The differential impact of climate interventions along the political divide in 60 countries

The study addresses how political polarization shapes climate change beliefs, policy support, and individual climate action worldwide, and whether widely used climate communication interventions affect liberals and conservatives differently. Prior work shows strong links between political ideology and climate beliefs, with liberals generally expressing greater belief in anthropogenic climate change and support for climate policies than conservatives. The authors examine whether this ideological polarization in beliefs translates into behavioral polarization (green act hypothesis) or whether belief–behavior incongruence emerges (green gap hypothesis). They also test whether eleven theoretically grounded interventions differentially influence beliefs, policy support, and an incentivized pro-environmental behavior across the ideological spectrum. Understanding these dynamics is crucial for designing effective, scalable interventions to increase climate mitigation efforts globally.

A substantial literature documents ideological divides in climate beliefs across many nations: liberals and left-leaning individuals more strongly endorse anthropogenic climate change than conservatives or right-leaning individuals (meta-analyses and cross-national studies in Australia, the UK, the US). Political identity salience can causally reduce conservatives’ climate beliefs and policy support. While attitudes often predict behavior, numerous factors (cognitive biases, perceived control, costs, social norms, efficacy) moderate this link, producing an attitude–behavior or green gap. Some evidence suggests liberals’ stronger pro-climate attitudes may not always translate into costly actions, while some climate skeptics (e.g., farmers) may still engage in pro-environmental practices. Interventions like scientific consensus messaging, moral foundations framing, and norm-based strategies have shown mixed and sometimes partisan-specific effects, with potential reactance among skeptics. This background motivates testing a broad set of interventions for differential ideological impact on beliefs, policy support, and behavior.

Design: A global, multi-country experiment in 60 countries (N=51,224 with ideology data) tested 11 climate interventions versus a no-intervention control. Interventions were crowdsourced, screened for feasibility and theoretical support, and expert-rated; top candidates were implemented in collaboration with proposing researchers. Participants were randomly assigned to one of 11 interventions or a control (reading a neutral literary passage). Measures: Climate beliefs were assessed via agreement (0–100) with four statements (α≈0.86–0.93 across mentions): necessity of action to avoid catastrophe, human causation, threat severity, and global emergency. Climate policy support captured agreement (0–100) with nine mitigation policies (α≈0.84–0.88). Climate action was measured behaviorally with the Work for Environmental Protection Task (WEPT): an optional, effortful screening task where each completed page planted a real tree (up to 8), creating actual environmental benefit at personal cognitive cost. The study funded 333,333 trees. Demographics (age, gender, education, income) and political ideology (0=liberal/left to 100=conservative/right; averaged social and economic scales, r=0.71) were collected; ideology was analyzed both continuously and via median split within country. Interventions: Eleven theoretically grounded interventions included dynamic social norms, work-together norms, effective collective action, psychological distance reduction, system justification framing, future-self continuity (letter from future self), negative emotions (doom-and-gloom facts), pluralistic ignorance correction (UN data), letter to a future generation, binding moral foundations framing, and scientific consensus messaging (99% of experts agree). Sampling/Procedure: Participants were largely recruited online via convenience/snowball sampling. Attention checks and WEPT demo ensured engagement and task comprehension. The survey was administered in the primary language of each country. Analysis: Hierarchical mixed-effects models accounted for clustering by item, participant, and country. Linear mixed models were used for beliefs and policy support; cumulative link mixed models for ordinal WEPT outcomes. Covariate-adjusted models included age, gender, income, and education. Bayes factors assessed support for null effects where relevant. Robustness checks (robust and weighted least squares mixed models) yielded identical conclusions. Time spent on the intervention phase was analyzed for its interaction with outcomes.

• Baseline polarization (control, N=4,302): Liberals believed more in climate change than conservatives (β≈-0.19, p<0.001; Cohen’s d≈0.35) and supported climate policies more (β≈-0.11, p<0.001; d≈0.27). These effects persisted after adjusting for demographics and when ideology was dichotomized.
• No behavioral polarization: Ideology did not predict number of trees planted on WEPT (β≈-0.001, p≈0.53; Bayes factor≈0.043 favoring the null ≈23:1). Results robust to covariates and binary ideology.
• Belief–behavior relationship differs by ideology: Beliefs and policy support predicted tree planting more strongly among liberals than conservatives (ideology×belief interaction p=0.009; ideology×policy interaction p=0.003). Standardized analyses showed actions were generally lower than beliefs, and conservatives’ actions were stronger than their beliefs relative to liberals, indicating a conservative-oriented green gap (conservatives acting despite lower belief).
• Interventions on beliefs: For liberals, six interventions increased beliefs: psychological distance reduction, effective collective action, future-self continuity, letter to future generation, system justification framing, and binding moral foundations. For conservatives, five increased beliefs: psychological distance reduction, effective collective action, future-self continuity, letter to future generation, and system justification. Thus, five interventions boosted beliefs across ideologies.
• Interventions on policy support: For liberals, five increased support: effective collective action, letter to future generation, future-self continuity, psychological proximity (reduced distance), and system justification. For conservatives, three increased support: effective collective action, letter to future generation, and future-self continuity; negative emotion messaging backfired (reduced support). Three interventions (effective collective action, letter to future generation, future-self continuity) increased policy support across ideologies.
• Interventions on behavior (WEPT): No intervention increased conservatives’ climate action. For liberals, scientific consensus increased tree planting; several interventions decreased action (letter to future generation, negative emotions, reduced psychological distance, work-together norms). For conservatives, eight interventions decreased action: effective collective action, future-self continuity, letter to future generation, negative emotions, pluralistic ignorance, reduced psychological distance, system justification, and work-together norms.
• Intervention time effects: Spending more time on the intervention phase generally predicted more trees planted, but effects varied: time increased action in scientific consensus and binding moral foundations, had no significant effect in dynamic norms, and decreased action in the other eight interventions—likely due to time budget constraints reducing time available for the WEPT.

The study demonstrates robust ideological polarization in climate beliefs and policy support worldwide, yet no corresponding polarization in an incentivized pro-environmental behavior. This dissociation is best explained by conservatives acting more than their beliefs would predict (a conservative-oriented green gap), rather than liberals failing to act on stronger pro-climate beliefs. Interventions showed outcome- and ideology-specific effects: several boosted beliefs and policy support across ideologies, but behavior proved harder to shift, particularly among conservatives. Scientific consensus messaging increased liberals’ action but had limited or negative effects for conservatives, aligning with prior mixed findings. Norm-based interventions, often considered state-of-the-art, did not reliably improve outcomes in this diverse global sample, suggesting limited generalizability of WEIRD-centric theories and the importance of cultural heterogeneity. Overall, the results inform targeted communication strategies and suggest that eliciting pro-environmental actions among conservatives may require approaches not framed explicitly as climate change solutions or that leverage alternative motives (e.g., purity, stewardship) without triggering reactance.

This global experiment clarifies that ideological polarization in climate beliefs and policy support does not necessarily translate into behavioral polarization in an incentivized climate action. The findings reveal a conservative-oriented green gap and identify interventions that can increase beliefs and policy support across ideologies (effective collective action, letter to future generation, future-self continuity) and a behaviorally effective approach for liberals (scientific consensus). Practically, practitioners can bolster conceptual outcomes broadly with certain interventions, but behavior change—especially among conservatives—remains challenging and may require alternative framings or mechanisms that do not focus on climate per se. Future research should: (1) test additional behaviors, including collective/systemic actions (e.g., advocacy, voting); (2) develop culturally grounded theories and tailor interventions to ideological audiences; (3) examine time and resource constraints in intervention delivery; and (4) explore belief-independent pathways to pro-environmental action.

• Behavioral outcome scope: Climate action was operationalized solely via the WEPT tree-planting task, a private, individual mitigation behavior; findings may not generalize to other behaviors, particularly collective or systemic actions (e.g., advocacy, voting, donations).
• Attrition: Overall attrition was 36.4% from initial completions to analyzed sample, typical for online studies but nonetheless a potential source of bias.
• Non-tailored interventions: Interventions were administered homogeneously and not ideologically targeted; tailoring could yield different or larger effects.
• Time constraints: Longer interventions may have reduced time available for the WEPT, contributing to observed decreases in action; generalizability to behaviors not constrained by time is uncertain.
• Cross-cultural heterogeneity: Substantial between-country variability suggests limited generalizability of some norm-based theories developed in WEIRD contexts; effect heterogeneity may obscure average treatment effects.
• Sampling: Predominantly online convenience/snowball sampling and lack of preregistered power analyses for country-level subsamples may limit representativeness.