Patterns of democratic backsliding in third-wave democracies: a sequence analysis perspective

ABSTRACT

Democracy has come under pressure worldwide, with growing concern over an apparent reverse wave of democratic backsliding at the global level. Bridging conceptual approaches and empirical research, this article investigates patterns of democratic backsliding in third-wave democracies. It applies a range of innovative sequence analysis techniques to the Varieties of Democracy dataset to provide a dynamic perspective on the evolution of different types of democratic safeguards against executive expansion. The resulting typology differentiates stable trajectories from different patterns of backsliding and sheds light on the diversity of backsliding processes that diverge in their shape, depth, and timing in respect to initial democratic transition. The findings contribute to broader debates on the nature of democratic backsliding and have important implications both for our theoretical understanding of the phenomenon and the practical responses devised to counter backsliding trends.

KEYWORDS:

• accountability
• autocratisation
• democratic backsliding
• sequencing
• sequence analysis
• third-wave democracies

Introduction

Democracies have come under pressure worldwide. The swift expansion of democratization from the mid-1970s onwards has given way to growing concern over an apparent reverse trend of “autocratisation” or “democratic backsliding” at the global level.¹ The term “democratic backsliding” denotes a process of gradual dismantling of domestic checks and balances generally carried out by an increasingly dominant executive.² At the same time, authors have cautioned against lumping together widely varying phenomena under the common term of “democratic backsliding,”³ advocating for a more differentiated approach in both conceptually and empirically.

This study responds to this call by introducing a novel sequence analysis perspective to disentangle democratic trajectories across third-wave democracies and distinguish specific patterns of democratic erosion. Exploring the diversity of backsliding trajectories, as well as differentiating them from instances of stable democratic performance, is relevant for several reasons. In conceptual terms, it pushes us to develop a clear delineation of the different types of democratic safeguards that determine the quality of a given democracy, thus overcoming the tautological definition of democratic backsliding as a “decline in democratic quality.”⁴ Empirically, our analysis explores the heterogeneity of democratic trajectories to produce a typology of distinct types of backsliding that can serve as a starting point for future studies investigating the causes leading to these distinct outcomes. Finally, in practical terms a better understanding of the shape and sequencing of backsliding trajectories can point to early warning signs of democratic erosion that may inform efforts to tackle such trends before they become entrenched.

Democratic backsliding has been defined as the “state-led debilitation or elimination of any of the political institutions that sustain an existing democracy.”⁵ In contrast to the earlier blatant attacks against democracy leading to democratic breakdown, democratic backsliding tends to take more subtle forms, whereby an elected government gradually erodes democratic safeguards to the point of dismantling them completely.⁶ To pinpoint how and to what extent such processes occur, we provide a multidimensional perspective on the evolution of democratic quality that surveys the evolution of three types of democratic safeguards over time: vertical safeguards relate to the formal electoral process and electoral turnout; diagonal safeguards comprise freedom of expression and association, and free media; and horizontal safeguards encompass an independent parliament and judiciary.

Our empirical analysis focuses on the democracies of the “third wave” of democratization that encompasses 79 countries that experienced democratic transition from 1974 onwards. Previous research has emphasized the distinct conditions in which third-wave democracies developed compared older democracies as well as the considerable variation in democratic outcomes among these groups of countries.⁷ While first-wave democratization episodes mostly concluded successfully and the second wave was dominated by failures, democratization episodes since the end of the Cold War have resulted in a comparable number of successful and failed attempts.⁸ This diversity makes third-wave democracies a particularly promising universe of cases in which to study the presence and unfolding of democratic backsliding.

Our research design leverages sequence analysis,⁹ a method used to enable a systematic description and comparison of time series with categorical data that is sensitive to the timing, order, and duration of states in a given sequence.¹⁰ This approach provides a novel bird's eye view perspective on the dynamics of democratization and its reversal. The particular multichannel variant of sequence analysis, which was developed to compare life course developments on different dimensions in parallel, matches our multidimensional understanding of democratic safeguards. We use the optimal matching algorithm to compare cases and build clusters of distinct types of democratic trajectories.

Our study makes several contributions to ongoing debates around democratic backsliding. First, it offers a multidimensional perspective on the evolution of democratic quality that pays explicit attention to the sequencing of backsliding processes. By delineating different types of democratic safeguards, it highlights the centrality of executive attacks against diagonal safeguards that dominates several types of backsliding trajectories and tends to drive broader assaults on democratic quality. Second, we build a typology of third-wave democracies that distinguishes stable from backsliding trajectories and nuances different patterns of backsliding. Our analysis shows that the phenomenon of backsliding follows no universal template. Instead, we differentiate instances of full democratic reversal from those where backsliding remains confined to individual dimensions of democracy or is ongoing. These insights have important implications for our theoretical understanding of the phenomenon and the development of appropriate responses to backsliding trends. Finally, our study illustrates the benefits of sequence analysis for investigating macro-political dynamics that may inform future research on processes of democratization, autocratisation, and beyond.

We begin by placing our analysis in the context of existing research on democratic backsliding and the sequencing of such processes. We then spell out the multidimensional understanding of democracy and describe the three types of democratic safeguards against executive expansion that underpin our analysis. The next section addresses our operationalization of these safeguards and explains the use sequence analysis methods and their benefits for studying backsliding dynamics. Our empirical findings pinpoint distinct clusters of third-wave democratic trajectories and build a typology that distinguishes different forms of stable and backsliding patterns. The discussion and conclusion highlight the broader insights that emerge from our analysis and set out the theoretical, practical, and methodological implications of our findings.

A sequencing approach to democratic backsliding

As democratic backsliding becomes increasingly widespread, a lively debate has emerged around the appropriate conceptualization and measurement of the phenomenon.¹¹ For some scholars, it represents a specific subtype of the broader term “autocratisation” that encompasses any form of development away from democracy.¹² A widely cited recent study contends that a “third wave of autocratisation” has begun as early as 1994.¹³ Generally speaking, the availability of more fine-grained democracy indices has facilitated the analysis of democratic backsliding in different contexts, be it the post-communist region,¹⁴ third-wave democracies¹⁵ or at the global level.¹⁶ In parallel, scholars have produced in-depth case studies on democratic backsliding in specific countries.¹⁷ Despite this growing body of literature, the precise nature and sequencing of backsliding remains an underexplored dimension.¹⁸ By assessing distinct patterns in the unfolding of backsliding trajectories, we respond to calls to study democratic backsliding “as a process, and not as an ‘event’ or an ‘instance.’”¹⁹

The question of sequencing is a long-standing concern among democracy scholars. Recent contributions on the temporal order of specific democratization steps have highlighted a historic shift in sequencing.²⁰ In contrast with “old” democracies, where the emergence of institutions tended to precede the gradual extension of suffrage, third-wave democracies were shown to have “started democratization backwards,”²¹ introducing free elections first and only subsequently developing the rule of law and civil society as well as horizontal checks and balances.²² Mainwaring and Bizzarro highlight the variation in democratic outcomes among third-wave democracies, distinguishing “democratic breakdowns, erosions, stagnations, advances, and cases where regimes have remained highly democratic without major advances” (emphasis in original).²³ Against this backdrop, our analysis proposes a multidimensional exploration of democratic trajectories in third-wave democracies that focuses specifically on the presence and shape of democratic backsliding among this group of countries.

Several recent contributions seek to disentangle how, when, and in which order backsliding processes unfold. Analysing the full range of countries contained in the Varieties of Democracy dataset, Coppedge points to two distinct pathways to democratic erosion, with one characterized by attacks against civil liberties and free media, and the other by an erosion of horizontal accountability.²⁴ Jee et al. offer a conceptual differentiation of the dimensions along which backsliding may occur, distinguishing the electoral arena, the erosion of constraints on executive power, and challenges to democratic politics by powerful non-political actors.²⁵ Surveying specific cases of autocratising countries on the basis of selected indicators of democratic decline, Maerz et al. highlight that backsliding sequences tend to begin with attacks against media freedom and civil society and eventually extend to an erosion of free and fair elections as the very core of democracy.²⁶ Finally, Kneuer inductively derives an actor-centred “logic of action” for democratic backsliding based on the Venezuelan case to show how the intentional erosion of civic freedoms is enabled by specific mobilization and legitimation strategies that allow an “erosion agent” to be elected into power in the first place.²⁷

The present study contributes a meso-level perspective to these debates by offering a theory-guided exploration of democratic trajectories in third-wave democracies. To do so, it develops a multidimensional conception of democracy that distinguishes different types of democratic safeguards, which we detail in the following. This approach allows us to go beyond describing broad patterns of democratic development by zooming into separate clusters of cases that represent distinct forms of backsliding, and establish their shared characteristics.

Studying democratic backsliding as a multi-dimensional process

To make the notion of democratic backsliding amenable to empirical study, it is necessary to start out by providing a clear understanding of the different dimensions of democracy that may come under attack during such processes.²⁸ We put forward a multidimensional understanding of democracy as a system that serves to limit citizens’ domination by their leaders through specific democratic safeguards.²⁹ Democratic backsliding, in turn, is a process that threatens to weaken or undermine such safeguards. This conceptualization reflects earlier definitions of democratic backsliding as a process of “executive aggrandizement”³⁰ or “incumbent takeover”³¹ via the gradual erosion of domestic checks and balances by an elected leader. Building on previous distinctions between different types of accountability,³² we differentiate three types of democratic safeguards: vertical safeguards concern the electoral process; diagonal safeguards focus on freedom of expression and mechanisms of contestation; and horizontal safeguards revolve around executive constraints.

Our approach resonates with previous attempts to locate empirically the deficiencies of democracies on specific dimensions. Dahl's understanding of polyarchies as “relatively (but incompletely) democratized regimes”³³ offered a pioneering attempt to overcome binary definitions of democracy by posing “more-or-less questions rather than whether-or-not questions.”³⁴ Merkel's concept of “embedded democracy”³⁵ differentiates full-fledged liberal democracies embedded in different “partial regimes” from a range of “diminished subtype[s] of democracy.”³⁶ The three-fold distinction of democratic safeguards we adopt most closely resembles a recent effort by Boese et al.³⁷ to visualize diverging authority patterns in a three-dimensional space that classifies polities according to the election of the executive, constraints on executive decision-making authority, and the extent of political participation.³⁸

Based on this conceptual approach to democratic backsliding, we define a series of states for each of the three democratic safeguards we distinguish. This so-called “alphabet,” in sequence analysis terminology, lies at the heart of our subsequent effort to discriminate patterns of democratic trajectories among third-wave democracies by evaluating the extent to which different democratic safeguards are developed, undermined, or not developed at all over time.³⁹ As shown in Table 1, we distinguish states based on the source from which the relative weakness or strength of a given safeguard originates. Structural weaknesses concern the formal underdevelopment of a certain dimension of the democratic system, whereas agency-based weaknesses stem from an actor-centred feature, be it a deliberate attack against a given safeguard conducted by executive actors, or the failure of voters to engage in a formally well-established electoral process. It is important to clarify that the different possible states we define for each of our three democratic safeguards do not represent a mere variation in degree, as may result from segmenting indices into different parts. Moreover, there is no expectation that democratic trajectories evolve along a specific order of states. Instead, our states discriminate in a largely qualitative fashion between distinct configurations that may characterize a given democratic safeguard in a given country-year.

Table 1. Conceptualization of democratic safeguards.

	Structure-focused	Agency-focused
Vertical safeguards	Weak vertical safeguards Intermediary vertical safeguards Established vertical safeguards	Disengagement Hollowing
	Strong vertical safeguards^a
Diagonal safeguards	Weak diagonal safeguards Intermediary diagonal safeguards Strong diagonal safeguards	Partial attacks on diagonal safeguards Resilient diagonal safeguards Dismantled diagonal safeguards
Horizontal safeguards	Weak horizontal safeguards Intermediary horizontal safeguards Strong horizontal safeguards	Weak parliament Weak judiciary Dismantled horizontal safeguards

^aNote that the state “strong vertical safeguards” combines structural and agency-focused elements to similar degrees and is therefore placed between the two categories. For further details, see section II of online appendix.

For vertical safeguards, we focus on the electoral process as the core of any democratic system. We differentiate states based on the quality of the formal set-up of elections, the presence of different forms of election manipulation, as well as the level of turnout as an indicator of citizens’ engagement in democratic processes. Two potential forms of weakness or deterioration in the quality of vertical democratic safeguards are distinguished: a first set concerns the electoral process itself, be it due to a lack of resources provided to electoral institutions or the presence of electoral manipulation. These states, which we describe as weak, intermediary, established or strong vertical safeguards, depending on the overall quality of the formal electoral process, concern free and fair elections as the core of a democratic system. A second set of states distinguishes the relative level of citizen engagement in elections as measured by turnout citizens’ disengagement from the democratic process, which has been described as a distinct subtype of democratic backsliding.⁴⁰ While low participation thus does not amount to a decline of democratic quality per se, eroding levels of turnout may make a democratic system more vulnerable to direct attacks on other dimensions. We distinguish here between disengagement and hollowing, which describe a formally intact electoral process characterized by low or very low turnout levels.

Our definition of diagonal democratic safeguards centres on the competitive character of a regime, and on citizens’ capacity to challenge and oppose leaders or specific decisions.⁴¹ To discriminate different possible states for this dimension, we examine the relative degree of freedom of expression, along with mechanisms of contestation such as media freedom and freedom of association, and possible government attacks against these mechanisms. For structure-focused states, we distinguish between weak, intermediary, and strong diagonal safeguards based on the relative development of formal diagonal safeguards in the absence of direct attacks by the government. Agency-focused states emphasize the presence and relative strength of government attacks against media freedom and/or freedom of association. Partial attacks happen when only some diagonal safeguards are concerned, resilient diagonal safeguards where strong attacks occur in the presence of relatively strong diagonal safeguards, and dismantled diagonal safeguards where strong attacks result in a substantial weakening of freedom of expression, media freedom, and freedom of association.

Finally, we examine horizontal safeguards in the form of institutional constraints upon executive power. These are often seen as the discriminating factor between developing and full-fledged democracies.⁴² Where horizontal safeguards are strong, authoritarian-leaning leaders will be limited in their ability to undermine checks and balances, which provide citizens with “an effective means of control over political institutions.”⁴³ We differentiate states according to the existence of a strong parliament as well as independent courts, or not.⁴⁴ For structure-focused states, we distinguish between weak, intermediary, and strong horizontal safeguards based on the relative strength of these non-executive actors in the absence of direct government attacks. Our agency-focused states discriminate between weak parliament (in the presence of a strong judiciary), weak judiciary (but reasonably strong parliament), and dismantled horizontal safeguards where both main kinds of non-executive actors are weakened following government attacks.

In sum, the proposed disaggregation of distinct sets of democratic safeguards restraining executive expansion serves to facilitate the analysis of common patterns of democratic backsliding within a broader typology of democratic trajectories of third-wave democracies. The following section describes the move from theoretical conceptualization to operational measurement of democratic safeguards, as well as the use of sequencing methods for our empirical analysis.

Research design and methods

Operationalizing democratic trajectories

The empirical analysis focuses on the 79 third-wave democracies as identified in the Varieties of Democracy (V-Dem) dataset,⁴⁵ with each country included from the moment of its first transition until 2020 (see Table A1 for list of countries included). The V-Dem dataset provides a large number of variables both in continuous form, i.e. formulated as having more or less of a range of democratic characteristics, and as ordinal indicators that indicate the presence or absence of specific democratic features, with in-between nuances. These two formats present distinct advantages and drawbacks over the alternative, binary format when it comes to the study of democratization.⁴⁶ Numeric and ordinal data facilitate the creation of different sub-indices, to which scholars can apply the linear modelling toolbox, including historical democratization curves, comparison of indices in scatterplots, factor analyses, and linear or ordinal regression modelling. For instance, the “cube of democracy” proposed by Boese et al. is conveniently displayed as a three-dimensional scatterplot, with countries or country-years appearing as clouds of cases.⁴⁷ Lindenfors et al. propose a specific approach to ordinal sequential data that includes causal assessment based on Bayesian dynamical systems.⁴⁸ Another recent analysis using V-Dem data employs pairwise domination analysis to assess the order of liberalization across a large number of “episodes of regime transformation.”⁴⁹

Our approach falls between continuous/ordinal and binary measurement approaches. We build democratic trajectories as multichotomous sequences that allow us to survey the evolution of the three dimensions of democratic safeguards. Using V-Dem data, a specific state is assigned to each country-year. We first convert a selection of 24 five-scale ordinal V-Dem variables, each contributing to one of the three dimensions, into as many two- or three-point variables (see Annex II for exhaustive aggregation rules). We then group these recoded variables into three higher-level alphabets of vertical, diagonal, and horizontal safeguards, as defined above. Importantly, as advised by Munck and Verkuilen, throughout the recoding, we remain “sensitive to the multitude of ways in which attributes might be linked and avoid the tendency to limit [ourselves] by adherence to defaults, such as additivity.”⁵⁰ In other terms, we keep a theoretical control over the meaning of each category, which in turn enables us to make sense of the resulting trajectories throughout all stages of the analysis: coding, processing, and interpretation.

Exploring holistic trajectories of third-wave democracies

To study democratic trajectories empirically, we leverage recent advances in sequence analysis (SA).⁵¹ SA can be understood as representing time series with categorical data: whereas time series analysis generally examines the evolution and correlation of strictly continuous variables on frequent time points, social and political sequences are typically categorical and observed at less frequent intervals. Applications of SA to the life course and other social scientific objects have demonstrated the added value of the method for the study of careers, and more generally trajectories and processes. SA's comprehensive approach to the duration, order, and timing of social processes gives it an edge for the study of trajectories over methods such as duration models, time series, or panel data analysis, which focus on one or two of the three time dimensions, and/or that cannot handle categorical data.⁵²

With few exceptions,⁵³ SA applications in Political Science have so far focused on the individual level, analysing careers of political⁵⁴ or economic elites⁵⁵ or voting trajectories.⁵⁶ One study of democratization has applied SA to the whole set of independent regimes in the post-World War II period in order to test Huntington's three-wave theory.⁵⁷ The resulting six-type classification highlights long-term trends in regime stability and reversals based on a single dimension of regime denominations, i.e. without engaging with detailed democratic features, as we do here. A recent study has used SA to investigate presidential term-limit reforms in Latin America and sub-Saharan Africa, showing how such reforms indicate a process of autocratisation or, alternatively, reflect democratic resistance to authoritarian tendencies.⁵⁸ However, political processes were again described by assessing a single dimension of constitutional policy changes.

In a first step of the analysis, we build holistic sequences of third-wave democracies and examine the data separately across the three dimensions, or channels: vertical, diagonal and horizontal safeguards. Such monochannel analyses remain limited in scope, but are important to validate a range of statistical parameters that are subsequently transposed at the multichannel level. Distribution plots render the yearly evolution of the different states from the alphabets. Figure 1 displays state distribution plots (d-plots) for the entire time period for all three dimensions separately. The graphs show the proportion of the different states in the sample for each year, giving a first indication of the considerable variation in democratic performance across countries. Following the important number of democratic transitions that occur around 1989/1990, we note the persistence of a large share of country-years – around a third – characterized by enduring structural weaknesses. For the remaining country-years contained in the sample, there is a broad spread across the remaining states. However, it is notable that the theoretical ideal-state – represented in the brightest colour for each dimension, i.e. light blue/yellow/red – only concerns a very low number of country-years over time. Importantly, this visualization obscures the individual dimension of trajectories and therefore does not offer any insight into the specific sequencing of states across individual countries. Nonetheless, it can provide a first insight into broad patterns of democratic development across our sample.

Figure 1. Monochannel distribution plots.

Based on these general evolutions, we contrast different monochannel groups of country cases by means of optimal matching (OM) and clustering algorithms (see Figures III-1/2/3 in Annex, and related interpretations). OM consists in treating all pairs of trajectories (I, J) using a minimal set of elementary operations of insertion, deletion and substitution of the states that compose I and J, in order to convert I into J. The sum of the “costs” of the operations is referred to as the dissimilarity between I and J. After having tested a range of algorithms and options (see Annex III), we decided to innovate compared to more common uses of sequences analysis in three respects: first, OM dissimilarities are calculated on the basis of sequences of transitions, instead of sequences of states,⁵⁹ thereby putting the accent specifically on backsliding events rather than more general trends. Second, OM dissimilarities are normalized in order to reduce the impact of uneven sequence lengths due to earlier versus later democratic transitions. Finally, we set comparatively high insertion-deletion costs, so as to minimize dissimilarities between sequences where backsliding events happen in similar order. For other parameters, we follow more established SA practices, including substitution costs based on transition rates, clustering with Euclidean metric, and hierarchical, ascending agglomeration of clusters.⁶⁰

In a second step, we assess the degree of sequential association between the three dimensions of backsliding, i.e. whether timing, order and duration are synchronized between channels. This is equivalent to checking the consistency of a list of survey items prior to conducting factor analysis. For this, we apply Pearson's and Spearman's correlation coefficients as well as Cronbach's alpha to the same monochannel dissimilarity matrices used above to cluster country cases.⁶¹ All scores appear sufficiently strong (see Table 2), which confirms that the dynamics inherent to the three separate dimensions of backsliding are closely entangled with one another. In particular, the high values of multichannel sequence analysis coefficients show that the dynamics internal to each channel will be well accounted for in the multichannel analysis, and high Cronbach's alphas mean that the three channels reflect the same construct to a good extent. Hence, it is justified to move on to a joint analysis of the three channels.

Table 2. Sequential correlations between dimensions of backsliding.

	Pearson's correlation (mean: 0.58)				Spearman's correlation (mean: 0.56)				Cronbach's alpha (P-C-E: 0.77)
	VS	DS	HS	mcsa	VS	DS	HS	mcsa	VS	DS	HS
Vertical safeguards (VS)	1				1
Diagonal safeguards (DS)	0.58	1			0.55	1			0.74
Horizontal safeguards (HS)	0.47	0.5	1		0.47	0.52	1		0.64	0.68
Multichannel sequence analysis (mcsa)	0.61	0.82	0.84	1	0.59	0.81	0.82	1

The degree of similarity between trajectories in the three channels is estimated by three complementary approaches: Pearson's r, based on scores, Spearman's ρ, based on ranks, and Cronbach's α, based on scale reliability. Statistics are applied to pairs of sequential dissimilarity matrices.

In the third step of our analysis, we establish a typology of democratic trajectories that is as comprehensive as possible, as it maximizes sequential homogeneity within clusters as well as sequential heterogeneity between clusters for the three channels jointly.⁶² Clustering statistics (Figures IV-1/2 in Annex) point to six to eight clusters. We settle on the eight-cluster solution, as the added level of detail reveals three clusters (5, 6 and 8) that carry crucial nuances in their democratic trajectories (see next section). Figure 2 materializes the distances between country trajectories and their grouping into clusters. For ease of reference, we provide the country names and corresponding country codes in Table 3 below.

Figure 2. Space of country trajectories. Multidimensional scaling⁶⁹ is applied to the multichannel dissimilarity matrix in order to extract the main dimensions (MDS1 and MDS2) that contrast country trajectories. The closer the points representing two countries, the more similar their trajectories. Most clusters appear very distinct, which indicates clear sequential contrasts. However some countries seem to be attracted towards neighbouring clusters, such as Ukraine, a member of Cluster 5, isolated in Cluster 1. The specificities of these countries would appear more clearly on subsequent MDS dimensions (3, 4, etc.), which cannot be represented on Figure 2, but will be taken into account in the clustering (see Figures 3 to 7 and IV-3 to IV-6 for details about the trajectories that compose the clusters).

Table 3. Country codes (used in Figures 2–6).

ALB	Albania	CPV	Cabo Verde	HRV	Croatia	MEX	Mexico	PRT	Portugal	THA	Thailand
ARG	Argentina	CZE	Czechia	HUN	Hungary	MKD	North Macedonia			TLS	Timor Leste
ARM	Armenia	DOM	Dominic. Rep.	IDN	Indonesia	MLI	Mali	PRY	Paraguay	TUN	Tunisia
BEN	Benin	ECU	Ecuador	IND	India	MNE	Montenegro	ROU	Romania	TUR	Turkey
BFA	Burkina Faso	ESP	Spain	KOR	Korea, Rep.	MNG	Mongolia	RUS	Russian	TWN	Taiwan
BGD	Bangladesh	EST	Estonia	LBN	Lebanon	MWI	Malawi	SEN	Senegal	TZA	Tanzania
BGR	Bulgaria	FJI	Fiji	LBR	Liberia	NAM	Namibia	SLB	Solomon	UKR	Ukraine
BLR	Belarus	GEO	Georgia	LKA	Sri Lanka	NER	Niger	SLE	Sierra Leone	URY	Uruguay
BOL	Bolivia, Plurinat. State			LSO	Lesotho	NIC	Nicaragua	SLV	El Salvador	VUT	Vanuatu
BRA	Brazil	GHA	Ghana	LTU	Lithuania	NPL	Nepal	SRB	Serbia	ZAF	South Africa
BTN	Bhutan	GRC	Greece	LVA	Latvia	PAN	Panama	STP	Sao Tome and Principe
CHL	Chile	GTM	Guatemala	MDA	Moldova, Rep.	PER	Peru	SUR	Suriname	ZMB	Zambia
COL	Colombia	GUY	Guyana	MDG	Madagascar	PHL	Philippines	SVK	Slovakia
COM	Comoros	HND	Honduras	MDV	Maldives	POL	Poland	SVN	Slovenia

Empirical findings: patterns of democratic trajectories

Based on the analyses described in the previous section, this section focuses on multichannel patterns, i.e. the unfolding of democratic trajectories across the three dimensions of democratic safeguards simultaneously. This cross-case analysis focuses on exploring the different clusters of democratic trajectories identified in Figure 2. Grouping different democratic sequences into clusters inevitably lumps together countries that share a certain number of similarities, but also diverge from one another on some accounts. By addressing each cluster identified by the sequencing algorithm in turn, we shed light on the diversity of democratic trajectories across third-wave democracies. We distinguish two broad types of trajectories: the first four clusters are characterized by stable trajectories, where initial democratic transition is followed either by a persistent weakness of democratic quality on one or more dimensions (Clusters 1-3) or democratic stability at an advanced level (Cluster 4). We qualify the second set of clusters (Clusters 5-8) as backsliding trajectories of differing quality and severity and explore their particular characteristics in more detail. Table 4 presents extracts of the transitions that best distinguish each cluster from the others. Figures 3–7 and IV-3 to IV-6 (in Appendix) visualize the exact democratic trajectories for each cluster by means of index plots, using the same name and colour convention introduced in Figure 1. For the sake of clarity and to demonstrate the added value of index plots, Cluster 5 is also displayed as a distribution plot as in Figure 1.

Table 4. Characteristic transitions per cluster.

Stable trajectories

Despite its increasing prevalence, democratic backsliding is not a universal phenomenon, and scholars have criticized excessive alarmism regarding the “end of democracy”⁶³ and cautioned against applying the concept all-too-readily to a large number of cases.⁶⁴ Our sequence analysis confirms this intuition, with almost half of the countries contained in our sample – 35 out of 79 – being characterized by overall stability rather than any degree of backsliding. We identify four distinct clusters that fall into this category.

The largest and geographically most diverse cluster is characterized by autocratic stagnation (see Figure IV-3). Despite some variation on the vertical dimension, what defines the cluster is an overall weak quality of all three democratic safeguards, with the most frequent transitions concerning a shift from resilient to dismantled diagonal safeguards as well as the alternation between weak and dismantled horizontal safeguards, with transitions frequent in both directions.

The second cluster, which our clustering algorithm sets clearly apart from all others, groups together just three countries whose trajectories are both remarkably stable and very similar. Characterized by established vertical safeguards, but weak diagonal and either weak horizontal safeguards (Solomon and Dominican Republic) or a weak parliament (Suriname), we describe this cluster as weak electoral democracies (see Figure IV-4).

The main common feature of Cluster 3 concerns the horizontal safeguards, leading us to qualify this group of countries as electoral democracies with weak parliaments (see Figure IV-5). Sao Tomé and Principe is the central sequence of this cluster, showing complete stability across all three safeguards over time. Malawi stands out due to the enduring weakness of the electoral process, but otherwise corresponds to the general pattern of stably medium-quality democracies contained in this cluster, with transitions from resilient to intermediary diagonal safeguards and from dismantled horizontal safeguards to weak parliament the most prevalent.

The final cluster of stable trajectories presents the opposite pattern to the three previous ones: we find here three well-developed democracies, all located in Southern Europe, which share strong diagonal and intermediary horizontal safeguards for most country-years contained in the analysis (see Figure IV-6). What clouds this strong performance however are developments on the vertical dimension, with the most discriminating transitions for this cluster concerning the shift from strong to disengaged and disengaged to hollow vertical safeguards, leading us to qualify the overall cluster as hollowing democracies. Besides, we see a sequence of weakening diagonal safeguards in Greece in the most recent years that appears to indicate that even long-standing consolidated democracies are not entirely immune to weakening on specific dimensions.

Backsliding trajectories

How do processes of democratic backsliding unfold? By adopting a meso-level perspective, our study set out to identify distinct backsliding patterns by locating deteriorations of democratic quality on one or more dimensions of democratic safeguards. Our empirical analysis produces a typology of backsliding patterns that diverge regarding both the shape and the degree of democratic decline among different groups of third-wave democracies. For each of the four clusters of backsliding trajectories, we describe the main sequential characteristics uniting their members, zoom into certain central sequences that illustrate the main features of the specific backsliding pattern, and address particular discriminating transitions based on Table 4.

The comparatively large and geographically diverse Cluster 5 contains sequences qualified as democratic reversal (see Figures 3 and 4). It partially overlaps with the autocratic stagnation cluster (see Figure IV-3), meaning that some sequences are close to the pattern described for this first cluster (such as Sierra Leone), whereas others see some positive developments during the early years following democratic transition, but eventually revert to weak performance across all three democratic safeguards (Croatia being a partial exception in this regard). Unlike other clusters, we also note a considerable diversity in the length of sequences contained in this group, ranging from 33 years since democratic transition for Nicaragua to just 10 years in Tunisia, adding to the heterogeneity that characterizes this cluster.

Figure 3. Democratic reversal (Cluster 5) (see legends in Figure 1).

Figure 4. Democratic reversal (alternative visualisation of Figure 3 as distribution plot).

Georgia, Serbia, and Tunisia have low distances to the centre and illustrate well the range of trajectories that can be classified as democratic reversals. Georgia experiences only short episodes during which democratic quality surpasses the weakest state, with the gradual deterioration particularly visible for horizontal safeguards, which go from structurally weak to the presence of direct attacks on the judiciary to the eventual dismantling of both independent judiciary and independent parliament. Serbia, in contrast, shows a good decade of democratic progress regarding vertical and horizontal safeguards immediately after its initial transition following the toppling of Slobodan Milošević in 2000. However, severe backsliding then occurs near-simultaneously across all three dimensions from 2012 onwards under the government of strongman Aleksandar Vučić, with attacks against the independent judiciary as well as diagonal safeguards eventually leading to the breakdown of the very electoral process at the core of the democratic system. Tunisia, finally, illustrates that a reversal of democratic fortunes need not always result in outright regime change: although the country experienced a decline in the quality of vertical safeguards and the enduring weakness of diagonal safeguards during its short democratic history, horizontal safeguards remain for now of sufficient quality to prevent the full breakdown of democracy. However, the recent dissolution of the Tunisian Parliament following lawmakers challenging the President's increasingly autocratic use of his powers is likely to end Tunisia's status as the only country of the Arab Spring that was able to conserve some of its newfound freedom.

Comparatively well-developed democratic safeguards characterize Cluster 6 that brings together eight countries from the Global South (see Figure 5). This cluster shows reasonably strong vertical and horizontal safeguards practically throughout the observed period, with backsliding trends almost exclusively concentrated on the diagonal dimension, leading us to describe the cluster as backsliding on diagonal safeguards. This cluster illustrates both the resilience of certain third-wave democracies, but also points to the weakening of media freedom and civic engagement as a potential early warning sign ahead of a more comprehensive backsliding process that encompasses also the institutional and electoral dimensions.

Figure 5. Backsliding on diagonal safeguards (Cluster 6).

Partial backsliding characterizes Cluster 7 (Figure 6), in which horizontal safeguards are well-developed and backsliding trends focused on the diagonal dimension, as indicated also by the partial overlapping of Clusters 6 and 7 shown in Figure 2. At the same time, the cluster shares some similarities with the hollowing democracies of Cluster 4 when it comes to the temporary presence of strong diagonal safeguards and trends of declining voter turnout that weaken the vertical dimension. In geographical terms, the cluster combines some of the most advanced third-wave democracies, including South Africa, Taiwan, as well as several post-Communist and Latin American countries. We qualify this cluster as partial backsliding with resilient horizontal safeguards.

Figure 6. Partial backsliding with resilient horizontal safeguards (Cluster 7).

The Czech Republic represents the most central sequence in the cluster, moving from strong vertical safeguards in the early years following initial transition towards declining voter turnout and even some structural weakening of the electoral process. Backsliding on diagonal safeguards occurs only in the most recent years, but drops to an outright weak level in the last two years included in our analysis, whereas horizontal safeguards remain stable at a well-developed level throughout the entire period. Brazil experiences a similar weakening of diagonal safeguards, with backsliding in this case however extending to horizontal safeguards. Both cases thus indicate that a successful initial democratization experience is no guarantee against backsliding at a later stage.

Our final cluster contains a group of countries we qualify as democracies under attack (see Figure 7). Its most defining feature is the presence of deteriorations across all three dimensions, with the most discriminating transitions being those from intermediary to weak vertical safeguards, intermediary diagonal safeguards to partial attacks, partial attacks to resilient diagonal safeguards, and weak judiciary to intermediary horizontal safeguards, all four of which appear in 43 per cent of the observed sequences (see Table 4).

Figure 7. Democracies under attack (Cluster 8).

The clustering algorithm identifies the Phillipines as the most central sequence of this cluster, showing considerable weakness on the diagonal dimension and lots of transitions on the horizontal safeguards dimension. However, the country stands out due to the enduring weakness of the electoral process, a dimension that sees more variation for other countries contained in the cluster. Another more unusual trajectory concerns Poland, which shows a strong performance across all three safeguards for a long period but has experienced a swift and comprehensive decline of democratic quality in the most recent years, rightly placing it among the democracies under attack despite its initially more successful course.

Discussion and conclusion

This article has proposed a novel approach to the diversity of democratic trajectories among third-wave democracies and in particular the distinct patterns of democratic backsliding among this group of countries. We differentiate broadly between stable and backsliding trajectories, and develop a more detailed typology based on the different clusters of cases contained in these two categories. Our analysis highlights divergent patterns regarding the shape, depth and breadth of backsliding, distinguishing instances of full democratic reversal or continued attacks across all safeguards from those where backsliding trends remain confined to specific dimensions of democratic safeguards while others remain broadly intact.

In conceptual terms, the meso-level analysis of the evolution of democratic safeguards over time allows us to explore the multi-dimensional nature of backsliding processes. Comparing the evolution of democratic quality across three kinds of safeguards against executive expansion, we pinpoint the key role of diagonal safeguards. These elements come most centrally under pressure from government attacks, confirming earlier findings by Maerz et al. that the repression of media freedom and civil society are often the first stage in backsliding processes.⁶⁵ In turn, the relative strength of horizontal constraints able to compensate for a weakening level of scrutiny by independent media and civil society appears as the main discriminating factor between different types of backsliding trajectories. This finding contrasts with Coppedge's assertion that the erosion of horizontal constraints constitutes a distinct pathway to backsliding. ⁶⁶

In sum, our analysis yields no universal template for backsliding. Instead, it suggests that diagonal safeguards are most susceptible to erosion and remain vulnerable, even where they were strongly present over a longer period. Horizontal safeguards, in turn, appear more resilient in general but, where dismantled, coincide more readily with a full democratic reversal extending to the vertical dimension and the electoral process itself. Our analysis thus provides empirical evidence for the incrementalism of backsliding, whereby degradations on one component precede assaults on others.⁶⁷ Overall, our sequential typology points to a temporal unfolding of backsliding that goes beyond the mere discrimination of distinct outcomes to unpack the multidimensional nature of backsliding processes and the diverging depth to which they proceed.⁶⁸ In doing so, we offer a corrective to the common tendency in the backsliding literature to lump together highly dissimilar cases.

Where do we go from here? The distinct clusters we identify among third-wave democracies beg the question of what sets countries apart in their democratic trajectories and can serve as starting points for a more systematic analysis of how different explanatory factors identified in the backsliding literature shape the onset and persistence of backsliding. Do different clusters correspond to distinct institutional set-ups that make them more or less resilient to democratic backsliding? Do they reflect divergent economic trajectories that open the door to executive expansion – for instance, in response to an economic crisis – or, on the contrary, facilitate durable democratic consolidation? Taking the reverse perspective, follow-up studies could also focus on the determinants of resilience: are the chances for re-democratization different across the different types of backsliding we identify?

Besides pointing to promising future research avenues, our findings on the range of backsliding patterns also have important practical implications when it comes to developing appropriate responses to backsliding. For one, they signal how a weakening of diagonal safeguards is often a precursor to more sweeping declines in democratic quality and should be read as an early warning sign. Responding promptly to government attacks against independent media or restrictions of civic spaces may therefore prevent the erosion of democratic quality from spreading towards further safeguards. At the latest once horizontal constraints come under pressure, decisive action is required to prevent the full breakdown of the democratic system. Where domestic actors are too weak to counter such developments alone, international actors – for instance, the European Union – need to step in to prevent a further erosion of democratic quality.

Finally, our empirical analysis offers an innovative application of sequence analysis to a macro-political phenomenon, illustrating its usefulness beyond the study of individual life course trajectories. By building multichannel sequences of categorical states, we are able to overcome some of the shortcomings of both binary and continuous measurements of democracy, and to offer a more detailed understanding of the nature, order, and timing of democratic trajectories and the unfolding of backsliding sequences. Our analysis can serve as a template to study the evolution of different dimensions of democratic quality, including levels of corruption or degrees of professionalization in the civil service. Another interesting avenue concerns the emergence of the rule of law, and the sequencing in which such processes occur. Ultimately, we demonstrate the benefits of applying sequence analysis tools to facilitate the systematic comparison of political sequences more generally, be it in the area of democratization and autocratisation or with regards to other system-level processes, such as the evolution of welfare states or electoral systems.

Acknowledgements

We are grateful to Staffan Lindberg, Mariana Llanos, Seraphine Maerz, Kelley Morrison, Svend-Erik Skaaning, Yuko Sato, and Gregor Zons for comments on earlier versions of the manuscript. We thank Laurenz Derksen for his excellent research assistance.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Additional information

Funding

Natasha Wunsch gratefully acknowledges funding from the Swiss National Science Foundation via the Ambizione Grant “Democratic Backsliding in Eastern Europe: Sequence, Strategies, Citizens” (grant number PZ00P1_185908 / 1).

Notes on contributors

Natasha Wunsch

Natasha Wunsch is Senior Researcher at the European Politics Group of ETH Zurich and Assistant Professor at Sciences Po, Centre d'études européennes.

Philippe Blanchard

Philippe Blanchard is Associate Professor at the University of Warwick.

Notes

1 Mechkova, Lührmann and Lindberg, “How Much Democratic Backsliding?”; Waldner and Lust, “Unwelcome Change”; Haggard and Kaufman, Democratic Regress in Contemporary World; Cassani and Tomini, “Reversing regimes and concepts”.

2 Waldner and Lust, “Unwelcome Change”; Bermeo, “On Democratic Backsliding”.

3 Lührmann and Lindberg, “Third wave of autocratization”; Cianetti and Hanley, “End of Backsliding Paradigm”.

4 See Jee, Lueders and Myrick, “Towards a unified approach”.

5 Bermeo, “On Democratic Backsliding,” 5–6.

6 Ibid.

7 Mainwaring and Bizzarro, “Fates Of Third-Wave Democracies”.

8 Wilson et al., “Successful and Failed Episodes”.

9 Ritschard and Studer, Sequence Analysis and Related Approaches.

10 Blanchard, “Sequence Analysis”.

11 Bermeo, “On Democratic Backsliding”; Mechkova, Lührmann and Lindberg, “How Much Democratic Backsliding?”; Cassani and Tomini, “Reversing regimes and concepts”; Lueders and Lust, “Multiple Measurements, Elusive Agreement”; Jee, Lueders and Myrick, “Towards a unified approach”.

12 Lührmann and Lindberg, “Third wave of autocratization”; Cassani and Tomini, “Reversing regimes and concepts”.

13 Lührmann and Lindberg, “Third wave of autocratization”; but see criticism in Skaaning, “Waves of autocratization”; Tomini, “Current autocratization debate”.

14 Bochsler and Juon, “Authoritarian footprints Eastern Europe”; Stanley, “Backsliding Away?”.

15 Mainwaring and Bizzarro, “Fates Of Third-Wave Democracies”.

16 Wilson et al., “Successful and Failed Episodes”; Maerz et al., “State of world 2019”.

17 Buštíková and Guasti, “Illiberal Turn or Swerve”; Bogaards, “De-democratization in Hungary”; Bieber, Authoritarianism in Western Balkans; Diamond, “Breaking Out Democratic Slump”; Aspinall et al., “Elites, masses, democratic decline”; Kneuer, “Unravelling democratic erosion”.

18 Gerschewski, “Erosion or decay?”, 12.

19 Tomini, “Current autocratization debate,” 7.

20 Edgell et al., “Institutional Order of Liberalization”; Wilson et al., “Successful and Failed Episodes”.

21 Rose and Shin, “Democratization Backwards,” 332–3.

22 Møller and Skaaning, “Regime Types Democratic Sequencing,” 153; Mechkova, Lührmann and Coppedge, “Mapping Sequences of Accountability,” 4–5.

23 Mainwaring and Bizzarro, “Fates Of Third-Wave Democracies,” 100.

24 Coppedge, “Eroding Regimes”.

25 Jee, Lueders and Myrick, “Towards a unified approach”.

26 Maerz et al., “State of world 2019”.

27 Kneuer, “Unravelling democratic erosion”.

28 Jee, Lueders and Myrick, “Towards a unified approach”.

29 Shapiro, State of democratic theory, 51.

30 Bermeo, “On Democratic Backsliding”.

31 Svolik, “Which Democracies Will Last?”.

32 Lührmann, Marquardt and Mechkova, “Constraining Governments”; Morlino, Changes for democracy, 200; O'Donnell, “Horizontal Accountability New Democracies”.

33 Dahl, Polyarchy, 8.

34 Shapiro, State of democratic theory, 51.

35 Merkel, “Embedded and Defective Democracies”.

36 Merkel and Croissant, “Good and Defective Democracies,” 202.

37 Boese et al., “Visualizing Authority Patterns”.

38 Gates et al., “Polity Duration, 1800-2000”.

39 Boese et al., “Visualizing Authority Patterns," 5.

40 Greskovits, “Hollowing and Backsliding”.

41 Dahl, Polyarchy.

42 Coppedge et al., “V-Dem Comparisons and Contrasts,” 21.

43 Morlino, “What is ‘Good’ Democracy?,” 13.

44 Morlino, Changes for democracy, 217.

45 Coppedge et al., “V-Dem Dataset v10”.

46 Ulfelder and Lustik, “Modelling Transitions,” 352–3; Lueders and Lust, “Multiple Measurements, Elusive Agreement,” 737–8.

47 Boese et al., “Visualizing Authority Patterns”.

48 Lindenfors et al., “Investigating Sequences”.

49 Edgell et al., “Institutional Order of Liberalization”.

50 Munck and Verkuilen, “Conceptualizing and measuring democracy,” 24.

51 Blanchard, Bühlmann and Gauthier, Advances in Sequence Analysis; Ritschard and Studer, Sequence Analysis Related Approaches.

52 Blanchard, “Sequence Analysis”.

53 E.g. Casper and Wilson, “Sequences to Model Crises”.

54 E.g. Jäckle, “Pathways to Karlsruhe”.

55 Blanchard, Dudouet and Vion, “Affaires zone euro”.

56 Buton, Lemercier and Mariot, “Household effect electoral participation”.

57 Wilson, “Analyzing Sequences Explain Democratization”.

58 Heyl and Llanos, “Sequences of presidential-term-limit reforms”.

59 Biemann, “A Transition-Oriented Approach”.

60 Gauthier, Blanchard and Bühlmann, “Introduction,” 5–11.

61 Piccarreta, “Joint Sequence Analysis”; Robette, ““Seqhandbook” R package”.

62 Pollock, “Holistic trajectories”; Gabadinho et al., “State Sequences in R”.

63 Lührmann and Lindberg, “Third wave of autocratization”.

64 Cianetti and Hanley, “End of Backsliding Paradigm”.

65 Maerz et al., “State of world 2019," 8–9.

66 Coppedge, “Eroding Regimes”.

67 Haggard and Kaufman, Democratic Regress Contemporary World.

68 Mainwaring and Bizzarro, “Fates Of Third-Wave Democracies”; Stanley, “Backsliding Away?”.

69 Piccarreta and Lior, “Exploring sequences”.