Article A / 2024 / 2025 issue
Conceptualizing and Measuring Support for Democracy: A New Approach
A broad democracy question leaves institutional commitments unclear. This study tests concrete questions across 19 samples and develops a shorter scale of support for liberal democratic institutions. It also shows why institutional support, satisfaction with government performance, and comparable national scores require separate attention.
Explanation, evidence, and limits
In everyday terms: saying “I support democracy” can mean different things to different people. This study asks more concrete questions about elections, rights and limits on leaders to find out what people support.
Suppose a survey asks whether democracy is a good system. One respondent imagines competitive elections. Another imagines an elected majority able to carry out its program. Another imagines courts and journalists able to challenge that majority. They can all answer yes while disagreeing about a concrete institutional conflict. An attractive word can also invite agreement without much reflection. Claassen and colleagues therefore ask what a democracy-support question actually measures.
Their target is support for liberal democracy: elections together with freedoms, equal legal rights, and constraints on government power. Here liberal describes institutional protections, not a left-wing party or an American partisan identity. An executive is the president, prime minister, or government making and implementing decisions. Constraints allow other institutions, including courts and parliament, to examine and sometimes block those decisions. Winning an election does not remove these constraints.
The authors start from eight components in the V-Dem definition: expression, association, voting rights, elected decision-makers, fair elections, judicial oversight, legislative oversight, and equality before the law. Concrete items make a stated conflict observable. As a different-topic analogy, asking whether a music association is “fair” does not tell us whether its members support open auditions, equal appeal rights, or a jury elected by members. A useful measure names its institutional target rather than trusting the appealing label.
Study details · four evidence anchors
The study in four receipts
- The question changes the measure
- The battery asks about specific rights and institutional limits. A yes to a general democracy label does not answer each of these questions.
- One commitment is an empirical question
- Factor analysis tests whether answers fit together. Some institutions connect less consistently with the shared pattern; the authors do not assume every respondent endorses every component together.
- A shorter scale has a coverage cost
- The seven-item version improves how consistently answers fit together but omits support for elected decision-makers. A convenient score cannot capture every institutional issue.
- A score is not an election result
- The study validates a measure of institutional support. It does not follow these respondents into an election or demonstrate that a particular answer determines an individual vote.
Table 3 lists 19 samples fielded during 2022–23. Portugal and Taiwan use probability samples; most others use online quota samples, while Turkey uses stratified online recruitment. The main models are fitted separately for each sample.
The 17-item models use ordinal confirmatory factor analysis and include an orthogonal acquiescence factor to account for a tendency to agree with differently framed questions. One- and two-dimensional models fit similarly; the full unidimensional model satisfies all three reported fit thresholds in six samples.
The concise set contains FreeExp2, FreeAssc1, UniSuff1, FFElect2, JudCnstr2, LegCnstr1, and EqLaw1. It covers seven of eight institutional components and omits elected decision-makers. Cronbach’s alpha is generally acceptable, with Peru just below .70; item selection may make later retest reliability lower.
Note 8 explicitly discusses non-equivalence across languages, samples, and wording. The full scale does not establish even configural equivalence. Do not present country mean differences as a validated ranking. In the main analyses sample weights are not used; weighted checks are discussed in the supplement.
Replications: https://doi.org/10.7910/DVN/ZVXV2K. Published-version PDF read in full; no correction located in checked publisher metadata and DOI searches on October 12, 2026 Seattle time.
How the researchers studied it
The researchers field 17 questions in 19 national samples. Most are online samples recruited using demographic targets rather than random selection from the population. They then use factor analysis: a statistical way to check whether answers to different questions move together as a common underlying commitment would predict. They compare a single commitment with separate electoral and liberal commitments.
They also investigate questions that do not fit the common pattern. Finally, they select seven questions that cover much of the concept while cohering more consistently. Testing a measure asks whether it captures the intended commitment; it does not itself test an election or make someone more committed.
How consistently answers fit together and coverage are different tests. Dropping an item can make answers move together more strongly while removing an important part of the concept. A within-context score may still be informative even when direct cross-country ranking is unjustified. Check how respondents interpret the item as well as how answers correlate.
What they found
A general commitment connects many answers, but particular institutions can have distinctive patterns. The shorter seven-question scale usually performs better than the full set as one combined measure. It leaves out elected decision-makers, whose questions did not fit consistently. Satisfaction with how democracy is working also differs from support for its institutions.
Where the claim stops
The samples and wording differ, and the study does not establish that national scores are directly comparable. Local conflict is a plausible interpretation of some unusual patterns, not a causal effect isolated by the surveys. The scale can improve diagnosis without revealing a person’s complete understanding or guaranteeing a vote.
Read Our Measurement Approach; Data and Research Design; Examining Item Heterogeneity; Deriving and Validating a Concise Battery; and note 8 on measurement equivalence.
Read first: Start with the explanation on this screen. Then check what was compared, what was found, and what remains uncertain. The journal article link is optional.
Free published-version PDF in the University of Glasgow repository · CC BY.
