A conjoint experiment on the content of national identity in Mainland China and Taiwan
Steven Denney · Leiden University
H. Christoph Steinhardt · University of Vienna
American Political Science Association Annual Meeting · Boston · September 2026 Shaping National Identity and the Good Citizen · Division 63
Do democracies and autocracies define the nation differently?
Situating the question
Ethnic and civic
Nations are imagined through descent or through residence, participation and shared principles. A continuum, and a contested one.Kohn 1955; Smith 1991; Brubaker 1992; Reeskens & Hooghe 2010; Bonikowski 2016
Institutions and preferences
The two tend to align and reinforce each other. Regime legacies show in attitudes decades later.Bisin & Verdier 2001; Bowles 1998; Alesina & Fuchs-Schündeln 2007; Pop-Eleches & Tucker 2017
Regimes and nationhood
Democracies are expected to define the nation in civic terms and autocracies in ethnic and deferential ones. The evidence is cross-national but rather thin.Nodia 1992; Tilly 1994; Kunovich 2009; Erhardt et al. 2021; Tudor & Slater 2021
Cross-national correlations cannot separate regime from culture.
National identity boundaries are less exclusionary in democracies
ISSP 2023, 29 countries, against the V-Dem liberal democracy index. OLS line with 95% band. Taiwan in blue. Mainland China is not in the ISSP.
A descriptive comparison of two regimes
What we know
Institutions and the preferences that sustain them tend to align.
What we ask
How do the criteria of belonging differ between (and within) Sinosphere countries?
Design
Preregistered conjoint
Cases
2
Respondents
5,028
Fielded 2025 to 2026 in quota-matched online panels.
Problems with observational data on national identity
Ceilings
When rated one at a time, often all (or most) criteria are important.Krosnick 1991; Sniderman & Grob 1996; Wright, Citrin & Wand 2012
Trade-offs
No way to disaggregate correlated items.Hainmueller et al. 2014; Denney & Steinhardt 2026
Measurement equivalence
Publics use rating scales differently, so a gap in ratings can be a gap in scale use.Davidov 2009; Bansak et al. 2021
Construct validity
Abstract items leave respondents to supply their own meaning.Bonikowski & DiMaggio 2016; Dawkins & Hanson 2024
Rarely used for national belonging and not before across regimes.
Common inheritance, one juncture, two paths
Mainland China n = 2,528
One-party state throughout. Patriotic education from 1994 tied the nation to the Party. Official narratives have grown more Han-centred.
Taiwan n = 2,500
Democratic since 1996. One-party rule until 1987. The 1997 curriculum reform shifted identification toward the island.
Zhao 1998; Wang 2012; Cantoni et al. 2017 · Chu & Lin 2001; Wang & Liu 2004; Chen et al. 2023 · Alesina & Fuchs-Schündeln 2007; Shi 2015
Navy bar marks the four preregistered hypothesis families.
Outcome
Which of these two is the truer guomin?
Estimation
Marginal means against even odds. Respondent-clustered errors. Preregistered on OSF.
Sample task
Which of these two is the truer guomin, the more genuine national?
Person A
Person B
Place of birth
Own country
Indonesia
Ethnicity
Han Chinese
Southeast Asian
Legal status
Permanent residence
Passport
Time in country
14 years
Most of life
Mandarin
Fluent
Limited
Patriotism
Patriotic, rarely expresses it
Patriotic, often expresses it
Political views
Prefers a strong leader
Values a free press
Social values
Conformist
Tolerant
Civic involvement
Attends government community events
Organizes petitions
Person A
Person B
Then a rating from 1 to 7 of each person. Open-ended questions as well.
What we expected
Ethnicity
Ancestry counts more under autocracy.
Civic participation
Self-organized participation is a plus in a democracy and costs in an autocracy.
Regime support
Democratic commitments is a plus in a democracy and cost in an autocracy.
Patriotism
Patriotism counts in both. Its public display counts more under autocracy.
We expected each regime to write its creed into the national ideal.
Ancestry, language and patriotism weigh almost the same.
Marginal means, bars from even odds, 90% and 95% intervals. The right column is Taiwan minus Mainland China with a 95% interval, pale when it covers zero.
What differs is whether political commitments count.
Political values differ and citizenship is valued differently.
Marginal means, bars from even odds, 90% and 95% intervals. The right column is Taiwan minus Mainland China with a 95% interval. The grey band bounds the mainland strong-leader and free-press effects within 2pp of no opinion.
Two publics organize belonging around different things.
Taiwan’s largest swings are political views and legal status. The mainland’s are patriotism and ethnicity, with Mandarin close behind.
Marginal means. Exploratory. Attributes with more levels have more room to swing.
The shared ethnic core is moving apart
Cohorts by the regime in place at about age 18, named as in the paper. Exploratory, FDR corrected, 90% and 95% intervals.
Two publics with one inheritance differ on the political boundary and are moving apart on the ethnic one.
Ancestry, language and patriotism weigh almost the same on both sides of the Strait.
Political commitments and citizenship count in Taiwan and barely on the mainland.
Across cohorts the ethnic core hardens on the mainland and softens in Taiwan.
[To explore] Open politics ties belonging to political beliefs; closed politics ties it to ancestry and identity.
Steven Denney · Leiden University · s.c.denney@hum.leidenuniv.nl H. Christoph Steinhardt · University of Vienna
Marginal means, bars from even odds, 90% and 95% intervals. The right column is Taiwan minus Mainland China. Every level is within 2pp of even odds. Petitions help in Taiwan and clear the correction. The mainland effects are signed as expected and do not.
Robustness
Logit
All 51 effects agree with the linear-probability estimates in sign and significance.
Rating outcome
Using the 1 to 7 rating rather than the choice, 23 of 25 Taiwan effects and 25 of 26 mainland effects agree in sign.
Weights
Reweighting to online and resident population targets moves no estimate outside its unweighted 95% interval.
Restricted design
Dropping the domestic-minority level, which does not name the same thing in both countries, widens the strongman difference to 12.9pp.
Two tests for a cohort explanation
A trait explains a cohort change only if it moves the preference and is more common among the young. Exploratory. Subgroup differences in marginal means for the Han level, respondent-clustered. Solid points survive a false-discovery correction. Taiwan has two such traits. The mainland has none we measure.
Concealment cannot be ruled out. Four things weigh against it.
Within the attribute
Mainland respondents distinguish among political levels. A pro-election profile is chosen 53 percent of the time, a strongman or free-press profile 49.
Reliability
Within-respondent reliability is 0.73 on the mainland and 0.74 in Taiwan, and no cohort differs from its country’s estimate.
Bounded, not absent
Equivalence tests bound the strongman and free-press effects within 2pp of the no-opinion reference, so the nulls are estimates, not missing data.
Design
Nine attributes vary at once, so no single criterion is transparently under test, and guomin names membership without naming either nation.
Not ruled out
Political selection into online panels. No external benchmark exists for the mainland. The Taiwan direct battery sits close to the ISSP estimates.
What the contested levels said
Domestic minority
A state-recognized ethnic minority on the mainland, a Taiwan Aboriginal in Taiwan. Not the same referent, so every cross-country contrast is re-estimated without it (Backup 2).
Prefers a strong leader
Prefers a strong leader unconstrained by elections. The other political levels are no opinion, prefers leaders elected one person one vote, and prefers a free press without government control.
Organizes petitions
Organizes others to petition the government. The other civic levels are stays home and takes part in government-organized community events.
Level descriptions as documented in the supplementary information (coding appendix). Slide labels are shortened.
← → navigate · R reset · F fullscreen · N notes · B blackout · P print preview