Are Federal Systems Better Than Unitary Systems

Are Federal Systems Better than Unitary Systems? ABSTRACT Much has been written about the putative virtues and vices of federal and unitary systems of government, but little empirical testing of the impact of such systems on the quality of governance has been conducted. Do federal or unitary systems promote better social, political and economic outcomes? The paper takes up a series of theoretical debates put forth by advocates of federalism, including competition among subnational governments, fiscal federalism, veto points, accountability, and the size of government. In each case, there is room for doubt about the practical impact of federalism on governance. The paper then conducts a series of cross-national empirical tests over several decades of the impact of unitary systems on fifteen indicators of political, economic and human development. In most cases, a strong empirical relationship between unitarism and good governance obtains. To the extent that these constitutional structures make a difference, unitary systems appear to hold distinct advantages over federal ones. Decentralization is, quite possibly, the dominant political trend of our time. Significant devolutions of authority from national to subnational levels have occurred in Africa (e.g., Cote d’Ivoire, Ghana), Asia (e.g., Bangladesh, India), Europe (e.g., Belgium, Britain, France, Italy, and Spain), and Latin America (e.g., Argentina, Colombia, and Mexico).1 Several new democracies have emerged with explicit constitutional guarantees for sub-state authorities (e.g., Russia and the European Union). Meanwhile, older federal polities such as Germany, India, the United States, and Switzerland continue intact. We are aware of no democracy that has moved from the constitutional status of “federalism” to that of its contrary, “unitarism.” As a consequence, most large democracies (calculated either in population and/or land area) are now constitutionally federal. Indeed, over 56% of the world’s democratic citizens (those living in states with multi-party competition) live in federal polities at the outset of the twenty-first century. In addition to a political fact on the ground, decentralization is an idea, and an increasingly popular one at that. At mid-century, the three dominant models of development—socialism, export-oriented industrialization, and import substitution—all featured a leading role for the national state. Big projects under central-state direction would pave the way for economic and human development in the third world. Now, the focus has shifted to local-level initiatives, micro-enterprise, NGOs, and the fostering of democratic institutions. The current vogue of decentralization is echoed in academic work by most economists and political scientists. Indeed, scattered evidence suggests that academics, politicians, and policymakers from across the political spectrum have embraced decentralization as a key to good governance. While the Old Right defended the prerogatives of the state, the New Right is stridently antistatist, preferring local democracy, entrepreneurial (“democratic”) capitalism, and constitutionally guaranteed property rights. On the Left, the centralist model associated with socialism and communism is increasingly under attack. In its place one finds renewed interest in democracy, accountability, citizen participation, civil society, social capital, and deliberation. Thus, the appeal of decentralization cuts across the usual left-right cleavage. Before endorsing the decentralized model of governance it behooves us to probe its conceptual underpinnings and empirical patterns more thoroughly. Despite widespread enthusiasm, relatively little research has attempted to probe critically the theory of decentralism or to test systematically the policy impact of decentralized systems of governance. In this paper we focus on one aspect of decentralization, the constitutional choice between federal and unitary structures of government. This feature of politics may or may not be emblematic of other aspects of the centralization/decentralization debate. Indeed, we surmise that constitutional decentralization is quite different, in purpose and in effect, from administrative or fiscal decentralization. In any event, the constitutional status of a polity is a central feature of its governance apparatus and thus a central feature of the overall debate, both empirically and theoretically. Bird and Vaillancourt (1998), Crook and Manor (1998), Elazar (1996), Montero and Samuels (2003), Rodden, Eskeland, and Litvack (2003). 1 2 We begin this paper with a review of federalist theory, the dominant normative theory of governance in our time. We argue that there are a priori reasons to doubt the efficacy of decentralized constitutional structures. In the second section of the paper we proceed to an empirical examination of this constitutional question. Here, we examine three broad policy areas—political development, economic development, and human development—operationalized across fifteen governance indicators. For each measurable outcome, we conduct crossnational tests in order to determine whether federal or unitary regimes are associated with better governance, controlling for other relevant factors. Our findings suggest that constitutional centralization leads to better governance, particularly in economic and human development. Contrary to most of the literature on this topic, we argue that there is a policy payoff to unitary systems. THE NORMATIVE THEORY OF FEDERALISM Unitarism is distinctly out of favor, so much so that there is no well-defined theory supporting a unitary vision of governance. Consequently, our theoretical discussion focuses on its polar opposite, federalism, where there is a venerable, albeit diverse, theoretical tradition to draw upon (for general surveys see Bakvis and Chandler 1987; Beer 1963; Davis 1978; Elazar 1995; Forrest 1988; Karmis and Norman 2005; Mogi 1931; Ornstein and Coursen 1992; Wachendorfer-Schmidt 2001; Watts 1999; Wheare 1963). “A constitution is federal,” writes William Riker (1964: 11), “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.” Following Riker, we understand this binary concept (federalism/unitarism) to refer to the constitutional features of a polity, not its administrative or fiscal arrangements. Thus, when we label a polity unitary we are saying that constitutional authority—sovereignty—is vested in the central (national) government, not that all decision-making occurs at the center or all money is raised or spent at the center. The crucial distinction is that power delegated from national to subnational bodies in a unitary polity may be retrieved. In a federal system, by contrast, regional authorities enjoy constitutional status; their power is inherent rather than delegated. (In the following discussion, we use the terms subnational unit, region, state, and territory interchangeably.)2 A panoply of diverse arguments has been offered to vindicate the federalist vision of good governance. Arguably, decentralist “theory” is more a series of interrelated arguments than a single, coherent theoretical apparatus. In any case, space constraints compel a selective treatment of these arguments. We 2 Granted, there are instances in which unitary countries “federalize,” so this is not an entirely exogenous factor. However, there is greater stickiness to this constitutional status than is evident in most purely administrative or fiscal policies. 3 focus on causal claims that have plausible ramifications for a wide range of policies and policy outcomes. We treat arguments that are more narrowly focused—i.e., on one specific policy domain—in cursory fashion, or not at all. Note that although the theoretical literature on federalism and decentralization is vast, a handful of relatively general, abstract claims recur. These concern a) competition among subnational governments, b) fiscal federalism, c) veto points, d) accountability, and e) the size of government. These are the core features of a nascent, normative theory of democratic federalism. Each of these mechanisms should have repercussions for a wide range of public policies; they are not limited to policies of a specific type. The remainder of this section presents, quite briefly, each of the arguments. (Although some of these arguments were originally developed in the context of debates over fiscal or administrative policy, they are nonetheless often applied to debates over constitutional structure and are thus relevant to the present discussion.) We endeavor to show here why there might be reason to doubt the conclusions put forward by the proponents of federalism. Our purpose here is not to present an opposing theory, a normative theory of unitarism. Rather, it is to call into question a fairly entrenched consensus among scholars and policymakers and to motivate the empirical tests to follow. Our hope is that in opening up this debate we may help pave the way for new theoretical frameworks that provide a better accounting of the relationships between constitutional structures and policy outcomes. CAVEATS Several caveats are necessary before we proceed. First, our theoretical discussion does not extend to government at the global level, where federal arrangements may be necessary and efficient. Our attention focuses on federalism and unitarism as constitutional arrangements for nation-states. Second, unless otherwise specified, all of the following arguments are intended to apply to outcomes obtained in the long run. Thus, whether federalism provides a useful transitional structure—either to a single consolidated state or to several independent (unitary) states—is not under consideration here. What matters in our theoretical discussion is what matters in the final analysis: does federalism or unitarism provide an optimal equilibrium outcome? Finally, there is a question of what counterfactuals are relevant to this discussion. Unlike other constitutional reforms (e.g., changes of executive type from presidential to parliamentary executives, or electoral system reform), the theory of federalism engages not only a question of constitutional structure but also a question of territory. This vastly expands the number and type of counterfactuals that one might consider. An existing federal state could be unified (governed under a unitary constitution). An existing unitary state could be federalized, and these federal units could be of any shape, size, and number. An existing set of unitary states could be federalized (as the European Union). An existing set of federal states could be brought into a larger federation. An existing set of federal states could be brought into a unitary 4 constitution. One might also disregard existing lines of sovereignty altogether, dividing up the world into unitary and/or federal states with no regard to existing boundaries. There is no limit to the number and type of federal/unitary possibilities if these two dimensions—a) constitutional structure, and b) territory—are allowed to vary freely. In principle, unitary and federal constitutions might be applied to states with any set of initial conditions—e.g., any location (contiguous or non-contiguous), any size (enormous or tiny), any composition (regardless of existing social identities), any socioeconomic standing (rich or poor), any set of natural resources (favorable or unfavorable), and any circumstance of state-formation (forcible or peaceful). This is one reason for the wide-ranging, ambient nature of the ongoing discussion over federalism. It is always possible to envision situations in which one, or the other, sort of reform would be unworkable. Indeed, no one would argue that federalism (or unitarism) is best in all circumstances and all occasions. As such, any theoretical formulation is intractable; it is not clear where it might apply. In order to establish sensible scope-conditions for a theory of federalism/unitarism focused on democratic governance we issue the following caveat. Counterfactuals are limited to situations in which reform (a change of constitutional structure or territory) might be achieved peaceably and with the general consensus of the inhabitants in a region. Recall that we are not concerned here with the efficacy of federalism/unitarism in maintaining peace, national integrity, or democracy. We are concerned, rather, with the quality of governance within an existing democratic state. In this context, we adopt a presumption of peaceful and consensual change to existing democratic constitutional structures.3 COMPETITION AMONG SUBNATIONAL GOVERNMENTS Following Tiebout (1956), federalists often argue that multiple subnational governments with distinct areas of competence simulate the structure of a free market, offering exit options to unhappy constituents and hence incentives to politicians and bureaucrats to provide superior governance so as to attract and retain constituents (Brennan and Buchanan 1980; Breton 1996; Buchanan 1995; Montinola et al 1995; Qian and Weingast 1997; Ribstein and Koboyashi 2006; Salmon 1987). There are many difficulties with this argument. It supposes, for example, that regional differentiation establishes the same set of incentives as localities (the focus of Tiebout’s work). The presumed mechanism— voting with one’s feet—is much more difficult if such a move entails relocation into a different state rather than an adjacent community. Granted, constituents may move from one state to another without legal hindrance; however, they are unlikely to incur the costs of such a move without a strong incentive to do so. To be sure, most of the unitary and federal states in our sample were not originally forged in a spirit of compromise, and we can anticipate that the conditions of its formation had some—possibly quite significant—effect on subsequent success (Stepan 1999). However, it is difficult to operationalize these matters when the processes of stateformation are so various, and occur over long stretches of historical time. The “peace and consensus” caveat refers therefore to an aspect of our theory that cannot be easily tested. But it is a sensible caveat, for it allows us to conceptualize the effects of constitutional structure on the quality of governance without entertaining extreme counterfactuals. 3 5 Thus, while there is some evidence of Tiebout effects at the local level, there is considerable debate over whether this theory extends to the subnational level (Dowding, John, and Biggs 1994). Many factors other than the quality of governance influence the movement of citizens from place to place. Thus, the entrance and exit of citizens may be a poor signal of the quality of governance in a state, even if one could measure these demographic shifts (a difficult matter in many parts of the developing world, where population surveys are typically infrequent and unreliable). Most important, it is not clear that government officials have an incentive to attract citizens from another state, or to maintain citizens currently residing in their own state. Thus, the presumed analogy between citizens and consumers is questionable. The analogy makes somewhat more sense when it concerns the movement of capital, which crosses borders more easily, is easier to track, is more likely to respond to the varying quality of governance from one region to another, and is presumably highly valued by politicians. However, it is not clear that the competition for capital alone is sufficient to improve the quality of governance from one region to another. Indeed, it may degrade the quality of governance insofar as corporations are offered incentives that, collectively, reduce a region’s tax base and hence its ability to provide public goods (Prud’homme 1995). Ultimately, both “leveling up” (Vogel 1997) and “leveling down” (Cai and Treisman 2000; Schneider 2002) scenarios stemming from inter-regional competition are plausible (for general discussion see Paul 1992). Additionally, with respect to the intertwined processes of policymaking and policy implementation, there are strong reasons to expect greater competence and probity at the center than at the periphery. Competent and highly trained bureaucrats generally prefer to work in major metropolitan areas. Civil service training and remuneration is often superior in the national bureaucracy. Most importantly, civil servants at the core of the national bureaucracy may be more strongly motivated by patriotic goals, and consequently less likely to see themselves as representatives of particularistic interests. For all these reasons, the quality of public service is likely to decline as one moves from center to periphery (Bardhan 2002: 189; Huntington 1968: 68; Prud’homme 1995; Tanzi 1996; Treisman 2003; Wade 1985). National initiatives may also be more difficult to implement by virtue of the labyrinthine nature of a federal bureaucracy, where different levels of powerholders, each with distinct constituencies, interests, and perspectives, must cooperate in order to make legislation a reality on the ground (Bardach 1977; Pressman and Wildavsky 1973). Finally, economies of scale may be more achievable in a unitary than in a federal system. FISCAL FEDERALISM Another oft-invoked argument for federalism derives from literature on “fiscal federalism” and “subsidiarity,” two concepts that are similar enough to be treated together. As stated by Oates (1999: 1121-2), the principle of decentralization rests on the assumption that Decentralized levels of government have their raison d’etre in the provision of goods and services whose consumption is limited to their own jurisdictions. By tailoring outputs of 6 such goods and services to the particular preferences and circumstances of their constituencies, decentralized provision increases economic welfare above that which results from the more uniform levels of such services that are likely under national provision. The basic point here is simply that the efficient level of output of a ‘local’ public good . . . is likely to vary across jurisdictions as a result of both differences in preferences and cost differentials. To maximize overall social welfare thus requires that local outputs vary accordingly. Yet, the theory of fiscal federalism, strictly interpreted, does not imply that federal states are superior to unitary states. Fiscal federalism is a theory of public administration, not of constitutional structure. It “lays out a general normative framework for the assignment of functions to different levels of government and the appropriate fiscal instruments for carrying out these functions” (Oates 1999: 1121). The theory has nothing to say, directly, about whether constitutional federalism is a good or bad form of government. Indeed, one proponent of fiscal federalism has acknowledged that fiscal federalism and constitutional federalism are orthogonal to one another (Weingast 2000: 8). From a fiscal-federalism perspective, the constitutionally sanctioned separation of national and subnational government introduces incentives for each that may, or may not, promote the public interest. Insofar as the provision of public goods happens to coincide with the borders of regional governments, federalism may be looked upon as an efficient mechanism (Olson 1969). However, there is little reason to assume that constitutional boundaries are propitiously drawn. Oates (1999: 1130) remarks that the sets of existing boundaries that one finds in the United States and in other constitutionally federal nation-states “are largely historically and culturally determined and . . . may make little sense in terms of the economic and geographical realities.” Insofar as the provision of public goods breaches the boundaries of regional governments, a federal arrangement may lead to an under- or over-provision of that good. If, for example, the repercussions attached to an action taken by one regional government are felt beyond the borders of that government one faces formidable coordination problems among two or more subnational governments (Inman and Rubinfeld 1997). This would include negative externalities, such as pollution created by one region that crosses boundaries and becomes a burden on neighboring states, or bureaucrats who “over-graze” the commons (Shleifer and Vishny 1993). It would also include positive externalities, such as state-supported education. Since educated workers have freedom of movement they may choose to work in neighboring states, thus providing a positive spillover that does not benefit the providing jurisdiction (thereby reducing the incentives to fund education) (Scharpf 1988). Perhaps the most important coordination issues involve revenues and expenditures. Sometimes, a given problem is circumscribed within a single area. So long as this area lies within an existing state or province, solving it may involve few externalities, positive or negative. However, if the revenue necessary to solve that problem is a) considerable and b) must be raised locally, it may not result in an efficient allocation of resources nationally. Poverty is an example of such a problem. Thus, fiscal federalists often argue that the optimal polity reserves macroeconomic policies and redistributive policies for the center (Oates 1999: 1121). 7 (At times, fiscal federalists sound like macroeconomic centralists.) In this case, one must wonder about the possibilities of maintaining centralized control over such policies if constitutional power is decentralized, as it is by definition in a federal polity. This brings us to a final issue. If there are clear-cut and rational rules for allocating policymaking decisions between center and periphery in a federal system, and if these rules can be maintained over time, then federalism may lead to better public policies, for all the reasons that advocates of fiscal federalism and subsidiarity suppose (Bermann 1994). One example of such a rule is the “hard budget” constraint imposed in some federal polities by national governments on subnational authorities (Montinola et al 1995; Oates 1999; Qian and Weingast 1997; Rodden 2002; Rodden et al. 2003). However, even in this best of all possible worlds one must be willing to assume that widely varying coordination issues can be effectively solved with simple and relatively police-able rules. It is not clear that there is a rule—analogous to hard budgets—that would regulate conflicts between national and subnational authorities over environmental and conservation issues and a wide swath of other contentious public policies. Even if such rules could be created, it is not clear that they could be effectively maintained under the terms of a federal constitution. Indeed, the very political factors that lead to coordination problems in the first place may also undermine the maintenance of effective rules, rules that necessarily impose benefits and burdens unequally across the populace. Thus, it seems unlikely that this Solonic device will work consistently in federal regimes. The most serious coordination problems are likely to arouse the greatest political resistance to the imposition of whatever formal rules might be devised to solve them. All national governments (excepting micro-states) must deal with the problem of how to allocate duties and responsibilities across regions and between central and local agencies. This is the problem that the theory of fiscal federalism is designed to solve (in a general sort of way). There is no a priori reason to suppose that a federal state will have an easier time developing and maintaining an efficient division of labor than a unitary state. Indeed, the constitutional rigidities of federalism, and the competing agendas of regionally ensconced politicians, might lead to less-than-optimal outcomes. Our argument is that because unitarism allows policymakers greater flexibility than a constitutionally constrained federal system, there is a better probability that (in a democratic system) they will assign a given policy decision to an appropriate governmental unit, thus producing a more efficient outcome (in utilitarian terms). To be sure, it does not assure that they will do so. But if there is any rationality to the democratic decisionmaking process, they should do so, overall, more often than under federalism. VETO POINTS Another perspective on federalism rests on the theory of veto points (Tsebelis 2000). Constitutional federalism creates a new set of actors with constitutional powers. Arguably, the regions in a federal system, or some combination of them, constitute a new set of veto points. Assuredly, they raise the political 8 threshold for any democratic change to the status quo. The multiplication of veto points means that a larger set of actors with independent bases of power and (somewhat) independent interests will have to agree on any policy. One argument for greater veto points, and higher barriers to change, seems secure. Since a federal system serves to institutionalize the status quo (the foregoing exceptions duly noted), this should enhance a government’s credible commitment to a policy, once that policy is established (Henisz 2000a; Keefer and Stasavage 2003; North 1993). Citizens and potential investors will have fewer worries about what changes the next administration might bring. The problem with this argument is that it presumes that the status quo is reasonably favorable—to property rights, to the provision of public goods, and so forth. If, on the other hand, the status quo is unfavorable—if the existing set of policies and institutions are corrupt, inefficient, or poorly devised—then the existence of multiple veto points will serve only to institutionalize a low-equilibrium trap. Actors within civil society as well as foreign investors will be convinced, because of a government’s constitutional inability to act, that existing pathologies are not likely to be overcome. Under such circumstances, we see credible commitment to bad policies. The veto-points argument might be saved if one could issue the following dictum: polities will adopt federal constitutions only after securing good government. However, such a dictum is fanciful. A realistic consideration of options suggests that we frame the matter by comparing possible changes in two kinds of multiple veto-points systems. In the first system, typified by Switzerland, the quality of governance is high and the barriers to change are high. In the second system, typified by Brazil, the quality of governance is low and the barriers to change are equally high. Now let us suppose a quasi-experimental intervention in which both polities adopt unitary constitutions. What are the attendant risks and benefits of a reduction in veto points in these two polities, holding all else constant? It seems less likely that a unitary Switzerland will experience a decline in the quality of government, though it will of course lose some degree of credible commitment (since it is now easier for policies to change). Political leaders in Brazil, on the other hand, will have much greater latitude to change policies, and these changes may improve the quality of governance (with some loss of credible commitment, as in Switzerland). Plausibly, the realistic opportunities for policy improvement in Brazil are greater than the realistic opportunities for policy decline in Switzerland. If so, the benefits of flexibility will outweigh theloss in credible commitment in both polities. The advocate of veto points might insist that the foregoing thought-experiment shows, at the very least, that Switzerland—if not Brazil—is better off with a federal structure than with a unitary structure. Good policies are locked in. But we may question even this minimal defense of multiple veto-points. The issue rests with a consideration of a much more difficult question: what kinds of policies are likely to be adopted in states with greater or lesser veto points? The existence of federal divisions may drive out bad ideas, including policies that are prejudicial to minority groups or specific regions and policies concocted in haste (without due deliberation). If so, resulting legislation is more likely to serve the general interest. As Hume 9 remarked, “In a large government, which is modeled with masterly skill . . . the parts are so distant and remote, that it is very difficult, either by intrigue, prejudice, or passion, to hurry them into any measures against the public interest” (quoted in Treisman 2000b: 9). Yet, while an increase in veto points may drive out bad ideas it may also drive out good ones. This is an elemental point, but an important one. On this fundamental issue, much depends upon one’s attitude to government. Is it a force for good or bad? Are new policies more likely to improve, or corrupt, society? For those who fear errors of commission a constitution with multiple veto points is attractive; for those who fear errors of omission it is not. ACCOUNTABILITY A fourth argument for federalism rests upon the ways in which this constitutional form may enhance political accountability and policymaking skills at local levels. Federal constitutions bring government closer to the people insofar as decisionmaking is decentralized, the personal vote (or regional vote) is stronger, and politicians are more beholden to local interests. This means that elective bodies may be more responsive to local needs and concerns, more flexible, and in a certain sense more democratic (Elazar 1987; Main 1974; Storing 1981). Citizens may have greater incentive to involve themselves in the affairs of state where the venue is small enough for the individual citizen to feel empowered (Eichenberger 1994). In any case, one can probably assume that local accountability will be higher wherever local (i.e., regional) authorities enjoy greater autonomy. That said, there is room for argument over whether local accountability actually brings in its train better governance overall (Prud’homme 1995). Consider the complications that arise when voters are asked to assign responsibility for policies across different levels of government. Voters tend to hold national and subnational officials accountable for different policy issues. This is reasonable insofar as these levels of government are responsible for different policies. But subnational policies often affect national outcomes in ways that are difficult for voters to understand. Thus, Wibbels (2000) points out that while economic policy reform is decentralized in federal systems subnational elected officials are not held accountable for these reforms because it is not an issue that voters connect with subnational governments. Consequently, Wibbles argues that the painful process of economic adjustment is made more difficult in the context of a federal system. It has also been argued that regional governments are prone to capture by special interests (Bardhan and Mookherjee 2000; Gibson 2004). Subnational governments are more likely to encompass small and homogeneous social groups and are apt to be managed by close-knit power structures (Dollard 1949). The smaller the region the more difficult it will be to sustain more than one faith, more than one major media outlet, more than one party, business chamber or civic association. The absence of rival social bases may preclude effective counterweights to official malfeasance. Where the major political, social, and economic leaders are incorporated into a relatively consolidated elite, opposition may be difficult to mobilize. 10 Following Madison, one might surmise that largeness of size and heterogeneity of constituency leads to greater transparency, greater publicity, and more effective political competition. If so, the ideal of political accountability depends upon political structures that may be difficult to maintain at local levels. Arguments about accountability inevitably turn on the organization of political parties, the quintessential linkage mechanisms in all representative democracies. Federalism generally has the effect of fragmenting party competition, leading to separate regional party systems with some degree of autonomy from national forces. Commonly, different parties are competitive at national and regional levels, as in India. Alternatively, the ideological composition of the same party may differ at national and regional levels, as in the United States. The first sort of diversity fragments the national party system; the second fragments each major party internally. Thus, in different ways federalism enhances localistic political allegiances, leading to weaker, more fissiparous national parties (Aucoin 1985b; Chandler 1987; Epstein 1964; Samuels 2003; Truman 1955). Federalism does not present an insuperable barrier to strong parties and coherent national/local party linkages, but it does make this outcome less likely. From the federalist’s perspective, this is a salubrious outcome, for over-centralized parties may insulate elites from popular control, confine and restrict the recruitment of new leaders (and hence minimize the circulation of elites), prevent insurgent reform movements within the parties, and ossify intra-party competition and accountability (Chandler 1987; Diamond 1969; Grodzins 1969; Herring 1940). The overcentralized nature of political parties has been identified with a “cartel” form of government in Europe (Katz and Mair 1996) and in Latin America (Crisp 2000). By the same token, in decentralizing power within political parties one may also decrease the level of competition across parties. Indeed, some have averred that intra-party and inter-party competition are inversely correlated (Schattschneider 1942). That is, the greater the level of intra-party competition among candidates and factions within parties, the more difficult it is for these parties to present a united front during elections and in post-election periods of governance. This absence of unity may impair accountability and depress the quality of governance, since individualistic behavior is generally oriented toward short-term, particularistic goals rather than long-term, solidaristic political goals. Federalist theory generally embraces a more individualistic political model, where relationships between constituents and their (personal) representative or their regional party are primary (Herring 1940). Unitarism embraces a partisan model, where relationships between constituents and their national party are primary (Ranney 1962; Schattschneider 1942). A priori, it is difficult to say very much about the viability of these disparate models of democratic governance. However, as an empirical observation one may offer the argument that the worst-governed polities are often those with the weakest political parties, not those with the strongest. Party institutionalization, insofar as this can be measured, seems to be correlated with good governance (Mainwaring and Scully 1995). THE SIZE OF GOVERNMENT 11 A final argument for federalism hinges upon the role of a federal constitution in constraining the growth and intrusiveness of government. By virtue of dividing political authority, democratic governments under a federal constitution should be smaller and should impose fewer regulatory burdens upon the market and upon civil society. These features of government are seen as salutary. Indeed, arguments for and against federalism hinge to a certain extent on basic assumptions about the state, the citizenry, the workings of the marketplace, and the nature of politics. Federalists generally associate good governance with small government. As Oates (1999: 1141) remarks, “one’s view of the role of intergovernmental competition clearly depends on how one views the operation of the public sector more generally.” It is probably true that federal institutions constrain government size, as measured by aggregate revenue or expenditures (Huber, Ragin and Stephens 1993; Swank 2002). However, it is much less clear that diminutive government should be regarded as a sign of good governance. Indeed, we find little or no correlation between the “bigness” of government (as measured by aggregate fiscal spending) and good governance in the realm of political, economic and human development.4 Thus, the second leg of this standard argument for a federalist constitution rests on dubious empirical grounds. “Big government” has both advantages and disadvantages. EMPIRICAL TESTS We have now reviewed five causal pathways commonly invoked by scholars to explain the beneficial effects of a federal constitution on the quality of governance in a democracy. Granted, much more could be said about each of these mechanisms, and other causal pathways might be added to this short compendium. Space constraints limit our attention to a small sample of theoretical arguments. However, the foregoing discussion incorporates those causal pathways whose policy purview is greatest—that is, they should apply across the board to all policies and policy outcomes (more or less). This is in keeping with our intent to treat federalism as a general theory of public policy, and does not preclude narrower interpretations (e.g., federalism as applied to monetary policy). In each case, we have shown reasons to doubt the veracity of federalist claims. While we cannot hope to settle basic philosophical debates about the proper role of government, we have argued here that many of the usual arguments for federalism involve dubious assumptions. Arguments based on presumed competition among subnational governments, fiscal federalism, multiple veto points, accountability, and big government do not offer conclusive reasons to suppose that federal systems will produce higher-quality governance in polities around the world. On balance, we suggest that the theoretical points reviewed above militate against federalist premises. Most important, we have provided theoretical fodder for an empirical In evaluating this relationship it is important that measures of government size include all expenditures—not just those classified as “consumption” of goods and services, a small subset of public spending. 4 12 investigation that is of great practical import. Should democratic regimes adopt federal or unitary systems of government? Most studies bearing on this question focus on the experience of one or several countries, usually those that are constitutionally federal. The strength of this literature is its attention to differences across regions, across federalisms, and through time. The weakness of this literature is its lopsided attention to a few countries (e.g., the United States, Canada, and Germany), the lack of variation on the chief independent variable of interest (since authors often focus only on federal cases), and the inability to tie constitutional structures to a wide range of policies and policy outcomes. Thus far, there have been few attempts to test the causal role of federalism on governance in a global setting, and these have been relatively narrow in their policy scope and robustness tests.5 Our approach is wide-ranging, including all countries and all years for which sufficient data are available and a variety of empirical indicators that measure the quality of governance. In taking a relatively comprehensive approach we do not mean to presume that federalism’s causal effects are, in fact, uniform across all polities and all policy areas. It is quite possible that this institution has no aggregate effect on the quality of governance (Riker 1964; 1969). Or perhaps its effect varies with the type of regime or political culture. Perhaps, that is, federalism offers a good constitutional solution to the dilemmas of politics under certain conditions (e.g., large population, large territory, great heterogeneity, or polarization).6 Conceivably, there are as many differences within as there are between each constitutional type. After all, federalism and unitarism are labels for highly variegated phenomena (Elazar 1991; Watts 1999). It is also possible that federalism has different effects along different policy dimensions—sometimes good, sometimes not so good. Note, however, that the theoretical justifications for and against federalism, as outlined in the previous section, rest upon causal mechanisms that are general in purview. If, for example, competition among subnational units leads to better public policies in federal regime then the causal effect of this mechanism should reach across a wide range of policy outcomes. The empirical range of a study should be commensurate with the empirical range of the theory under discussion. We are compelled to employ a wide range of governance indicators and a global sample of country-cases so that results can be compared across policies and policy dimensions. Another reason for the broad scope of this study is more narrowly methodological. We consider it 5 On budget deficits see Rodden (2002) and Rodden et al. (2003). On inflation see Treisman (2000a). Treisman (2000b) analyzes a somewhat larger set of outcomes in an unpublished paper that focuses on corruption, vaccination rates, and literacy rates. Treisman’s results are equivocal, but generally favorable to federalism. However, his study is only marginally about federalism, per se. Instead, Treisman’s concern is with the broader topic of decentralization, variously conceptualized. Given this broader focus, it is not surprising that he includes a smaller number of governance outcomes, specification tests, and robustness tests. Perhaps the most important difference between that study and the present one is that Treisman measures federalism as a contemporary, rather than historical, phenomenon (see discussion below). 6 Our tests (not shown) reveal no strong interaction effects between unitarism and a variety of measures of size and heterogeneity. 13 important to take multiple cuts at the material for the simple reason that any individual result (i.e., for a specific policy or polity outcome) may be spurious. Regardless of the care one exercises in the choice of control variables and tests for robustness it is still possible that empirical correlations with a given outcome variable are driven by unmeasured factors or measurement error. The cross-country regression is a blunt empirical tool, and few results are robust across all specifications (Levine and Renelt 1992; Sala-I-Martin 1997). However, if we observe a fairly consistent pattern across a wide range of dependent variables and specifications, greater confidence is warranted. Thus, for substantive and methodological reasons we prefer a wide-angle lens on the big-picture question of federalism and governance. The theoretical gain is greater and the empirical insight more robust. CLASSIFYING FEDERAL/UNITARY REGIMES As stated, we understand federalism as an institutionalized division of authority between national and subnational units. Since this sharing of sovereignty takes a variety of forms we utilize three coding categories: 0 = federal (elective regional legislatures plus constitutional recognition of subnational authority), 1 = semifederal (where there are elective legislatures at the regional level enjoying important policymaking power but in which constitutional sovereignty is reserved to the national government), and 2 = unitary.7 Since our interest is in the causal effect of constitutional structures we must be concerned not only with a country’s current status but also its history. Some countries have longer histories of federalism or unitarism than others, and in this sense have a longer experience with one or the other system. This is because countries have been independent, and democratic, for different lengths of time (and also because countries occasionally make small changes in their constitutional apparatus). Thus, although France and Botswana are both unitary, France has been unitary for a much longer period of time. Although the United States and India are both federal, the US has been federal for a much longer period of time. These histories, we assume, matter for politics and policymaking and therefore must be incorporated into any empirical analysis. We suppose that it takes time for political institutions to exert an appreciable effect on governance outcomes. A country switching from a unitary to a federal system or establishing a federal system in a newly democratic setting should not expect to see immediate, dramatic changes in the quality of governance. Instead, these effects are likely to cumulate over time as the new institutional rules begin to influence expectations and behavior. To capture this historical argument empirically we create a new variable that draws upon the annual (“raw”) scores for unitarism described above. We assume that history matters (i.e., a country’s long-run experience with constitutional institutions is what helps drive governance outcomes), but that recent history 7 Principal sources employed in coding: Alvarez (1999), Derbyshire and Derbyshire (1996), Elazar (1991), Hicken and Kasuya (2001), McHenry (1997), Watts (1997), The Database of Political Institutions (Beck et al. 2000), The Political Reference Almanac (polisci.com). An appendix, located on the journal web site, shows actual codings for all minimally democratic countries in the year 2000. 14 matters more. Thus, we calculate a moving, weighted sum of the annual scores for unitarism, beginning in 1901 and ending in the observation year. Weights capture long-term historical patterns while giving greater weight to more recent years. For example, a country’s score in 1996 is the weighted sum of its scores from 1901 to 1996. Its score in 1998 is the weighted sum of its scores from 1901 to 1998, and so on.8 Country-years figure in this weighted summation process so long as a country surpasses a minimum threshold of democracy during that year. (Recall that we are concerned with democratic governance; unitarism has a quite different meaning within authoritarian settings, if it has any meaning at all.) We employ a relatively low threshold of democracy because we wish to include as many plausible cases as possible in our analysis and because we believe that the distinct dynamics triggered by federalism or unitarism should be in evidence in any situation where there is a modicum of multi-party competition. Thus, we include a countryyear in the weighted summation process and in the following analysis if it obtains a score greater than zero (on a scale ranging from -10 to 10) on the Polity2 measure of democracy in a given year.9 With federalism/unitarism coded as a historical process, it becomes necessary to identify an operative hypothesis. One can either measure years of unitary and democratic rule or years of federal and democratic rule. Given own theoretical suppositions, drawn in part from the discussion in the previous section that gives reason to doubt the theory of federalism, we construct the empirical analysis as a test of unitarism. 8 The weights used change progressively by the observation year used in the analysis. For the observation year 1980, for example, a country’s raw score in 1901 is weighted by 1/80, its score in 1902 by 2/80, its score in 1903 by 3/80, until finally reaching a weight of 1 (80/80) in 1980. Each of these weighted, annual scores is summed for a single country into a cumulative score for a given observation year. (For the observation year 1981, the weighting denominator would be 81, and so on.) The formula for the weighting schemes is as follows: Let S be the raw score, and W the weighted score. Then Wt = t ' s ( 1900 $ ! %& t ( 1900 "# * S s s =1901 For the observation years 1980 and 2000, for example, the weighting schemes would be: 1980 W1980 = # s "1900 & 1 2 79 80 S1979 + S1980 ( * Ss = S1901 + S1902 + ...+ 80 ' 80 80 80 80 s=1901 ) %$ 2000 W 2000 = # s "1900 & 1 2 99 100 S1901 + S1902 + ...+ S1999 + S2000 ( * Ss = ' 100 100 100 100 100 s=1901 ) %$ If a country is non-democratic (receiving a Polity2 score of less than 0) in a given year, or if a country is not formally sovereign during that year, it receives a score of 0 for that year. 9 Results are robust when a higher democracy threshold is employed (e.g., Polity2>4). Note that in this process ! of historical coding there is no way to avoid conflating constitutional structure with regime-type (democratic/authoritarian), since the fact of federalism/unitarism makes sense only in a democratic context. In the subsequent analysis, we attempt to parse out the independent effects of regime-type with a separate historically-weighted variable that measures the degree and extent of democracy over the previous century. Results are robust with and without this control variable. 15 DEPENDENT VARIABLES In order to test the empirical relationship between unitarism and the quality of governance we must identify a set of outcome indicators that are measurable, valid crossnationally, and reasonably clear in their normative implications (good or bad for to the public interest).10 Length limitations preclude an extensive discussion of our choice of outcome variables (for further discussion see Gerring and Thacker forthcoming). Suffice to say that we attempt to include a wide array of governance outcomes: those that are survey-based (“soft”) and those that are directly measurable (“hard”), those that measure processes (e.g., the workings of the bureaucracy) and those that measure policy effort or policy outcomes. There is no single best way to measure the quality of governance; our only recourse is to extend the empirical purview to a wide range of plausible measures. Outcome measures fall roughly into three policy areas: political development, economic development, and human development. Indicators of political development include two measures of corruption control (one from Political Risk Services [Howell 1998] and another from the World Bank (Kaufmann, Kraay and Mastruzzi 2003). We also employ a measure of bureaucratic quality (data from PRS). Finally, we include three additional governance indicators provided by the World Bank (Kaufmann, Kraay and Mastruzzi 2003): government effectiveness, political stability, and rule of law. In each case, higher scores suggest better governance in the area of political development. We test six indicators of economic development. The number of telephone mainlines per 1000 inhabitants captures an important dimension of economic infrastructure and communication (data from World Bank 2003). Import duties as a percentage of imports provide one measure of trade protection, a negative governance outcome (ibid.). Trade openness, measured by the value of total trade divided by gross domestic product, is an outcome-based measure of a country’s trade policy (ibid.). (Although the degree of the impact of trade policies on growth and development is contested [see Krueger 1995; Rodrik 1995], the preponderance of work among economists suggests that open trade leads to greater prosperity.) To measure the overall investment climate, we use Euromoney’s country risk ratings, which are based on polls of economists and political analysts and supplemented by quantitative data such as debt ratios and access to capital markets (Euromoney 2004). The World Bank’s “regulatory quality” indicator offers a measure of the quality of the economic policymaking framework of countries around the world (Kaufmann, Kraay and Mastruzzi 2003). Finally, our broadest economic measure captures the level of prosperity within a country, as measured by real per capita GDP (World Bank 2003).11 Except in the case of import duties, higher values There is unlikely to be complete agreement over the best indicators of good government (one person’s good outcome may be another’s bad outcome), and we do not pretend to settle (or even adequately address) such issues here. We simply choose a set of outcomes that represent a wide range of views on the topic and gauge the relationship between those outcomes and unitarism. Even if there is not unanimous agreement that trade openness, for example, is in the broad public interest, it is worth knowing what the effect of constitutional structures on trade openness is. 11 We impute missing GDP data in the World Bank dataset using Penn World Tables (Heston, Summers and 10 16 indicate better economic goverance. Three variables capture the level of human development in societies around the world. The infant mortality rate (per 1000 live births, age 0 to 1) is a critical health and quality-of-life indicator (see Gerring, Thacker and Moreno forthcoming; data from World Bank 2003). Encapsulating the entire life cycle are estimates of life expectancy (World Bank 2003). Illiteracy measures the percentage of the population that is unable to read and write a sentence about daily life in their native language (World Bank 2003).12 Lower scores for infant mortality and illiteracy and higher scores for life expectancy indicate better governance performance. EMPIRICAL STRATEGY Our method of analysis is largely contingent upon the nature of the evidence at hand. Where data are not comparable over time, as with the World Bank governance indicators, we employ a between-effects estimator. This approach averages the results of a series of cross sections, with data drawn from four adjacent time periods (1996-2002). Where comparable time-series data are available, we employ a time-series-cross-section (TSCS) format. Although preferable to simple cross-sections, this sort of analysis is subject to simultaneous spatial and temporal difficulties. We cannot employ a unit-based fixed-effect research design to address spatial issues, such as unobserved heterogeneity, because our causal variable, Unitarism, does not vary sufficiently from year to year. Its movement through time is “sluggish” (Wilson and Butler 2003; see also Beck 2001, 285; Beck and Katz 2001, 492-93; Wooldrige 2002, 286). We do, however, employ a set of regional fixed effects and a geographically weighted version of the dependent variable (see below) to help remedy spatial problems. With respect to temporal issues, we employ a statistical correction for first-order autocorrelation, and a timetrend variable to control for possibly spurious correlations between heavily trended dependent and independent variables. Our TSCS regressions employ Newey-West standard errors, which assume a heteroskedastic error distribution and are a TSCS equivalent of Huber/White/sandwich, or “robust,” standard errors (Newey and West 1987). While Newey-West is a common approach in economics, it is less frequently used in political science. We employ it here because it achieves the above goals and is somewhat less computationally expensive than the alternatives. In any event, results are quite similar in other formats (e.g., with a PraisWinsten feasible generalized least squares approach with panel corrected standard errors and an AR1 correction for autocorrelation). The literature on governance suggests a wide array of political, economic, geographic, demographic, Aten 2002). 12 We interpolated data to fill in missing years based on existing trends for the infant mortality, life expectancy, and illiteracy variables. Each of these variables demonstrates a consistent, clear pattern or trend that facilitates this process to help avoid the bias inherent in missing data. 17 and cultural factors as plausible control variables. While we cannot measure all of these influences easily in a global setting, we include all the standard controls plus several not-so-standard ones that, we surmise, may also play an important role. A country’s regime history is likely to influence the quality of governance (Gerring et al.: forthcoming). Thus, we include a variable that captures the cumulative democratic history of each country over the course of the twentieth century, measured as the sum of a country’s total scores on the Polity2 democracy/authoritarian index, counting all years since 1900 and transforming the total by the natural logarithm in order to correct for non-linearities. We anticipate that a longer history of democratic rule should improve a country’s governance. A country’s level of economic development is also likely to be strongly associated with good governance. We control for this in standard fashion, by including GDP per capita (in logarithmic form). We measure this once, in 1960, rather than on an annual basis, to alleviate concerns about endogeneity between GDP and our various outcomes (World Bank 2003).13 We include regional controls for Africa, Asia and Latin America/Caribbean to capture whatever cultural, demographic, geographic, historical, and socioeconomic factors might be common to those regions. Dummy variables for Africa and Latin America/Caribbean are expected to reflect lower levels of governance in those regions compared to others, while expectations for Asia are somewhat mixed but generally more favorable. We anticipate that a significant period of socialist rule (LaPorta et al. 1999) has negative effects on our political and economic development indicators (with the exception of telephone mainlines, with which we anticipate a positive relationship) and positive effects on human development. Having an English legal origin is often thought to promote good governance (LaPorta et al. 1999). Other colonial systems are thought to have been uniformly bad for governance outcomes in the developing world and hence are encapsulated in this variable. To the extent that countries farther from the equator have better governance, latitude (absolute value, scaled to 0-1, logarithm, LaPorta et al. 1999) should correlate with better outcomes. Expectations for ethnic (and linguistic) fractionalization (Alesina et al. 2002) are more tentative; however, heterogeneity is generally expected to hamper the quality of governance in a country. To the extent that having a large population (total population, logarithm, World Bank 2003) makes certain governmental tasks more difficult, population might be expected to diminish governance quality (except in the case of investment rating, where expectations are ambiguous because population may serve as a proxy for market size). Distance (in thousands of km) from the nearest financial center (Tokyo, New York, or London) is intended to capture the negative impact of geographic distance from the ‘cores’ of the international economy. Oil (millions of barrels per day per capita) and diamond (rescaled to billions of metric carats per year per capita) production levels capture the ‘resource 13 The alternative to the use of a base year is the employment of instrumental variables in place of per capita income, but there are no viable instruments that can feasibly be used in this particular case. We anticipate that any remaining endogeneity—likely quite minimal—biases the results against our hypotheses, by “soaking up” some of the variation that might otherwise be attributed to Unitarism. Results obtained from using contemporaneous annual scores for per capita GDP, rather than the base year, yield the same substantive conclusions as those presented here. 18 curse’ (Humphreys 2004). Yet, these resources also provide sources of revenue and wealth. As such, expectations are mixed.14 We also include a control variable that measures the average value of the dependent variable across all countries, weighted by the inverse of the geographic distance (in kilometers) of each country from the country in question. (In the case of GDP per capita, we weight the average value of the dependent variable by each country’s share of trade with the observed country, rather than by the inverse of the geographic distance between the countries.) Countries lying close to one another may display similar values for extraneous reasons (culture, geography, diffusion, and so forth). Thus, we anticipate a positive sign for this variable. The inclusion of this variable in all regressions should help minimize possible spatial autocorrelation in the sample. We employ additional variables in selective regressions, as appropriate. We include Protestants in the analysis of most political development outcomes and Muslims in the estimations for various human development outcomes (both are measured as a percent of the total population). Prior research suggests that a Protestant heritage may improve state capacity (Gerring and Thacker 2004), while having a large Muslim population may impede human development (Moon 1991). Linguistic fractionalization (Alesina et al. 2002) substitutes for ethnic fractionalization in our analysis of illiteracy. Because there exists no standard benchmark model for any of these dependent variables we conduct two tests for each dependent variable. The first is a “full” model, including all variables discussed above. The second is a reduced-form model, omitting most controls that do not pass the threshold of statistical significance (p<0.10 in two-tailed tests), in the expected direction.15 We retain the geography-weighted control, the time-trend (for TSCS models), democracy stock, and GDP per capita variables in all models, regardless of statistical significance, because of our strong expectation that these variables capture important and otherwise unobserved effects.16 RESULTS Table 1 presents the results of a series of tests of the empirical relationship between Unitarism and various governance outcomes. In each case, the model fit is quite good, with F values significant at better than the 0.0001 level and R2/pseudo-R2 ranging from 0.48 to 0.90.17 Some indicators measure the export value of these last two items as a percent of all exports or of GDP. We believe that this confuses two issues—the extent of natural resources in a country and the degree of its economic development or export orientation, which is implicit in the denominator. Since it is the first, not the second, that we wish to measure we employ a “raw” measure of natural resources. 15 We retain a small number of control variables that come close to this threshold but that do not quite achieve it, if their deletion interferes with the results for other control variables. In no case did the deletion of inclusion of such a variable affect our Unitarism variable meaningfully. 16 Descriptive statistics for all variables can be found in an appendix posted on the journal web site. 17 R2 is an ordinary-least-squares concept and our TSCS regressions use a generalized least squares estimator. 14 19 Unitarism appears to be associated with higher levels of political development, but the relationship is not robust across all dependent variables and specifications. Models 1 and 2 show Unitarism to be correlated with lower levels of corruption control in the TSCS format (PRS); however, the between-effects format (models 3-4) reveals no statistically significant relationship. Unitarism is strongly related to bureaucratic quality in the full model (5), though not in the reduced-form model (6). It correlates negatively with political stability in the full model, but not in the reduced. Results for government effectiveness and rule of law are inconclusive. Results for various indicators of economic development are much stronger. Unitarism is associated (at the ninety-five percent level of confidence or better) with better telecommunications infrastructure, lower import duties, greater trade openness, higher regulatory quality, and higher levels of per capita GDP, across both full and reduced-form models. It is significant at the ninety percent confidence level in the reduced form model for investment rating. Results for our three measures of human development are also encouraging. Unitarism is significantly associated with lower infant mortality and illiteracy rates. Results for life expectancy are strongly significant in the reduced form but not in the full model. Table 2 summarizes findings for all regression tests reported in Table 1. It appears that unitary systems hold distinct advantages over federal ones across a wide range of indicators of political, economic and human development. In only one case—the full-form model for political stability—do federal structures appear to offer an advantage in good governance. Results for Unitarism are especially strong for economic and human development. We suspect that the weakness of the political development results is due at least in part to the inclusion of the democracy stock control variable, which has a predictably strong relationship with our indicators of political development. Tests that excluded that control variable yielded stronger findings for Unitarism (results not reported). In addition to considering the strength of the relationship between Unitarism and good governance in terms of statistical significance, it is worth considering the substantive strength of the relationship as well. A hypothetical scenario is instructive here. Employing the coefficients for Unitarism from those full-form models in Table 1 in which Unitarism emerges statistically significant (and keeping all control variables constant), we find that fifty years of fully unitary rule (1951-2000, in this example) are associated with an improvement of 0.15 points on the seven-point scale of both corruption control (PRS) and bureaucratic quality, fifteen percent more telephone mainlines, a five percent reduction in import duties as a percentage of GDP, fifteen percent more trade openness, a increase of 0.23 in the index of regulatory quality (a little less than one-third of a standard deviation), more than a seven percent improvement in per capita income, and In order to report a measure of fit for those estimations, we calculate a pseudo-R2 equal to the square of the correlation between the dependent variable and its predicted values in each equation. Note that the use of the geography-weighted and time trend control variables likely inflates the R2 and pseudo-R2 values obtained here. We report them as a measure of fit for the interested reader, without placing much substantive emphasis on them. 20 greater than a seven percent reduction in both infant deaths and illiteracy.18 On the other hand, the same period of unitary rule is associated with a 0.38 reduction (just under one half of a standard deviation) in political stability. Thus, while the substantive impact of Unitarism may range from modest (e.g, for corruption control, bureaucratic quality and regulatory quality) to moderate (e.g., for telephone mainlines and trade openness), it shows an important relationship with a wide variety of different outcomes. In assessing unitarism’s true impact, it is critical that we need to consider its cumulative effects across a quite broad spectrum of governance issues, rather than focusing on a single outcome. Several robustness issues arise here. Our vector of controls might be broadened to include a measure of income inequality. However, when we introduce a Gini coefficient measure of inequality to the model we find that it has little effect on the coefficient for Unitarism and inconsistent effects on various governance outcomes. Moreover, poor data coverage for various measures of inequality means that their inclusion would dramatically reduce the number of observations available for each analysis. Thus, we do not include income inequality as part of our core model. Numerous additional specification tests (not reported) included such variables as a) the number of years of sovereignty (logged), b) decade dummies (to further control for time effects), and c) alternative measures of fractionalization. None had an appreciable effect on Unitarism or on the outcomes under investigation in these tests. We also experimented with the exclusion of various controls; again, the results were robust. Another robustness issue concerns the inclusion of possibly influential country-cases. Federal/unitary systems are not equally distributed throughout the universe of democratic polities; considerably more countries are unitary than are federal. It is possible, therefore, that one or two influential cases of federalism are driving the results reported in Table 1. In order to test for this possibility, we removed each long-term federal case (seriatim) from the sample and re-ran the models. (Countries that are federal for a short period are unlikely to influence the results since there are fewer observations for these cases.) These model-alterations revealed no significant effects on our theoretical variable of concern, Unitarism. A fourth issue concerns the inclusion of observations during years when a country is only marginally democratic. Recall that we adopt a low threshold for democracy (Polity2>0), under the assumption that a minimal level of multi-party competition is sufficient to set in motion whatever causal effects federalism/unitarism might have on the quality of governance. However, it is possible that the effects of unitarism (or federalism) in a marginally democratic society are different than their effects in a fully democratic society. Thus, we re-tested all the models in Table 1, this time including only years in which a 18 The coefficients for the logged dependent variables (telephone mainlines, trade openness, GDP per capita, IMR and life expectancy) reported in Table 2 measure the effect of a 1-unit change in the independent variable on those outcomes as a proportion change in the dependent variable. Thus, a 1-unit change in the independent variable results in a change in the dependent variable of 100*β % (Wooldridge 2002). 21 country scored more than 5 on the Polity2 scale (which extends from -10 to +10). We find little substantive difference in the results reported here. Fifth, a variety of different methods of computing our historical measure of unitarism might be considered. In order to assess the possibility that our results might be dependent on one particular method of calculating this measure, we tested two alternative historical measures of the concept. Using an historical “stock” measures of unitarism with one percent and five percent depreciation rates yielded no notable changes in the substantive interpretations of the results.19 The results presented here do not appear to depend on one specific measurement technique. Of course, the assignment of the “treatment” in observational research is not random. It may be that unitary systems are more likely to be adopted where prospects for good governance are otherwise more propitious, in which case our key variable may be proxying for other, un-measured factors. In order to compensate for this potential identification problem we employed a series of instruments for Unitarism in two-stage least squares estimations of the same set of governance outcomes as shown in Table 1. Results so obtained are robust, thus providing some assurance that the effects reported here are not simply the product of a non-random choice of political institutions.20 Even so, we do not have a great deal of confidence in the two-stage models. All of the instruments that we could imagine violate at least one of the assumptions of instrumental-variable analysis (Reiss 2003): they either correlate poorly with Unitarism or they correlate with the error term (i.e., they are probable causes of good/bad governance in the contemporary period). Thus, though the instrumental-variable results are supportive, we do not report them here. In any case, we think it unlikely that the choice of constitutional institutions reflects a country’s future prospects for good (bad) governance. To be sure, whether a country becomes unitary or federal depends partly on a country’s colonial heritage, its size and heterogeneity, and on patterns of government that obtain in a regional or historical context. However, these exogenous influences are relatively easy to model and appear as controls in all our regression tests. Other factors influencing constitutional choice are more or less stochastic and do not seem to accord with a country’s proclivity to good or bad governance. In some instances, for example, federal institutions have been chosen because of their anticipated success in resolving conflict among heterogeneous groups (e.g., Canada, India, Switzerland, the United States). In other instances, unitarism has been viewed as the cure for precisely the same set of conflicts. This is the approach taken by all currently unitary states, whose populations were once—and in many cases remain—fractious and diverse (e.g., France). In short, it all depends. It is not the case, therefore, that federalism is chosen only in instances of high conflict or great heterogeneity. See the website Appendix for a summary of these results. Chosen instruments for Unitarism include democracy stock (logged), latitude, ethnic fractionalization, religious fractionalization (Alesina et al. 2002), Western Europe (dummy), state history, social conflict (a compilation of measures from Marshall 1999), instability (a compilation of measures from Banks 1994), and population size (1960, logged). 19 20 22 This raises another possibility. Perhaps unitarism is a symptom of successful statebuilding, rather than a cause. This, however, flies in the face of many countries’ experiences. Federal constitutions are associated with strong nation-states in Switzerland and the United States. And Unitary constitutions have been found in less successful states in Spain, Italy, and the United Kingdom, as witnessed by recent trends toward greater devolution and persistent regional dis-harmony. One must also consider the fact that constitution-makers generally have notoriously short timehorizons. They are usually interested in installing a system that will benefit them personally, their parties, or their constituencies. In this respect, the type of constitution a country arrives at is the product of a highly contingent political battle, with little or no bearing on a country’s long-term governance potential. We think it reasonable to regard a country’s choices among constitutional institutions as a largely stochastic phenomenon with respect to the outcomes of interest in this study: long-term patterns of good or bad governance DISCUSSION The tenor of most academic and policy-related work on federalism is favorable to federalism. However, we find in this study that there are theoretical grounds for doubting the federalist argument, and strong empirical evidence supporting unitarism. If there is a superior system of governance under democratic rules we propose that it is characterized by a centralized constitutional system with sovereignty monopolized by a national government, rather than one in which national and subnational governments share power. To be sure, there may be circumstances in which federalism serves as a useful expedient—a temporary waystation towards a unitary state—as in Iraq or the European Union in the near future (Lawson and Thacker 2003). However, we surmise that in the long run, all else being equal, most polities will be better off if they are able to form (with the general consent of the populace) unitary constitutional arrangements, either by fission or fusion. We should reiterate that what we have analyzed in this paper is a constitutional interpretation of unitarism/federalism. We do not intend to address directly the relative virtues and vices of fiscal or administrative decentralization (for a review of alternative conceptualizations see Schneider 2002; Treisman 2000b). Indeed, we surmise that constitutional centralization does not in any way preclude other aspects of governmental decentralization. Sweden, for example, is constitutionally unitary but has engineered a high level of administrative and fiscal decentralization. Thus, our findings pertain to the “classical” understanding of federalism (a la Montesquieu, Madison, and Riker), and only indirectly to contemporary writings on fiscal federalism (e.g., Oates 1999). Much more research on this topic is needed. We have spent relatively little time discussing the complex question of causal mechanisms. Why are longstanding unitary regimes associated with better governance? Our theoretical discussion suggests a variety of causal pathways. It could be because 23 competition among subnational governments derogates the quality of public policy and bureaucratic efficiency. It could be because the policies prescribed by fiscal federalism are, as a practical matter, difficult to sustain, in which case either there is no credible commitment or the policies adopted and maintained are suboptimal. It could be because multiple veto points hamper the flexibility and strength of governments, particularly in the developing world, where low equilibriums are the norm. It could be, finally, because democratic accountability works toward better governance only when mechanisms of accountability are relatively clear and centralized (in the hands of party leaders and bureaucratic chefs), rather than devolved to local-level relationships between circumscribed regional units and individual politicians. Decentralized accountability, if constitutionally entrenched, may encourage the “capture” of government bodies by powerful constituencies or particularistic interests. The question of causal pathways is enormously complex; hence, we are reticent to speculate upon which particular causal paths may be determinate. However, the strength of the empirical relationship between unitarism and good governance sets a clear agenda for future research on the role of constitutional structures in promoting good governance. 24 Table 1: Results POLITICAL DEV. Corruption control (PRS) Corruption control (WB) Bureaucratic quality Government effectiveness Political stability Rule of law (+ = good gov.) (+ = good gov.) (+ = good gov.) (+ = good gov.) (+ = good gov.) (+ = good gov.) 1 2 3 4 5 6 7 8 9 10 11 12 Unit aris m 0.002** (0.001) 0.003** (0.001) 0.001 (0.002) 0.002 (0.002) 0.002*** (0.001) 0.001 (0.001) 0.001 (0.002) 0.001 (0.002) -0.005** (0.002) -0.004 (0.002) 0.001 (0.002) 0.002 (0.002) Geo-weighted dep. var. 0.002 (0.009) 0.009 (0.006) 0.0001 (0.018) 0.024 (0.016) 0.029*** (0.008) 0.025*** (0.005) 0.008 (0.018) 0.033** (0.015) 0.019 (0.031) 0.020 (0.029) 0.026 (0.020) 0.037** (0.016) Trend -0.031*** (0.007) -0.027*** (0.007) 0.011** (0.005) 0.012** (0.005) Democracy stock (ln) 0.568*** (0.163) 0.256** (0.119) 1.147*** (0.178) 1.011*** (0.140) 0.572*** (0.119) 0.589*** (0.098) 0.977*** (0.178) 0.920*** (0.130) 0.866*** (0.227) 0.744*** (0.167) 0.929*** (0.179) 0.829*** (0.137) GDP pc (ln) (1960) 0.490*** (0.045) 0.460*** (0.033) 0.166*** (0.042) 0.182*** (0.037) 0.495*** (0.037) 0.483*** (0.033) 0.201*** (0.042) 0.224*** (0.036) 0.160*** (0.056) 0.160*** (0.047) 0.215*** (0.042) 0.218*** (0.038) Africa -0.150 (0.167) -0.204 (0.174) -0.409*** (0.133) -0.333*** (0.120) -0.163 (0.174) 0.245 (0.245) 0.001 (0.173) Asia 0.258* (0.141) 0.030 (0.163) 0.434*** (0.113) 0.514*** (0.089) 0.054 (0.163) 0.129 (0.224) 0.306* (0.163) 0.269** (0.133) Latin America/ Caribbean -0.935*** (0.118) -0.251* (0.129) -0.570*** (0.097) -0.641*** (0.077) -0.250* (0.129) 0.020 (0.191) -0.127 (0.128) -0.173* (0.099) Socialism 0.494*** (0.178) 0.142 (0.170) 0.259** (0.119) 0.051 (0.170) 0.118 (0.218) 0.188 (0.170) English legal origin -0.361*** (0.084) -0.104 (0.108) 0.482*** (0.063) -0.071 (0.109) -0.168 (0.141) 0.001 (0.108) Latitude (ln) -0.073 (0.051) 0.309*** (0.072) Ethnic fract. 0.200 (0.182) Population (ln) (1960) -0.070*** (0.021) Distance fin. center 0.034* (0.018) Oil production per capita -0.740*** (0.282) Diamond production pc 0.042** (0.020) Protestant (%) 0.009*** (0.001) 0.008*** (0.001) 0.007*** (0.002) Constant -0.184 (1.261) 1.598 (1.115) Estimator Observations Countries Sample period R2/pseudo-R2 Prob > F NW NW 1218 97 -0.846*** (0.089) -0.044** (0.020) -0.582** (0.286) 1244 99 1984-2001 0.64 0.64 0.0000 0.0000 -0.095* (0.051) 0.264*** (0.072) 0.009 (0.206) 0.021 (0.137) 0.162 (0.206) -0.522* (0.306) -0.019 (0.024) 0.079*** (0.017) 0.017 (0.024) -0.135*** (0.037) -0.011 (0.022) 0.005 (0.012) -0.021 (0.022) -0.013 (0.029) -0.032 (0.022) -0.358 (0.611) -0.402* (0.221) -0.307 (0.612) 0.093 (0.743) -0.254 (0.610) 0.060* (0.033) 0.076*** (0.018) 0.059* (0.033) 0.028 (0.042) 0.079** (0.033) 0.006*** (0.002) 0.002* (0.001) 0.003 (0.002) -8.203*** (1.219) -7.739*** (0.851) -8.026*** (0.863) -6.535*** (0.734) -7.818*** (1.218) -7.327*** (0.774) -3.949** (1.551) BE BE NW NW BE BE BE 311 118 0.345*** (0.057) 0.447*** (0.058) -0.261*** (0.097) 323 123 1996-2000 0.83 0.80 0.0000 0.0000 1218 97 1253 100 1984-2001 0.73 0.72 0.0000 0.0000 328 118 0.285*** (0.058) 341 123 1996-2000 0.80 0.78 0.0000 0.0000 0.360*** (0.106) 294 106 0.439*** (0.083) 0.260*** (0.075) 0.305*** (0.060) 0.055 (0.205) -0.125*** (0.033) 0.003 (0.024) 0.006*** (0.002) 0.004** (0.002) -3.422*** (1.062) -7.487*** (1.226) -6.763*** (0.830) BE BE BE 297 107 1996-2000 0.64 0.61 0.0000 0.0000 321 119 333 124 1996-2000 0.82 0.80 0.0000 0.0000 ECONOMIC DEV. Telephone mainlines (ln) Import Duties Trade openness (ln) Investment rating Regulatory quality GDPpc (ln) (+ = good gov.) (- = good gov.) (+ = good gov.) (+ = good gov.) (+ = good gov.) (+ = good gov.) 13 14 15 16 17 18 19 20 21 22 23 24 Unit aris m 0.002*** (0.001) 0.003*** (0.001) -0.066*** (0.008) -0.058*** (0.008) 0.002*** (0.0004) 0.002*** (0.0004) 0.021 (0.015) 0.025* (0.015) 0.003** (0.002) 0.004** (0.002) 0.001** (0.0004) 0.001** (0.0004) Geo-weighted dep. var. -0.001** (0.001) -0.001 (0.001) -0.032* (0.018) -0.017 (0.015) 0.00001 (0.001) 0.002* (0.001) 0.005 (0.005) 0.007 (0.005) 0.022 (0.017) 0.041*** (0.013) 0.038*** (0.014) 0.037*** (0.014) Trend 0.051*** (0.002) 0.051*** (0.002) -0.091*** (0.031) -0.121*** (0.028) 0.009*** (0.001) 0.009*** (0.001) 0.340*** (0.085) 0.264*** (0.085) 0.015*** (0.001) 0.015*** (0.001) Democracy stock (ln) 0.306** (0.146) 0.284** (0.133) 3.995*** (1.344) 5.881*** (0.878) -0.011 (0.048) -0.128*** (0.043) 15.266*** (2.324) 16.699*** (2.089) 0.648*** (0.169) 0.493*** (0.128) 0.464*** (0.121) 0.455*** (0.096) GDP pc (ln) (1960) 0.737*** (0.068) 0.845*** (0.054) -3.394*** (0.482) -3.716*** (0.309) 0.003 (0.014) -0.009 (0.010) 9.184*** (0.588) 9.109*** (0.551) 0.197*** (0.040) 0.176*** (0.036) 0.790*** (0.063) 0.774*** (0.051) Africa -0.849*** (0.144) -0.696*** (0.100) 3.759** (1.643) 2.624* (1.479) -0.162*** (0.050) -4.654*** (1.700) 0.126 (0.166) -0.204** (0.100) -0.277*** (0.068) Asia -0.322** (0.145) 1.928* (1.071) 0.074 (0.051) 8.215*** (1.836) 10.242*** (1.782) 0.164 (0.157) 0.148 (0.094) Latin America/ Caribbean 0.190** (0.075) 1.490* (0.850) -12.933*** (1.419) -11.112*** (1.149) 0.308** (0.129) 0.056 (0.051) Socialism 0.502*** (0.146) -9.748*** (2.237) -8.755*** (2.154) 0.206 (0.164) -0.281*** (0.099) English legal origin 0.022 (0.100) -0.012 (0.038) -0.313*** (0.033) -0.256*** (0.023) -4.252*** (1.159) 0.289*** (0.051) 0.109* (0.062) -0.229 (0.495) 0.071*** (0.026) Latitude (ln) 0.301*** (0.055) 0.303 (0.586) -0.036* (0.020) 2.327*** (0.851) Ethnic fract. -0.072 (0.120) -2.057 (1.286) 0.405*** (0.049) -1.807 (2.067) Population (ln) (1960) -0.060*** (0.013) -0.076*** (0.011) -0.678*** (0.182) -0.246*** (0.007) -0.218*** (0.007) 2.815*** (0.245) Distance fin. center -0.052*** (0.010) -0.063*** (0.012) 0.415*** (0.093) -0.045*** (0.005) -0.050*** (0.004) -0.959*** (0.232) Oil production per capita -0.525*** (0.164) -0.654*** (0.133) -2.995* (1.628) 0.146 (0.099) Diamond production pc 0.068*** (0.014) -0.312* (0.180) 0.022*** (0.007) Constant -6.292*** (0.717) -7.125*** (0.667) 28.045*** (8.766) 6.917 (5.546) 7.095*** (0.348) 7.774*** (0.305) -185.110*** (16.681) -186.648*** (16.070) -6.383*** (1.156) Estimator Observations Countries Sample period R2/pseudo-R2 Prob > F NW NW NW NW NW NW NW NW BE 2129 126 0.700*** (0.158) 2.030*** (0.652) 2189 129 1964-2000 0.79 0.77 0.0000 0.0000 1411 97 0.514*** (0.088) 1436 98 1970-2000 0.52 0.49 0.0000 0.0000 2521 126 0.094*** (0.025) 2609 131 1961-2001 0.67 0.62 0.0000 0.0000 0.605 (1.021) 0.217*** (0.046) 0.206*** (0.045) 0.175 (0.199) -0.269*** (0.075) -0.273*** (0.076) 2.697*** (0.222) 0.050** (0.024) -0.023*** (0.007) -0.017** (0.007) -0.970*** (0.221) -0.018 (0.021) -0.048*** (0.005) -0.046*** (0.006) 1.111 (2.389) -0.529 (0.566) 0.455*** (0.166) 0.464*** (0.170) 1.541*** (0.210) 0.054* (0.032) 0.134*** (0.007) 0.133*** (0.007) -4.283*** (0.768) -1.834*** (0.454) -1.710*** (0.407) BE NW NW 1544 122 3.252*** (0.753) 1579 126 1982-2001 0.80 0.80 0.0000 0.0000 0.173** (0.071) 332 119 0.169*** (0.053) -0.313*** (0.106) 345 124 1996-2000 0.71 0.67 0.0000 0.0000 2522 124 2522 124 1961-2000 0.90 0.90 0.0000 0.0000 HUMAN DEV. IMR (ln) Life expectancy (ln) Illiteracy (ln) (- = good gov.) (+ = good gov.) (- = good gov.) 25 26 27 28 29 Unit aris m -0.001*** (0.0004) -0.001*** (0.0004) 0.0001 (0.0001) 0.0002*** (0.0001) -0.001** (0.001) -0.001* (0.001) 30 Geo-weighted dep. var. 0.002** (0.001) 0.002** (0.001) 0.00001 (0.0001) -0.0002* (0.0001) 0.056*** (0.008) 0.058*** (0.008) Trend -0.032*** (0.001) -0.031*** (0.001) 0.003*** (0.0002) 0.004*** (0.0002) -0.020*** (0.002) -0.021*** (0.002) Democracy stock (ln) -0.600*** (0.073) -0.631*** (0.066) 0.034*** (0.012) 0.015** (0.007) -0.524*** (0.125) -0.481*** (0.123) GDP pc (ln) (1960) -0.312*** (0.035) -0.309*** (0.023) 0.047*** (0.006) 0.045*** (0.005) -0.753*** (0.031) -0.776*** (0.025) Africa 0.414*** (0.071) 0.436*** (0.055) -0.192*** (0.015) -0.204*** (0.013) 0.220*** (0.085) 0.262*** (0.067) Asia -0.121 (0.085) Latin America/ Caribbean 0.254*** (0.051) Socialism -0.447*** (0.078) English legal origin 0.026** (0.012) -0.135 (0.094) 0.277*** (0.039) 0.013 (0.008) 0.539*** (0.088) 0.615*** (0.065) -0.503*** (0.082) 0.018 (0.012) -2.042*** (0.157) -2.012*** (0.148) -0.014 (0.028) -0.013** (0.006) -0.335*** (0.052) -0.336*** (0.052) Latitude (ln) -0.057* (0.035) 0.018*** (0.006) 0.016*** (0.005) Ethnic fract. 0.480*** (0.062) 0.493*** (0.052) -0.064*** (0.011) -0.063*** (0.012) Population (ln) (1960) 0.031*** (0.008) 0.017** (0.007) -0.007*** (0.001) -0.005*** (0.001) Distance fin. center 0.042*** (0.005) 0.042*** (0.006) 0.001 (0.001) 0.061*** (0.008) Oil production per capita -0.414*** (0.120) 0.008 (0.025) -0.305 (0.280) Diamond production pc -0.001 (0.010) -0.003 (0.005) 0.010* (0.006) Muslim 0.005*** (0.001) 0.005*** (0.001) -0.001*** (0.0001) -0.001*** (0.0001) Linguistic fract. -0.069 (0.043) 0.018 (0.011) 0.062*** (0.008) 0.009*** (0.001) 0.008*** (0.001) 0.690*** (0.093) 0.690*** (0.093) Constant 11.532*** (0.410) 11.976*** (0.412) 3.479*** (0.064) 3.596*** (0.044) 11.692*** (0.880) 11.852*** (0.815) Estimator Observations Countries Sample period R2/pseudo-R2 Prob > F NW NW NW NW NW NW 2633 126 2682 129 1961-2001 0.82 0.80 0.0000 0.0000 NW=TSCS regression with Newey-West BE=Between-effects regression. 2634 125 2652 127 1961-2001 0.75 0.74 0.0000 0.0000 2401 108 2438 109 1961-2001 0.85 0.85 0.0000 0.0000 standard errors and a correction for first-order autocorrelation. Standard errors in parentheses * p<10%;** p<5%;*** p<1% Sample is limited to country-years that are minimally democratic (Polity2>0) Table 2: Summary Results OUTCOME: POLITICAL DEVELOPMENT Corruption control (PRS) Corruption control (WB) Bureaucratic quality Government effectiveness Political Stability Rule of Law ECONOMIC DEVELOPMENT Telephone mainlines Import Duties Trade openness Investment rating Regulatory quality GDPpc HUMAN DEVELOPMENT Infant mortality Life expectancy Illiteracy MODEL: Full Reduced ++ ++ +++ -- +++ +++ +++ ++ ++ +++ ++ +++ +++ +++ + ++ ++ +++ +++ + +++, ++, and + denote statistical significance at .01, .05 and .10 levels in the expected direction. ---, --, and - denote statistical significance at .01, .05 and .10 levels in the unexpected direction.