Climate Justice Norms

Subjects will either receive (i) information about how the average person reduced CO2 emissions or (ii) how the highest-income individuals reduced CO2 emissions or (iii) no information.

To generate different numerical values of CO2 reductions, we will vary the reference year from 2008 to 2009. This will allow us to provide causal evidence on whether subjects react to the degree of correction.

Thus, there are six experimental conditions.

-Control 2008: No information.
-Control 2009: No information.
-Average 2008: The average income group reduced CO2 emissions since 2008 by -22%.
-Rich 2008: The top 10% income group reduced CO2 emissions since 2008 by -22%.
-Average 2009: The average income group reduced CO2 emissions since 2009 by -16%.
-Rich 2009: The top 10% income group reduced CO2 emissions since 2009 by -15%.

2026-10-12

2026-12-31

-Donations to Future Cleantech Architects (green tech innovation) and Atmosfair (Carbon offsetting).
-Policy support for green policies.

The donation decisions are incentivized: a random subset of respondents is drawn, and their chosen allocation is paid out using real money.

We measure respondents’ support for green policies by asking whether they support or oppose each of the following policy measures on a five-point Likert scale: i) mitigation measures aimed at achieving climate neutrality; ii) a cap-and-trade system (ETS), iii) government efforts to accelerate the development of clean technologies, iv) more coordinated or collective approaches to further reduce CO2 emission across sectors and regions, such as implementing carbgon taxes, and v) adaptation strategies such as increased recycling, upgrading to energy-efficient air conditioners, or installing heat pumps. Our primary outcome will be an index of respondents’ support. As a secondary analysis, we will examine treatment effects on support for separate policy measures.

- Regional heterogeneity with respect to important economic, environmental, and cultural variables (e.g., income inequality, per capita emissions, disaster experience)

- Heterogeneous treatment effects with respect to prior beliefs regarding CO2 emissions (belief updating).

- Heterogeneous treatment effects with self-perception (income ladder position & regional identification as (EU, national, subnational level).

- Heterogeneous treatment effects with respect to climate skepticism and norms related to climate support.

We first elicit demographics and self-perceptions regarding income and regional identification.

Subjects are then randomly assigned to one of the experimental conditions.

We then elicit priors about changes in CO2 emissions since 2008 or 2009, depending on the experimental condition. Next, subjects will either receive (i) information about how the average person reduced CO2 emissions or (ii) how the highest-income individuals reduced CO2 emissions or (iii) no information.

Subjects can then decide how to split 150€ among themselves or Future Cleantech Architects (Green Tech Innovation) and Atmosfair (Carbon offsetting). The decision is implemented for a randomly selected subset. In addition, we measure policy support on a five-point scale.

We measure the following main behavioral mechanisms:
• Change in belief in trends (“In the next 10 years, by what percentage do you think CO₂ emissions in the EU will change compared to today?”)
• Self-efficacy (“I feel confident that I can personally take actions that can contribute to fighting global warming”) • Economic feasibility (“CO₂ emissions can be achieved without harming the economy in the longer run.”)
• Fairness norms (“Wealthy individuals in the EU are doing their fair share to reduce carbon emissions.”)
• Injunctive norms (“Most other people think someone like me should reduce their carbon emissions.”)
• Static relative descriptive norms about emission levels (Which group do you think produces higher average annual CO₂ emissions per person?, follow-up: “By how much do the absolute CO₂ emissions of people in group A differ from group B?”)
• Cross-belief updating (receiving info about the rich changes beliefs about changes among the low-income households)

Hypotheses:
H1: Subjects underestimate the CO2 reduction change, in particular those of the top 10% income earners.
H2: Correcting subjects’ misperceptions (underestimation) increases donations and policy support.
H3: Subjects respond more strongly when learning about CO2 emission changes of the top 10% earners.
H4: Causal effect of correction level: A larger correction (treatments using the 2008 reference year) leads to larger donations and stronger policy support.
H5: Main heterogeneity analysis: Donations and policy support increase with larger corrections w.r.t. prior beliefs, stronger identification with references group (income or regionally), and climate skepticism.

This experiment is one module of a larger survey fielded in the 27 EU member states. A second module on social norms and their misperceptions (injunctive and descriptive norms, regional and national) is pre-registered separately (link to be updated). Both modules share the same sample and are fielded in one questionnaire; the norms module comes first, and the assignment of conditions in the two modules is independent.

Not available

The survey company will conduct the randomization via computer. Randomization will be stratified by sub-national region.

Individual respondent.

No

Not applicable (no clustering).

The survey is fielded online across all 27 EU member states, with respondents recruited to be representative at the subnational level (NUTS regions). The regional unit is the NUTS-2 level in most countries and the NUTS-1 level in Belgium, France, and Greece (and in the single-region states Cyprus, Latvia, Luxembourg, and Malta); a few small countries constitute a single regional unit. Samples are drawn to be representative by sex, age, and income at the national level, with the aim of achieving representativeness within each region (subnational level) as well; regional number of observations are partly based on Eurobarometer Flash 539; national and regional quotas is based on EUROSTAT data. The total target sample is 60,765 respondents across 209 regional units, an average of about 291 per region. Twenty-four countries are drawn from representative national online panels; Malta, Cyprus, and Luxembourg are recruited via non-probability social-media (Niche) sampling and are flagged separately. In the social-media (Niche) samples (Malta, Cyprus, Luxembourg), the donation decision (primary outcome) is hypothetical; respondents are informed accordingly. We will use the following exclusion criteria: attention check, bot check (honeypots, adversial question, response time). To test the data collection pipeline, data quality, and study logistics, we will conduct a pilot with 600 participants in Germany. If no substantive changes to the study design are required, the pilot sample will be included in the main analysis.

About 10,128 respondents per arm (60,765 ÷ 6):

Control 2008: about 10,128
Control 2009: about 10,128
Average 2008: about 10,128
Rich 2008: about 10,128
Average 2009: about 10,128
Rich 2009: about 10,128

Pooled across reference years: Control about 20,255, Average about 20,255, Rich about 20,255.

At an alpha level of 5% and 80% power, we can detect the following effect sizes using a two-sided t-test. When comparing 20,255 vs. 40,510 subjects (e.g., for H2) the MDE is 0.031. When comparing 20,255 vs. 20,255 subjects (e.g., for H3) the MDE is 0.036.