Science by Consensus: Eliciting Citizens’ Research Funding Priorities

2025-05-22

2026-05-22

We will include the following outcome variables in the sample of citizens (C), policymakers (P), and experts (E).

1. Allocation to R&D: Continuous measure of the desired share of the total government spending on R&D (C, E, P)
2. 1(Allocation to R&D > current): Binary measure that takes a value of 1 if a participant’s desired share allocated to R&D is greater than the current allocation (3%) and 0 otherwise (C,E,P)
3. Social returns to R&D: Continuous measure of the beliefs about the social returns to R&D (C,E,P)

In specifications involving Citizens’ desired R&D spending, we will use quantile regressions to compare the change in the median view.

02/06/2025 Update: This version of the pre-registration was registered before collecting data from the Policymakers and Citizens samples. We have made two changes in this version. First, in the Policymakers' sample, we no longer plan to invite non-US policymakers to the study. Second, for Citizens, we plan to modify the treatments within one of the treatment dimensions. These changes are reflected below.

We conduct our study using samples of U.S. Citizens, Policymakers, and “Experts.”

For the Citizens’ sample, we will recruit 2,000 U.S. respondents using the data collection provider Bilendi. We will include quotas to ensure that our sample is representative of the U.S. population based on age, gender, income, education, and region. We exclude participants from the study when the demographic quotas have been met. We also exclude participants who fail our attention checks. Among those who completed the survey, we will exclude those who attempted to take the survey multiple times.

Policymakers are recruited via the Policymakers Lab at the University of Warwick, which hosts a panel of 200 policymakers working primarily in central governments around the world, and through our professional networks. All 67 U.S. policymakers part of the lab will be invited to participate in our study. In addition, we plan to reach out to about 50 Policymakers through our networks and invite them to participate in our survey. We expect a completion rate of 30-50%.

Finally, our sample of Experts consists of member heads at the various programs at the National Science Foundation (NSF), members and non-resident fellows at the Institute for Progress, economists who have participated in the NBER Science of Science Funding Summer Institutes from 2019 to 2025, and economists who have participated in the NBER Innovation Summer Institutes from 2014 to 2025. These Experts are invited via email to participate in our study. We aim to invite about 750 experts and expect about a third of them to complete the survey. In case we find difficulties with recruitment, we will reach out to editors at Nature and Science and experts at various funding agencies, such as the European Research Council (ERC), UK Research and Innovation (UKRI), Dutch Research Council (NWO), and Swiss National Science Foundation (SNSF).

Citizens’ Study

We measure Citizens' desired R&D spending by asking them to indicate the percentage of the total government spending that should be allocated to R&D. We randomize participants into various treatments that vary the available information and the nature of the decisions. In the first treatment dimension, we vary whether respondents learn information about government spending on other categories (e.g., defense, social security). The second treatment dimension varies whether respondents are asked to allocate the government budget only to R&D or to several budget functions, including R&D. Prior to eliciting Citizens’ desired R&D spending, we elicit their prior beliefs about R&D spending. In treatments where Citizens are only asked about their desired R&D spending, we only elicit Citizens' perceived R&D spending. In treatments where Citizens are asked about their desired spending for several categories, we elicit their perceptions of spending for these categories.

To understand the mechanisms driving Citizens’ desired R&D spending and distribution, we include both open-ended and closed-ended questions. The close-ended questions capture various mechanisms revolving around the social returns to R&D, the distribution of R&D, and beneficiaries of the returns from R&D. We also include questions on sociodemographics, such as region of residence, age, and education.

We elicit Citizens’ desired R&D spending under counterfactual scenarios, which vary the returns to R&D, spending of the R&D budget, and the beneficiaries of the returns to R&D.

Finally, we ask Citizens about their perceived and desired allocation of the R&D budget to various domains, such as Health, Defense, and Energy.

Experts’ Study

We elicit Experts’ desired R&D spending and compare their responses to those of citizens. We also include other questions, such as beliefs about the returns to R&D and beliefs about the median Citizen’s desired R&D spending.

Policymakers’ Study

We expect to find differences between Citizens’ desired R&D spending and the actual R&D spending. To further understand the reasons for these differences, we conduct a study with Policymakers. We will include open-ended and close-ended questions to understand Policymakers’ views about the bottlenecks involved in implementing Citizens’ desired R&D preferences in policies. Some of these close-ended questions include desired R&D spending, beliefs about the returns to R&D, and beliefs about the median Citizen’s desired R&D spending.

Software-based randomization (Qualtrics)

Individual

No

Our final sample will include 2000 Citizens recruited from Bilendi. We plan to invite about 117 Policymakers via the Policymakers Lab and through our professional networks to participate in the survey, and expect a completion rate of 30-50%. Finally, we plan to invite about 750 Experts to participate in the survey, and expect about a third to complete the survey.

Same as the above, since the design is not clustered.

As explained above, Citizens are randomized into treatments in a 2x2 between-subjects design. The sample of 2000 Citizens would be randomized into one of these four cells in equal proportions. There are no randomizations in the sample of Policymakers and Experts.