Public Acceptance of International Redistribution in High-Income Countries

There are various independent treatments with several branches in each survey. The survey design is best described in the survey flows (cf. analysis plan).
The questionnaire is uploaded to the repository. The survey’s structure will be as depicted in the survey flow, where underneath rectangles represent random branches. When relevant, the survey randomizes the order of response items. The questionnaires are almost identical across countries. Deviations from the main questionnaire are the following:
- Some questions are not asked in certain countries where they are not relevant (e.g. vote is not asked in Russia nor Saudi Arabia; support for slavery/colonization reparations is only asked in former colonial or slave states: France, UK, Germany, Spain, Italy, U.S.).
- Some questions are asked differently in Russia as we cannot use the software Qualtrics as in the other countries. In particular, questions involving sliders are not asked.
- In many questions, figures (e.g. income brackets or tax revenues) and specificities of the questions (e.g. the country name or list of countries covered by a policy) are adapted to the country. Country-specific elements (including quotas) are specified in the file sources.xlsx.

2025-04-15

2025-10-09

Variables
Manipulated variables
Attached is the provisional codebook. It contains the variable names, labels and levels.
Treatment variables start with "variant_".
There will be some small adjustments between these codebooks and the final datasets (more variables, and variables more exhaustively labelled).

Measured variables
The measured variables are best described in the questionnaires, attached in Qualtrics original format and Word exported format.
Country-specific values are inserted automatically (by Javascript) from sources.xlsx (sheets features, Income, and Policies) and replace corresponding HTML tags.

Indices
We will define the variable share_solidarity_supported as the share of realistic policies supported (among 10).
We may define indices by taking the z-score of the explanatory factor of the index variables. For example, we may do that for the variables of support.

Hypotheses
H1: Global redistribution policies help a political program to be preferred
The conjoint analysis asks respondents their preference between two political programs (they can also express indifference), with policies (or absence of policy) taken at random in each of five policy domains.

H1a: No positive effect of “Cut development aid” in the conjoint analysis.
H1b: Positive effect of “International tax on millionaires with 30% financing healthcare and education in low-income countries” in the conjoint analysis.


H2: Global issues are given substantial priority when allocating a budget
I ask respondents for their preferred allocation among five categories (there are two random branches: one where the five categories are fixed, one where they are taken at random out of 13 categories). I expect global issues to remain at a similar level of priority as previously found in Fabre et al. (2025), substantial though lower than most national issues. Namely, I expect an average allocation at two thirds of the average amount, that is 13.33% (66.66% of 20%).

H2: Allocation of at least 13.33% of the budget to each global issue.


H3: Global redistributive policies are supported by a majority
I test the support for various global redistributive policies, from realistic ones to radical ones.
When the questions are preceded by a treatment, the majority support is generally tested on the control group. I will also test an alternative specification where the control group is extended to treatment branches with no treatment effect (at a conservative 20% threshold).

H3a1, …, H3a10: Majority support for each realistic globally redistributive policy.
H3b: Majority support for the Global Climate Scheme.
H3c1, …, H3c9: Majority support for a global wealth tax, for a NCQG at least as high as the current amount ($26 billion in grant-equivalent), for a sustainable future, for both version a radical tax (on the global top 1% or top 3%), for global convergence in GDP per capita, for a global sustainability movement, for “my taxes should solve global problems”.


H4: Lower support but still majority support for international policies when fewer countries participate
I expect support to significantly decrease with partial (rather than universal) participation, though to remain in the relative majority.

H4a1, …, H4a6: Majority support for an international (non-global) wealth tax (two participation scenarios) and for an international climate scheme (four participation scenarios).
H4b1, …, H4b6: Lower support for an international policy under partial participation.


H5: Absence of warm glow
To test whether people support global redistribution only for as long as its implementation seems unlikely (H5a), I randomly treat respondents with information about ongoing negotiations on globally redistributive policies. Following the information, I expect their belief that global redistribution is likely to be higher. Warm glow would be identified if the number of realistic global redistribution policies supported were then lower. I will test the effect of information on support both directly and through a two-stage least squares model, with the belief in likelihood of global redistribution as the endogenous variable.
To test the effect of moral substitute (H5b), before the question on the support for the Global Climate Scheme, I randomly ask (or not) to the respondents how much they would like to donate to a reforestation charity, should they win the $100 lottery they are enrolled in. If the donation acts as a moral substitute, support should be lower for those who are offered this treatment.

H5a: Information increases the belief in the likelihood of global redistribution but does not reduce support for realistic global redistribution policies.
H5b: Being offered the possibility to donate to a reforestation charity (in case of lottery win) does not reduce support for the Global Climate Scheme.

Randomization
We use stratified randomization using Qualtrics.

Randomization is done at this individual level.

No

11 countries (each country corresponds to a cluster).

The intended sample sizes are as follows: * U.S.: 3,000 * Japan: 2,000 * Russia: 1,000 (We are still unsure that we will be able to run a survey in Russia.) * Saudi Arabia: 1,000 * Europe: 5,000, including: - France: 798 - Germany: 1048 - Italy: 756 - Poland: 500 - Spain: 603 - Switzerland: 469 - UK: 826

All treatments are independent and uniformly distributed, i.e. the sample is split in branches of equal sizes (in expectation).

Sample size rationale As I hypothesize that answers will be homogenous across European countries, to get better power in regressions, I will consider Europe as a block in the statistical analyses, weighing each country by its population. Therefore, our smallest samples will be Saudi Arabia and Russia (n=1,000) while our samples will be large in Europe (n=5,000) and the U.S. (n=3,000). In experiments where we will split the sample into two (H5a), three (H4b, H5b) or four (H1, H4a) random sub-samples of equal size. The minimum detectable effects (at 5% significance level and power 80%) are given by the following R code (cf. also https://www.evanmiller.org/ab-testing/sample-size.html): # MDE in a regression between 2 dummies. n is the total sample size (assuming 2 subsamples), y=50% is the average outcome between subsamples # Note that these calculations are conservative as they assume a two-sided test, while the tests can generally be one-sided as the hypothesis is directional. mde <- function(n, alpha = .05, beta = .8, y = .5) return((qnorm(1 - alpha/2) + qnorm(beta))*sqrt(4*y*(1-y))/sqrt(n)) mde(12000) # .025 mde(6000) # .036 - Worst MDE for all countries taken jointly mde(3000) # .05 mde(2500) # .06 - Worst MDE for Europe mde(2000) # .06 mde(1500) # .07 - Worst MDE for the U.S. mde(1000) # .09 - Worst MDE for Japan mde(666) # .11 mde(500) # .125 - Worst MDE for Saudi Arabia, Russia mde(333) # .15 mde(250) # .18 - Worst MDE for Poland, Switzerland This shows that our worst MDEs are 0.125 (treatments with 4 branches in Saudi Arabia and Russia where n=1,000), while the MDEs are 6-7% for our main analyses in the U.S. and Europe. This is a good compromise between precision and cost. Furthermore, for stated support results (H3), we will be able to detect significantly majority support at the country level as soon as support is above 55.8% in our smallest sample (Switzerland) or above 52.3% in our largest one (the U.S.), which is precise enough (cf. https://www.stat.ubc.ca/~rollin/stats/ssize/n1a.html).

Pre-registration plan.pdf

Using an original survey of 12,000 respondents representative of eleven high-income countries, I examine public support for global redistribution and climate policies. Although global inequality is not a salient concern, political programs that address it are more likely to be preferred. In every country, majorities accept nearly all global policies tested, including those that would redistribute 5 percent of global income or entail personal costs for respondents. Survey experiments demonstrate the robustness of support. In particular, an information treatment shows that support for global policies causally increases among respondents who perceive them as likely; an effect opposite to warm glow.

Public Acceptance of International Redistribution in High-Income Countries, Adrien Fabre, 2025.

https://prod.wid.world/www-site/uploads/2025/11/WorldInequalityLab_WP2025_26_-Public-International-Redistribution-in-High-Income-Countries_Final.pdf