theoretical perspective. World Politics 54(2): 245–276.

22.

Diaz-Cayeros

Alberto

. 2004. The centralization of fiscal authority: An empirical assessment of Popitz’s Law. Paper presented at the annual meeting of the American Political Science Association, Chicago, 2 September.

23.

Dubin

Jeffrey A.

Rivers

Douglas

. 1990. Selection bias in linear regression, logit and probit models. In Modern Methods of Data Analysis, eds Fox

John

Long

Scott J.

, pp. 410–442. Newbury Park, CA: Sage.

24.

Efron

Bradley

. 1981. Nonparametric estimates of standard error: The jackknife, the bootstrap and other methods. Biometrika 68(3): 589–599.

25.

Elkins

Zachary

Sides

John

. 2007. Can institutions build unity in multiethnic states? American Political Science Review 101(4): 693–708.

26.

Fearon

James D.

2002. Ethnic structure and cultural diversity around the world: A cross-national data set on ethnic groups. Paper prepared for delivery at the annual meeting of the American Political Science Association, Boston, 29 August–1 September.

27.

Fearon

James D.

Laitin

David D.

2000. Development of group concentration values for the MAR data set. Paper prepared for presentation at the second semi-annual LiCEP meeting, University of Pennsylvania, 20–22 October.

28.

Fearon

James D.

Laitin

David D.

2002. Group concentration and civil war. Paper prepared for delivery at the annual meeting of the American Political Science Association, Boston, 29 August–1 September.

29.

Fearon

James D.

Laitin

David D.

2003. Ethnicity, insurgency, and civil war. American Political Science Review 97(1): 1–17.

30.

Filippov

Mikhail

Ordeshook

Peter

Shvetsova

Olga

. 2004. Designing Federalism: A Theory of Self-Sustainable Federal Institutions. Cambridge: Cambridge University Press.

31.

Freedman

David A.

Sekhon

Jasjeet S.

2010. Endogeneity in probit response models. Political Analysis 18: 138–150.

32.

Glazer

Nathan

. 1983. Federalism and ethnicity: The American solution. In Ethnic Dilemmas 1964–82, ed. Glazer

Nathan

, pp. 274–292. Cambridge, MA: Harvard University Press.

33.

Griffiths

Ann L.

Nerenberg

Karl

, eds. 2005. Handbook of Federal Countries. Montreal: McGill-Queen’s University Press.

34.

Gurr

Ted Robert

. 1993. Minorities at Risk: A Global View of Ethnopolitical Conflict. Washington, DC: United States Institute of Peace Press.

35.

Gurr

Ted Robert

. 2000. Ethnic warfare on the wane. Foreign Affairs 79(3): 52–65.

36.

Hadenius

Axel

Teorell

Jan

. 2007. Pathways from authoritarianism. Journal of Democracy 18(1): 143–156.

37.

Hale

Henry E.

2004. Divided we stand: Institutional sources of ethnofederal state survival and collapse. World Politics 56(1): 165–193.

38.

Hale

Henry E.

2005. The makeup and breakup of ethnofederal states: Why Russia survives where the USSR fell. Perspectives on Politics 3(1): 55–70.

39.

Hausman

Jerry

. 1978. Specification tests in econometrics. Econometrica 46(6): 1251–1271.

40.

Heckman

James J.

1976. The common structure of statistical models of truncation, sample selection and limited dependent variables and a simple estimator for such models. Annals of Economic and Social Measurement 5(4): 475–492.

41.

Henisz

Witold J.

2002. The institutional environment for infrastructure investment. Industrial and Corporate Change 11(2): 355–389.

42.

Horowitz

Donald

. 1985. Ethnic Groups in Conflict. Berkeley, CA: University of California Press.

43.

Kohli

Atul

. 2004. India: Federalism and the accommodation of ethnic nationalism. In Federalism and Territorial Cleavages, eds Amoretti

Ugo M.

Bermeo

Nancy

, pp. 281–300. Baltimore, MD: Johns Hopkins University Press.

44.

Kymlicka

1998. Is federalism a viable alternative to secession? In Theories of Secession, ed. Lehning

Percy

, pp. 111–150. London: Routledge.

45.

Lake

David A.

Rothchild

Donald

. 2005. Territorial decentralization and civil war settlements. In Sustainable Peace: Power and Democracy After Civil Wars, eds Roeder

Philip G.

Rothchild

Donald

, pp. 109–132. Ithaca, NY: Cornell University Press.

46.

Lemco

Jonathan

. 1991. Political Stability in Federal Governments. New York: Praeger.

47.

Lipset

Seymour M.

1963. Political Man. Garden City, NY: Anchor.

48.

McGarry

John

. 2002. Federalism (federation) as a method of ethnic conflict regulation. Working paper to be presented at the conference on From Power-Sharing to Democracy: Post-Conflict Institutions in Ethnically Divided Societies. London, Ontario, 9–10 November.

49.

McGarry

John

O’Leary

Brendan

. 2005. Federation as a method of ethnic conflict regulation. In From Power Sharing to Democracy, ed. Noel

Sid

, pp. 263–296. Montreal: McGill-Queen’s University Press.

50.

Monfardini

Chiara

Radice

Rosalba

. 2008. Testing exogeneity in the bivariate probit model: A Monte Carlo study. Oxford Bulletin of Economics and Statistics 70(2): 271–282.

51.

Nordlinger

Eric A.

1972. Conflict Regulation in Divided Societies. Cambridge, MA: Center for International Affairs, Harvard University.

52.

Panizza

Ugo

. 1999. On the determinants of fiscal centralization: Theory and evidence. Journal of Public Economics 74(1): 97–139.

53.

Rabushka

Alvin

Shepsle

Kenneth A.

1972. Politics in Plural Societies: A Theory of Democratic Instability. Columbus, OH: C.E. Merrill.

54.

Riker

William H.

1964. Federalism: Origin, Operation, Significance. Boston, MA: Little, Brown.

55.

Rivers

Douglas

Vuong

Quang H.

1988. Limited information estimators and exogeneity tests for simultaneous probit models. Journal of Econometrics 39(3): 347–366.

56.

Roeder

Philip G.

1991. Soviet federalism and ethnic mobilization. World Politics 43(2): 196–232.

57.

Roeder

Philip G.

2007. Where Nation-States Come From. Princeton, NJ: Princeton University Press.

58.

Roeder

Philip G.

2009. Ethnofederalism and the mismanagement of conflicting nationalisms. Regional and Federal Studies 19(2): 203–219.

59.

Rothchild

Donald

Roeder

Philip G.

2005. Dilemmas of state-building in divided societies. In Sustainable Peace: Power and Democracy After Civil Wars, eds Roeder

Philip G.

Rothchild

Donald

, pp. 1–25. Ithaca, NY: Cornell University Press.

60.

Saideman

Stephen M.

Lanoue

David J.

Michael

Campenni

Stanton

Samuel

. 2002. Democratization, political institutions, and ethnic conflict: A pooled time-series analysis, 1985–1998. Comparative Political Studies 35(1): 103–129.

61.

Snyder

Jack

. 2000. From Voting to Violence: Democratization and Nationalist Conflict. New York: Norton.

62.

Stepan

Alfred

. 1999. Federalism and democracy: Beyond the U.S. model. Journal of Democracy 10(4): 19–34.

63.

Teorell

Jan

Holmberg

Sören

Rothstein

. 2008. The Quality of Government Dataset, version 15May08. University of Gothenburg: Quality of Government Institute.

64.

Toft

Monica Duffy

. 2003. The Geography of Ethnic Violence: Identity, Interests, and the Indivisibility of Territory. Princeton, NJ: Princeton University Press.

65.

Watts

Ronald L.

1996. Comparing Federal Systems in the 1990s. Kingston, ON: McGill-Queen’s University Press.

66.

Watts

Ronald L.

1998. Federalism, federal political systems, and federations. Annual Review of Political Science 1: 117–137.

67.

Wilde

Joachim

. 2000. Identification of multiple equation probit models with endogenous dummy regressors. Economics Letters 69: 309–312.

68.

Woodward

Susan L.

1995. Balkan Tragedy: Chaos and Dissolution After the Cold War. Washington, DC: Brookings Institution.

69.

Wooldridge

Jeffery M.

2002. Econometric Analysis of Cross Section and Panel Data. Cambridge, MA: MIT Press.