How regional predictions of heat events in the coming summer and their forward-attribution to climate change affect support for adaptation and mitigation measures as well as attribution of guilt across generations. (third wave)

All participants receive graphical risk assessments on the likelihood of heat waves in their region (100km x 100km grid) in the summer (1st June to 31st August 2025). The treatment group in addition receives similar information for the previous summer.
Before the information is provided, all participants see a tutorial illustrating what the numbers and their graphical representation mean.

Heat waves are captured by two measures: the number of hot days (maximum above 30°C) and tropical nights (minimum above 20°C).
The risk assessment for each measure is done by presenting a graphical representation of the median, the 75th percentile and the 90th percentile for the number of events in a 30 member ensemble of seasonal forecasts taken from the operational seasonal forecast system GCFS2.0 by the German Weather Service (DWD).

2025-06-02

2025-06-13

We elicit a set of revealed preference outcomes on willingness to adapt or support adaptation and mitigation policies.

Revealed preference outcomes:
The first three are elicited using lotteries with one winner per 500 participants.
Willingness-to-pay for an air fan suitable for use in bedrooms. (medium-run selfish adaptation)
Willingness-to-pay for a 100€ voucher at an ice-cream vendor of the participant's choice. (short-run selfish adaptation)
Willingness-to-donate to plant a tree in a major German city. Participants choose how much of a 100€ prize they would donate. (long-run altruistic adaptation)
Vote among all members of an experimental condition (Control/Treatment group) to either receive 70ct each or retirement of 10 allowances from the European Union Emission Trading System using the 'buy, bank, burn' mechanism proposed by Gerlagh & Heijmans (2019, Nature Climate Change) and implemented by ForTomorrow gGmbH. (mitigation)

We elicit expectation for the number of heat events in the summer 2026, stated-preference support for adaptation measures, trust in and emotional responses to the risk assessments, trust in long-run climate predictions as well as open-ended responses to how they will use this information as well as which information participants would regard to be most useful.

Measure of trust in seasonal prediction/risk assessment. (5-point Likert scale)
Support for five adaptation measures elicited on 5-point Likert scales
a) mandatory insurance against natural hazards for all landlords
b) more space for and planting of urban trees to reduce heat in cities
c) Subsidies for the installation of air conditioning in private homes and offices
d) provision of reuasable water bottles for homeless people free of charge
f) Heat-warning app that also provides advice on how to protect oneself

There is exogenous between-subject variation in the information received.

Randomly assigned experimental conditions (Control/Treatment) vary in whether only the risk assessment for heat waves in the coming summer is given (control) or whether this is supplemented by the same information for the past summer (treatment). The latter serves as a reference point and increases the salience of how the coming summer might differ from the last one experienced.

Geographical variation is induced by providing participants with regional forecasts based on 100km x 100km grid cells. They are assigned to the relevant grid cell based on the post codes they provide. Grid cells differ in up to 12 dimensions, i.e. the 2 heat wave measures (hot days / tropical nights) represented by three values (median, 75th percentile and 90th percentile) each as well as the same for the previous summer (2 x 2 x 3 = 12).

H1 (anticipated heat):
For the entire sample, support for adaptation and mitigation measures increases in the predicted likelihood of the extent of heat waves in the summer 2025.
a) measured by the sum of the median, 75th percentile and 90th percentile for hot days and tropical nights
b) for the six individual data points provided to all participants

H2 (reference point):
For the treatment group, support for adaptation and mitigation measures increases in the predicted likelihood of the extent of heat waves in the summer 2025 relative to those predicted in 2024.
a) measured by the sum of the differences between the 2025 and 2024 versions of the median, 75th percentile and 90th percentile for hot days and tropical nights
b) for the six individual differences between the 2025 and 2024 versions of the variables

H3 (salience):
Using the same variable(s) as in H2 for the entire sample, the impact of these variable(s) is stronger for the treatment group than the control group.

randomization done in office by a computer

individual

No

two:
about 2,000 participants that have already completed waves 1 and 2
about 1,000 new participants

3,000 participants from the Bilendi online panel The newly recruited ones reflect the general adult population in Germany w.r.t. age, gender, and share of high school diploma (Abitur). The was true for those recruited for wave 1 of the study, but attrition over the 12 month since we started wave 1 might have affected group composition.

1,000 each among those that have completed waves 1 and 2
500 each among the newly recruited

All power calculations are based on standard two-sided tests at α = 0.05 and 80% power. We draw on data from waves 1 and 2 for those primary outcome variables that we have used before. For the ice cream voucher, we do not have any prior data and hence do not conduct an ex-ante power analysis for this outcome variable. • Vote for mitigation (binary): In wave 1, around 45% of participants opted for the mitigation policy. Based on this baseline, detecting a treatment effect of +5 percentage points (from 45% to 50%) at α = 0.05 and 80% power requires about 3,130 observations in total. Our planned sample size of 3,000 (1,500 per treatment) should be therefore nearly sufficient for this effect size. • Air fan choice (binary): Approximately 23% of participants chose the air fan in wave 1. To detect a +5 pp effect (from 23% to 28%) at 80% power and 5% significance, about 978 observations are needed, which is well within our planned sample. • Donation to plant trees: In wave 2, the average donation ranged between 36€ and 37€, with a standard deviation of roughly 28€. Based on this, detecting a mean difference of 3€ requires a sample of around 1,845 participants. With more than 3,000 planned observations, we should be well covered here.