It s all about State Structure: New Findings on Revolutionary Origins from Global Data

Transcription

1 Homo Oeconomicus 21(3): (2004) It s all about State Structure: New Findings on Revolutionary Origins from Global Data Jack A. Goldstone School of Public Policy, George Mason University, Fairfax, VA, USA. ( jgoldsto@gmu.edu) Ted Robert Gurr Dept. Government and Politics, University of Maryland, USA ( tgurr@cidcm.umd.edu) Monty G. Marshall Center for International Development and Conflict Management, University of Maryland, USA ( mmarshall@cidcm.umd.edu) Jay Ulfelder Science Applications International Corp, McLean, VA, USA ( jay_ulfelder@stanfordalumni.org) Abstract Revolutions have been attributed to economic failures, to the effects of modernization on autocracies, and to the vulnerability of certain regime types. Using global data from 1955 to 2001, this paper examines a wide range of political upheavals, and demonstrates that regime type is overwhelmingly the dominant factor behind revolutions, ethnic wars, and the overthrow of democratic regimes. However, the effect of regime type is not a simple function of the degree of democracy or autocracy. Rather, certain kinds of autocracies and democracies are highly unstable, while others kinds are stable, depending on the detailed institutions and nature of political participation under those regimes. JEL Classification H11 Keywords revolutions, democracy, institutions, conflict, stability 2004 Accedo Verlagsgesellschaft, München. ISBN ISSN

2 430 Homo Oeconomicus 21(3): 1 28 (2004) 1. Introduction In the enormous literature on revolutions, three major approaches have dominated the discussion of revolutionary origins. The first approach echoes the catch-phrase, It s the economy, stupid. In particular, the J-curve theory (Davies 1962) and relative deprivation theories (Gurr 1970; Feierabend, Feierabend, and Gurr 1972) argued that either a sudden decline in economic performance, or frustration with poor economic conditions relative to expectations raised by knowledge of conditions in richer countries, lie behind explosions of revolutionary discontent. From this approach, we would expect revolutions to be most common in states that experienced either long-term poverty or shortterm declines in macro-economic performance. 1 The second major approach focused on political culture and historical experience, drawing from what was then identified as modernization theory. In this view, as poor countries become moderately richer, a tension arises between political aspirations for participation by middle and professional classes, and the exclusionary institutions of autocratic states (Johnson 1966, Huntington 1968). Autocracies in modernizing nations thus should be the main locus of revolutionary conflicts. A third major approach emphasized the specific form and function of state institutions. Whether autocracies or democracies, states in this view are not all alike. Some autocracies are personalistic, others are run by totalitarian parties. Some are wholly dictatorial, others allow some limited political competition. Democracies also differ. Some might guarantee free competition and fully open political expression; but other democracies exhibit only some of these characteristics, and may retain some fairly authoritarian traits a relatively unconstrained chief executive who effectively selects his or her own successor; a prominent role for military or religious authorities in choosing or guiding elected leaders; the exclusion of certain groups from politics or from power; or severe limits on freedom of speech or assembly. Pointing to these differences, and drawing on Tocqueville s (1955) insights about reforming regimes being the most vulnerable to revolutionary crisis, this approach argued that regimes are most prone to revolution when their political institutions are under stress and undergoing reform (Skocpol 1979). If one were to generalize this, one could argue that the re- 1 Gurr (1970) pointed out that relative deprivation can not only be a matter of economic grievances, but can also derive from sudden exclusion or deprivation of political standing or status. Thus elites who lose or fail to attain expected status or power can also experience relative deprivation and harbor revolutionary discontents; e.g. indigenous colonial elites who can never rise to real power as long as colonial powers remain in control. However, this angle really focuses on state structure, the third major approach noted below.

3 J.A Goldstone, T.R. Gurr, M.G Marshall, and J. Ulfelder: It s All About State Structure 431 lationship between state authority and revolutionary conflict should follow an inverted U-shape curve, with the risks of revolution low in states that are fully authoritarian or fully democratic, but higher in intermediate, partially democratic regimes (Tilly 1978). So far, decisions about which of these approaches is most correct has rested mainly on analysis and comparison of compelling case studies (Goldstone 1980). Today, the third approach arguing that revolutions are the result of vulnerable state institutions and state-elite conflicts is dominant (Goldstone 2003, Goodwin 2002). Yet the precise characterization of which institutions are most vulnerable remains somewhat ad hoc, based on shifting sets of case studies. We are still lacking an empirical analysis of broad, global data that can identify the precise regime types that are prone to collapse, and moreover that can test and identify the relative impact of regime type versus economic and modernization factors on the risks of state upheaval. In other words, we still have no large-n statistical analysis for knowing which of these views is most accurate (Lieberson 1991). This paper addresses these problems by drawing on new research by the State Failure Task Force, a group of academic scholars and government officials who have collaborated to collect and analyze data on political regimes and conflict around the world from 1955 to We present the following seven findings, which we believe have major significance for policies aimed at promoting democracy and political stability around the world: 1. Full democracies are, in fact, the most stable kind of regimes. However, full democracies are not significantly more stable than full autocracies. Democrats and dictators alike can create stable states. 2. The most unstable and dangerous regimes are partial democracies, regimes that combine elections and some other elements of democracy with significant aspects of authoritarian rule. These regimes are in fact far more prone to experience revolutions, ethnic wars, and violent regime changes than any other kinds of regimes. Especially worrisome, such regimes are the most rapidly increasing type of regime in the last two decades. 3. Not all full or partial democracies are alike, however. We designate countries as full or partial democracies based on three factors: the extent to which the selection of political leadership is open and competitive; the degree to which political competition is non-violent, non-coercive, and unrestricted; and the extent to which executive power is limited by broader government institutions (each factor is rated on a separate 7 to 10-point scale). Different combinations of these factors can create the same overall democracy scores, but have very different degrees of political stability. Certain full democracies,

4 432 Homo Oeconomicus 21(3): 1 28 (2004) which we call weak full democracies, are in fact highly unstable. Certain partial democracies, which we call strong partial democracies, are almost as stable as the most stable full democracies, and more stable than some kinds of autocracies. In general, the fate of democracies is not merely a smooth curve based on degree of democracy, but instead varies according to specific characteristics of their political institutions. 4. Not all autocracies are alike, either. Autocracies that allow some political competition are more prone to violent crises than fully authoritarian regimes. 5. Economic factors do affect stability. Countries with lower levels of infant mortality, and with economies more open to international trade, do have lower rates of political crises. However, these effects are generally an order of magnitude smaller than the institutional effects of regime type. That is, the impact of getting the institutions right on the risks of violent political crisis is generally five to ten times as large as the impact of levels of poverty or trade. 6. Colonial heritage and the ethnic composition of a society have almost no effect on political stability, once economic development and political institutions are taken into account. 7. If we focus our attention on sub-saharan Africa, the effects of regime type on stability are even stronger, and again many times larger than that of economic variables. Even in this supposedly tribal region, a country s ethnic composition has no direct impact on stability. Only when ethnic differences are combined with active political discrimination against particular groups does political stability decline. On the whole, these results are hopeful for pro-democratic policies. They tell us that we can identify specific characteristics of institutions that promote stability and democracy at the same time. Even certain kinds of partial democracies can be fairly stable, it seems, if they have the right mix of institutions. However, these results also point out the hazards of an ill-founded policy of democratization. Simply loosening up an autocratic regime, or installing the wrong mix of democratic institutions, can sharply increase the likelihood that a country will face violent political turmoil. 2. The state failure task force and its approach to analyzing political crises From the late 1970s through the early 1990s, U.S. policy-makers found themselves reacting to a spate of largely unexpected events: revolutions in Iran and Nicaragua; civil wars and state collapse in Lebanon, Afghanistan, Somalia, and

5 J.A Goldstone, T.R. Gurr, M.G Marshall, and J. Ulfelder: It s All About State Structure 433 Yugoslavia; genocide in Rwanda; and the break-up of the Soviet Union. The State Failure Task Force (SFTF) was initiated in 1994 as a non-partisan effort to develop new models to guide policy-makers in understanding and forecasting these varied forms of political instability. Although the SFTF was initiated by the White House and is funded by the federal government, it consists mainly of academic scholars and statisticians outside of the government. Indeed, the SFTF was encouraged to bring in academic expertise as needed to examine a wide range of topics, and literally dozens of experts in revolution, ethnic conflict, terrorism, democracy, and various regions of the globe have been involved in SFTF discussions. In addition, the SFTF operates strictly on an open and non-classified basis. All data used by the Task Force were drawn from open sources (e.g. the U.S. Census Bureau, the World Bank, the U.N. and other similar agencies) or developed by Task Force members themselves, and the Task Force does not engage in any classified work. 2 The first step in the Task Force s research program was the development of a comprehensive list of all state failures from 1955 onward. Reflecting the diverse concerns of the government community, this list includes four types of severe political crises: Revolutionary Wars, Ethnic Wars, Violent or Adverse Regime Changes, and Genocides and Politicides. 3 As the list of historical events was developed, it was presented to various regional experts to ensure that no important events were omitted. Consequently, we believe this is the most complete listing of all kinds of adverse political crises in this period. The SFTF has employed a wide range of methodological techniques to identify those factors associated with the onset of state failure, including logistic regression, neural networks, Markov processes, and event-history analysis. The work of the Task Force is guided by two distinct, but related goals : 1) to identify factors consistently associated with the onset of state failure, and 2) to develop models that accurately predict the risks of state failure in particular countries. The forecasting objective faces inherent problems of predicting extremely rare 2 In addition to the authors of this paper, other leading contributors to the work of the SFTF have included Robert Bates, Robert Chen, Geoff Dabelko, David Epstein, Barbara Harff, Colin Kahl, Marc Levy, Michael Lustik, Tom Parris, Pamela Surko, Alan Unger, and Mark Woodward. In addition, many more scholars and government officials have contributed to the examination of particular topics and regions. For more information on the State Failure Task Force, see the State Failure website: 3 Revolutionary Wars, Ethnic Wars, and Violent Regime Changes are events involving at least 1,000 total deaths due to conflict with government authorities and at least 100 such deaths per year; adverse regime changes are sudden anti-democratic regime shifts that drop a country at least six steps on the Polity Democracy/Autocracy Scale (this typically replaces a democratic regime with a non-democratic one.) Genocides and Politicides involve government targeting of specific communities for destruction, with no precise death thresholds.

6 434 Homo Oeconomicus 21(3): 1 28 (2004) events; on average, out of roughly 150 countries in the world with populations larger than 500,000, there are only three new state-failure events in any given year. In light of this problem, the SFTF forecasting models have achieved reasonable overall success, correctly identifying 75 80% of the cases of state failure and non-failure in historical data sets, based on data from two years before the onset of such events. In this essay, however, we focus on results obtained in pursuit of the first goal: identifying factors consistently associated with the onset of state failures. For this task, fairly standard methods have yielded results that are statistically significant and robust across many different sets of data. To carry out its analysis, the SFTF has used a technique common in epidemiological analysis of risk factors for diseases that occur rarely and in diverse populations, such as cancer. According to this technique, subjects who are victims of a given disease are matched with a randomly drawn control set of disease-free individuals, and the two populations are then examined to identify factors consistently associated with a higher risk of the illness (Rothman 1986). In similar fashion, the SFTF matched its list of historical state failures with several control sets selected at random from among countries that were free of state-failure events for at least a seven-year period. The state failure cases and the control sets were then examined to identify factors consistently associated with a higher risk of state-failure events. To ensure that observations on countries experiencing failures were not contaminated by the failure events themselves, and to enhance the utility of its models for forecasting, the Task Force profiled failed states using data drawn from two years prior to the onset of a state failure. This advance snapshot was then compared with similar data for non-failure states in the same year. 4 To obtain a wide range of controls, each state-failure case was matched with three cases of non-failure. Controls were also matched by region, to ensure that the control set had the same proportion of cases from the Americas, Asia, Europe, sub-saharan Africa, and the North Africa/Middle East region as the state-failure set. 5 4 The Task Force also looked at data from one year before the onset of crisis events, and five years before. Identification of crisis events from data one year before onset was only slightly higher (3% more cases accurately identified). Identification from data five years prior was somewhat worse (10% fewer cases accurately identified) as this of course misses changes in data in the four years prior to the crisis. Still, the same causal pattern emerges with all three lag specifications. Also, it is important to note that when we say we look at data two years prior to a crisis onset, we do not mean only using data from that specific year. Rather, we mean using all information available up to two years prior to the crisis, such as for example changes in the rate of inflation or economic growth over the previous three to five years. 5 The same country can appear in both the control sets and the roster of failed states, and in both

7 J.A Goldstone, T.R. Gurr, M.G Marshall, and J. Ulfelder: It s All About State Structure 435 To identify factors associated with the onset of state failure, the Task Force assembled a large database of variables describing a wide range of political, economic, demographic, social, and environmental conditions. At present, this dataset includes more than 1,300 variables for all countries of the world from 1955 to Some coverage, unfortunately but not surprisingly, is poor. For many environmental and social factors, comprehensive global monitoring either does not exist or is only available for the most recent decade or two or for certain countries. Even for many standard economic and demographic variables (such as GDP per capita and infant mortality), for many nations and time periods, we must rely on estimated, smoothed, or interpolated data. At the same time, the Task Force has invested heavily in screening data, cross-checking data from various sources, and subjecting data to expert review, in order to obtain the most complete and reliable country-year data that it is possible to obtain. Acknowledging these issues, we have tried to identify models that are parsimonious, robust and highly significant; tenuous results based on extensive use of flimsy data are of little value. Of course, we recognized that violence in any one country might be the result of a prior history of violence in that country, or of violence in surrounding countries. We therefore looked for such relationships across our cases. Perhaps somewhat surprisingly, we found that prior histories of state failure had no effects on the odds of future revolutionary wars and violent or adverse regime change. Prior state failure events only increased the odds of future crises for ethnic wars and genocides or politicides. However, violence in two or more neighboring countries was always associated with a substantially higher risk of failure in a given country. Thus the results presented below are adjusted to account for the impact of war or violent crises in a country s neighbors, but (unless otherwise specified) not for the impact of prior state failure events. 6 cases more than once, if it has experienced several separate state failure events, or long stretches of stability. To ensure that subsequent state failures were indeed distinct events and not merely the continuation of a prior crisis, second (or third) events in a given country had to be preceded by at least five failure-free years. In addition, to ensure that control cases reflected conditions in truly stable states, any country-year used as a source of data for the control set had to be preceded by at least two failure-free years and followed by at least four failure-free years. This ensured that no countries were chosen as controls that had either just emerged from, or were about to descend into, a crisis. 6 Statisticians will realize there is also a special kind of error serial autocorrelation that arises from the fact that risks of failure in a given year strongly depend on conditions in the prior year; that is, peace tends to beget peace, and war to beget war. However, because of the case-control method used, the data being analyzed do not contain strings of observations drawn from consecutive years in the same country. Rather, both controls and problem cases are observed only in a specific year (the year two years prior to onset of the failure event). In addition, for each state failure, the data are drawn from a year of peace preceded by at least three prior years of peace (otherwise we would not be observing a new event onset). It is, of course, still possible that failure events more than three years

8 436 Homo Oeconomicus 21(3): 1 28 (2004) 3. Surprising results The SFTF has tested literally scores of variables to see if they have any association with a higher incidence of state failure. We originally expected that because state failure is a complex, many-sided process, no simple model would have much success in identifying the factors associated with failure onsets. Rather, we expected that it would require widely different models to identify the factors associated with failure onset for different regions, and for different kinds of events. Moreover, we assumed these models would need to be complex, relying on many independent variables, reflecting both their levels and rates of change, and their interaction in varied combinations. It was to our considerable surprise that these expectations turned out to be wrong. The Task Force s analysis has identified some differences across regions and types of state failure, but these differences have generally proved quite minor. Even more surprising, we have found that relatively simple models, involving just a handful of variables and no complex interactions, accurately classify 75-80% of the state-failure onsets and stable countries in the historical data. This is perhaps the most significant general finding of the Task Force s research: relatively simple models can identify the factors associated with a broad range of political violence and instability events around the world. This result is surprising, in part, because country and area experts have plausibly and often convincingly identified a host of factors as key drivers of instability in specific historical cases factors such as sudden jumps in inflation, excessive government debts, poor economic performance, excessive military spending, or ethnic heterogeneity. None of these factors, however, seems to underlie political stability or instability across the board. In fact, we believe the origins of political crises can best be understood by turning the problem on its head, asking what factors are necessary for a state to sustain stability despite the various problems economic, political, social it might encounter. The parsimony of the SFTF models seems best explained by assuming they are identifying factors associated with stability or resilience; absent those factors a host of particular incidents or conditions can trigger violent political crises. One might take this as a generalization of Tolstoy s homily about families to nations: All stable nations resemble one another; each unstable nation is unstable in its own way. 7 prior affect the risk of new events; it was this relationship that was explicitly tested and found to be non-significant for revolutionary wars and violent or adverse regime change. 7 The original is from Anna Karennina: All happy families resemble one another; each unhappy family is unhappy in its own way.

9 J.A Goldstone, T.R. Gurr, M.G Marshall, and J. Ulfelder: It s All About State Structure Regime types and state failure Given the considerations discussed at the start of this paper, one critical question confronting the SFTF was whether democratic government, and regime type more generally, was one of the key factors leading to political stability. The answer we found was a perplexing yes and no. While regime type was indeed the most important factor in the likelihood of a major political crisis, there was no smooth curve moving from full autocracy to full democracy. Rather, each type of regime autocracies, partial democracies, and full democracies could display different degrees of stability, depending on the precise mix of institutions that it exhibited. Political scientists often rely on the Polity IV dataset, developed and maintained by Monty Marshall, Keith Jaggers, and Ted Gurr, to identify countries as democracies or autocracies (2003). The Polity dataset characterizes regimes along three dimensions: the degree of openness and electoral competitiveness in the recruitment of the chief executive; the degree of institutional constraints on the authority of that chief executive; and the degree to which political competition is unrestricted, institutionalized, and cooperative rather than restricted, violent or factionalized. 8 Countries with open, competitive elections, strong and independent judiciaries and legislatures, and broad and well-institutionalized political participation score high in democratic characteristics; countries with dictatorial leaders, rubber-stamp legislatures, impotent courts, a highly restricted franchise, and factionalized political systems in which one communal group excludes others from power, score low on democratic but high on autocratic characteristics. Of course, many regimes exhibit a mix of characteristics, and the Polity scale captures this variety by combining these three dimensions into a scale spanning a range of 21 points, from a maximum democratic score of +10, to the maximum autocratic score of 10. Polity scores correspond fairly closely with other measures of governance, such as the Freedom House Index of Political Liberties (2000), but they generally offer more comprehensive coverage and provide more information on the specific characteristics of the regime in question. Full democracies are countries in which democratic characteristics are strongly dominant; they have scores of 8 to 10 on the Polity scale. Autocracies countries in which autocratic characteristics are dominant have scores ranging from 0 to 10. Although some autocracies may allow limited political com- 8 Competition is factionalized when the major political parties are not open to all and focused on particular policies, but identify themselves with particular ideological, ethnic, religious, or regional communities and are too polarized to work together, instead pursuing only goals for advancing that community.

10 438 Homo Oeconomicus 21(3): 1 28 (2004) petition, a dictator or one-party regime typically controls political outcomes. Historically, most countries have fallen into one of these two categories. The remaining countries exhibit a mix of autocratic and democratic characteristics, in which the democratic features predominate, but only slightly. These countries, with scores of 1 to 7 on the Polity scale, we label partial democracies. Partial democracy is often considered a transitional status, a brief stopping point along the path from autocracy to full democracy. Sometimes this is true. But far more often, partial democracies slide back into autocracy, or simply remain partial democracies, sometimes for very long periods. The longest-lived partial democracy in the twentieth century was the apartheid regime in South Africa, which lasted as a partial democracy for 84 years before becoming a full democracy in If we were to rate the political regime of Great Britain, it would have emerged as a partial democracy in 1688, following its Glorious Revolution, and remained one for nearly two hundred years, until the extension of voting rights with the various Reform Bills of the 19 th century. But these are the exceptions, and most partial democracies are short-lived. On average, those founded since 1955 have lasted only about seven years before backsliding to autocracy (far more common) or moving toward full democracy. However, in recent years partial democracies have become an increasingly common type of regime. At the end of 2001, some 41 regimes around the world were partial democracies, comprising roughly one-quarter of all states with populations larger than one half-million. The expansion of partial democracies is especially striking in the last decade. Since 1991, the percentage of autocracies among the world s states has fallen sharply, from 62% to 34%, but the percentage of full democracies has increased much less dramatically, from 30% to 38%. Almost all of the increase in democracy in last dozen years has been due to an increase in the percentage of partial democracies. Partial democracies are also staying around longer: as of 2001 many of those regimes were nearly a decade old, and some notably, Honduras, Malaysia, Paraguay and Sri Lanka had already endured as partial democracies for more than 20 years. The increased prevalence of partial democracy is not entirely good news. In a study of newly emerging and transitional democracies, Jack Snyder and Edward Mansfield (1995) found that such regimes often relied on bellicose nationalism to gain popular support and were thus more likely to become embroiled in international wars. In his widely read essay on illiberal democracies, a category that generally corresponds to our definition of partial democracies, Fareed Zakaria (2003) warned that populist, elected regimes which fail to respect civil liberties pose a significant and lasting threat to freedom. Our findings raise a further concern, namely, that partial democracies are exceptionally vulnerable to violent political crises and state failure. In the years from 1955 to 2001, only 14.6% of all country-years represented cases of partial

11 J.A Goldstone, T.R. Gurr, M.G Marshall, and J. Ulfelder: It s All About State Structure 439 democracy, yet more than one-third of all state failures (33.9%) during that period occurred in partial democratic regimes. 9 In other words, revolutions, ethnic wars, and violent regime changes were more than twice as likely to occur in partial democracies as one would expect if all regime types were equally prone to political violence. This is a significant clue to the origins of state failures. However, it is only a beginning. First, nearly two-thirds of state failure events still occurred in autocracies and democracies. Thus we need to look more closely into the precise features of those regimes that conduce to crises. We suspect that not all autocracies, nor all partial democracies, nor even all full democracies, have the same failure risks. Second, we need to look at other causal factors. It could be that the real causes of state failure are excessive population density, or poor economic performance, or ethnic divisions, and that partial democracies just happen to be cursed with an unfair share of these (or perhaps these factors prevent countries from easily moving to full democracy.) To test these hypotheses, we need to examine multi-variate models of state failure. We investigate both of these possibilities in the next two sections. 5. A closer look at regime types The Polity IV data are usually used to generate a single measure, the 21-point scale of autocracy to democracy. But the Polity dataset contains much more information that that. The component variables used to generate that 21-point scale provide detailed information on the processes of selecting leaders, the degree to which political institutions constrain executive authority, and the degree to which political competition is open and regulated by well-functioning institutions. We tend to think of democracy mainly in terms of the first of these three elements free and open elections for a chief executive and other government leaders. However, America s Founding Fathers were particularly concerned with the latter two elements. (In fact they sought to limit executive recruitment somewhat 9 We can examine the frequency with which countries have experienced partial democracy by counting the fraction of total country-years in which such regimes are observed. We count every year we observe a country with a particular kind of regime a country-year of that type. Since specific countries have gone into and out of existence during this period, we first need to add up the total number of country-years observed from 1955 to 2001 (6,131). Of these country-years, a majority were autocracies (55.8%), and more than a quarter were democracies (27.4%). Another 2.2% were years in which countries were experiencing foreign occupation, or a collapse of central authority, or were undergoing a regime transition, and thus had no clearly definable regime in power. 14.6% of country-years were classed as partial democracy.

12 440 Homo Oeconomicus 21(3): 1 28 (2004) by the invention of the Electoral College, which was originally intended to ensure that only individuals who had come to the attention of the national political elite could become president). To constrain executive authority, the appointed Supreme Court has the ability both to veto the acts of the elected legislature and President and to frame the scope of the latter s powers should they appear to threaten or contravene the Constitution. The President is dependent solely on the legislature to approve all government taxation and borrowing. The Founders design for Congress with a proportionately represented House and a regionally, not proportionally represented Senate was developed mainly to curb factionalism by making broad coalitions necessary to pass legislation, rather than to conform to the ideal of equal representation for all. In addition, the Founders were deeply concerned with institutionalizing political activity, calling for exceptionally frequent elections (every two years for the House, every four years for President) and rigorously guarding the rights of assembly and free speech. The SFTF found a deep wisdom in these concerns. It is ironic that while policies aimed at spreading democracy tend to focus on the first element ensuring open and competitive recruitment of chief executives by free elections it is the latter two elements that are the critical determinants of whether democratic regimes are resistant to political violence. This result became sharply apparent once we recoded and subdivided the three major regime types, using information in the Polity IV data on the constraints on executive authority and the degree of cooperative and institutionalized competition found in particular regimes. The Polity data ranks executive constraints on a scale from 1 (unlimited executive authority) to 7 (the executive has parity or subordination to other branches of government). The Polity data categorizes political competitiveness on a 10-point scale, from suppressed or restricted political competition at the low end, through coercive and factionalized competition in the middle ranges, to fully institutionalized and cooperative political competition at the high end. 10 Full and partial democracies cluster in the middle to upper half of these scales; autocracies cluster in the lower half. However, not all full democracies reach the highest categories of executive constraints (step 7, parity or subordination) and institutionalized political competition (steps 9 and 10 on the Polity competition scale), and not all autocracies show the lowest values for political competition. Note that the Polity scores do not reflect whether a political regime is presidential or parliamentary, based on two-party competition or proportional party representation. Rather, the Polity scores are based on coders assessment of the 10 Institutionalized and cooperative competition means that there are clear and recognized rules that govern competition for power and over policy and resources, and that different groups accept and play by those rules.

13 J.A Goldstone, T.R. Gurr, M.G Marshall, and J. Ulfelder: It s All About State Structure 441 functionality of political institutions with regard to executive selection, constraints and political competitiveness. Some presidential systems, such as that of the United States, place severe constraints on the executive. Other presidential systems, such as that of Mexico during the decades of dominance by the Party of Institutionalized Revolution (PRI), imposed almost none. Thus presidential and parliamentary systems of various configurations can receive high or low scores in the Polity scheme; what matters is how the institutions work to regulate access, authority, and political competition. Where executive authority is only mildly constrained, and political participation is substantially restricted, we consider democratic tendencies to be weak. Where executive authority is effectively constrained and political competition is institutionalized and smoothly functioning we consider democratic tendencies to be strong. These dimensions are found to vary somewhat independently of the degree to which countries select their leaders by open elections. For example, we identify full democracies as those countries with open and competitive elections, relatively constrained chief executives, and at least moderately well-institutionalized political competition. Yet there remains variation among this group. We apply the label strong full democracies to those full democracies that have highly institutionalized and cooperative political competition, and a chief executive who is subordinate to or on par with other government institutions. Virtually all of the advanced industrialized countries meet these conditions, as do a smattering of other countries around the world, including India, Costa Rica, and Botswana. By contrast, weak full democracies are full democracies that have achieved one but not both of these conditions. Though elections are open and competitive, such countries still have either an overly powerful chief executive or somewhat restricted or factionalized competition. (Countries with both such flaws could not be considered full democracies). Contemporary examples include Romania and Guatamala, where executive authority is not fully constrained, and Bolivia and the Moldova, where political competition is still notably factionalized or weakly institutionalized (e.g., regional elites have extensive influence on voting through vote-buying or other forms of local patronage). The same dimensions can be used to distinguish among partial democracies. Partial democracies generally have some limitations on access or competition in selecting their leaders. However, some are more successful than others in dealing with executive authority or institutionalizing political competition. According to our coding scheme, a strong partial democracy is one that has achieved at least one of these two conditions (executive parity or highly institutionalized and cooperative political competition). A weak partial democracy is one that, despite having some democratic characteristics, such as periodic open and competitive elections, meets neither of these two conditions. Turkey and Croatia provide

14 442 Homo Oeconomicus 21(3): 1 28 (2004) Percent of Total Country-Yrs Regime Type Figure 1 contemporary examples of strong partial democracy. The former has strong limits on executive authority; the latter has, in the past several years, displayed open and institutionalized competition. Nigeria and Bangladesh, with highly factional political competition and a less-than-fully constrained executive, exemplify what we mean by weak partial democracies. Even in autocracies where executive authority by definition is never subordinate or on par with that of other elements of the government we found varying degrees of political competitiveness. In most cases, autocrats either suppress political competition, or manufacture or control it to suit their ends. In some autocracies, however, the regime allows a limited capacity for independent political competition, often over seats in a legislature that wields no real authority. These regimes we labeled (clearly but perhaps a bit awkwardly) autocracies with some political competition. Contemporary examples include Angola, Jordan, and Tunisia. These refinements, which involve combining the 21-point Polity scale classification of autocracies, partial democracies, and full democracies with more detailed information on executive constraints and political competition, yield six distinct regime types. Figure 1 shows the frequency with which these regime types appeared throughout the world from 1955 to 2001.

15 J.A Goldstone, T.R. Gurr, M.G Marshall, and J. Ulfelder: It s All About State Structure 443 Autocracies (regime type 1) were by far the most frequently observed regime type, comprising almost one-half (49.6%) of all country-years. Strong full democracies (type 6) were the next most frequently observed regime type, at 21.6 %. Slightly more than one-quarter (26.6%) of all country-years represented the various intermediate regime types: autocracies with some political competition (type 2), weak and strong partial democracies (type 4), and weak full democracies (type 5). As noted above, the percentages of intermediate regime types have grown markedly in recent years. In 2001, out of 156 countries in the world with a population larger than 500,000, only 22% were autocracies; 11% were autocracies with some political competition, 16% were weak partial democracies, 10% were strong partial democracies, another 10% were weak full democracies, and 27% were strong full democracies. (Some 4% were in an interregnum.) Put another way, in 2001 nearly half of all regimes were of an intermediate type, nearly double the historical average. The implications of this situation for future state failures are troubling. Figure 2 compares the incidence of the various regime types between 1955 and 2001 (the dark bars) with the share of state-failure events that occurred in each regime type during that period (the lighter bars). Only for autocracies and full democracies is the percentage of state failures less than their observed percentage of country-years. For most of the intermediate regime types autocracies with % of total Country Yrs., % of State Failures Regime Type Figure 2

16 444 Homo Oeconomicus 21(3): 1 28 (2004) some political competition, weak partial democracies, and weak full democracies the rate of state failures is higher than their share of country years; indeed, for weak partial democracies (type 3) it is remarkably so. These high-risk regime types account, in total, for more than half (53.9%) of all the state-failure events observed from 1955 to 2001, even though they comprise less than one-quarter (21.4%) of the observed country-years. Weak partial democracies alone account for more than one-quarter (27.7%) of all state-failure onsets, even though they represent less than 10% of observed country-years. There is, however, a modest surprise in the data concerning strong partial democracies. These regimes have about the same share of state failures as would be expected from their share of total country years; in this respect, they appear to be more stable than either autocracies with some political competition or weak full democracies. To sum up, our research highlights two facts that, taken together, suggest a potentially volatile mix for the future. Fact one: intermediate regime types in general, and partial democracies in particular, have grown increasingly common in recent years. Fact two: with the exception of strong partial democracies, these intermediate regimes have historically shown exceptionally high rates of state failure. 6. Economics, demography, or politics? The relative frequencies of state failure for different types are suggestive, but regime type might still not be as important as it appears. It could be that, once we look at other factors levels of economic development, rates of economic growth, population size or density, ethnic diversity, and the like we would find that these economic and demographic factors are really driving political stability, and that they just happen to be associated with particular regime types as well. We have, in fact, tested this argument, and our results clearly indicate that the character of a country s political institutions are the single most-important factor shaping the relative risks of state failure. Figure 3 shows the odds ratios for various factors in a multivariate model of state failure, looking at all states and all state failures from 1955 to Odds ratios show how the odds of state failure vary in response to different levels of a particular variable, holding all other variables constant. For categorical variables, such as regime type, the odds ratio gives the odds of state failure for countries in any given category compared to the odds of state failure for a baseline category, all other things being equal. Thus the odds ratio for weak partial democracies gives the odds of state failure for such regimes relative to the odds of state failure for a baseline category, in this case, strong full democracies. For continuous variables, such as infant mortality, we present the odds ratio that

17 J.A Goldstone, T.R. Gurr, M.G Marshall, and J. Ulfelder: It s All About State Structure 445 Variables Figure 3 Infant Mortality Trade Openness Population Density Total Population Ethnic Composition Rate of Ec. Growth Weak Partial Dem Weak Full Dem Autocracy w/comp Strong Partial Dem compares the odds of state failure for countries at the 75th percentile of all countries on that variable, with the odds of state failure for countries at the 25th percentile, again all other things being equal. An odds ratio close to 1 means that different levels of the independent variable have relatively little impact on the odds of state failure. An odds ratio significantly greater than 1 indicates that the variable has a measurable impact on the odds of state failure. The larger the odds ratio, the larger the effect. What is immediately apparent from Figure 3 is the overwhelming impact of regime type, even when economic and demographic variables are included in the analysis. To our surprise, factors such as the rate of economic growth, population size and density, and ethnic composition had no significant effect on the risk of state failure. 11 Infant mortality and trade openness were indeed significantly 11 Ethnic composition was measured as the probability that any two randomly drawn individuals

18 446 Homo Oeconomicus 21(3): 1 28 (2004) associated with risks of state failure, and the effects of these variables flow in the expected direction: countries with lower infant mortality and higher trade openness are less vulnerable to state failure, all other things being equal. 12 For both of these factors, the relative odds of state failure between countries at the top 75 th percentile and the bottom 25 th percentile was about double (odds ratio approximately 2). Though substantial, this effect was far smaller than the effect of regime type. To be sure, to cut the risk of state failure in half would be a worthy reduction, if it could be easily achieved. The problem is that lowering a country s infant mortality rate from among the highest to among the lowest in the world, or changing its involvement in international trade to a similar degree, is no small task. While measurable progress in these areas often can be readily achieved through foreign aid or improved policies, it takes a very large swing in a country s global position on these variables to create a 50% reduction in its odds of state failure. By contrast, the effects of variations in regime type are much larger. Indeed, controlling for economic factors hardly dents the enormously higher risks confronting the intermediate regimes. 13 The odds ratio for weak partial democracies is over 17, roughly ten times higher than the odds ratios associated with high levels of infant mortality or low levels of trade openness. At roughly 7, the odds ratio for both weak full democracies and autocracies with some political competition is only half as large as the ratio for weak partial democracies, but it is still quite high compared with the economic and demographic factors in the model. However, the odds ratio for strong partial democracies, though significantly above 1, is in the same range as economic factors, at about 2. In short, the three high-risk regime types autocracies with some competition, weak full democracies, and especially weak partial democracies are associated with the greatest increases in relative risk. Of course, the SFTF has tried to improve upon these results, testing dozens of additional variables in numerous combinations. Among the other factors we have examined are the age structure of the population (e.g. whether there was a youth bulge ), the rates of government spending and taxation, inflation and foreign aid, trade in specific commodities, urbanization, and many more. We have will be from the same ethnic group; the value of this index is close to 1 for ethnically homogeneous societies and close to 0 for highly heterogeneous societies. 12 To normalize for the steady global decline in deaths per 1,000 live births over the past 50 years, infant mortality was measured relative to the global median in a given year. Trade openness was measured as imports plus exports over GDP, the standard practice in the social sciences. 13 The reference category in the odds ratios reported here for the various regime types are strong full democracies and autocracies with no competition, which have roughly the same level of risk when other factors are controlled.

19 J.A Goldstone, T.R. Gurr, M.G Marshall, and J. Ulfelder: It s All About State Structure 447 also looked for non-linear and interactive effects. Through all of these permutations, no additional variable, alone or in combination with others, has significantly improved the ability of the model to discriminate between control cases and state failures. In fact, the simple model presented here (along with a factor for the presence of violent conflict in neighboring states) accurately identified 75% of the state failures and 76% of the controls in the historical data. Moreover, this level of accuracy was repeated on two additional randomly drawn control sets, and for smaller samples drawn from just the pre and post Cold War eras. The import of these results is straightforward: the difference between a weak partial democracy and a strong partial democracy appears to be an increase in the relative odds of state failure by over 700% for the weak form of this type of regime. Similarly, the relative odds of state failure are some 700% higher in an autocracy with some competition than an autocracy with none. Even for full democracies, the weak form of this regime exhibits relative odds of state failure some 700% higher than for the strong form. These are notable effects. 7. Are things different in Africa? Even though in the preceding analysis we have controlled for region, and for economic development (via infant mortality and trade), it might be argued that we are still missing something important. That is, the global model could simply be picking up on some hidden ways in which rich and democratic countries in Europe and North America are far more stable than poorer and less democratic countries, for example in sub-saharan Africa. Thus the global model might produce reasonable inferences about the stability of, say, Canada compared to Ethiopia; but it might be of little value in explaining, or guiding us in ameliorating, political violence among poorer countries themselves. To test this possibility, the SFTF set out to develop a separate model to identify state failures from 1955 to 2001 within sub-saharan Africa, the world s poorest region and, according to our research, the one most frequently beset by state failure. Figure 4 displays the odds ratios derived from this model, using only cases of state failure and control sets drawn from within this region. Looking at this figure, it is immediately clear that even within this potentially exceptional part of the world, the same factors stand out. To be sure, the model differs in some details. Because most countries in Africa have high levels of infant mortality, this factor is no longer statistically significant. If we take infant mortality as an index of economic development, then while differences between rich and poor countries were a factor in the global model of state failures, the more modest differences in development within sub-saharan Africa apparently play no significant role in shifting the relative odds of state failure in that region.