Keep the (social) distance! Turnout and risk perception during health crisis

The study examines how COVID-19-related risk perception affects electoral participation, focusing on Romania’s December 2020 parliamentary elections conducted during the pandemic’s second wave. It asks: (RQ1) To what extent does perceived infection risk deter voting? and (RQ2) How did COVID-19 affect turnout in a new democracy? Grounded in risk perception theory (risk-as-feelings vs. risk-as-analysis), the authors argue that heightened fear of infection raises the perceived costs of voting, thereby lowering participation. The case is salient because Romania held elections amid high case counts and limited procedural adaptations, with potential contagion effects from postponed local elections. The article aims to contribute evidence from a post-communist, new democracy context, assessing both individual-level intentions and aggregate turnout under pandemic conditions.

Research on turnout links participation to cost–benefit calculations (Downs; Aldrich) and to psychological factors like risk perception and affect (Slovic; Loewenstein). Crises can shift costs and benefits, creating a trade-off among infection risk, accountability motives, and personal vulnerability. Prior studies show mixed crisis effects on turnout and incumbent support, with evidence that disease outbreaks (Ebola, Spanish Flu) can depress turnout or alter vote shares. During COVID-19, multiple studies reported reduced turnout where cases/deaths were higher and effects of lockdowns on incumbent support. Risk-as-feelings is particularly relevant for health hazards; media exposure and perceived protection (e.g., via vaccines) shape risk perceptions. The literature on Eastern Europe and new democracies is thinner; this study addresses that gap by analyzing Romania’s 2020 elections.

Design: Two complementary models. Model 1 (individual level) uses a national survey (IRES) from June 2020 to test whether perceived infection risk reduces intention to vote (H1). Model 2 (aggregate) uses county-level data to test whether higher infection prevalence reduces turnout (H2). Data (Model 1): IRES national survey (June 2020). Variables: dependent—intention to vote; key predictors—perceived risk of infection, perception of election optimal conditions, preference to postpone elections; controls—age (recoded: 18–35, 36–50, 51–65, 65+), gender (female=1), residence (urban/rural), education (elementary/medium/higher), prior voting, vote option formed. N=898. Estimation (Model 1): Binary logistic regression: logit(Intend to vote) = β0 + β1(Higher risk perception) + β2(Postpone election) + β3(Election optimal condition) + β4(Age over 65) + β5(Female) + β6(Education dummies) + controls. Robust SEs; additional controls to mitigate endogeneity concerns; LRT used to assess added variable relevance. Data (Model 2): 42 Romanian counties (including Bucharest). Turnout in the 2020 parliamentary election from Permanent Electoral Authority (registered voters, votes cast, valid votes; turnout = present voters / registered). COVID-19 intensity from UBB-FSEGA Romanian Economic Impact Monitor: cumulative infected per population over the four months before election day (Aug 5–Dec 5, 2020) to align with similar risk environment as in the survey period. Controls from National Institute of Statistics: active population rate, share aged 60+, education level; also computed changes 2016–2020 where applicable. Estimation (Model 2): OLS with robust SEs. Dependent variable: county turnout rate (2020). Key predictor: percent of population infected (cumulative, 4-month window). Controls: active population (%), share aged 60+, education (%).

Model 1 (individual level, logistic regression):

• Higher risk of infection: coefficient −0.38 (SE 0.09), p<0.01—greater perceived risk significantly reduces intention to vote. Authors interpret as ~0.7 percentage-point decrease in voting probability per unit increase in perceived risk.
• Election optimal conditions: 0.46 (0.20), p<0.05—perceiving conditions as optimal increases intention to vote.
• Preference to postpone elections: −0.84 (0.21), p<0.01—those preferring postponement are less likely to intend to vote.
• Political behavior: vote option formed 1.27 (0.18), p<0.01; voted in last elections 1.31 (0.25), p<0.01—both strongly increase intention.
• Controls: female −0.26 (0.16), p<0.10—women less likely to intend to vote; age over 65: 0.08 (0.19), ns; education −0.38 (0.13), p<0.05; urban 0.24 (0.16), ns. N=898; Log-likelihood −494.29; AIC 1008.59. Model 2 (aggregate county level, OLS, N=42):
• Percent infected: −2.873 (0.336), p<0.01—1 percentage-point increase in the infected share reduces turnout by about 2.87 percentage points.
• Active population (%): 0.001 (0.00), ns; Age 60+: −0.013 (0.009), ns; Education (%): 0.003 (0.007), ns. Constant 0.388 (0.050), p<0.01. Overall: Both models support H1 and H2—higher perceived or actual COVID-19 risk is associated with lower electoral participation. Women appear more risk-averse in this context; age 65+ not a significant predictor of intention to vote despite higher clinical risk.

Findings address the research questions by showing that perceived infection risk (risk-as-feelings) meaningfully raises the subjective costs of voting, reducing intended participation (RQ1). At the macro level, higher local infection prevalence correlates with lower turnout (RQ2), consistent with a health-risk mechanism rather than standard sociodemographic determinants. The results align with literature on crisis-driven turnout declines and risk perception effects. They suggest that during health crises, voters may prioritize self-protection over civic duty, with gendered differences indicative of greater risk aversion among women. The lack of significant age effects on intention to vote implies that older voters did not exhibit higher voting aversion despite elevated clinical risk, possibly due to norms of electoral discipline. These insights are particularly relevant for newer democracies where institutional trust and adaptive electoral procedures may be limited, underscoring the need for risk-mitigating electoral arrangements (e.g., safer in-person protocols, alternative voting methods).

The study demonstrates that COVID-19 risk perception depresses intended participation (individual level) and that higher infection prevalence reduces turnout (aggregate level) in Romania’s 2020 parliamentary elections. It extends crisis-turnout research to a post-communist, new democracy, showing that during health emergencies voters’ behavior shifts from rational cost–benefit to affect-driven risk avoidance. Contributions include dual-level evidence (individual and county), gender-based differences in voting intention, and contextual insights for electoral management during crises. Future research should deploy refined causal designs (e.g., instruments, panel or event-study approaches), multilevel datasets (regions/counties/localities), and richer health-system indicators (cases, deaths, ICU capacity) to better disentangle mechanisms and address endogeneity. Improved local-level data access would enable more granular analyses of turnout under varying health risks and policy responses.

• Cross-sectional design emphasizes covariation; cannot establish causality.
• Potential endogeneity and bidirectionality between risk perceptions and voting intention; omitted variables may bias estimates.
• Aggregate model may omit other factors (e.g., GDP growth, institutional trust, migration) that also affect turnout, though robustness checks yielded similar results.
• Data constraints: lack of disclosed local-level COVID-19 and health system metrics (e.g., ICU beds) limit finer-grained analyses.
• Context specificity to Romania’s 2020 elections may limit generalizability without comparative evidence across new democracies.