Democracy's delivery gap

The gap is larger than the ballot box

Free and fair elections matter greatly to respondents. They are not, however, the only standard against which people judge democracy. Equal treatment by courts, protection against poverty, and a government's response to income differences also appear in the democratic expectations recorded by the European Social Survey.

Across the 18 countries with usable designs in both rounds, giving each country equal weight, the poverty-protection gap averaged 3.94 points in Round 10. The corresponding election gap was 1.69 points. Those are differences between two ratings with different anchors, not quantities of democracy that institutions have objectively failed to supply. They nevertheless reveal a substantial mismatch between the importance attached to a principle and its perceived realization.

A shared scale does not make the questions identical

The first question asks how important a principle is for democracy in general. The second asks how fully a related statement applies in the respondent's country. Both use numbers from zero to ten. The first measures a normative expectation; the second measures a perception of practice. Subtracting them is a transparent descriptive convention, not proof that they are interchangeable cardinal scales.

The comparison is paired within each respondent. Someone who answers only the importance question cannot contribute an importance rating to a gap whose performance rating comes from somebody else. The same eligible people determine both displayed means and their difference, and the interval for that difference accounts for the covariance between their answers.

France makes the distinction concrete. For protection against poverty in Round 10, the paired mean importance rating is 8.71, while perceived performance is 4.40. The resulting gap is 4.31 points, with a pointwise 95% interval from 4.16 to 4.47, based on 1,939 paired respondents. Neither the difference nor its precision tells us why respondents made those judgments.

The exceptional cell is as informative as the dominant pattern. In Czechia, minority-rights performance is rated 7.43, above the importance rating of 7.07. The paired difference is -0.37 points, with a 95% interval from -0.55 to -0.18. Calling that "overdelivery" would smuggle a normative judgment into the arithmetic. The data show two differently framed responses, not whether minority rights are objectively too extensive, sufficient, or secure.

The other 197 of 198 latest-round country-principle cells have positive differences and positive lower bounds in their individual 95% intervals. These are pointwise intervals, not a simultaneous guarantee covering the entire matrix. The near-universal numerical gap is therefore a finding about the response pattern and its definition, not a license to rank national democracies.

Large shortfalls can persist while their direction changes

Two rounds allow a comparison, not a continuous history. Round 6 was fielded in 2012/13; the integrated Round 10 file covers 2020-22. Different people were interviewed, participating countries changed, and the later fieldwork included video interviews. The historical comparison here holds the usable country set constant and retains the question-level definitions.

The equal-country poverty-protection gap narrowed from 4.46 to 3.94 points. For media freedom, it widened from 0.81 to 1.54 points. The component ratings help interpret the contrast. For poverty protection, mean importance changed little, while perceived performance improved. For media freedom, mean importance rose and perceived performance fell. A single composite score would blur these different movements before the reader had a chance to examine them.

France's poverty-protection gap scarcely moved in point-estimate terms: 4.26 in Round 6 and 4.31 in Round 10. The estimated change is +0.05 points, with a 95% interval from -0.21 to +0.31. The interval does not establish exact stability, but it does not support a precise claim of widening either. The gap's level and its change answer different questions.

Country-principle changes also cut in both directions: 84 point estimates increase and 78 decrease among the 162 matched comparisons. These counts describe signs, not the number of statistically established changes. The exact table retains uncertainty and Benjamini-Hochberg adjusted p-values across the full 162-comparison family for each measure. Sorting point estimates does not identify an exact national ordering.

The respondent sample is part of the result

Pairwise completion preserves more observations than requiring every person to answer all 18 comparable questions. It can also mean that different principles represent slightly different subsets of respondents. The complete-response comparison asks how much that selection changes the result; it does not treat complete responders as a more representative population by definition.

For France's poverty-protection gap, the four prespecified restrictions produce point estimates between 4.31 and 4.34 after rounding. The conclusion of a large difference survives these particular restrictions. That is a bounded robustness finding, not proof against every form of selection or measurement bias. Other countries and principles retain their own estimates and sample sizes.

Face-to-face restrictions exclude video and unknown modes rather than treating them as equivalent interviews. Since mode was not randomly assigned in this comparison, any resulting difference combines mode with selection and fieldwork circumstances. It cannot be labeled a causal survey-mode effect.

What this evidence does not prove

The ESS measures residents' answers, not an independent institutional audit. Sampling intervals do not include translation differences, nonresponse bias, response styles, or the possibility that people use the two scales differently. Source-questionnaire comparability is documented; independent retranslation of every national questionnaire and measurement invariance are not claimed.

Nor can these two repeated cross-sections establish the effect of the pandemic, a government, an institutional reform, or an individual's changing beliefs. Denmark and Israel remain unavailable in the Round 6 analysis because the frozen design supplements failed the required identity checks. No missing design is repaired by inventing clusters or dropping uncertainty.

The useful conclusion is more specific than democratic decline in the abstract. Citizens distinguish elections, equal treatment, protection, representation, and accountability. Their expectations and assessments do not move together. For policy readers, the next task is to investigate the institution and principle behind a particular mismatch, then test that explanation against evidence beyond the survey. The matrix locates the question; it does not supply the cause.