Climate and Societal Change (CASC): the year in review

Jan Pfänder

22 October 2026

Who we are

Viktoria Cologna

Jan Pfänder

new members

Tobia Spampatti

Charlotte Kukowski

Zahra Rahmani Azad

This year

Strengthening Trust in Climate Scientists: a megastudy

Alienation and Trust in Climate Scientists

The Silicon Sample Benchmark

Where we are headed

1. Strengthening Trust in Climate Scientists: a megastudy

Research team

Jan Pfänder

Viktoria Cologna

Madalina Vlasceanu

Sara Constantino

Advisory board

Mike S. Schäfer

Mario Scharfbillig

Naomi Oreskes

Hugo Mercier

Edward Wile Maibach

James Druckman

Cameron Brick

John C. Besley

Sebastian Berger

Marlene Altenmüller

+ 42
co-authors

105

intervention proposals from an open call

interventions selected by blinded expert review

26,951

US adults, randomised to an intervention or a control text

outcomes: trust, belief, policy support, a donation, a newsletter signup

2 weeks

follow-up survey, 11 to 14 days later

105 proposals: Jan’s figure (2026-10-21); the DGPs deck says 107, the Hannover deck, the MPG abstract and the Ambizione text say 105.
N = 26,951 is n_distinct(response_id) in trust_climate_scientists/data/experiment/cleaned.rds (checked 2026-10-21). Two opt-in panels (CloudResearch, Bovitz), quotas on gender x age and gender x race. About 1,000 per intervention, 2,000 control.
13 outcomes: primary 12-item trust measure (0 to 100); secondary single-item trust and distrust, funding perceptions, policy role of scientists, institutional trust, AMS donation (lottery, 0 to 10 USD), Hayhoe newsletter signup; tertiary belief, concern, policy support, individual behaviours.
Follow-up: top-5 interventions plus control, same participants, 11 to 14 days after the experiment.
Interventions ranged from reading about climate science to a conversation with an LLM chatbot.

Half of the interventions strengthened trust in climate scientists.

But the interventions did not shift belief in climate change, policy support, or behaviour.

And two weeks later, the effects on trust had faded.

Data collection complete. Preprint and submission planned for February 2027.

2. Alienation and Trust in Climate Scientists

We tested whether structural factors make people feel alienated from science: living far from universities, a low household income, being underrepresented among scientists.

They do.

Paper with Sara Constantino, Hugo Mercier, Madalina Vlasceanu and Viktoria Cologna; pre-treatment data of the megastudy (N = 26,951, quota-matched to the census). Preregistered, correlational.
Alienation along four dimensions (spatial, social, institutional, informational), measured as perceived (survey items) and structural (distance to postsecondary institutions, household income, demographic representation among scientists).
Abstract: participants who lived further away from postsecondary institutions perceived greater spatial alienation, participants with lower household income greater social alienation, and participants underrepresented among scientists (gender, race) in some cases greater institutional and social alienation.
Greater perceived alienation was in turn associated with lower trust (analysis/results/rq2.rds, standardized estimates: spatial -0.45, informational -0.28, social -0.17, institutional +0.09), within Republicans, Independents and Democrats alike.

And people who feel alienated from science trust climate scientists less.

A lack of trust need not reflect individual irrationality. It may, in part, reflect structural exclusion from science.

Preprint on 1 November, journal submission end of November.

3. The Silicon Sample Benchmark

Core team

Jan Pfänder
Eawag

Yara Kyrychenko
Cambridge

Dirk U. Wulff
MPI for Human Development

Sara Constantino
Stanford

Jon Roozenbeek
Cambridge & VU Amsterdam

Viktoria Cologna
Eawag

Miriam Remshard
Cambridge

+ 56
co-authors

expert teams

AI-based approaches, time-stamped before the data were revealed

21,600

human participants to predict

16 × 13

interventions × outcomes: 208 treatment effects

tiers: full silicon sample, cell means, effect forecasts

Predictions correlate with the human results, with wide variation across teams.

Silicon samples can predict reasonably well which attitudes an intervention moves, but not how large the effect will be, or how answers will be distributed.

Preprint and public online event in December, submission in January.

4. Where we are headed

Directions

From designing single messages to sustaining exposure to science: conversations with AI, science content in people’s information environments
Translating what works into practice with stakeholders