A 27-country test of communicating the scientific consensus on climate change

The study addresses a persistent gap between the near-universal scientific consensus that human-caused climate change is occurring (97–99.9% in the literature) and public underestimates of that consensus, which can reduce belief in climate change, worry, and support for action. Prior research—largely from the United States and other Western, high-income countries—indicates that communicating the “97% consensus” can correct misperceptions and modestly shift beliefs and worry, though effects on policy support are inconsistent. To extend generalizability, this preregistered experiment tests two messages across 27 countries: a classic consensus message (97% of climate scientists agree that human-caused climate change is happening) and an updated message that also conveys that 88% of climate scientists (surveyed IPCC authors) agree climate change is a crisis. The research aims to: (1) evaluate whether each message reduces misperceptions and increases beliefs, worry, and support for public action versus a control; (2) assess whether the updated message yields added benefits for crisis belief, worry, and public action versus the classic message; and (3) examine moderators at the individual (message familiarity, trust in climate scientists, political ideology) and country (individualism–collectivism, power distance) levels. Findings inform theory on norm-based communication across cultures and provide practical guidance for scalable climate communication.

A substantial body of experimental work shows that simple messages about the scientific consensus increase perceptions of consensus and, to a smaller extent, beliefs and worry about climate change. Meta-analyses report large effects on correcting consensus misperceptions (Hedges’ g ≈ 0.56) and small effects on proclimate attitudes (g ≈ 0.09–0.12), with limited evidence for direct effects on support for public action. Prior studies are concentrated in the United States and a few Western countries, with rare exceptions (e.g., Japan, Germany), leaving cross-cultural generalizability unclear. The Gateway Belief Model proposes that perceived scientific consensus influences beliefs and worry, which in turn affect policy support. Evidence on moderation by political ideology and trust in science is mixed, with some studies showing larger effects among conservatives or higher-trust individuals, while others show similar effects or no moderation. Cultural factors such as collectivism and power distance could plausibly shape responsiveness to expert norms, but prior evidence beyond the U.S. context is sparse.

Design: Preregistered, multi-country, online, between-participants experiment with three conditions: (1) classic consensus message (“97% of climate scientists agree that human-caused climate change is happening”); (2) updated consensus message (classic plus “88% of climate scientists agree that climate change constitutes a crisis”); and (3) control message (unrelated norm: “97% of dentists recommend brushing your teeth twice per day”). Participants were randomly assigned (double-blind). Pre/post measurement of perceived scientific consensus and crisis agreement used an “estimate and reveal” approach to highlight gaps and enhance salience, with pre-intervention estimates controlled in analyses. Sampling and setting: Data were collected 27 July–4 August 2023 from convenience samples in 27 countries across six continents (N = 10,527 analyzed). Recruitment used researcher networks, social media, mailing lists, and Prolific supplements in Canada and Mexico. Inclusion criteria: ≥18 years, resident of a target country, fluent in survey language. Ethics approvals obtained (University of Amsterdam FMG-1123; University of Porto 2023/06–12); informed consent collected. Median completion time: 6.33 minutes. Measures: Outcomes included perceived scientific consensus (0–100%) and crisis agreement (0–100%); confidence in each estimate (0–100; exploratory); beliefs on: climate change reality (1–7), human causation (1–7), climate change as a crisis (1–7); climate change worry (1–7); and support for public action (1–7). Moderators: message familiarity (with 97% and 88% statements; 1–7), trust in climate scientists (1–7), and political ideology (0=left to 10=right). Demographics included age, gender, education, urbanicity, student status, and others. Translation followed forward–back translation with fidelity checks and pretests. Exclusions and missing data: Participants failing an attention check (n=1,032) or finishing in <2 min (n=0) were excluded; additional fraud/bot checks applied to paid samples per Qualtrics indicators. Forced responses minimized missingness; specific responses coded as missing where appropriate and handled per-analysis. Analysis: Bayesian model-averaging with mixed-effects (participants nested within countries). Linear mixed models for continuous outcomes (perceived consensus, crisis agreement) and mixed-effects cumulative probit models for ordinal outcomes (beliefs, worry, public action). Models included demographic covariates (age, gender, university degree, political ideology) and averaged over random-slope structures to allow for cross-country heterogeneity. Priors were informed by prior meta-analytic effects (e.g., d≈0.10 for ordinal outcomes; g≈0.55 for consensus correction). Complementary frequentist random-effects meta-analytic summaries reported Cohen’s d and between-country heterogeneity (τ) with CIs. Preregistration and materials/code/data available on OSF (https://osf.io/z6quh/). Deviations included reordering a comprehension check to preserve moderator validity and not measuring subjective income to keep the survey brief.

Misperceptions (pre-intervention): Across 27 countries (N=10,527), the 97% scientific consensus on human-caused climate change was underestimated by an average of −12.11 percentage points (95% CI −12.43, −11.80); 72.2% (95% CI 71.3, 73.0) underestimated it. Underestimation ranged from −20.91% (China) to −7.54% (Germany). The 88% scientific agreement that climate change is a crisis was slightly underestimated on average by −4.14% (95% CI −4.47, −3.81); 44.5% (95% CI 43.6, 45.5) underestimated it. Classic message vs control: Strong evidence that perceived scientific consensus increased (BF10 ≈ 2.01×10^12; d=0.47, 95% CI 0.41, 0.52), with some between-country heterogeneity (τc≈0.06). Belief in climate change (BF10=25.51; d=0.06, 95% CI 0.01, 0.12) and human causation (BF10=467.86; d=0.10, 95% CI 0.04, 0.15) increased slightly. Worry increased modestly (BF10=5.03; d=0.05, 95% CI −0.01, 0.10). No evidence for a direct effect on support for public action (BF10=0.62; d=0.02, 95% CI −0.03, 0.08). Exploratorily, belief that climate change is a crisis increased (BF10=35.80; d=0.06, 95% CI 0.01, 0.11). Updated message vs control: Perceived reality consensus increased (BF10 ≈ 2.12×10^12; d=0.47, 95% CI 0.41, 0.52) and perceived crisis agreement increased (BF10 ≈ 1.54×10^5; d=0.23, 95% CI 0.16, 0.31), with small heterogeneity (τc≈0.05 and 0.15, respectively). Effects on beliefs (reality d≈0.07; causation d≈0.09), worry (d≈0.04), and public action (d≈0.02; BF10=0.65) mirrored the classic message. Evidence for increasing crisis belief was weak (BF10=1.80; d=0.04, 95% CI −0.01, 0.09). Updated vs classic: No added benefit on crisis belief (BF10=0.09; d=−0.02, 95% CI −0.07, 0.03), worry (BF10=0.11; d=−0.01, 95% CI −0.06, 0.04), or public action (BF10=0.25; d=0.00, 95% CI −0.04, 0.05). However, participants reported higher confidence in their agreement estimates after the updated message (exploratory; BF10 ≈ 4.90×10^8; d=0.44, 95% CI 0.38, 0.51). Moderators (vs control): Message familiarity strongly moderated effects—messages were more effective among those less familiar (classic on reality consensus BF10 ≈ 1.43×10^16; updated on reality consensus BF10 ≈ 2.53×10^22 and on crisis agreement BF10 ≈ 2.76×10^10). Lower trust in climate scientists and more right-leaning ideology were associated with larger updates for the updated message (reality consensus BF10 ≈ 3.34×10^5 and 2.52×10^5; crisis agreement BF10 ≈ 2.60×10^6 and 3.10×10^16), consistent with greater initial misperceptions and thus more room to update. For the classic message, there was evidence against stronger effects among high-trust individuals (BF10=0) and weak evidence that effects were slightly larger among right-leaning individuals (BF10=1.89). Additional analyses showed larger classic-message effects among those with lower pre-intervention consensus perceptions (BF10 ≈ 6.87×10^21) and, to a lesser extent, greater increases in belief in reality (BF10=12.47), but not for causation, worry, or public action. Country-level moderators: No convincing evidence that individualism–collectivism or power distance moderated effects; analyses were likely underpowered for country-level moderation. Overall: Consensus messaging robustly reduces misperceptions across countries and modestly increases beliefs and worry, with no direct shift in public action support. The updated message offers no advantage over the classic message beyond boosting confidence in perceived scientific agreement.

The findings confirm that communicating expert consensus is an effective, scalable approach to reduce misperceptions about climate change across diverse international samples. The classic 97% message substantially corrected perceptions of the scientific consensus and produced small, consistent improvements in beliefs and worry, aligning with prior meta-analyses. The lack of a direct effect on support for public action suggests that consensus cues may need to be complemented by additional strategies or operate via indirect pathways (as posited by the Gateway Belief Model) to influence policy preferences. The updated message, adding the 88% “crisis” agreement, did not outperform the classic message on key attitudinal outcomes, likely because public perceptions of crisis agreement were already relatively accurate and because perceived dissent (88% vs 97%) may blunt persuasion. Moderation patterns suggest consensus messaging is non-polarizing and may be especially useful for audiences with greater initial misperceptions (lower familiarity, lower trust in climate scientists, more right-leaning), consistent with Bayesian updating accounts. Cross-country heterogeneity in effect sizes was small to modest, and exploratory analyses did not identify reliable country-level cultural moderators, highlighting the need for larger cross-national datasets to detect such effects.

This large, preregistered, 27-country experiment demonstrates that brief scientific consensus messages reliably reduce misperceptions and modestly increase beliefs and worry about climate change, but do not directly increase support for public action. Adding an explicit “crisis” agreement statistic does not enhance attitudinal outcomes beyond increasing confidence in perceived agreement. The intervention is most impactful among individuals less familiar with the message and those with higher initial misperceptions, including people with lower trust in climate scientists and more right-leaning ideologies. Future work should: (1) test indirect pathways (e.g., Gateway Belief Model) with pre/post belief measures; (2) evaluate effects on specific mitigation and adaptation policy preferences; (3) investigate strategies to communicate high scientific confidence about impacts and urgency; and (4) employ representative and larger multi-country samples to assess country-level moderators and optimize targeting.

- The study focused on direct effects and did not measure all pre-intervention beliefs, limiting tests of indirect pathways proposed by the Gateway Belief Model. - Effects on beliefs and worry were nominally small, though consistent with meta-analytic estimates; practical significance may depend on scalability and targeting. - Convenience, social-media-based sampling skews toward younger, more educated, more liberal, and more female participants, potentially affecting generalizability; however, effect sizes align with prior meta-analyses using representative U.S. samples. - Hard-to-reach populations (e.g., those offline) may be less exposed to such messaging in real-world campaigns. - Uncertain extent of between-country heterogeneity and insufficient power to detect moderation by country-level cultural dimensions (individualism–collectivism, power distance); results on country-level moderators are tentative. - Single-item measures, particularly for support for public action, may limit sensitivity; future studies should assess specific mitigation and adaptation policies.