Experimental Design
This is a pre-registered information-provision experiment with an embedded conjoint policy-choice task, fielded on a sample of US adults recruited online.
Respondents first report their prior belief about the share of total annual US greenhouse-gas emissions generated by the richest 1%, on a 0–100 slider, followed by a confidence item. They are then randomly assigned, with equal probability, to one of three arms: a pure control, which receives no statistic, and two treatment arms, each of which receives one accurate statistic on the top-1% emission share drawn from a published, peer-reviewed study, together with its source. The two treatments differ in the accounting frame used to attribute emissions to households: a consumption-based frame (Starr et al., Ecological Economics, 2023) and an ownership-based frame (Chancel & Rehm, Nature Climate Change, 2026). Treatment texts are harmonised across arms in length, structure and register: each names the source, states the attribution rule in one sentence, and gives a concrete household-level example.
A third treatment arm, using an income-based accounting frame (Starr et al., PLOS Climate, 2023), is specified in the pre-analysis plan but is not fielded in this wave, as its implementation depends on funding beyond the committed base budget. It will be added in a subsequent wave if that funding is secured.
All respondents then complete a conjoint policy-choice task. They see eight pairs of federal carbon-tax packages built from randomised attributes — annual household cost, use of revenue, and whether an additional tax applies to the richest households — and select their preferred package or the status quo of no carbon tax. Choice is the primary dependent variable.
After the conjoint, all respondents report a posterior belief. To limit anchoring, the posterior is framed as the share attributable to the remaining 99% rather than to the top 1%. Secondary outcomes include a policy-builder task in which respondents assemble their own preferred carbon tax, two Likert items on general support for a federal carbon tax with and without an additional tax on the richest households, and an incentivised allocation of a $50 lottery prize between the respondent and two charities.
Within the control arm, the timing of prior-belief elicitation is itself randomised: priors are elicited either before or after the conjoint task. This allows us to assess whether eliciting priors primes attention to inequality independently of any informational content.