For much of the 20th century, democracies that failed usually did so with tanks rolling through capitals. In the 21st century, the threat looks different. Elected leaders in countries from Hungary to Venezuela, from the Philippines to the United States, have chipped away at democratic institutions from the inside: attacking the press, weakening courts, sidelining legislatures, and casting doubt on the legitimacy of elections.
This pattern, often called democratic erosion or backsliding, has spread to roughly two dozen countries since 2000. A new analysis in PNAS asks what these countries have in common, and arrives at a consistent answer: income inequality.
A question that had gone untested
Political scientists have long suspected that economic inequality and democratic fragility are linked. Unequal societies tend to be more polarized, and polarization has been identified as a feature of nearly every case of erosion. Surveys in the United States show decades-long declines in public trust in institutions, which some scholars connect to widening income gaps.
But suspicion is not evidence. Earlier research on military coups produced mixed results on whether inequality played a central role, and the current wave of erosion had not been tested systematically against economic data. The problem, as Eli G. Rau of Tecnológico de Monterrey and Susan Stokes of the University of Chicago note, is partly one of measurement. Backsliding is slow, stealthy, and easy to confuse with ordinary executive overreach, making it hard to build a dataset that researchers can run through statistical models.
Rau and Stokes built their study on recent work by political scientist Melis Laebens, who developed a method to identify erosion cases using expert-coded indicators from the Varieties of Democracy (V-Dem) project. Laebens defines erosion as a significant decline in two kinds of accountability: vertical (the constraints voters and civil society place on governments) and horizontal (the constraints coequal branches, like courts and legislatures, place on the executive). Using this approach, the authors identified 23 spells of democratic erosion across 22 countries between 1995 and 2020.
What the numbers show
The authors then ran logistic regressions on country-year data, testing whether a country’s Gini coefficient, a standard measure of income inequality where higher numbers mean greater inequality, predicted whether that country was experiencing erosion in a given year. Their sample covered 92 countries that were continuously democratic during the study period, plus Venezuela, which began the period as a democracy and ended it as an autocracy.
The association was strong. In the most equal countries, the predicted probability of erosion in any given year sat in the single digits. In the most unequal, it rose above 30%. To put numbers on the gap: a country as equal as Sweden (Gini of 26.4 in 2017) had a predicted erosion risk of about 4%. A country as unequal as the United States (Gini of 38.4) had a predicted risk of 8.4%. South Africa, the most unequal democracy in the dataset, carried a predicted risk of 31%.
Using only three pieces of information, a country’s Gini coefficient, its GDP per capita, and the year of observation, the model correctly distinguished eroding from non-eroding country-years about 80% of the time.
Which inequality matters?
The Gini coefficient collapses an entire income distribution into a single number, which can hide very different patterns. To probe further, the authors drew on the World Inequality Database to examine income and wealth shares held by the top 1%, the top 10%, and the bottom 50% of the population.
Across 12 additional models, the pattern held. The larger the share of income or wealth flowing to the top, the higher the risk of erosion. The larger the share going to the bottom half, the lower the risk. Wealth concentrations at the very top turned out to be slightly more predictive than income inequality, but the differences were small. As Rau and Stokes put it, inequality matters “quite generally, regardless of the specific structure of the unequal distribution.”
What didn’t predict erosion
Several factors that shaped democratic stability in earlier eras did not perform the same way here.
National wealth, measured as GDP per capita, had a weaker and less consistent relationship with erosion than inequality did. In some models poorer countries appeared slightly more vulnerable, but the effect was not robust across specifications.
The age of a democracy, a reliable predictor of resistance to military coups, showed no protective effect against erosion. This finding held even when the United States, the world’s oldest democracy, was removed from the analysis. The authors describe this as a stark departure from earlier patterns of democratic breakdown.
State capacity, measured through indicators of bureaucratic quality, also showed no consistent effect, though the authors caution that this concept is notoriously hard to measure.
Polarization as a link in the chain
When the authors added a measure of political polarization to their models, polarization did emerge as a significant predictor of erosion, consistent with a broad literature in the field. But adding polarization did not eliminate the effect of inequality. Both variables added predictive power that the other did not capture.
The authors interpret this as evidence that polarization is one pathway, but not the only pathway, from inequality to erosion. In their reading, unequal societies generate resentment toward elites and cynicism toward institutions, which aspiring autocrats can harness in various ways, sometimes by stoking partisan hatred, sometimes by directing grievance toward immigrants or minority groups, sometimes by simply arguing that corrupt institutions are not worth defending.
They also point to evidence of contagion. Including year of observation as a variable, they found that erosion became more common over time, net of other factors, consistent with the idea that backsliding leaders learn from and imitate one another.
What the study does and does not claim
The authors are careful about the limits of their analysis. Identifying erosion cases relies on expert coding, and some scholars have raised concerns about such measures, though V-Dem is generally regarded as the highest-quality source available. The results hold when the authors substitute a slightly different set of backsliding cases used by other researchers, when they switch among four different sources of Gini data, when they use wealth rather than income measures, and when they add country fixed effects to absorb stable national characteristics.
The authors also note that once in power, backsliding leaders sometimes do reduce inequality. Left-populist leaders like Evo Morales in Bolivia and Rafael Correa in Ecuador oversaw drops in income inequality, and even right-wing backsliders have sometimes favored more generous social spending than competing conservative parties. That complicates any simple narrative about what these leaders do with the economic grievances that helped bring them to power.
What the study does argue is that the erosion of democracy in wealthy, long-established countries like the United States is not a freak occurrence. The researchers interpret their results as evidence that inequality is a structural risk factor, and that policies aimed at narrowing income and wealth gaps may have democratic as well as economic consequences. More research, they note, is needed to pin down the exact mechanisms and to identify interventions that can halt or reverse the process once it begins.
