Special analysis · Bayesian HGLM · Peer review pending
Does democracy protect scientific integrity?
We modeled retractions per 1,000,000 publications against the EIU Democracy Index, adjusting for GDP per capita, control of corruption, government effectiveness, and international collaboration in a Bayesian hierarchical negative-binomial model. The country-year association is weak and negative — directionally consistent with more democratic countries retracting less, but far from deterministic and sensitive to how publication volume is handled.
Pearson r
—
Democracy IRR
—
rate ratio, multivariate
Corruption-control β
—
control of corruption (WGI)
Countries
—
in current analysis
Figure 1
Democracy index vs. retraction rate, by country
FullFlawedHybridAuthoritarian
Loading…
Reading the chart. Each point is a country-year, sized by total publication output and colored by the dataset’s EIU regime classification. The vertical axis is logarithmic because rates span five orders of magnitude; country-years with no retractions sit on the zero line. Rates are per 1,000,000 citable documents, so tiny-output countries can post very high values from a handful of retractions — read the cloud, not the extremes.
Figure 2
Retraction rate over time, by regime
AuthoritarianHybrid regimeFlawed democracyFull democracy
Loading…
Reading the chart. The axis is logarithmic; without it the two authoritarian spikes flatten every other line. Rates start close together in 2006 and separate from about 2015, with authoritarian states running roughly two to three times the others by 2025. The 2010 and 2023 peaks are single large publisher-led cleanups concentrated in a few countries, not step changes in underlying behavior, and 2024 to 2025 will keep rising as retractions continue to be issued for those years.
Average retraction rate by region
Loading regional data…
Figure 3
Effect-size ladder
Loading effect estimates…
Figure 4
Retraction rate by regime type
Loading…
Pooled across — years. Full democracies sit at — per million against — for authoritarian states.
Robustness
Sensitivity analyses
Loading…
Largest contributors by volume
| Country | Regime | Retractions | Rate / 1M |
|---|---|---|---|
| Loading… | |||
How democracy correlates with covariates
Loading…
Country-year panel · highest rates
Country-years with at least 1,000 citable documents, so the ranking is not just the smallest publishers.
| Country | Region | Regime | Democracy | Rate / 1M | Publications |
|---|---|---|---|---|---|
| Loading… | |||||
Caveats
What this doesn’t prove
- Correlation is not causation. Regimes vary on many axes; we control for some but not all confounders.
- Retraction detection may be higher in democracies with more transparent journals, partially confounding the observed gradient.
- Publication volume estimates use OpenAlex, which under-counts some non-English literature.
- The model treats country-years as exchangeable conditional on covariates; serial correlation within countries is partially addressed via random effects.