Authoritarian backsliding and the concentration of political power

ABSTRACT This article introduces the concept of authoritarian backsliding as a class of strategies for the concentration of incumbent political power in hybrid regimes. Such actions include manipulating elections, violating civil liberties, creating an extremely uneven playing field for the opposition, and reducing the institutional constraints on executive power. While often falling short of a full regime change, backsliding can significantly alter the level of political competition in a country and reduce the quality of its political life. This article develops a theoretically-grounded strategy to identify and measure backsliding events since 1989, showing that they have been much more common than is typically appreciated. The article also shows the utility of the concept of backsliding for better understanding regime stability. Using cross-national analysis of backsliding events from 1989–2004, we find that threats such as opposition electoral gains or economic crises in resource-dependent regimes create incentives for authoritarian backsliding.


Introduction
In recent years, scholars have noticed a decline in the quality of democracy in many regions of the world, often referring to such events as authoritarian "pushback" 1 or "autocratization". 2 How do we understand these trends? Are they cases of regime change or something else? Importantly, a large part of the decline in democratic quality has occurred in countries that were far from the ideal type of liberal (or even electoral) democracy to begin with. Instead, much of this movement has occurred in so-called hybrid regimes, where limited freedoms and the formal procedures of competitive elections are paired with institutions or practices that create an "uneven playing field" that strongly disadvantages the opposition.
These hybrid regimes, which became widespread in the years after the Cold War, have been the subject of a great deal of research in recent years. Contemporary work has focused on variation in the types of nondemocratic regimes, 3 examining which institutional and policy arrangements were most stable, 4 and the behaviour of various actors in these regimes. 5 Perhaps unsurprisingly, the driving interest behind these studies has usually been a desire to know what makes some regimes more susceptible to democratization than others. This "democratizing bias" 6though understandablehas left the field largely blind to other dynamics within hybrid regimes. While imperfect electoral competition has created a stable equilibrium and even facilitated liberalization in some hybrid regimes, in other cases elections have become markedly less competitive over time. Indeed, in countries from Russia and Albania to Tanzania and Zambia, incumbents have increased their domination of the political space and cracked down on previously permitted opposition, media freedoms, and political rightsa process we call "authoritarian backsliding". This process is different from the types of democratic failure that have previously been studied. Why do some hybrid regimes experience this type of authoritarian backsliding, while others do not? Given the supposed stability of such arrangements, why do some incumbents attempt to close off previously open political space?
Until recently, few scholars had taken on this question directly. While some focused on variations in the level of repression across regimes, 7 there has been very little explanation of variation in the level of competition across time within countries. Those who have asked such questions have typically focused most on liberalization 8 or on the emergence of hybrid regimes. 9 Scholars have yet to generate reliable cross-regional studies of increasing repression within hybrid regimes. We define this phenomenon of authoritarian backsliding as a decrease in the competitiveness (or potential for competitiveness) of the electoral playing field due to increasing concentration of power in the hands of the incumbent executive, relative to other actors. The importance of authoritarian backsliding emerges in studies of one or a few cases 10 and increasing attention has been called to the phenomenon in recent years, 11 but we still lack a way to compare backsliding crossnationally and robust theories to explain the conditions under which it is likely to occur.
Building on a review of the existing literature, we suggest that backsliding is above all a reactionary process. Based on the observation that hybrid regimes most often emerge as a result of deliberate regime choices by incumbents facing various challenges, 12 we suggest that authoritarian backsliding is the result of conditions that either enable the incumbent to more easily concentrate power or that increase the risks to the incumbent of not concentrating power. In this sense, our proposed framework integrates both structural and strategic explanations. While any number of "facts on the ground" could serve similar functions in triggering backsliding, examining backsliding as a distinct phenomenon provides ample opportunities for hypothesis testing. This allows for examination of the relationship between changes in the structural and strategic context and variation in the quality of electoral regimes. To illustrate this utility, we use our measure of backsliding to revisit two literatures on regime liberalization. Our results indicate that events often identified as opportunities for liberalization can also motivate incumbents to engage in authoritarian backsliding. Specifically, we find the likelihood of backsliding to be increased by economic crises in natural resource-dependent contexts, as well as by growing electoral threats from opposition parties. These findings provide examples of how our understanding of backsliding can enrich and concretize studies of regime dynamics.

Authoritarian backsliding
Perhaps because of the broader bias towards liberalization as a topic of study, authoritarian backsliding has lacked both conceptual rigour and a comprehensive means of identifying backsliding episodes. To clarify what precisely backsliding is (and is not), we begin by presenting a conceptual discussion of how backsliding should be defined. Based on this conceptualization, we then offer a systematic approach for identifying backsliding episodes. This approach is intended as a comprehensive way to identify members of a class of phenomena that, while varying in their particular forms, share enough core features to be considered part of the same family. 13

Defining backsliding broadly
We define authoritarian backsliding as a decrease in the competitiveness of the electoral playing field, which results from an increase in the incumbent executive's concentration of power. This can take the form of expanding formal or informal executive powers, such as abolishing term limits or granting discretion over wider areas of the budget. It can also involve increasing restrictions on institutions or actors with the ability to challenge the executive, including further repression of opposition parties, suppression of existing media, or interference with civil society groups. Moreover, backsliding often occurs in the defining areas of hybrid regimes: unfair elections become even more rigged, uneven playing fields become even more tilted, violations of civil liberties become harsher, and whatever constraints exist to check the executive become even less effective. While backsliding in an electoral authoritarian regime may sometimes result in a regime that is more appropriately classified as fully authoritarian due to the removal of all space for meaningful political competition, it does not always do so. 14 Two features of this definition are particularly important. First, authoritarian backsliding represents a shift in a country's standard level of political manipulation or in the formal institutions governing access to, and the use of, power. There is temporal variation in the level of observable violence and the amount of direct coercion utilized by any regime. 15 Even incumbents in relatively benign hybrid regimes may engage in increased manipulation or media intimidation in the months leading up to an election. This would not, however, register as backsliding on its own. The concept focuses instead on the overall level of manipulation or coercion within a regime. Thus, our relatively "benign" regime would be coded as experiencing backsliding only where the incumbent has greater capacity to engage in manipulation than in prior electoral cycles or chooses to use previously un-tapped manipulative capacity.
Second, this definition is not purely about electoral practices themselves. Instead, it takes a more holistic approach to the political playing field and considers the way in which the underlying political structure drives the relative competitiveness of elections. Electoral practices are important sites of power concentration, but they may also be derivative of other factors. For example, an incumbent who has already effectively neutralized an emerging civil society may be less likely to face protests challenging stolen elections. A decline in the quality of elections in this particular case is thus only made possible by a prior concentration of power. Alternatively, an incumbent may quietly consolidate control over national media outlets and potential sources of campaign funding in order to avoid having to directly manipulate the results of elections. In this hypothetical case, consistency in the quality of election day processes masks a more general concentration of political power. By expanding our lens beyond elections themselves, 16 our approach to backsliding allows us to avoid conflating the phenomenon (the concentration of power) with its possible effects, such as particular electoral outcomes.
Authoritarian backsliding vs the failure of democratic consolidation Prior scholarship on shifts towards greater regime repression has tended to focus on the failure of democratic institutions to constrain incumbents. "Backsliding" in this sense is treated as a decline in democratic quality or the movement away from democracy as a regime type. The conceptual outcomes that these studies address include the use of repression, 17 the erosion of democratic quality, 18 the failure of democratic consolidation, 19 or the change from a democratic to an authoritarian regime. 20 While the outcomes under examination and the methods used to study them vary, these studies all share a conceptual focus on the erosion of democracy.
Although this approach is valuable for understanding the failure of democratic consolidation, many of these studies take for granted the existence of effective constraints on incumbent action, which are deemed fundamental to democracy. Thus, arguments about economic crises as moments of unique opportunity for incumbent power-grabs, 21 or the vulnerability of presidential systems, 22 assume that there is some sort of minimally democratic equilibrium that is being disturbed. When dealing with hybrid regimes, however, this assumption largely does not apply. Indeed, the literature on electoral authoritarianism emphasizes the absence of these sorts of constraints. Backsliding in these regimes is thus a different puzzle than the decline in democratic quality among minimally established democracies. What leads incumbents to engage in backsliding when they already enjoy a dominant position? This conceptual distinction highlights the fundamental difference between authoritarian backsliding as we have defined it and the sorts of "autocratization" that other scholars have examined.

Identifying backsliding
The concept of backsliding involves ideas that are fairly hard to quantify: power, its concentration, and the capacity of actors to compete in a potentially manipulated environment. Despite these challenges, the qualitative literature on hybrid regimes has successfully engaged with many of these issues and has identified key arenas in which electoral regimes may be subject to manipulation and repressive state behaviour. To identify backsliding cross-nationally, we elaborate on this existing literature to generate a series of four dichotomous measures that capture the primary components of authoritarian backsliding.
Because our definition is necessarily broad, we require a theoretical guide to distinguish the types of concrete policies or outcomes that should qualify as authoritarian backsliding. For this, we rely on the important contributions of Steven Levitsky and Lucan Way. 23 Levitsky and Way identify four criteria by which they judge regimes to be competitive authoritarian. 24 First and most fundamentally, the regime must not qualify as "full authoritarian"elections cannot be "effectively meaningless" and repression cannot be "so severe that major civic and opposition groups cannot operate in the public arena". 25 Once this condition is met, regimes are classified as competitive authoritarian if they meet any one of the following conditions: unfair elections, violation of civil liberties, or an uneven playing field. 26 Levitsky and Way thus offer a framework for examining power concentration in electoral regimes: electoral fairness, civil liberties, the evenness of the political playing field, and formal constraints on executive power. 27 In order to identify cases of backsliding, we generate composite measures for each of these conditions from existing data sets and then produce a dichotomous measure that takes a value of 1 if backsliding has occurred in that particular area. A positive value on any of these measures in a given year indicates that a backsliding episode has occurred.

Unfair elections
Perhaps the most fundamental aspect of hybrid regimes is the manipulation of their elections. While these regimes combine formal democratic institutions (such as elections) with informal practices that reduce their competitiveness, the manner in and degree to which they are unfairly tilted towards the incumbent varies across both countries and time. Incumbents can bar major opposition candidates from running, manipulate (or falsify) parts of the election itself, or prevent the opposition from effectively campaigning through harassment or legal restrictions. An increase in any of these behavioursall of which reduce the competitiveness of the election and politics as a wholeis considered backsliding.
Authoritarian backsliding of this nature is exemplified by the 2000 presidential election in Georgia. Between 1995 and 2000, President Eduard Shevardnadze was able to consolidate his control over Georgian political life via an institutionally powerful presidency with broad powers over appointments and few checks from the legislature. 28 While Georgia's previous presidential election in 1995 was far from perfect and was conducted amidst serious capacity shortcomings, 29 observers were especially critical of the polling process in 2000. Changes to the electoral laws in the weeks leading up to the election reduced the transparency of the election commission and the legal avenues for appeal. 30 The election itself was marred by allegations of institutional bias, ballot box stuffing, and fraud reported by both domestic and international observer teams. 31 In order to identify this form of backsliding cross-nationally, we rely on data from the National Elections Across Democracy and Autocracy (NELDA) Dataset. 32 To code backsliding in the area of unfair elections, we rely on three of NELDA's dichotomous variables: "Were opposition leaders prevented from running?", "Before elections are there significant concerns that elections will not be free and fair?" and "Is there evidence that the government harassed the opposition?" Each question is assigned a value of 1 if the answer is yes and a value of 0 if the answer is no and the values for each executive election are summed. 33 We code backsliding as having occurred if there is an increase in the total value of the sum, relative to the previous executive election. 34 Violations of civil liberties A second area in which incumbents can reduce the constraints on their own power involves the civil liberties protections available to the public at large. Here again, the condition identified by Levitsky and Way is multifaceted: Is there harassment of the media? To what degree are political association and free speech restricted? Are there violent attacks on opposition or government critics? Authoritarian incumbents who have already neutralized opposition parties, for example, may still face a critical media or civil society. Actions taken to reduce their ability to hold the government accountable would be considered backsliding. Importantly, however, restrictive actions taken in the area of civil liberties must be sustained to be considered backsliding. Isolated attacks on journalists or repression of a single rally do not rise to the level of backsliding. Instead, backsliding occurs where there is a significant (and broadly applied) reduction in civil liberties.
While backsliding along these lines is frequent, the experience of Senegal is particularly illustrative. While Senegal experienced some political liberalization in the run-up to the 1993 election, 35 the playing field was far from fair and the years immediately following saw reports of increased political imprisonment and the use of torture by law enforcement. 36 While some of this backsliding behaviour was related to a low-level insurgency in the Casamance region, there was also a major crackdown on opposition leaders in early 1994. While Senegal's opposition parties did manage to unseat the incumbent party in 2000 for the first peaceful alternation of power since independence, the backsliding in civil liberties protections in 1994 temporarily halted the trajectory of reforms made in the preceding years.
To systematically identify episodes of authoritarian backsliding related to civil liberties protections, we again turn to prominent cross-national data sets. The parameters identified by Levitsky and Way pertaining to civil liberties protections are not all captured by any one source. Accordingly, we rely on three measures from two separate data sets. First, we take the Freedom of the Press status from Freedom House 37 and assign numerical values from 0 to 2 for "Free," "Partly Free," and "Not Free" status assignment. Second, we take the Association measure from the Cingranelli-Richards Human Rights Dataset (CIRI), which is also coded on a scale from 0 to 2 and indicates the degree to which government policies restrict freedom of association. 38 Third, we use the CIRI data's Physical Integrity Index as a measure of the level of attacks on political opposition. This variable measures the government's use of torture, political imprisonment, extrajudicial killing, and disappearances on a scale from 0 to 8. Each of these three measures is converted to a fraction and added to create a scale of 0 to 3. Backsliding is coded as an increase in this scale of 0.5 or more over the preceding year. 39

Uneven playing field
The third area in which electoral regimes' competitiveness can vary is the degree to which all parties have access to the "fundamentals" of electoral politics: state institutions, media coverage, and financial resources. Nearly every electoral authoritarian regime has politicized state institutions, biased media, and an inequitable distribution of resources available to political parties. Backsliding occurs when the bias favouring the incumbent sees an increase over prior election cycles. 40 Authoritarian backsliding in the form of an increasingly uneven playing field is related to that which takes the form of unfair elections, but the two are not identical. The backsliding in Mozambique in the lead-up to the 1999 national elections illustrates the distinction. In the preceding years, the incumbent Frelimo party oversaw both reconstruction efforts after a major civil war and a broad programme privatizing state assets that was tilted in favour of its supporters. 41 Reports in 1999 indicated a perception that Frelimo was misusing recovering state resources for campaign purposes, that the electoral management bodies lacked transparency, and that media was particularly biased in its favour, 42 Frelimo thus enjoyed both a growing monopoly over the state and an expanded base of private support, reinforcing its position as the dominant party.
The "playing field" aspect of electoral regimes is, in some ways, the broadest of Levitsky and Way's criteria. It includes questions regarding the equality of access to funding sources, the neutrality of state institutions, and the equality of access to media. By its very nature, this type of backsliding is difficult to observe. To identify episodes of an increasingly uneven playing field, we rely on the Data on International Election Monitoring data set, 43 which investigates the quality of elections based on the reports of election observation missions. Specifically, we rely on variables for the quality of the media environment, complaints on the conduct of the electoral commission, and the overall structural environment in which the election was conducted. The values of these variables are summed and any increase from one executive election to the next is coded as an episode of authoritarian backsliding.

Executive constraints
To Levitsky and Way's three conditions of competitive authoritarianism, we add a fourth and final measure of authoritarian backsliding. Because our concept of backsliding is not solely limited to the electoral arena, we include consideration of changes in Polity IV's Executive Constraints measure. 44 This component of the Polity index measures the degree to which executives are constrained by other actors in the government. This measure allows us to account for extra-electoral institutional changes that enable the incumbent to act with greater freedom. Examples of backsliding in this area would include constitutional amendments that grant the executive expanded power over the legislature or sustained manipulation of a formerly independent judiciary.
Such changes were implemented in 2003 in Tajikistan. Though many of the institutional constraints on the executive in Tajikistan originated in a 1997 peace agreement that concluded the country's civil war, a referendum in 2003 pushed through a number of constitutional changes that removed limitations on President Emomali Rahmonov. According to official results, 93% of voters approved an expanded term limit for the already powerful presidency and, with the phrasing of the new law, Rahmonov was given the opportunity to continue governing for up to 17 more years. 45 These changes allowed for the continuation and intensification of executive dominance in Tajik politics.
To summarize, we measure backsliding with reference to Levitsky and Way's three fundamental characteristics of hybrid regimes (unfair elections, violations of civil liberties, and an uneven playing field), and we add a fourth (low levels of executive constraints). Backsliding is measured not based on the absolute values on any of the indicators, but rather on major changes from prior values. Where a regime experiences a more repressive shift in any of these measures, we consider backsliding to have occurred. 46 While many of the types of backsliding are related, both conceptually and empirically, this approach allows us to identify events that qualify as authoritarian backsliding, whatever particular form it takes in different contexts. 47 This approach for the period 1989-2004 yields 81 distinct episodes of authoritarian backsliding, which occurred across 45 different countries, of which 27 experienced multiple backsliding episodes. 48 Twenty-one countries that met the criteria to be considered hybrid regimes experienced no backsliding. 49 While there were only six backsliding episodes in the early years of the study period, by 1993 the phenomenon was much more common, with an average of 6.25 backsliding episodes per year between 1993 and 2004, as reflected in Figure 1. Backsliding is also a regionally diverse phenomenon, with at least one backsliding episode occurring in every region but East Asia and the industrialized regions of Western Europe and North America. As Figure 2 demonstrates, backsliding events were particularly concentrated in sub-Saharan Africa and the countries of Eastern Europe and the former Soviet Union, a trend that follows the distribution of hybrid regimes in general.
Our measure of backsliding is not perfect, and it may be subject to a number of objections. Some could argue that our measure is too broad and encompasses too many distinct phenomena. Indeed, we acknowledge that backsliding can take any number of forms. Authoritarian incumbents may seek to increase their independence  vis-à-vis the legislature by pushing for constitutional amendments, or they may seek to increase their independence vis-à-vis civil society by harassing independent media or restricting freedoms of association. Either of these actions increases the incumbent's power relative to potential opposition and reduces the overall competitiveness of the political system, and so both could be considered authoritarian backsliding. We are both agnostic as to the relative severity of these diverse strategies and unwilling to arbitrarily dictate a checklist of specific behaviours that "count". Accordingly, we believe any measure of backsliding ought to be inclusive and grounded in the theoretical literature examining the fundamental features of electoral authoritarian systems. 50 This leads to a second challenge, namely that since comprehensive measures of democracy already exist, our coding may unnecessarily complicate the process of identifying authoritarian backsliding. However, summary measures of democracy from Freedom House or Polity IV are actually too inclusive to identify backsliding. Each of these measures includes a range of variables, only some of which are focused on the degree to which power is concentrated in the executive. Measures relating to political participation, for example, are beyond the scope of any operationalization of backsliding. 51 To use changes in a composite measure of democracy as a proxy for backsliding would measure a sort of "reverse democratization" that is a much broader process than the one under consideration.

Why does backsliding occur?
Much of the literature on hybrid regimes emphasizes the deliberate nature of limited openings of political space. 52 The limited political competition that is the defining feature of hybrid regimes, though importantly influenced by opposition actors' behaviour, is often a strategy of power retention for incumbents facing the circumstances of the post-Cold War world. Increases in repression can also be deliberate strategies of incumbents. 53 The level of competition within a hybrid regime (closely related to repression) is thus not random, nor are changes in this level. Where an incumbent's interests are served by the status quo, he or she has little incentive to pay the costs associated with clamping down on competition even further. 54 Backsliding thus requires that incumbents adjust their calculations of the costs and benefits of consolidating power, usually reflecting a change in the circumstancesor at least a perception of such a changethat conditions political actors' behaviour. Since each electoral regime exhibits different institutional constraints and each incumbent relies on a different configuration of resources and coalitions for support, triggers for authoritarian backsliding are likely to vary widely. Incumbents may feel threatened by a surprisingly strong electoral performance by the opposition or a regional wave of popular protest, as in the Colour Revolutions or Arab Spring. Foreign aid may be made conditional on political reform, or an economic downturn may raise popular dissatisfaction. Any of these phenomena might be enough to motivate an incumbent to engage in backsliding. Approaching backsliding as a deliberate reaction to new pressures provides a common framework for investigating these varied causal pathways.

Economic crises, opposition gains, and regime dynamics
Is this conceptualization of backsliding and theoretical framework of its causes empirically useful in understanding the dynamics of electoral regimes? In addressing this question, we revisit the literature on liberalization to show how this approach can provide a nuanced and concrete understanding of incumbent behaviour by examining the effects of economic performance and opposition electoral performance on the likelihood of backsliding. The analysis demonstrates the ways in which both structural and strategic variables can influence an incumbent's incentives to engage in backsliding.
Economic crises are a common focus in studies of regime transitions 55 and the relationship between natural resource wealth and regime type has been the subject of a large body of work and a great deal of debate. Some studies have found that natural resources promote authoritarianism 56 or reduce political competition, 57 while others find the effect to be more mixed, with some evidence that resource revenues actually strengthen democracy in some contexts. 58 Many of these studies, however, focus on transitions between autocracy and democracy or on changes in aggregate continuous measures of democracy. In the context of hybrid regimes more specifically, the key question is whether resource-dependence constrains incumbents' options for remaining in power. Several studies have found that incumbents who rely on state assets to retain their positions become vulnerable to demands for liberalization when access to those assets is diminished either by privatization 59 or economic crises. 60 The common hypothesis underlying these studies is that liberalization is more likely when incumbents lose their monopoly on politically and economically valuable resources.
Yet manipulation of economic and state patronage is not the only way that incumbents retain their position. As our discussion of backsliding shows, there are multiple avenues by which an incumbent can attempt to (re)consolidate dominance over the political sphere. While losing control over economic assets likely increases the pressure on an incumbent from would-be competitors, it also creates incentives for the incumbent to reinforce his or her position and reduce the space available for political competition.
Other scholars have looked to the opposition as a key source of liberalization. Even in hybrid regimes, opposition parties do sometimes make significant gains. In Zambia, for example, the ruling Movement for Multi-Party Democracy saw its share of seats in the National Assembly decline from 87% in the 1996 elections to just 46% in 2001. Some have thus argued that a key reason repeated elections promote liberalization is the opportunity they provide opposition parties to strengthen themselves. 61 Other studies have found that opposition electoral coalitions promote post-election liberalization 62 and that opposition unity can pressure incumbents to hold fairer elections. 63 Yet it has also been found that the level of electoral competition in a non-democratic system is not consistently related to regime breakdown. 64 Opposition parties do not become stronger in a vacuum. One of the reasons that many electoral authoritarian regimes survive as long as they do is the "menu of manipulation" available to them. 65 Just as losing control over economic assets may incentivize an incumbent to reinforce his or her position by other means, a strengthening opposition may trigger backsliding as the incumbent attempts to prevent even more robust challenges in the future.

Methods and data
To test the hypotheses that backsliding becomes more likely when incumbents' positions are threatened by structural or strategic challenges, we analyse time-series cross-national data using logistic regression with a cubic polynomial for time, following Carter and Signorino. 66 The sample includes all hybrid regimes for the period 1989-2004 and the dependent variable is the initiation of a backsliding event. This approach is equivalent to a survival model, treating a regime's level of competition as the status quo that "survives" until backsliding occurs. 67 The first hypothesis regarding threats to incumbents' economic assets is tested via two primary independent variables, each lagged by one year: the economic growth rate and the nation's income from natural resource rents, taken as a percentage of gross domestic product (GDP). 68 We also include an interaction term between these two key variables. The hypothesis derived from the literature discussed above is that governments whose resource base diminishes should be more likely than other incumbents to engage in backsliding behaviour in an attempt to reduce the threat from political competition. Thus, while relatively high levels of rents may provide the incumbent with the resources to control the political playing field, it is changes in the availability of these resources that are likely to affect backsliding. Thus, an incumbent facing an economic crisis in a highly resource-dependent state would have the greatest incentives to attempt to concentrate greater control over political competition.
The second hypothesis is that gains by the opposition will motivate an incumbent to pursue backsliding strategies in an attempt to "rebalance" the playing field in his or her favour. This is tested by including the dichotomous variable Opposition Gains, which indicates whether or not the opposition's performance in the most recent legislative or executive election represented a gain for the opposition over prior elections. 69 While the dependent variable is coded exclusively in terms of executive elections, the legislature is often a key arena for opposition. Gains in either type of election may thus motivate incumbents to reduce the space for political competition.
In addition to these main variables, we include several controls. Since the literature indicates that wealthier democracies are less likely to move towards authoritarianism, 70 we control for the country's GDP per capita. 71 Additionally, because our coding strategy includes several components that are coded only in executive election years, we control for that as well. 72 Finally, since some countries experience more than one backsliding event during the sample period, we include a count variable of the number of previous backsliding events. 73 Results: economic performance and resource dependence Table 1 presents the regression results. 74 Model 1 shows the results without an interaction term, while Model 2 includes the interaction between the economic growth rate and the relative importance of natural resources to the country's economy. While the regression table is difficult to interpret on its own, the results of Model 2 as displayed in Figure 3(a) and 3(b) provide greater clarity. Figure 3(a) shows the predicted probability of authoritarian backsliding at different levels of economic growth in a hypothetical country in which natural resource rents are equivalent to just 1% of GDP, which is roughly equivalent to Georgia in 2001 or the Dominican Republic in 1996. In such countries, we see a positive but fairly small effect of economic growth.
The findings are more interesting when compared to the results portrayed in Figure  3(b). This figure presents the probability of authoritarian backsliding in a hypothetical country with rents from natural resources amounting to at least 25% of GDP, which is roughly equivalent to Russia in 2003. Comparing Figure 3(a) and 3(b), we see that highly resource-dependent countries are more likely to experience backsliding during tough economic times than are low-resource countries. To be more concrete, countries whose prior-year natural resource rents amounted to at least 25% of GDP and whose economies experienced negative growth had a mean predicted probability of backsliding of roughly 0.34, based on Model 2. Countries facing similar negative growth but with resource rents amounting to less than 5% of GDP had a mean predicted probability of 0.09, indicating that a country such as Papua New Guinea, with high resource dependence, is three times more likely than Togo (with much lower resource dependence) to experience backsliding during tough economic times.
Countries with a high dependence on natural resources are also more likely to experience backsliding in times of economic weakness than they are during times of high economic growth. A highly resource-dependent country (>25% of GDP) experiencing a negative economic growth rate has a mean predicted probability of backsliding of 0.34, compared to 0.10 when the country is experiencing positive economic growth. For the most resource-dependent countries, the probability of authoritarian backsliding is thus nearly three times higher during tough economic times. 75 How do we square this with previous findings in the literature suggesting that incumbents who lose control over autonomous resources are more likely to be forced to liberalize? The loss of such resources, at least in the case of natural resource income, is destabilizing to an incumbent's ability to remain in power under a particular configuration of formal and informal institutions and practices. Yet the results of this destabilization are indeterminate. Our findings indicate that incumbents' hands are not tied when they lose control over some part of the system that had previously sustained them in power. While this may force them to accept liberalization in some cases,  Notes: Table 1 presents the results of a logistic regression with the initiation of a backsliding event as the dependent variable. Consecutive years of backsliding are dropped from the sample and repeated non-consecutive events are treated as new periods. All variables except Opposition Gains, Executive Election, Prior Backslides, and the cubic polynomial are lagged by one year. Robust standard errors in parentheses, clustered by country. ***p<0.01. **p<0.05. *p<0.1.
the results in Table 1 and Figures 3(a) and 3(b) indicate that incumbents who could previously depend on natural resource rents to reinforce their position may turn to more coercive or manipulative measures when those resources diminish during periods of economic crisis. If an incumbent can no longer afford to buy the support of key constituencies, outright electoral fraud or targeted coercion of the opposition may  Table 1. Shaded area indicates 95% confidence intervals; Figure 3(b) illustrates the predicted probability of backsliding in countries with prior-year resource rents amounting to 25% of GDP, based on Model 2 in Table 1. Shaded area indicates 95% confidence intervals.
become appealing strategies of political survival. While the particular strategic choices may vary across countries or over time, such behaviours reduce the level of competition within a political system and thus all qualify as authoritarian backsliding under our definition. In this sense, fraud, coercion, and co-option can be thought of as substitutes. Our results suggest that when an incumbent's ability to engage in co-option deteriorates, other avenues for power retention are available and likely to be taken.

Results: opposition gains
The results indicate that it is not only changes in incumbents' assets that lead them to pursue new strategies of power concentration, but also changes in their strategic position. When the opposition demonstrates an ability to improve its performance at the polls, this has a significant and positive effect on the likelihood of authoritarian backsliding. While the logit coefficients themselves are difficult to interpret, an examination of the marginal effects reveals that opposition electoral gains increase the probability of backsliding by 0.076. With an average baseline probability of 0.131, this represents a 58% increase in the likelihood of backsliding. The case of Burkina Faso in 2004 illustrates the process. After the 2002 National Assembly elections drastically reduced the ruling party's share of seats, the incumbent government took several steps to regain the upper hand, including returning to a proportional representation electoral system widely viewed to advantage the ruling party. Cabinet seats formerly allotted to opposition parties were also reallocated to members of the ruling party, despite their increased voice in the National Assembly. A number of meetings and demonstrations by civil society groups were also forcibly broken up. 76 These findings suggest that elections provide information to incumbents that they use to "recalibrate" their efforts to limit political competition. While prior research has found that opposition leaders have the capacity to foster liberalization via their strategic choices 77 and that the repeated experience of elections empowers the opposition to be more effective, 78 our findings suggest that this may be a double-edged sword. Where the opposition is successful in gaining votes, but not successful enough to win elections, incumbents have the motivation to engage in backsliding activities to secure their positions. 79

Conclusion
Backsliding, though relatively rare in the immediate post-Cold War period, has become a frequent occurrence in electoral regimes since 1989. It has occurred in countries from almost every geographic region and in every year in the period under investigation. Some hybrid regimes never experience backsliding, while others see a consistent decline in political competition over long periods as new backsliding events occur in all their various forms. The received wisdom that such regimes are a stable type, distinct from closed autocracies and liberal democracies, may be true, but it belies a sort of everyday volatility in how these regimes govern. By looking at changes in the qualities of hybrid regimes, rather than pure shifts in regime type, we can come to a more concrete understanding of how they function. This is especially important because contestation within hybrid regimes is a constant process of interaction between incumbents, opposition leaders, and their respective constituencies. An economic crisis in a resource-dependent state might damage the popularity of the incumbent and offer an opportunity for opposition parties to capitalize on widespread discontent, but it may also generate incentives for higher levels of political repression. Opposition parties who make significant electoral gains may find themselves in an improved bargaining position vis-à-vis the government, but they may also face added challenges as the incumbent attempts to regain his or her dominant status.
Two implications for future scholarship emerge from our findings. First, the foregoing discussion highlights the value of studying concrete aspects of political contestation, without taking them out of context. Cross-national scholarship has tended to focus either on individual regime behaviours such as the use of violence against opposition supporters or on broad shifts in regime type, often measured by changes in Polity IV or Freedom House scores. The theoretical and empirical contributions of this study suggest that there is also value in a middle approach. As scholars of hybrid regimes have long noted, incumbents seeking to remain in office can select from a "menu of manipulation". 80 The selection of, and abstention from, various strategies are likely interrelated, and examining how incumbents trade off among these options is a potentially fruitful avenue for future research.
Second, the question of whether authoritarian backsliding is effective has not yet received adequate attention. Just as economic crises may lead either to liberalization or to a repressive response, backsliding strategies may lead either to entrenchment of the incumbent or to redoubled challenges from opposition supporters. Nearly half of the countries in the sample experienced more than one backsliding event in the period under review. Is it because incumbents learn that backsliding is effective and are thus incentivized to continue reducing the competitiveness of the political arena? Or is it because backsliding is frequently ineffective and incumbents find themselves resorting to ever-more desperate measures to retain power?
These questions highlight the importance of charting a middle path between studies of regime change writ large and studies of individual tactics for power concentration. Political competition within electoral regimes varies significantly over time and these fluctuations have important implications for the quality of elections, the protection of human rights, and the prospects for democracy. In introducing a precise conceptualization of authoritarian backsliding, we offer a means of thinking about the concentration of political power that is both concrete and comprehensive. By examining patterns of backsliding and testing its empirical utility as an outcome of interest, we find that backsliding does offer a meaningful way to categorize and investigate a constellation of political behaviours that restrict political competition. Finally, by acknowledging and accounting for the substitutability of these behaviours, we have improved our ability to understand the circumstances and strategic choices that lead to their use. This approach helps to illuminate the processes underlying a type of power concentration that significantly affects the quality of political life in countries around the world.
28. Devdariani, "Georgia"; and King, "Potemkin Democracy." 29. National Democratic Institute for International Affairs, "Pre-Election Report." 30. OSCE, "Republic of Georgia." 31. Ibid., 7, 17. 32. Hyde and Marinov, "Which Elections Can be Lost." 33. We include only elections in which the office of the incumbent leader is being contested, including presidential elections and parliamentary elections where there is no separate popular election for head of government. While elections for the executive leadership are meaningful across electoral systems, the importance of legislative elections varies widely. Without a separate indicator to weigh the power of legislatures vis-à-vis executives, we cannot reliably use the change in quality from a legislative election to an executive election as a measure of the fundamental quality of the system as a whole. 34. We use the aggregate score in this way in order to account for possible trade-offs among strategies of manipulation within the category. A similar approach is taken for the other forms of backsliding. 35. Beck, "Senegal's 'Patrimonial Democrats.'" 36. Amnesty International, "Senegal: Mass Arrests"; and Amnesty International, "Senegal: Comments." 37. Freedom House, "Freedom of the Press." 38. Cingranelli, Richards, and Clay, "CIRI." The Human Rights Dataset is coded so that lower values indicate more restrictive policies. Accordingly, we invert their measures for both the "freedom of association" measure and the "physical integrity index" measure. 39. The findings in Table 1 are not driven by the selection of this cut point for the civil liberties backsliding measure. Reducing the cut point from 0.5 to 0.375 does not alter the core findings. 40. Both Unfair Elections and Uneven Playing Field are coded by executive electoral cycle rather than by year. The issues at stake in these conditions relate specifically to the conduct of elections, and coding them by year would generate a misleading set of false positives in which standard, cyclical increases in opposition harassment in the pre-election period would be mistaken for the much more robust changes underlying our concept of backsliding. 41. Pitcher,Party Politics,"Mozambique,[14][15][16]26. 43. Kelley,Monitoring Democracy. 44. Marshall, Gurr, and Jaggers, "Polity IV." 45. Political Instability Task Force, "Tajikistan." 46. We have excluded cases of military coups d'état from our list. These are generally accepted as a type of regime change, rather than a within-regime shift. Coup d'état data was taken from the Coup d'État Events List (Marshall and Marshall, "Coup d'État Events"). 47. The Supplemental Appendix provides a list of all cases of backsliding in hybrid regimes for the period 1989-2004. Due to time limitations in the data used to code backsliding, our list does not include backsliding episodes that begin after 2004. While attention to the phenomenon has grown in the wake of prominent events occurring after 2004, Figures 1 and 2 demonstrate that backsliding had been a relatively common occurrence in countries around the world for at least the preceding decade. We acknowledge the limitations of our data, but political events in countries from Venezuela to Zimbabwe that occurred after our sample period suggest that our findings likely retain external validity. 48. A number of countries also oscillate between liberalization and backsliding on several of our measures, helping to account for the repeated number of backsliding events in our relatively brief time period. A complete investigation of this phenomenon is beyond the scope of this study, but provides a potentially fruitful avenue for future work. 49. We operationalize hybrid regimes as those systems in which there are multiple parties (at least one of which is not aligned with the party in power), as coded by Cheibub, Gandhi, and Vreeland ("Democracy and Dictatorship"), and in which the overall Polity IV score is between −5 and 5, corresponding to the Polity definition of "anocracy." Alternative codings are discussed in the Supplemental Appendix. 50. It is also true that several of these types of backsliding are endogenous to one another.
Reductions in institutional executive constraints give an incumbent latitude to reduce the space for civil liberties, for example. For our purposes, this does not change our analysis of