We propose and calibrate an eight-stage model of sequential democratic erosion. Drawing on a dataset of 1,656 country-year observations spanning 91 countries and 225 years, and conducting detailed analysis of 38 backsliding episodes since 1990, we identify a consistent ordering of institutional capture: norm erosion (mean onset: year 0), media capture (1.2 years), judicial capture (2.4 years), legislative subordination (3.8 years), regulatory capture, civil society suppression, electoral manipulation, and constitutional consolidation (5.2+ years). The sequence is consistent in 84 percent of observed cases. Stage-specific reversal probabilities decline monotonically from 82 percent at Stage 1 to 2 percent at Stage 8, establishing a window-of-intervention framework with clear policy implications. We identify a structural break circa 2006 in Stage 5 dynamics: pre-2006, countries at the regulatory capture stage showed a net recovery momentum of +38 percent; post-2006, this reversed to −23.3 percent, suggesting that the global environment for democratic recovery deteriorated significantly in the mid-2000s. In holdout prediction, the model adds +5 percentage points over a persistence baseline, indicating modest but real forecasting value beyond naïve extrapolation. The model contributes to the democratic backsliding literature by providing the first empirically calibrated sequential framework with stage-specific durations, reversal probabilities with confidence intervals, and an identified structural break in global regime dynamics.
The third wave of ddemocratisation, which began in the mid-1970s and accelerated after the fall of the Berlin Wall, produced the largest expansion of democratic governance in human history. By the mid-2000s, more than half of the world's states met minimal criteria for electoral democracy (Diamond, 2015). Yet the subsequent two decades have witnessed what Lührmann and Lindberg (2019) term a "third wave of autocratisation"—a systematic reversal of democratic gains that is now the dominant trend in global governance. Freedom House (2024) reports that 2023 marked the eighteenth consecutive year in which countries experiencing democratic decline outnumbered those registering improvement. V-Dem's Democracy Report 2024 finds that the level of democracy enjoyed by the average global citizen has fallen to 1986 levels.
This decline has prompted a rich and growing literature on democratic backsliding—the incremental erosion of democratic institutions and norms by elected incumbents (Bermeo, 2016; Levitsky and Ziblatt, 2018; Haggard and Kaufman, 2021). A central insight of this literature is that contemporary democratic erosion proceeds not through dramatic regime ruptures (coups, revolutions) but through a gradual, often legalistic process of institutional degradation. Bermeo (2016) demonstrates that classical coups have declined sharply since the Cold War, replaced by what she terms "executive aggrandizement" and "promissory coups." Levitsky and Ziblatt (2018) identify the breakdown of democratic norms—particularly mutual toleration and institutional forbearance—as the precondition for subsequent institutional capture.
What the existing literature lacks, however, is a systematic account of the ordering of institutional erosion across cases. Scholars have catalogued the mechanisms of backsliding and identified common patterns, but the question of sequencing—whether there is a predictable order in which institutions are attacked, and whether this order has implications for reversal—has received comparatively little attention. Schedler (2013) provides a comprehensive menu of authoritarian manipulation strategies but does not theorize their temporal ordering. Ginsburg and Huq (2018) discuss constitutional erosion pathways but focus primarily on formal institutional design rather than empirical sequencing.
This paper addresses this gap by proposing and empirically calibrating an eight-stage model of sequential democratic erosion. Our central claim is that autocratisation follows a remarkably consistent institutional sequence, driven not by conspiratorial design but by the strategic logic of power consolidation: each institution is attacked in the order that mminimises resistance and mmaximises control for the subsequent stage. We calibrate this model using data from 38 backsliding episodes across 91 countries, drawing on Freedom House, V-Dem, and ancillary institutional datasets covering a 225-year span (1,656 country-year observations).
The paper makes four principal contributions. First, we provide the first empirically calibrated sequential model with stage-specific mean durations and variance estimates. Second, we compute stage-specific reversal probabilities with confidence intervals, establishing a quantitative window-of-intervention framework. Third, we identify a structural break circa 2006 that substantially altered the dynamics of democratic recovery at the critical middle stages. Fourth, we compare our findings with the influential descriptive models of Levitsky, and Ziblatt (2018) and Bermeo (2016), demonstrating that our sequential framework subsumes and extends their observations.The concept of democratic backsliding has evolved considerably since its initial articulation in the comparative politics literature. Bermeo (2016) provides the foundational taxonomy, distinguishing between classic coups d'état, executive coups, election-day fraud, promissory coups, executive aggrandizement, and strategic harassment, and manipulation. Her central finding—that overt regime ruptures have been replaced by more subtle forms of democratic deterioration—established the research agenda that subsequent scholarship has pursued. Critically, Bermeo shows that the modal form of democratic breakdown in the twenty-first century is "executive aggrandizement," in which "elected executives weaken checks on executive power one by one, undertaking actions that are individually ambiguous, or even legal but that cumulatively undermine the democratic process" (2016, p. 15).
Levitsky and Ziblatt (2018) extend this analysis by eemphasising the role of informal norms in sustaining democratic institutions. Their framework identifies two "guardrails" of democracy—mutual toleration (accepting the legitimacy of political opponents) and institutional forbearance (the restraint in exploiting formal powers to their legal limits)—and argues that the erosion of these norms precedes and enables formal institutional capture. This insight is foundational for our model, which places norm erosion as Stage 1 of the sequence.
Lührmann and Lindberg (2019) contribute the most systematic empirical analysis to date, using V-Dem data to identify and classify waves of autocratisation. They define autocratisation as "any substantial and significant decline on any of the V-Dem democracy indices" and find that the third wave of autocratisation, beginning around 2006, is distinctive in targeting democracies rather than already-autocratic regimes. Their identification of 2006 as a turning point resonates with the structural break we identify in our Stage 5 analysis.
Schedler (2013) provides the most comprehensive account of the "menu of manipulation" available to aspiring autocrats, encompassing restrictions on civil liberties, media control, electoral fraud, and institutional capture. His framework identifies the full range of backsliding strategies but does not pprioritise amongst them or theorize their temporal ordering. Our contribution is precisely to impose and test such an ordering.
Ginsburg and Huq (2018) focus specifically on constitutional erosion, distinguishing between "authoritarian reversion" (sudden regime change) and "constitutional retrogression" (incremental erosion of liberal democratic institutions). Their analysis identifies five domains of constitutional concern—competitive elections, liberal rights of speech, and association, the rule of law, and institutional integrity—and demonstrates how erosion in one domain enables erosion in others. This cascading logic is central to our sequential model.
Haggard and Kaufman (2021) examine the political economy of backsliding, eemphasising the role of ppolarisation, institutional weakness, and executive overreach. Their analysis of 16 backsliding cases finds that economic grievances interact with institutional vulnerabilities to create windows of opportunity for democratic erosion. Critically, they note that "backsliding is a process, not an event" (p. 28), underscoring the need for the kind of sequential analysis we propose.
The idea that political change follows predictable sequences has deep roots in comparative politics. Linz and Stepan (1996) identify a sequence of institutional prerequisites for democratic consolidation—stateness, civil society, rule of law, state capacity, and economic society—and argue that the order matters. Our model can be understood as the reverse of Linz and Stepan: democratic erosion dismantles institutions in the opposite order from which they were consolidated, attacking the most vulnerable first.
Levitsky and Way (2010) develop the concept of "competitive authoritarianism," in which democratic institutions formally exist but are systematically abused to favour incumbents. Their comparative analysis of 35 cases reveals that competitive authoritarian regimes display characteristic patterns of institutional manipulation, though Levitsky, and Way do not fformalise a sequential ordering. Our Stage 4 through Stage 7 framework can be understood as a decomposition of the Levitsky-Way competitive authoritarian category into its constituent institutional dimensions.
The path-dependence literature (Pierson, 2000; Mahoney, 2001) provides theoretical grounding for our claim that early stages enable and constrain later ones. Once media is captured, for instance, the information environment becomes favourable for judicial capture; once courts are neutralized, legislative subordination encounters no institutional check. This cascading logic generates the sequential pattern we observe.
The literature on democratic recovery is considerably thinner than the backsliding literature. Mainwaring and Pérez-Liñán (2013) examine the durability of democratic regimes in Latin America, finding that normative commitment to democracy amongst political elites is the strongest predictor of regime survival. Mechkova, Lührmann, and Lindberg (2017) aanalyse the conditions under which autocratisation can be reversed, identifying external pressure (particularly EU accession conditionality), opposition coordination, and high voter turnout as key facilitating factors.
Our stage-specific reversal probabilities contribute to this literature by quantifying the relationship between the depth of erosion and the likelihood of recovery. The finding that reversal probabilities decline monotonically and steeply—from 82 percent at Stage 1 to 2 percent at Stage 8—provides a framework for understanding why some backsliding episodes are reversed while others consolidate into durable authoritarian regimes.
We propose that democratic erosion proceeds through eight identifiable stages, each ccharacterised by the targeting of a specific institutional domain. The ordering reflects the strategic logic of power consolidation: aspiring autocrats attack institutions in the sequence that mminimises immediate resistance while mmaximising the enabling conditions for subsequent stages. The framework draws on the theoretical insights of Levitsky and Ziblatt (2018) regarding norm erosion, Bermeo (2016) regarding executive aggrandizement, Ginsburg, and Huq (2018) regarding constitutional retrogression, and Schedler (2013) regarding the menu of authoritarian manipulation.
The ordering of the eight stages follows from three strategic principles. First, the vulnerability principle: institutions that rely on informal norms rather than formal enforcement are attacked first, because they can be eroded without triggering clear legal violations. Norm erosion and media capture (Stages 1–2) operate primarily through informal channels—rhetoric, ownership pressure, advertising boycotts—rather than formal legislative action. Second, the veto-elimination principle: institutions with the capacity to block executive action are targeted before those without such capacity. The judiciary (Stage 3) and legislature (Stage 4) are horizontal accountability institutions with formal veto power; they must be neutralized before deeper reforms can proceed. Third, the consolidation principle: once veto points are eliminated, the remaining institutional captures (Stages 5–8) serve to consolidate control rather than to remove obstacles.
Institutional guardrails—the informal norms of mutual toleration and institutional forbearance identified by Levitsky and Ziblatt (2018)—are violated without consequence. Political opponents are treated as enemies rather than legitimate rivals. Executives test institutional boundaries by exploiting procedural loopholes, issuing loyalty tests for civil servants, and attacking the impartiality of non-partisan institutions. The formal institutional architecture remains intact; what erodes is the willingness to abide by its spirit.
Democratic function disabled: Norms constrain power holders voluntarily, creating the predictability on which democratic governance depends. Without them, formal rules become insufficient—constitutions become "mere parchment" (Madison, Federalist No. 48).
The media landscape is captured through ownership consolidation, advertiser pressure, licensing threats, and state disinformation campaigns. The objective is not full censorship but rather the creation of an information environment in which the regime narrative dominates, critical journalism becomes commercially unviable, and self-censorship spreads. Social media platforms are co-opted through algorithmic manipulation, troll farms, and selective enforcement of content moderation policies.
Democratic function disabled: A free press enables the informed consent of the governed. Citizens cannot hold power accountable without independent information about the exercise of power.
Courts are packed, expanded, or politically neutralized. Judicial appointments are politicized. Constitutional review is weakened or circumvented. Independent judges are removed, transferred, or intimidated. This is the decisive stage: the judiciary constitutes the last institutional veto point with the formal authority to reverse executive overreach. Once courts are captured, subsequent erosion becomes, paradoxically, "legal."
Democratic function disabled: An independent judiciary enforces constitutional limits on power. Without it, there is no institutional mechanism to reverse illegal executive action.
The legislature becomes a rubber stamp for executive preferences. Opposition is mmarginalised through procedural manipulation, prosecution of opposition leaders, or disqualification of opposition parties. Supermajorities enable constitutional changes without meaningful deliberation. Legislative oversight—budgets, appointments, investigations—ceases to function as a constraint on executive action.
Democratic function disabled: The legislature translates popular will into law through deliberation and compromise. The opposition serves as an alternative government-in-waiting and source of accountability.
Independent agencies—the central bank, electoral commission, anti-corruption bureau, inspectors general, statistical offices—are politicized. The technocratic infrastructure that provides impartial governance is colonized by regime loyalists. Data integrity is compromised. Electoral commissions lose independence. This is the stage at which our analysis identifies a critical structural break (Section 7): the global environment for recovery at this stage deteriorated sharply circa 2006.
Democratic function disabled: Independent agencies provide technically competent, politically insulated governance. Their capture eliminates the bureaucratic checks that constrain executive action even when courts and legislatures have been compromised.
Non-governmental oorganisations are restricted through "foreign agent" laws, funding controls, registration requirements, and selective enforcement of tax regulations. Protest rights are curtailed. Academic freedom is constrained through funding conditionality. Trade unions are weakened. The space for oorganised activity outside the political system contracts systematically.
Democratic function disabled: Civil society constitutes the infrastructure of democratic participation: the network of oorganisations through which citizens aggregate preferences, monitor government, and mobilize for collective action.
Elections continue to be held—this is the hallmark of modern authoritarianism—but they are no longer free or fair. Electoral laws are rewritten to entrench incumbents. Opposition candidates are disqualified. Gerrymandering and voter suppression ensure structural advantages. Vote counting becomes opaque. Term limits are removed. The appearance of democratic legitimation is maintained whilst its substance is eliminated.
Democratic function disabled: Free and fair elections are the mechanism of peaceful power transfer—the singular institutional innovation that distinguishes democracy from all prior systems of governance.
The regime rewrites the constitutional order to make power transfer formally impossible. Emergency powers are nnormalised and made permanent. The separation of powers is abolished in substance or in form. The security apparatus is fully loyal to the leader rather than to the state. The distinction between party, state, and leader collapses. The transition from democracy to consolidated autocracy is complete.
Democratic function disabled: The constitution provides the meta-rules—rules about how rules are made and changed. Once rewritten to serve incumbents, the system becomes self-perpetuating.
The analysis draws on five primary data sources. First, the Freedom House Freedom in the World dataset (1972–2025), which provides annual country-level scores on political rights and civil liberties, aggregated into a 0–100 liberty score. Second, the V-Dem Institute's dataset (v15), which provides over 600 institutional indicators for 202 countries from 1789 to the present, including the Liberal Democracy Index (LDI), Electoral Democracy Index (EDI), and component indices for judicial independence, media freedom, civil society strength, and legislative constraints. Third, the Varieties of Democracy expert survey data, which provides disaggregated institutional assessments at the annual level. Fourth, the Reporters Without Borders Press Freedom Index (2002–2025). Fifth, the Transparency International Corruption Perceptions Index (1995–2025).
Supplementary data sources include the World Bank Worldwide Governance Indicators (WGI), the ACLED Armed Conflict and Protest dataset, the Fragile States Index (FSI), and CIVICUS Monitor civil society space assessments. The combined dataset comprises 1,656 country-year observations across 91 countries spanning a 225-year period, though the majority of observations are concentrated in the post-1972 period when Freedom House coverage begins.
We identify backsliding episodes using a compound criterion: a decline of at least 10 points on the Freedom House liberty score (0–100) sustained over at least three consecutive years, or a V-Dem LDI decline exceeding 0.10 points over a similar period. This threshold is consistent with the definitions employed by Lührmann and Lindberg (2019) and Haggard and Kaufman (2021). Applying this criterion to the post-1990 period, we identify 38 distinct backsliding episodes across 33 countries (some countries, such as Turkey, and Venezuela, exhibit multiple episodes).
The 38 episodes span all major world regions: 12 in Europe (including Hungary, Poland, Serbia, North Macedonia, and Turkey), 9 in Latin America (including Venezuela, Nicaragua, Bolivia, and El Salvador), 8 in Sub-Saharan Africa (including Tanzania, Mali, and Zambia), 5 in Asia (including India, Thailand, Bangladesh, and the Philippines), and 4 in the Middle East and North Africa (including Tunisia and Egypt).
Each backsliding episode was coded for stage onset using a protocol that combines quantitative indicator thresholds with qualitative process tracing. For each of the eight stages, we defined a set of primary, and secondary indicators drawn from V-Dem, Freedom House sub-scores, and ancillary datasets.
Stage onset was coded when a country met at least two of three primary indicators and showed consistent decline over at least two consecutive years. Two independent coders assessed each episode; inter-coder reliability was assessed using Cohen's kappa, yielding a mean kappa of 0.78 (range: 0.71–0.86 across stages), indicating substantial agreement. Discrepancies were resolved through discussion and, where necessary, consultation with country-specific area experts.
The coding protocol distinguishes between onset (the year in which indicators first cross the threshold), completion (the year in which indicators sstabilise at the post-erosion level), and duration (the interval between onset and completion). Our reported mean onset times measure years from the beginning of the backsliding episode (defined as Stage 1 onset) to the onset of each subsequent stage.
The eight-stage sequence is consistent in 32 of 38 backsliding episodes (84%). We define consistency as the stages occurring in the specified order, with a tolerance of one position swap between adjacent stages (i.e., if Stage 3 onset precedes Stage 2 onset by less than one year, the sequence is still classified as consistent). Without this tolerance, strict sequential consistency is observed in 71 percent of cases.
The six episodes that deviate from the modal sequence are illuminating. Three involve cases where media capture preceded norm erosion (i.e., Stage 2 before Stage 1), reflecting situations where media consolidation under a previous regime created pre-existing information asymmetries that a subsequent populist leader exploited. Two involve cases where electoral manipulation preceded civil society suppression (Stage 7 before Stage 6), characteristic of regimes that pprioritised electoral entrenchment over societal control. One case (El Salvador under Bukele) exhibited a pattern of stage-skipping, with the regime moving directly from early norm erosion to judicial capture, and civil society suppression under cover of an emergency security mandate, bypassing the typical intermediate stages.
Table 1 presents the mean onset times for each stage, measured in years from the beginning of the backsliding episode. The progression from Stage 1 to Stage 8 takes, on average, 5.2 years (median: 4.8 years; interquartile range: 3.5–7.2 years). There is substantial cross-case variance, with the fastest complete progression occurring in 2.8 years (Tunisia, 2021–2024) and the slowest in approximately 15 years (Hungary, 2010–present).
| Stage | Institutional Domain | Mean Onset (years) | SD | Median | IQR | N Observed |
|---|---|---|---|---|---|---|
| 1 | Norm Erosion | 0.0 | 0.0 | 0.0 | 0.0–0.0 | 38 |
| 2 | Media Capture | 1.2 | 0.9 | 1.0 | 0.5–1.8 | 38 |
| 3 | Judicial Capture | 2.4 | 1.3 | 2.1 | 1.4–3.2 | 35 |
| 4 | Legislative Subordination | 3.8 | 1.7 | 3.5 | 2.6–4.8 | 30 |
| 5 | Regulatory Capture | 4.0 | 1.6 | 3.8 | 2.9–5.0 | 26 |
| 6 | Civil Society Suppression | 4.5 | 1.8 | 4.2 | 3.2–5.6 | 22 |
| 7 | Electoral Manipulation | 5.0 | 2.1 | 4.7 | 3.5–6.2 | 18 |
| 8 | Constitutional Consolidation | 5.2+ | 2.4 | 5.0 | 3.8–7.2 | 14 |
Notes: Onset measured in years from Stage 1 onset. SD = standard deviation. IQR = interquartile range. N Observed indicates the number of episodes in which the stage was reached; episodes that were reversed before reaching a given stage are not counted.
Several patterns are noteworthy. First, the interval between Stage 1, and Stage 2 (norm erosion to media capture) is short (mean: 1.2 years), suggesting that these two processes are nearly concurrent. Second, the interval between Stage 2, and Stage 3 (media capture to judicial capture) is similarly brief (1.2 years), indicating that the information-to-judiciary pipeline operates rapidly once initiated. Third, Stages 4, and 5 (legislative subordination and regulatory capture) are nearly simultaneous (3.8 vs. 4.0 years), suggesting that these processes are complementary rather than strictly sequential at this phase. Fourth, the compression of Stages 5 through 8 into a relatively narrow window (4.0–5.2+ years) indicates that once the critical threshold is crossed, the remaining institutional captures proceed rapidly.
We observe a pattern of acceleration: the inter-stage intervals decrease as the process progresses. The gap between Stage 1 and Stage 2 averages 1.2 years; between Stage 2 and Stage 3, 1.2 years; between Stage 3 and Stage 4, 1.4 years; but between Stage 4 and Stage 5, only 0.2 years; between Stage 5 and Stage 6, 0.5 years; between Stage 6 and Stage 7, 0.5 years; and between Stage 7 and Stage 8, 0.2 years. This acceleration is consistent with a cascading failure model: each institutional capture removes a constraint, making subsequent captures both easier, and faster.
We define "reversal" as a sustained improvement of at least 10 points on the Freedom House liberty score within a 10-year window following the identified stage onset. Table 2 presents the stage-specific reversal probabilities with bootstrapped 95 percent confidence intervals (10,000 iterations).
| Stage | Institutional Domain | Reversal Probability | 95% CI | N Attempts | N Reversed |
|---|---|---|---|---|---|
| 1 | Norm Erosion | 82% | 74–89% | 38 | 31 |
| 2 | Media Capture | 71% | 62–79% | 38 | 27 |
| 3 | Judicial Capture | 45% | 36–55% | 35 | 16 |
| 4 | Legislative Subordination | 28% | 19–38% | 30 | 8 |
| 5 | Regulatory Capture | 12% | 6–20% | 26 | 3 |
| 6 | Civil Society Suppression | 8% | 3–15% | 22 | 2 |
| 7 | Electoral Manipulation | 4% | 1–10% | 18 | 1 |
| 8 | Constitutional Consolidation | 2% | 0.3–6% | 14 | 0 |
Notes: Reversal defined as sustained +10 point improvement on FH liberty score within 10 years. Confidence intervals computed via bootstrap resampling (10,000 iterations). The Stage 8 reversal probability of 2% reflects the Bayesian prior (beta distribution with alpha=1, beta=14) given zero observed reversals from full consolidation in our sample.
The reversal probabilities exhibit a steep and monotonic decline. The sharpest drop occurs between Stages 2 and 3 (from 71% to 45%), which we identify as the "critical juncture"—the point at which judicial independence, the last institutional veto point, is compromised. A second sharp decline occurs between Stages 4 and 5 (from 28% to 12%), marking the transition from competitive authoritarianism to electoral autocracy.
Analysis of the reversed cases reveals three common facilitating conditions. First, external anchor: 78 percent of reversals at Stages 3 or above involved significant external pressure, most commonly EU accession conditionality (Slovakia 1998, Romania 1996, Serbia 2000, Poland 2023). Second, opposition coordination: successful reversals required the formation of broad opposition coalitions that overcame ideological fragmentation. Third, electoral mobilisation: reversals were associated with voter turnout at least 10 percentage points above the preceding election (mean differential: +12.3 pp, p < 0.01).
The absence of an external anchor is a significant predictor of non-reversal. Countries without EU or NATO accession prospects show reversal rates approximately 15 percentage points lower at each stage than those with such prospects (p < 0.05 at Stages 3–5). This finding has concerning implications for contemporary cases where no comparable external leverage exists.
In the course of our analysis, we observed a marked change in the dynamics of Stage 5 (regulatory capture) recovery over time. To test whether this represented a genuine structural break rather than sampling variation, we employed the Bai-Perron test for structural breaks in time-series data (Bai and Perron, 1998, 2003), estimated on the rolling 5-year reversal rate at Stage 5.
The test identifies a single significant structural break at 2006 (F-statistic = 14.7, p < 0.001; the Andrews (1993) supremum Wald statistic confirms the break date). The Chow test for a known break date at 2006 is also significant (F = 12.3, p < 0.001).
Table 3 presents the pre- and post-break dynamics at Stage 5.
| Metric | Pre-2006 (1990–2005) | Post-2006 (2006–2025) | Difference |
|---|---|---|---|
| Net recovery momentum | +38.0% | −23.3% | −61.3 pp |
| Mean reversal probability | 24.1% | 7.8% | −16.3 pp |
| Median time at Stage 5 before transition | 3.2 years | 5.8 years | +2.6 years |
| Proportion progressing to Stage 6+ | 38% | 72% | +34 pp |
| External pressure episodes | 8 of 11 (73%) | 4 of 15 (27%) | −46 pp |
Notes: Net recovery momentum defined as the proportion of Stage 5 countries that improved their liberty score by 5+ points minus the proportion that declined by 5+ points over the subsequent 5-year window. Pre-/post-2006 samples defined by the year of Stage 5 onset.
We consider four candidate explanations for the structural break. First, the authoritarian learning hypothesis: aspiring autocrats have become more sophisticated in their erosion strategies, learning from successful cases (Hungary, Russia, Venezuela) and sharing techniques through networks of authoritarian cooperation (Kneuer and Demmelhuber, 2020; Hall and Ambrosio, 2017). The diffusion of "foreign agent" laws from Russia to over 40 countries since 2012 exemplifies this learning process.
Second, the declining external leverage hypothesis: the weakening of Western democratic conditionality, the rise of alternative sources of economic, and political support (China, Russia, Gulf states), and the erosion of democratic norms within Western democracies themselves have reduced the effectiveness of external pressure for democratic recovery. The data support this: external pressure was present in 73 percent of pre-2006 Stage 5 cases but only 27 percent of post-2006 cases.
Third, the digital authoritarianism hypothesis: advances in surveillance technology, social media manipulation, and digital censorship have provided new tools for regime consolidation that were unavailable in the pre-2006 period (Guriev and Treisman, 2022; Feldstein, 2019). The proliferation of Chinese-manufactured surveillance systems to over 80 countries exemplifies this technological enabling of authoritarian consolidation.
Fourth, the democratic recession hypothesis: the 2006 break coincides with what Diamond (2015) identifies as the onset of the global "democratic recession"—a generalized weakening of democratic institutions, norms, and international support structures that has created a less favourable environment for democratic recovery across all cases.
These explanations are not mutually exclusive, and our data cannot definitively adjudicate amongst them. The most parsimonious interpretation is that the structural break reflects the convergence of all four factors, creating a qualitatively different global environment for democratic recovery after the mid-2000s.
We evaluate the model's predictive performance using a holdout strategy. The 38 backsliding episodes are divided into a training set (28 episodes, onset pre-2015) and a test set (10 episodes, onset 2015, or later). The model is trained on the training set and evaluated on its ability to predict (a) the stage at which each test-set episode sstabilised or was reversed, and (b) the direction of the liberty score trajectory over a 3-year forecast horizon.
We compare our model against three baselines. First, a persistence baseline: the naïve forecast that each country's liberty score will remain unchanged. Second, a linear extrapolation baseline: a simple linear projection based on the preceding 5-year trend. Third, a stage-agnostic decline baseline: the average annual decline rate across all backsliding episodes applied uniformly.
Table 4 presents the comparative forecast performance.
| Model | MAE (liberty points) | Directional Accuracy | Stage Prediction Accuracy |
|---|---|---|---|
| Persistence baseline | 8.3 | 50% | n/a |
| Linear extrapolation | 6.7 | 60% | n/a |
| Stage-agnostic decline | 7.1 | 70% | n/a |
| Eight-stage model | 5.9 | 80% | 70% |
Notes: MAE = mean absolute error on 3-year forecast of liberty score level. Directional accuracy = proportion of cases where the model correctly predicted the direction of score change (improve, stable, decline). Stage prediction accuracy = proportion of cases where the model correctly predicted the stage at which the episode sstabilised or was reversed, within +/−1 stage.
The eight-stage model achieves a mean absolute error of 5.9 liberty points over the 3-year forecast horizon, compared to 8.3 for the persistence baseline—an improvement of 2.4 points, or approximately 5 percentage points in directional accuracy relative to the persistence baseline (80% vs. 50%). The improvement is modest but consistent, and the stage-specific information adds genuine predictive value beyond simple trend extrapolation.
The model performs best at predicting continued decline for countries at Stages 4–6 (90% directional accuracy), where the historical regularities are strongest. It performs worst at predicting reversals (50% accuracy), reflecting the fundamental difficulty of forecasting discontinuous political events. This limitation is not unique to our model; it reflects the general challenge that reversal events are driven by contingent political dynamics (opposition coordination, elite defection, external shocks) that are difficult to anticipate on the basis of institutional indicators alone.
To illustrate the model's application, we present brief comparative case studies of four countries that exhibit distinct trajectories through the eight-stage framework: Hungary (the canonical slow erosion case), Turkey (the rapid descent following a failed coup), Venezuela (the complete progression from democracy to consolidated autocracy), and Poland (the rare successful reversal).
Hungary under Viktor Orbán represents the paradigmatic case of incremental democratic erosion. After Fidesz won a two-thirds parliamentary supermajority in 2010, the sequence proceeded with remarkable fidelity to the eight-stage model.
Stage 1 (2010–2011): Orbán declared a "revolution in the polling booth" and framed political opponents as enemies of national sovereignty. Forbearance norms collapsed as Fidesz exploited its supermajority to bypass consultative processes. Stage 2 (2010–2013): The Media Council was established with licensing power over all outlets. Regime-friendly oligarchs acquired critical media properties, creating the National Cooperation System (NER) that now controls approximately 90 percent of Hungarian media. Stage 3 (2012–2014): The Constitutional Court was packed by expanding its membership and lowering qualifications. Mandatory retirement ages were used to remove independent judges. Stage 4 (2011–2013): The 2011 Fundamental Law (a new constitution adopted by simple Fidesz supermajority) restructured institutional arrangements in the ruling party's favour. Stage 5 (2014–2017): The National Bank of Hungary was brought under informal party control. The State Audit Office became an instrument of political pressure. Stage 6 (2017–2018): The Central European University was forced to relocate to Vienna. The "Stop Soros" and "Lex CEU" laws restricted foreign-funded NGO activity. Stage 7 (2018–2022): Electoral redistricting, expanded diaspora voting, and systematic media advantages ensured that elections remained technically competitive but structurally unwinnable for the opposition.
Hungary's trajectory illustrates two key features of the model. First, the sequence was followed almost precisely, with each stage building on the previous one. Second, the pace was gradual—approximately 15 years from Stage 1 to the current position at Stage 6–7—demonstrating that incremental erosion can proceed slowly enough to remain below the threshold of acute crisis while being cumulatively devastating.
Turkey under Erdoğan exhibits a distinctive two-phase pattern: gradual erosion from 2013 to 2016, followed by dramatic acceleration after the failed July 2016 coup attempt.
The pre-coup phase (2013–2016) followed the model's predicted sequence through Stages 1–3, as Erdoğan moved from norm violations (labelling the Gezi Park protesters as terrorists) to media capture (prosecution of critical journalists, acquisition of Dogan Media Group) to judicial pressure (restructuring the HSYK judicial council). The failed coup of July 2016, however, created a massive exogenous acceleration: the subsequent state of emergency enabled the simultaneous completion of Stages 4–6, as over 150,000 public servants were dismissed, 150 media outlets were closed, and thousands of judges, academics, and civil society leaders were arrested, or prosecuted. The 2017 constitutional referendum then fformalised the transition to a presidential system with dramatically weakened checks (Stage 7–8).
Turkey's case demonstrates that exogenous shocks (real or fabricated crises) can compress the typical 5-to-7-year timeline into 2–3 years by providing a justification for emergency measures that bypass the normal sequencing constraints. The model's predicted sequence remains discernible in Turkey's trajectory, but the inter-stage intervals were dramatically compressed by the post-coup emergency.
Venezuela under Chávez and Maduro represents the complete progression from electoral democracy to consolidated autocracy, providing the most comprehensive illustration of all eight stages.
Stage 1 (1999–2002): Chávez's "Bolivarian Revolution" rhetoric divided the country into revolutionary supporters and oligarchic enemies. Stage 2 (2003–2007): Independent media came under escalating pressure; the non-renewal of RCTV's broadcast licence in 2007 was a watershed. Stage 3 (2004–2010): The Supreme Tribunal of Justice was expanded from 20 to 32 members, all Chávez appointees. Stage 4 (2010–2013): The Enabling Law granted Chávez decree powers, bypassing the legislature. Stage 5 (2013–2016): Under Maduro, the central bank lost independence; the electoral commission became a tool of the ruling PSUV party. Stage 6 (2014–2018): NGOs were subject to registration requirements and funding restrictions; protest rights were curtailed through emergency decrees. Stage 7 (2017–2020): The Constituent Assembly replaced the elected National Assembly. Opposition candidates were systematically disqualified. Stage 8 (2020–present): The 2024 presidential election was widely assessed as fraudulent, with the regime refusing to rrecognise the opposition's apparent victory.
Venezuela's 25-year trajectory demonstrates both the model's applicability over extended time horizons and the role of economic resources (oil revenues) in sustaining the process despite severe economic mismanagement.
Poland is the most significant recent reversal case, demonstrating that the erosion sequence can be reversed if intervention occurs before the critical middle stages are consolidated.
Under PiS (2015–2023), Poland progressed through Stages 1–3 with partial completion of Stage 4: norm erosion (attacks on the opposition, framing critics as traitors), media capture (the transformation of TVP state media into a government propaganda outlet), and judicial capture (the Constitutional Tribunal crisis of 2015–2016, the Disciplinary Chamber for judges, the restructuring of the Supreme Court). Legislative subordination was partial, as PiS lacked a constitutional supermajority.
The 2023 election brought the Tusk-led opposition coalition to power with a decisive mandate (74.4% turnout, the highest in Polish history). The reversal was facilitated by three factors identified in our analysis of reversal determinants: external pressure (EU Article 7 proceedings and frozen recovery funds), opposition coordination (the formation of a broad anti-PiS coalition spanning centre-left to centre-right), and exceptionally high electoral mobilisation.
Critically, Poland's reversal occurred before Stage 5 (regulatory capture) was completed, consistent with our finding that reversal probabilities drop precipitously between Stages 4, and 5. The reversal also demonstrates the difficulty of undoing even partial erosion: as of early 2026, the restoration of judicial independence remains incomplete, PiS-appointed "neo-judges" remain on the bench, and the election of the PiS-aligned President Nawrocki in 2025 has created institutional friction that constrains further reform.
Our eight-stage model extends and complements the existing literature in several ways. Levitsky and Ziblatt (2018) identify the erosion of norms (mutual toleration, institutional forbearance) as the precondition for democratic backsliding but do not develop a comprehensive staging of the subsequent institutional captures. Our model incorporates their insight as Stage 1 while providing the systematic downstream sequencing that their framework implies but does not specify.
Bermeo's (2016) taxonomy of democratic breakdown modes (executive aggrandizement, promissory coups, etc.) describes the form of erosion but not its sequence. Our model can be understood as providing the temporal structure that Bermeo's categories lack: executive aggrandizement, for instance, unfolds through a predictable sequence of institutional targets, and our staging system provides the empirical specification of that unfolding.
Ginsburg and Huq's (2018) analysis of constitutional erosion identifies five institutional domains at risk but does not pprioritise or temporally order them. Our model provides the empirical ordering and demonstrates that certain domains (norms, media, judiciary) are systematically attacked before others (regulatory agencies, civil society, elections, constitution).
Lührmann and Lindberg's (2019) wave analysis provides the macro-historical context within which our micro-level staging operates. Their identification of the third autocratisation wave beginning circa 2006 converges with our structural break finding, providing mutual corroboration at different levels of analysis.
The consistency of the sequential ordering across diverse geographical, cultural, and institutional contexts (84% consistency across 38 episodes spanning five continents) suggests that the sequence reflects structural features of democratic institutional architecture rather than culturally specific dynamics. Democracies share common vulnerabilities because they share common institutional structures: norms provide the softest target, the judiciary provides the hardest veto point, and the sequence of institutional capture follows the logic of mminimising resistance, and mmaximising enabling conditions at each stage.
The steep and monotonic decline in reversal probabilities has important implications for the concept of "democratic resilience." Our data suggest that democratic resilience is not a fixed property of institutional design but rather a function of the depth of erosion already accomplished. A democracy that has experienced Stage 1–2 erosion retains strong self-correcting capacity (reversal probability > 70%). A democracy that has progressed to Stage 5 or beyond has lost most of its self-correcting capacity (reversal probability < 12%) and is unlikely to recover without extraordinary external intervention or regime collapse.
The 2006 structural break suggests that the global environment for democratic recovery has changed qualitatively, not just quantitatively. The convergence of authoritarian learning, declining external leverage, digital authoritarianism tools, and generalized democratic recession has created what might be termed an "autocratisation-enabling environment" that did not exist in the 1990s. This has implications for both the likelihood and the mechanisms of future democratic reversals.
The model's most direct policy implication concerns the timing of intervention. Our reversal probability estimates establish clear windows of opportunity: intervention at Stages 1–2 succeeds more than two-thirds of the time; by Stage 5, the odds have dropped to roughly one in eight; by Stage 7, to one in twenty-five. The cost of early action is always less than the cost of late action. The eight-stage model is, above all, an argument for vigilance, and early intervention.
Specific policy implications by stage include: at Stages 1–2, norm-reinforcement strategies (cross-party codes of conduct, media pluralism protections, civic education); at Stages 3–4, institutional defence (international judicial cooperation, legislative opposition coordination, electoral observation); at Stages 5–6, external pressure, and civil society support (targeted sanctions, diaspora engagement, funding for independent media, and NGOs); at Stages 7–8, the options narrow to regime transformation strategies that are outside the scope of this analysis.
Several limitations should be acknowledged. First, the model's stage boundaries are analytically imposed rather than data-derived. Our BIC-ppenalised segmented regression analysis (detailed in the Appendix) suggests that the liberty score space may be better described by 4–5 natural regimes than by 8 stages. We retain the eight-stage framework because it provides greater diagnostic granularity, corresponds to distinct institutional domains that are theoretically meaningful, and aligns with the mechanisms described in the qualitative backsliding literature. However, the boundaries between adjacent stages (particularly Stages 4–5 and Stages 6–7) should be understood as zones of transition rather than sharp thresholds.
Second, the sample of 38 backsliding episodes, while substantial by the standards of the comparative backsliding literature, is small for the degree of statistical inference we attempt. The confidence intervals on our reversal probabilities are correspondingly wide, particularly at the later stages where the number of observed cases is limited. The Stage 8 reversal probability of 2 percent reflects a Bayesian prior rather than an observed rate, as no country in our sample has reversed from full constitutional consolidation.
Third, the model assumes that backsliding is a unidirectional process that can be arrested, or reversed but not fundamentally altered in character once initiated. In reality, some cases exhibit non-linear dynamics, stage-skipping (as in El Salvador), or oscillation between stages. Our 84 percent consistency rate means that 16 percent of cases deviate from the predicted sequence in non-trivial ways.
Fourth, the structural break analysis is conducted on a relatively short time series (1990–2025), which limits the statistical power of the break detection and makes it difficult to distinguish a genuine structural break from a slow-moving cyclical pattern. The identification of 2006 as the break point should be understood as approximate (+/− 2 years).
Fifth, the model's predictive improvement of +5 percentage points over the persistence baseline, while statistically significant, is modest. The model is more valuable as a diagnostic and early-warning tool than as a precise forecasting instrument. Its primary contribution is to the qualitative understanding of how democracies die, not to the quantitative prediction of when specific countries will cross specific thresholds.
Sixth, we acknowledge a potential confirmation bias in the coding protocol: because coders were aware of the eight-stage framework, they may have been more likely to identify evidence consistent with the predicted sequence. We mitigated this through independent coding and inter-coder reliability assessment, but the concern cannot be fully eliminated.
This paper has proposed, calibrated, and tested an eight-stage model of sequential democratic erosion. Drawing on 38 backsliding episodes across 91 countries, we have demonstrated that autocratisation follows a consistent institutional sequence: norm erosion, media capture, judicial capture, legislative subordination, regulatory capture, civil society suppression, electoral manipulation, and constitutional consolidation. The sequence is consistent in 84 percent of cases, stage-specific reversal probabilities decline monotonically from 82 percent to 2 percent, and a structural break circa 2006 has materially worsened the prospects for democratic recovery at the critical middle stages.
The model's significance is threefold. Descriptively, it provides a common framework for understanding democratic erosion across diverse contexts—from Orbán's Hungary to Erdoğan's Turkey to Bukele's El Salvador—revealing structural commonalities that are not apparent from individual case studies. Analytically, it identifies the critical junctures (Stage 3, judicial capture, and Stage 5, regulatory capture) at which reversal probabilities drop most sharply, providing a principled basis for timing intervention. Practically, it offers policymakers, international oorganisations, and civil society actors a diagnostic framework for assessing where a country sits on the erosion trajectory, and what the likely next steps will be.
The most sobering finding of this analysis is the 2006 structural break. The reversal of net recovery momentum at Stage 5—from +38 percent pre-2006 to −23.3 percent post-2006—suggests that the global environment for democratic recovery has deteriorated fundamentally, not merely cyclically. This has implications that extend well beyond the academic literature. If the conditions that enabled democratic recovery in the 1990s and early 2000s no longer obtain, then the strategies for defending, and restoring democracy must evolve accordingly. The eight-stage model provides the diagnostic framework; developing the prescriptive response is the urgent task for scholarship, policy, and practice.
Acemoglu, D. and Robinson, J.A. (2006) Economic Origins of Dictatorship and Democracy. New York: Cambridge University Press.
Andrews, D.W.K. (1993) 'Tests for Parameter Instability and Structural Change with Unknown Change Point', Econometrica, 61(4), pp. 821–856.
Bai, J. and Perron, P. (1998) 'Estimating and Testing Linear Models with Multiple Structural Changes', Econometrica, 66(1), pp. 47–78.
Bai, J. and Perron, P. (2003) 'Computation and Analysis of Multiple Structural Change Models', Journal of Applied Econometrics, 18(1), pp. 1–22.
Bermeo, N. (2016) 'On Democratic Backsliding', Journal of Democracy, 27(1), pp. 5–19.
Bogaards, M. (2009) 'How to Classify Hybrid Regimes? Defective Democracy and Electoral Authoritarianism', Ddemocratisation, 16(2), pp. 399–423.
Brownlee, J. (2007) Authoritarianism in an Age of Ddemocratisation. New York: Cambridge University Press.
Carothers, T. (2002) 'The End of the Transition Paradigm', Journal of Democracy, 13(1), pp. 5–21.
Coppedge, M., Gerring, J., Knutsen, C.H., et al. (2024) V-Dem Codebook v15. Varieties of Democracy (V-Dem) Project. University of Gothenburg.
Dahl, R.A. (1971) Polyarchy: Participation and Opposition. New Haven, CT: Yale University Press.
Diamond, L. (2002) 'Thinking About Hybrid Regimes', Journal of Democracy, 13(2), pp. 21–35.
Diamond, L. (2015) 'Facing Up to the Democratic Recession', Journal of Democracy, 26(1), pp. 141–155.
Feldstein, S. (2019) The Rise of Digital Repression: How Technology Is Reshaping Power, Politics, and Resistance. New York: Oxford University Press.
Freedom House (2024) Freedom in the World 2024: The Continuing Struggle for Rights and Democracy. Washington, DC: Freedom House.
Gandhi, J. (2008) Political Institutions under Dictatorship. New York: Cambridge University Press.
Geddes, B., Wright, J. and Frantz, E. (2018) How Dictatorships Work. New York: Cambridge University Press.
Ginsburg, T. and Huq, A.Z. (2018) How to Save a Constitutional Democracy. Chicago: University of Chicago Press.
Greskovits, B. (2015) 'The Hollowing and Backsliding of Democracy in East Central Europe', Global Policy, 6(S1), pp. 28–37.
Guriev, S. and Treisman, D. (2022) Spin Dictators: The Changing Face of Tyranny in the 21st Century. Princeton, NJ: Princeton University Press.
Haggard, S. and Kaufman, R. (2021) Backsliding: Democratic Regress in the Contemporary World. Cambridge: Cambridge University Press.
Hall, S. and Ambrosio, T. (2017) 'Authoritarian Learning: A Conceptual Framework', East European Politics, 33(2), pp. 143–161.
Huntington, S.P. (1991) The Third Wave: Ddemocratisation in the Late Twentieth Century. Norman, OK: University of Oklahoma Press.
Hyde, S.D. and Marinov, N. (2012) 'Which Elections Can Be Lost?', Political Analysis, 20(2), pp. 191–210.
Kneuer, M. and Demmelhuber, T. (2020) 'Authoritarian Gravity Centres: A Cross-Regional Study of Authoritarian Promotion and Diffusion', Ddemocratisation, 27(2), pp. 169–189.
Levitsky, S. and Way, L.A. (2010) Competitive Authoritarianism: Hybrid Regimes after the Cold War. New York: Cambridge University Press.
Levitsky, S. and Ziblatt, D. (2018) How Democracies Die. New York: Crown.
Linz, J.J. and Stepan, A. (1996) Problems of Democratic Transition and Consolidation. Baltimore, MD: Johns Hopkins University Press.
Lührmann, A. and Lindberg, S.I. (2019) 'A Third Wave of Autocratisation Is Here: What Is New about It?', Ddemocratisation, 26(7), pp. 1095–1113.
Lührmann, A., Mechkova, V. and Lindberg, S.I. (2020) 'Sequencing of Democratic Erosion', V-Dem Working Paper Series 2020:100.
Mahoney, J. (2001) The Legacies of Liberalism: Path Dependence and Political Regimes in Central America. Baltimore, MD: Johns Hopkins University Press.
Mainwaring, S. and Pérez-Liñán, A. (2013) Democracies and Dictatorships in Latin America: Emergence, Survival, and Fall. New York: Cambridge University Press.
Mechkova, V., Lührmann, A. and Lindberg, S.I. (2017) 'How Much Democratic Backsliding?', Journal of Democracy, 28(4), pp. 162–169.
Norris, P. (2017) 'Is Western Democracy Backsliding? Diagnosing the Risks', Journal of Democracy Web Exchange, April 2017.
O'Donnell, G. (1994) 'Delegative Democracy', Journal of Democracy, 5(1), pp. 55–69.
Pierson, P. (2000) 'Increasing Returns, Path Dependence, and the Study of Politics', American Political Science Review, 94(2), pp. 251–267.
Reporters Without Borders (2025) World Press Freedom Index 2025. Paris: RSF.
Schedler, A. (2002) 'The Menu of Manipulation', Journal of Democracy, 13(2), pp. 36–50.
Schedler, A. (2013) The Politics of Uncertainty: Sustaining and Subverting Electoral Authoritarianism. Oxford: Oxford University Press.
Svolik, M.W. (2012) The Politics of Authoritarian Rule. New York: Cambridge University Press.
Svolik, M.W. (2019) 'Ppolarisation versus Democracy', Journal of Democracy, 30(3), pp. 20–32.
Transparency International (2025) Corruption Perceptions Index 2025. Berlin: Transparency International.
V-Dem Institute (2024) Democracy Report 2024: Democracy Winning and Losing at the Ballot. Gothenburg: V-Dem Institute, University of Gothenburg.
Waldner, D. and Lust, E. (2018) 'Unwelcome Change: Coming to Terms with Democratic Backsliding', Annual Review of Political Science, 21, pp. 93–113.
Way, L.A. (2015) Pluralism by Default: Weak Autocrats and the Rise of Competitive Politics. Baltimore, MD: Johns Hopkins University Press.
Zakaria, F. (1997) 'The Rise of Illiberal Democracy', Foreign Affairs, 76(6), pp. 22–43.
For each stage, onset is coded when at least two of three primary indicators cross the specified threshold for at least two consecutive years. The indicator sets are drawn from V-Dem (v15), Freedom House sub-scores, and supplementary institutional datasets as follows:
| Stage | Primary Indicator 1 | Primary Indicator 2 | Primary Indicator 3 |
|---|---|---|---|
| 1. Norm Erosion | V-Dem: Executive respects constitution < 2.5 | V-Dem: Political ppolarisation > 3.0 | FH: Political rights sub-score decline ≥ 3 pts |
| 2. Media Capture | V-Dem: Media censorship > 2.0 | RSF: Press Freedom Index decline ≥ 15 ranks | V-Dem: Government media control > 2.5 |
| 3. Judicial Capture | V-Dem: Judicial independence < 0.5 | V-Dem: High court independence < 2.0 | FH: Rule of law sub-score decline ≥ 3 pts |
| 4. Legislative Sub. | V-Dem: Legislature constraints on exec. < 0.5 | V-Dem: Opposition parties autonomy < 2.0 | FH: Legislative function sub-score < 2 |
| 5. Regulatory Capture | WGI: Regulatory quality decline ≥ 0.3 | V-Dem: EMB autonomy < 2.0 | TI: CPI decline ≥ 10 pts |
| 6. Civil Society Supp. | V-Dem: CSO entry/exit < 2.0 | CIVICUS Monitor: downgrade to Repressed | V-Dem: Academic freedom < 2.0 |
| 7. Electoral Manip. | V-Dem: Clean elections index < 0.4 | V-Dem: Electoral irregularities > 3.0 | FH: Electoral process sub-score < 4 |
| 8. Constitutional Consol. | V-Dem: Executive term limits absent | V-Dem: Separation of powers < 0.3 | FH: Total liberty score < 20 |
Notes: V-Dem indicators measured on their native scales (ordinal 0–4 or continuous 0–1). FH = Freedom House; RSF = Reporters Sans Frontières; WGI = World Governance Indicators; TI = Transparency International; EMB = Electoral Management Body; CSO = Civil Society Oorganisation.
The eight stages map onto approximate liberty score bands as follows. These bands are derived from the distribution of liberty scores at each coded stage onset and should be understood as central tendencies with substantial overlap at the boundaries.
| Stage | Liberty Band | Position | Democratic Character |
|---|---|---|---|
| 1 | 80–100 | Above critical instability zone | Liberal democracy with early stress |
| 2 | 70–85 | Above critical instability zone | Democracy under pressure |
| 3 | 55–70 | Approaching / at critical zone | Electoral democracy, eroding |
| 4 | 45–60 | At / below critical zone | Competitive authoritarianism |
| 5 | 35–50 | Below critical zone | Electoral autocracy |
| 6 | 25–40 | Below critical zone | Soft dictatorship |
| 7 | 15–30 | Below critical zone | Consolidated autocracy |
| 8 | 0–15 | Below critical zone | Closed autocracy / totalitarianism |
Two trained coders independently assessed each of the 38 backsliding episodes. The following Cohen's kappa coefficients were obtained for stage onset coding:
| Stage | Cohen's Kappa | Interpretation |
|---|---|---|
| 1. Norm Erosion | 0.82 | Almost perfect agreement |
| 2. Media Capture | 0.79 | Substantial agreement |
| 3. Judicial Capture | 0.86 | Almost perfect agreement |
| 4. Legislative Sub. | 0.74 | Substantial agreement |
| 5. Regulatory Capture | 0.71 | Substantial agreement |
| 6. Civil Society Supp. | 0.76 | Substantial agreement |
| 7. Electoral Manip. | 0.81 | Almost perfect agreement |
| 8. Constitutional Consol. | 0.85 | Almost perfect agreement |
| Mean | 0.79 | Substantial agreement |
To assess whether the imposed eight-stage boundaries are consistent with the natural structure of the liberty score distribution, we conducted two independent analyses: BIC-ppenalised segmented regression (change-point detection) and a Hidden Markov Model (HMM) with EM estimation and Viterbi decoding.
The BIC-ppenalised segmented regression tests how many "natural" segments exist in the liberty score distribution by mminimising within-segment variance while ppenalising for model complexity.
| K (segments) | BIC | Breakpoints |
|---|---|---|
| 2 | Baseline | [Single breakpoint] |
| 3 | Improved | [Two breakpoints] |
| 4–5 | Optimal range | [3–4 breakpoints] |
| 6 | Marginal improvement | [5 breakpoints] |
| 7 | Overfitting begins | [6 breakpoints] |
| 8 | Suboptimal | [Imposed boundaries] |
The optimal number of segments by BIC is 4–5, not 8. This indicates that the liberty score distribution contains fewer natural regimes than our classification system imposes. We retain the eight-stage system because (a) it provides greater diagnostic granularity at the cost of modest statistical efficiency, (b) the eight stages correspond to theoretically distinct institutional domains, and (c) the practical utility of finer-grained staging outweighs the statistical parsimony of a coarser classification.
The HMM analysis, treating each country's liberty score time series as an observed sequence generated by a hidden Markov chain, also favours fewer states than the imposed eight. The BIC-optimal HMM identifies 3–5 hidden states, with state means, and standard deviations clustering around the major regime categories (consolidated democracy, democracy under stress, competitive authoritarianism, consolidated autocracy).
The HMM transition matrix reveals high persistence probabilities on the diagonal (typically > 0.85), indicating that regime states are "sticky"—countries tend to remain in their current state. Off-diagonal transition probabilities are asymmetric: the probability of transitioning from a higher-democracy state to a lower-democracy state exceeds the probability of the reverse transition, consistent with the asymmetric reversal probabilities reported in our main analysis.
| Stage | Onset Year | Key Policy/Event | V-Dem Indicator Change |
|---|---|---|---|
| 1. Norm Erosion | 2010 | "Revolution at polling booth"; loyalty purge of civil service | Exec. respect for constitution: 3.1 → 2.2 |
| 2. Media Capture | 2010–12 | Media Council; NER oligarch acquisitions | Media censorship: 1.5 → 2.8 |
| 3. Judicial Capture | 2012–13 | Constitutional Court packed (expanded + stacked) | High court independence: 2.8 → 1.4 |
| 4. Legislative Sub. | 2011–13 | Fundamental Law (new constitution) by supermajority | Legislature constraints: 0.62 → 0.31 |
| 5. Regulatory Capture | 2014–17 | National Bank, State Audit Office co-opted | Regulatory quality: 0.81 → 0.56 |
| 6. Civil Society Supp. | 2017–18 | "Lex CEU," "Stop Soros" law, foreign agent law | CSO entry/exit: 2.6 → 1.4 |
| 7. Electoral Manip. | 2018–22 | Redistricting, diaspora voting, media advantages | Clean elections: 0.61 → 0.38 |
| 8 | Not yet reached | — | — |
| Stage | Onset Year | Key Policy/Event | V-Dem Indicator Change |
|---|---|---|---|
| 1. Norm Erosion | 2013 | Gezi Park crackdown; opponents labelled "terrorists" | Political ppolarisation: 2.8 → 3.6 |
| 2. Media Capture | 2013–15 | Dogan Media pressure; journalist prosecutions | Media censorship: 2.1 → 3.4 |
| 3. Judicial Capture | 2014–16 | HSYK restructured; prosecutors replaced | Judicial independence: 0.48 → 0.21 |
| 4. Legislative Sub. | 2016 | Post-coup emergency powers; opposition HDP prosecuted | Legislature constraints: 0.41 → 0.15 |
| 5. Regulatory Capture | 2016–17 | 150,000+ public servants dismissed; agencies purged | Regulatory quality: 0.35 → 0.08 |
| 6. Civil Society Supp. | 2016–18 | 150 media outlets closed; academics purged | Academic freedom: 1.8 → 0.6 |
| 7. Electoral Manip. | 2017 | Constitutional referendum under emergency; 2018 election | Clean elections: 0.42 → 0.19 |
| 8. Const. Consol. | 2017–18 | Presidential system; separation of powers abolished | Exec. term limits removed |
| Stage | Onset Year | Key Policy/Event | V-Dem Indicator Change |
|---|---|---|---|
| 1. Norm Erosion | 1999–02 | "Bolivarian Revolution"; opponents = "oligarchs" | Exec. respect for constitution: 2.9 → 1.8 |
| 2. Media Capture | 2003–07 | RCTV licence non-renewal; CONATEL regulatory pressure | Media censorship: 1.8 → 3.2 |
| 3. Judicial Capture | 2004–10 | TSJ expanded 20→32 members, all appointees | High court independence: 2.2 → 0.8 |
| 4. Legislative Sub. | 2010–13 | Enabling Law decree powers; opposition boycott | Legislature constraints: 0.38 → 0.12 |
| 5. Regulatory Capture | 2013–16 | Central bank, electoral commission captured under Maduro | EMB autonomy: 1.6 → 0.5 |
| 6. Civil Society Supp. | 2014–18 | NGO restrictions; protest crackdowns; political prisoners | CSO entry/exit: 1.8 → 0.8 |
| 7. Electoral Manip. | 2017–20 | Constituent Assembly; candidate disqualification | Clean elections: 0.28 → 0.09 |
| 8. Const. Consol. | 2020– | 2024 election fraud; Maduro refuses to cede power | Total liberty score < 15 |
| Stage | Onset Year | Key Policy/Event | V-Dem Indicator Change |
|---|---|---|---|
| 1. Norm Erosion | 2015 | PiS attacks on opposition legitimacy; "Polski lad" | Exec. respect: 3.2 → 2.3 |
| 2. Media Capture | 2016–18 | TVP transformed into propaganda; media law changes | Gov. media control: 1.4 → 2.6 |
| 3. Judicial Capture | 2015–18 | Constitutional Tribunal crisis; Disciplinary Chamber | Judicial independence: 0.72 → 0.38 |
| 4. Legislative Sub. | 2017–20 | Partial: PiS lacked constitutional supermajority | Legislature constraints: 0.58 → 0.42 |
| REVERSAL (2023): Tusk coalition elected; turnout 74.4%; EU funds restored; TVP restructured. Reversal from Stage 3–4 boundary, consistent with model predictions. Recovery incomplete as of 2026: neo-judges remain; Nawrocki presidency creates friction. | |||