Intervention (Hidden)
Survey takers will answer 10 questions about income transfers to US families. Choice sets differ in terms of the income of the families and cost of transfer, both of which will be randomized across questions and across recipients.
This means that people are exogenously exposed to varying levels of income and transfer. This random treatment variation allows us to estimate people's elasticity redistributive preferences with respect to these two variables.
We are attaching the full survey as a document to this pre-registration.
Update June 27 2025: We have now collected the data and are planning to run a second wave that tests whether redistributive preferences and beliefs about efficiency effects of taxes are malleable by information treatments. Specifically, our Wave 2 Survey will look just like Wave 1 but randomize the following information treatments:
1. Information about the average Elasticity of Taxable Income (ETI) that we found in our Meta Analysis.
2. Information about one selected ETI that is particularly high. We do not state that this ETI is high relative to the literature but we simply tell subjects that one of the estimates we found is equal to this value.
3. Information about the ETI of a worker earning $40,000 annually versus the ETI of a worker earning $100,000.
4. Information about the redistributive preferences that of other American households that we surveyed in Wave 1.
Subjects will be randomized into treatments 1-3 and a control group that receives no information on the ETI. Then we cross-randomize treatment 4 that provides information about redistributive preferences. Our interventions are designed to study how beliefs about i) the ETI and ii) other people's redistributive preferences shape our estimates of social marginal welfare weights.
Update – September 19, 2025
After completing the second experimental wave, we now aim to examine whether the optimal income tax schedule estimated from the data aligns with people’s stated preferences. We also seek to test whether distrust in the government contributes to a gap between individuals’ preferences for taxation and their behavior in our experiment. In addition, we want to understand whether framing transfer choices as budget-neutral influences redistributive preferences.
In our third survey wave, we will modify the transfer choice set. While it remains similar to the one used in earlier waves, it will now include the possibility of “taking away” funds from taxed families whose contributions finance transfers. Each family will receive a standard bonus endowment that is strictly larger than any taxes owed, so no family will ever end up owing money to the researcher.
We will also introduce a question on participants’ hypothetical income tax schedule, illustrated as net payments to the government by families with different incomes. Finally, we will measure people’s trust in government, specifically whether they believe tax revenues are ultimately used for transfers or risk being lost in the bureaucratic process. We will then use this measure as a predictor of lower support for income taxation.
Update - January 14, 2026
In wave 3 of this experiment we found a substantial degree of "government aversion": Strikingly, many people explicitly prefer giving less money directly to poor recipients over choosing options that deliver more to the poor but also more to the Treasury. We explore the mechanism of this behavior in wave 4. Specifically, we are interested whether 1) subjects believe that the government wastes money, e.g. by overpaying bearucrats and politicians and 2) that the money not donated to the government and families is money saved to the researcher's budget (i.e., us), which then may have other alternative productive use. To speak to these mechanisms, we have three main treatments and a control group. Different from the prior waves, all group will now make hypothetical rather than consequential transfers to families and the government, because some of the treatments would be impossible to implement in an incentivized way. Below we list the treatments:
1) Government is perfectly efficient: In this treatment we ask people to imagine the government would be perfectly efficient and we also highlight the public goods that tax revenue typically finances.
2) No alternative use of research budget: In this treatment we ask people to imagine that any money they save to the researcher (by choosing lower transfers) would not have any productive use but instead would disappear.
3) Highlighting alternative uses: In this treatment, we explicitly highlight to people that if they choose lower transfers and taxes that they save funds for our research budget which has alternative productive use for future research projects.
4) Control group: No informationt treatment. All else equal to the other groups in terms of transfer decisions.
Finally, we also make a small adjustment to our stated preference elicitation of the optimal tax schedule that we elicited in the previous wave. Previously we did not explicitly state whether the income taxes were federal taxes or also state and local taxes. Since we are interested in total tax burden from income taxes, we now ask subjects explicitly for the sum of these three income taxes (federal, state, and local). This is not randomized but serves just as an adjustment to our previous waves. We also add various questions on attitudes toward the government.
Update – July 11, 2026
In wave 5, we address two questions left open by the previous waves: first, how robust the "government aversion" documented in waves 3 and 4 is to the design of the choice environment, and second, whether the value people attach to government funds can be measured directly, without the structural assumptions of our baseline model. We aim for 2,000 respondents, randomized into three separate arms in proportions 5:4:6.
Arm 1 (consequential, approx. n = 670). Participants make 15 choices between two transfers to the same U.S. family: option A is $1,000 tax free; option B is $2,000 taxed at rate r, where the taxed amount is donated by us to the U.S. government through the Treasury's official Pay.gov portal. We vary the family's income ($20,000, $100,000, $500,000) and the tax rate (25, 45, 52, 65, 80 percent) within subject. As in the earlier consequential waves, one randomly selected choice of one randomly selected participant is implemented, with the recipient family drawn from the Prolific pool and matched on income. The design identifies, without functional-form assumptions, the rate at which participants trade off dollars to a family against dollars to the government. Choosing B at rates above 50 percent reveals a positive willingness to pay for government revenue (at r = 52 percent, the price is 26 dollars of government revenue per dollar of family income forgone), while choosing A at rates below 50 percent reveals a willingness to pay to keep funds away from the government, since option B then gives more to both the family and the government. Choice shares across the rate grid trace the distribution of the value of government funds relative to family transfers, including its negative range.
Arm 2 (with potential policy consequentiality, approx. n = 530). Participants choose one of three complete after-tax income schedules, displayed as tables of after-tax income at eight income levels: an approximation of the current U.S. tax and transfer system, the optimal schedule implied by revealed preferences from our earlier waves, and the optimal schedule adjusted for the estimated degree of government aversion. Column order is randomized. We cross-randomize an information treatment: half of the participants see the current U.S. schedule explicitly labeled as such, while the control group sees unlabeled columns. This tests whether people prefer the aversion-adjusted optimum over the status quo and whether identifying the status quo shifts choices. After the policy choice, participants state their preferred average tax rates using the slider instrument from waves 3 and 4. Choices are consequential in that they will be compiled into an anonymized, aggregate report sent to the House Committee on Ways and Means and the Senate Committee on Finance, which participants are told upfront.
Arm 3 (hypothetical, approx. n = 800). Participants make 10 pairwise transfer choices as in the previous waves, with the same income randomization ($5,000 to $500,000); we additionally include zero-income families with 5 percent probability, described as having no employed adult or adults unable to work. We cross-randomize four between-subject factors: stakes (all amounts multiplied by 1 or 10), horizon (a one-time payment versus the same amounts paid once a year, every year, permanently, like a permanent tax policy), framing (transfers given to a family versus payments due to a family from which the government takes a share as tax), and recipient (a single family versus a group of 100 families sharing the amounts). This arm tests whether the tax-wedge coefficient underlying our government-aversion estimate is robust to stakes, permanence, give-versus-take framing, and the number of beneficiaries. The primary regression is the one specified in the January 14, 2026 update, run by factor cell and with factor interactions; standard errors clustered by respondent.
All arms conclude with belief questions (e.g., what the government would do with money received through Pay.gov, whether donating to the government resembles paying federal taxes), the government-attitude battery from wave 4, and demographics. Randomization is done by computer when the survey loads; arm assignment is at the respondent level and choice sets vary at the respondent-question level.