Choosing the right COVID-19 indicator: crude mortality, case fatality, and infection fatality rates influence policy preferences, behaviour, and understanding

The study investigates whether the choice of COVID-19 mortality indicators—crude mortality rate (CMR), case fatality rate (CFR), and infection fatality rate (IFR)—changes people’s policy preferences, attitudes, and behaviours, and whether people understand these indicators. Motivated by public discourse where different metrics can produce conflicting narratives (e.g., use of CFR versus deaths per capita), the authors examine how presenting different indicators affects support for preventive policies and personal compliance. The purpose is to identify which indicator is most effective at promoting containment-supportive preferences and behaviours and to assess public comprehension of these commonly used statistics during a pandemic.

The paper relates to several literatures. First, on communication effects in pandemics, prior work shows that how COVID-19 information is framed influences attitudes, policy preferences, and behaviours (Sotis et al., 2020; Garcia et al., 2020; Van Bavel et al., 2020; Arora and Grey, 2021; Kootsounidis and Rojo, 2020). Second, health literacy research (CDC definition; Nutbeam, 2008; Paasche-Orlow et al., 2005; Paakkari and Okan, 2020) documents that limited health literacy impairs decision-making and disease prevention, a challenge heightened during COVID-19. Third, messaging and political communication studies show wording and leadership cues affect compliance (Ajzenman et al., 2020; Plohl and Musil, 2020), with broader context on capacity and the need for behaviour change (Moghadam et al., 2020; Van Bavel, 2020). The paper also connects to research on risk perception, culture, and compliance (Huynh, 2020a; Huynh, 2020b) and to literature on valuation of life and economic implications of mortality (Rice and Cooper, 1954; Colley, 1976; Weisenthal, 1980; Viscusi, 2008; Power, 2020).

Design: Double-blind experiment approved by the Nanjing Audit University IRB. Participants: US residents recruited via Amazon Mechanical Turk (n = 1,199 as stated in text; descriptive tables report Ns around 1,900). Data were collected on Qualtrics. To avoid bias from forced responses, participants were not required to answer every question; no attention checks were included, consistent with cited methodological trade-offs. Randomization and groups: Participants were randomly assigned to one of three treatment groups or a control. Treatments presented information using one of three mortality indicators: crude mortality rate (CMR), infection fatality rate (IFR), or case fatality rate (CFR). The control group received no indicator information. The sample was generally balanced across groups, with some differences noted for income and political identification densities. Procedure: All participants first viewed a screen containing sufficient data to calculate CMR, IFR, and CFR before outcome measurement, to mimic realistic exposure without requiring deep statistical processing as would be unusual in typical media consumption. Outcome measures then captured policy preferences, attitudes, and behaviours. Understanding assessment: Comprehension was tested with three tasks: (1) a calculation task using provided data for Singapore to compute the assigned mortality rate (identify the correct numerator/denominator and perform the division to select the correct answer from four options); (2) a definition recognition task to select the correct definition of the assigned rate; (3) a second calculation task using data for an imaginary disease structured like the first task. Covariates and demographics: Standard demographics were collected: age (mean ~38 years), gender (about 42% female), income distribution, education (majority bachelor’s degree), political preferences (self-identified Democrats and Republicans), and COVID-19 news exposure frequency. Descriptive statistics and balance by treatment are reported (Tables 1–2). Statistical analysis: Outcomes were analyzed using ordinal logit models. Treatment indicators (CMR, IFR, CFR) were included, with models estimated both with and without controls (demographics, news exposure). Fixed effects or controls for individual characteristics (e.g., age) were included as specified. Effects are reported as changes in probabilities or percentage points for support of policies and behaviours. Additional models examined worry about health and economic crises (Table 4) and behavioural outcomes (Tables 5–6).

• Presenting crude mortality rate (CMR) increased support for containment policies: individuals were 26.9 percentage points more likely to support a law enforcing mask use compared to control.
• Presenting case fatality rate (CFR) decreased support for banning indoor events by 24.7 percentage points when models ignored controls; this effect’s significance depended on inclusion of news-related controls, suggesting heterogeneity by attention to COVID-19 news.
• Concern about future crises: CMR information was the strongest at increasing worry about an economic crisis relative to control, with positive and significant effects when controls and fixed effects were included (e.g., Table 4 shows significant coefficients for CMR in economic crisis worry models). IFR showed negative associations with health crisis worry under some specifications.
• Behavioural effects showed mixed and sometimes counterintuitive patterns: CMR information tended to incentivize attendance of small gatherings; IFR exposure led individuals to stay at home less; CFR exposure decreased propensity to use a mask.
• Understanding was limited across all groups: participants exhibited significant misunderstanding of COVID-19 indicators, struggling with both definitions and calculations for CMR, IFR, and CFR despite being provided necessary data.
• Overall, CMR appears more effective for eliciting support for policies and elevating concern relevant to containment, whereas IFR may reduce behavioural compliance, potentially due to confusion or misinterpretation of what IFR represents.

The findings support the hypothesis that the type of mortality indicator used in communication meaningfully affects attitudes, preferences, and behaviours, and reveal that public understanding of these indicators is low. CMR likely exerts stronger effects because it contextualizes risk at the population level, providing a clearer, more tangible frame for severity and potential economic impact. In contrast, IFR is harder to estimate and understand, possibly interpreted by laypeople as comparable to familiar illnesses like seasonal flu, reducing perceived severity and compliance. The results also suggest moral hypocrisy dynamics: information can increase approval of preventive policies without inducing personal compliance. Political identity and media narratives may moderate effects, with evidence that news consumption patterns interact with CFR impacts on support for event bans. Cultural and risk-perception factors (e.g., uncertainty avoidance, risk aversion) likely shape responsiveness to indicators and compliance with behaviours such as social distancing and staying at home. These findings highlight the importance of clear, accessible metrics and messaging to shape both policy support and personal preventive behaviours during a pandemic.

The study shows that indicator choice in COVID-19 communications matters: using CMR generally strengthens support for containment policies and increases concern about broader impacts, while IFR can undermine behavioural compliance, and CFR has mixed effects on policy support and behaviours. The work challenges claims of no link between communication strategies and support for measures, aligning instead with evidence that beliefs about pandemic risk influence policy preferences and behaviour. Policy recommendations include: (i) regularly providing clear definitions of COVID-19 indicators to address information gaps; (ii) favouring CMR in communications during normal pandemic periods to normalize protective behaviours (e.g., mask use, sanitization); (iii) using CFR strategically to enhance public understanding of the pandemic’s gravity when seeking support for limits on social events. The authors note implications for vaccination messaging, where misunderstanding of efficacy can reduce willingness to vaccinate, and framing personal health risks can improve compliance. Future research should examine different populations and pandemic phases to assess how efficacy of indicator-based communications varies over time and across contexts.

Methodological limitations include potential omitted variable bias and possible inclusion of non-randomized participants. The sample is drawn from US MTurk users, limiting generalizability and introducing potential partisanship effects. The design intentionally avoided forced responses and omitted attention checks, which may affect data quality. Reported Ns in descriptive tables differ from the initial recruitment figure. Effects may vary by pandemic phase and across countries or cultural contexts.