Measuring Democratic Backsliding by Little & Meng,

Overview of the Global Democratic Backsliding Narrative

• General Narrative: Current popular and academic discourse asserts that the world is experiencing a significant period of democratic decline.

  • Media Examples:

    • New York Times headline (August $19$, $2022$ ):

      • How Democracy Is Under Threat Across the Globe.

    • Washington Post slogan adopted in $2017$:

      • Democracy Dies in Darkness.

  • NGO and Academic Reports:

    • Freedom House ($2022$): Opened its annual report, stating,

      • Global freedom faces a dire threat.

    • Varieties of Democracy (V-Dem) ($2023$): Claimed that global democracy levels for the average citizen in $2022$ are down to $1986$ levels, effectively wiping out $35$ years of progress.

Defining and Conceptualising Democratic Backsliding

• Definition by Bermeo ($2016$): Democratic backsliding or erosion is defined as

  • the state-led debilitation or elimination of any of the political institutions that sustain an existing democracy.

• Definition by Waldner and Lust ($2018$): A broader view that backsliding can occur in any regime type. In democracies, it is a decline in quality; in autocracies, it is a decline in democratic qualities of governance.

• Three General Categories of Backsliding:

  1. Shifts from liberal democracy toward electoral democracy.

  2. Shifts from democracy to autocracy (regime breakdown).

  3. Shifts from institutionalised autocracy to personalist autocracy.

• Regime Breakdown vs Erosion: The literature often muddies the distinction between these concepts. The authors focus on average levels across all countries to capture both positive gains (democratisation) and negative shifts (backsliding).

Subjective vs. Objective Measurement of Democracy

• Subjective Indicators:

  • Process: Rely on the judgment of expert coders to answer evaluative questions (e.g., was this election free and fair?).

  • Example (V-Dem): Experts rate elections on a $0$-to-$4$ scale. A score of $0$ means is fundamentally flawed; a score of $4$ means largely has only unintentional errors.

  • Issues: Coders often disagree. Disagreement frequency is non-trivial, with the average standard deviation across coders ranging from $20\text{\%}$ to $25\text{\%}$ of the scale (approx. $1$ point on a $0$-to-$4$ scale).

  • Measurement Models: Models like Pemstein et al. ($2018$) adjust for harshness or leniency but assume these biases are constant over time.

• Objective Indicators:

  • Definition: Based on factual, observable data rather than opinion. A litmus test is whether multiple experts with the same information would reach the same conclusion (Cheibub, Gandhi, and Vreeland $2010$ ).

  • Example: Whether the incumbent party lost an election, the percentage of the population with legal suffrage, or the number of journalists jailed.

Empirical Findings: Electoral Outcomes

• The Crucial Test of Democracy: As per Przeworski ($1991$), the most important feature is that incumbent parties lose elections. Leaders taking undemocratic actions (banning opposition, controlling media) aim to prevent these losses.

• Turnover Rates: Data from the National Elections Across Democracy and Autocracy (NELDA) dataset shows that the rate of ruling-party and individual-leader turnover has remained fairly constant since the late $1990\text{s}$.

• Winner Dominance: Using the Database of Political Institutions (DPI), the authors analysed the vote share of presidential winners and seat shares of winning legislative parties.

  • Trend: Winner vote and seat shares have generally decreased over the last few years, suggesting winners are becoming less dominant, not more.

• Competitiveness Indices:

  • Multiparty Index: Based on whether opposition was allowed and voters had a choice. Increased from $0.7$ to near $0.95$ in the $1990\text{s}$ and has remained stable for the last $20$ years.

  • Process-Violation Index: Meant to capture major violations (e.g., suspending previous elections, incumbents violating term limits). This index has remained rare and stable for two decades.

  • Frequency of Elections: The sheer number of executive and legislative elections held globally has not declined.

Empirical Findings: Executive Constraints

• Formal Rules: Expanding data from Meng ($2020$), the authors tracked constitutional rules for term limits, succession procedures, and leader dismissal.

  • Observation: These rules saw a marked increase after the Cold War. Over the past decade, they have remained flat or increased slightly.

• Term-Limit Evasion: Data from Versteeg et al. ($2020$) identifies $234$ instances from $2000$ to $2018$ where an incumbent reached a limit.

  • Attempt Rate: A serious attempt to evade limits occurred in $60$ cases ($26\text{\%}$ of observations).

  • Success Rate: $34$ of the $60$ attempts that succeeded.

  • Trend: There is no clear upward trend in the frequency of successful or unsuccessful term-limit evasions.

Empirical Findings: Media Freedom and Attacks on Journalists

• Data Source: Committee to Protect Journalists (CPJ) database ($1992$-$2021$).

• Journalists Jailed: Shows a clear negative trajectory. There has been a significant increase in jailed journalists since around $2000$, including those specifically jailed for anti-state activities.

• Journalists Murdered: Shows a positive trajectory. After a peak in the late $1990\text{s}$ and early $2000\text{s}$, the number of journalists killed for their work has decreased steadily since $2008$.

• Interpretation: The authors suggest dictators may be shifting to softer tactics (using the judicial system to jail) rather than the harder tactic of murder (Guriev and Treisman $2020$).

The Aggregate Objective Index

• Components: The authors created a summary index by normalizing and averaging variables:

  • suffrage (V-Dem)

  • vote/seat shares (DPI)

  • party tenure (truncated at $20$ years, DPI)

  • legislative/executive competitiveness (DPI)

  • incumbent loss (NELDA)

  • multiparty index (NELDA)

  • process-violation index (NELDA)

  • existence of term limits/succession/dismissal rules (Meng).

• Aggregate Trend:

  • Unweighted: The index is at or near its historical peak in $2020$, showing no evidence of an aggregate global decline.

  • Population-Weighted: Shows a decrease starting around $2018$.

  • The India/China Effect: The population-weighted decline is almost entirely driven by the two most populous nations. When India and China are omitted, the trend mirrors the unweighted, stable graph.

Explaining the Discrepancy

• Mechanism 1: Time-Varying Coder Bias:

  • Common Shocks: Increased media attention on backsliding (especially in the US during the Trump presidency) may cause experts to perceive undemocratic behaviour as more common or severe than in previous years.

  • Rising Standards: Coders in $2023$ may apply stricter standards than those in $2015$.

  • Motivated Beliefs: Experts may want to see patterns that fit a popular global narrative.

• Mechanism 2: Strategic Substitution:

  • Subtle vs Blatant: Leaders might shift toward subtle, undemocratic actions (detectable by experts but not by objective indices) because the cost of blatant actions has increased.

  • Counter-argument: If leaders were successfully using subtle backsliding to insulate themselves, we should see an increase in incumbents winning or staying in power longer, but the objective data show the opposite.

Media and Academic Coverage Analysis

• Media Spikes: New York Times hits for democratic backsliding or democratic erosion surged around $2008$ and spiked massively in the mid-$2010\text{s}$.

• Academic Spikes: Google Scholar hits for these terms are exploding, while the term democratic transition is declining.

• Data Reporting: Reports of backsliding events (Gottlieb et al. $2022$) more than doubled from $2000$ to a peak in $2018$.

Conclusion and Recommendations

• The Extraordinary Claim: The authors argue that claiming massive global decline while objective indicators are stable is an extraordinary claim that requires stronger evidence than currently exists.

• Future Research:

  • Move beyond a few high-profile cases (e.g., Hungary, Venezuela) which may create a availability heuristic bias.

  • Investigate why journalist jailing is up while murders are down.

  • Focus on more objective measures of media freedom (e.g., state ownership percentage).

  • Rigorously account for potential time-varying bias in expert coding practices.