Preferences over the Timing of Redistribution Policies

In the project we study whether subjects exhibit differences in their willingness to redistribute between two workers depending on when the workers would be told about the policy and when the policy would be implemented relative to the work event.

To understand the impact of timing of information and intervention in redistribution on preferences for redistribution, we adapt the current experimental methodology in the fairness literature to allow for differences in timing. In the primary portion of the experiment, we collect subjects’ (henceforth called spectators) redistribution decisions that will truthfully be carried out on future subjects, called workers. We will randomly vary the timing of when the workers will be told about the redistribution and when the redistribution will occur relative to the work event. In other words, we have two dimensions, each with two treatments. The first dimension is the timing of redistribution information going to the worker:
i) Pre-work Info: Workers will be told of the spectators’ decision to redistribute before they work.
ii) Post-work Info: Workers will be told of the spectators’ decision to redistribute only after they work.
The second dimension is the timing of the redistribution, relative to the work event:
i) Pre-work Intervention: Workers will never find out what their earnings would have been without redistribution.

ii) Post-work Intervention: Workers find out what their earnings would have been without redistribution.

The spectators thus face one of 4 treatments – pre-work info/pre-work intervention; pre-work info/post-work intervention; post-work info/pre-work intervention; and post-work info/post-work intervention – which vary the information provided to spectators about when the workers will learn about the redistribution (info dimension) and whether they will learn about what they would have received without redistribution (intervention dimension).

2025-06-09

2025-11-30

We collect the following primary outcome:
- Willingness to redistribute: This is defined as the spectators willingness-to-destroy (WTD) for redistribution.

Given the literature on post-work spectator redistribution (Bartling et al., 2018), the effect of recipients’ reference points on spectator redistribution (Charité et al., 2022), and loss aversion for others (Füllbrunn and Luhan, 2015), we construct the following hypotheses:

Hypothesis 1: spectators’ WTD is greater when workers will be informed of redistribution before working.
Hypothesis 2: spectators’ WTD is greater when the intervention will be implemented before the work event.

For Hypothesis 1, we acknowledge that there is the potential for a competing hypothesis, that the spectators’ WTD will be lower when workers will be informed of the redistribution before working. This is based on the fact that reducing the inequity in the payments may reduce worker productivity (de Bresser and Knoef, 2021). If spectators recognize this (as suggested by Andre (2024)), and value this productivity they may be less likely to redistribute if workers would be informed before the work event.

We will also consider the interaction of our two treatments.

In our primary analysis, we will remove individuals who have multiple switch points. In robustness analysis we will include them, testing both using the first and last switch point.

We will collect information about spectators’ perceptions of workers’ use of AI on the task and how that affected their choice of redistribution. In some analysis we will analyze the data overall controlling for this variable; furthermore, we will run heterogeneity analysis where we are powered to do so.


References
Andre, Peter. "Shallow meritocracy." Review of Economic Studies (2024): rdae040.
Bartling, Björn, Alexander W Cappelen, Mathias Ekström, Erik Sorensen, and Bertil Tungodden. 2018. “Fairness in Winner-Take-All Markets” NHH Dept. of Economics Discussion Paper No. 08/2018.
Charité, J., Fisman, R., Kuziemko, I., Zhang, K., 2022. Reference points and redistributive preferences: Experimental evidence. J. Public Econ. 216, 104761.
De Bresser, Jochem, and Marike Knoef. 2021. “Preferences for Income Redistribution: A New Survey Item and Experimental Evidence”. CentER Discussion Paper Nr. 2021-035.
Füllbrunn, Sascha C., and Wolfgang J. Luhan. 2017. “Decision Making for Others: The Case of Loss Aversion.” Economics Letters 161 (December): 154–56.

We additionally will measure the following spectator outcomes:
- Free Redistribution Rate: the fraction of spectators that choose to redistribute at zero cost
- Anticipated Perceived Fairness: what spectators anticipate is perceived as fair by workers under different redistribution amounts and starting point values
- Anticipated Emotion: how spectators anticipate workers will feel under different redistribution amounts and starting point values
- Anticipated Effort: the level of effort from workers anticipated by the spectator about under different redistribution amounts and starting point values

We additionally will measure the following worker outcomes:
- Production: the count of tasks completed by the workers
- Perceived Fairness: workers’ perception of fairness of the outcome
- Emotion: workers’ self-reported emotion at the end of the study
- Anticipated Perceived Fairness: what workers’ anticipate they would perceive as fair under different redistribution amounts and starting point values
- Anticipated Emotion: how workers anticipate they would feel under different redistribution amounts and starting point values
- Anticipated Effort: the level of effort workers anticipate they would exert under different redistribution amounts and starting point values

See above

Our design aims to measure the impact of timing of information and intervention on spectators’ preferences for redistribution.

The design consists of two stages.

In stage 1, we will recruit subjects from Prolific to act as “spectators”. These spectators will be informed about the workers’ task that will occur in stage 2. The task, counting zeros, will be described and the number of correct answers to the counting zeros task will be called points. They will be told that workers will be paired, and when they work on their task they will receive a bonus (in addition to base payment and completion fee) based on who got more points. For the purposes of this document, we will call the worker who got more points the “winner” and the other worker the “loser” (in the case of ties, random chance will determine which worker is the winner) – however, in the experiment this language will not be used. The bonus of the winner will be $0.20 for each point earned by both themselves and their partner. The loser will receive no bonus, just their entry fee and completion fee.

We will then tell the spectators that they have the opportunity to intervene in the payment procedure for the workers. If the spectator chose to redistribute, the winner’s bonus would be $0.20 for each point the winner earned, and the loser’s bonus would be $0.20 for each point the loser earned. The spectators will know that they will have a 1-in-4 chance of being matched with a worker pair and their choice being implemented for that worker pair.

Spectators’ decisions for redistribution will be made using a BDM methodology. There will be multiple rows with two options, A and B, and spectators have to choose between A and B for each row. Option B will always be to not redistribute. Option A will be to redistribute at a cost to the workers, ranging from nothing being taken away from the workers up to the entire base payment ($1) being taken away from the workers.

Spectators will be randomized into one of 4 treatments, based on what they are truthfully told will be the timing of when the workers would be informed of the redistribution and when the redistribution would occur relative to the work event:
● Pre-Work Info, Pre-Work Intervention: workers will be informed of the redistribution and the cost of redistribution prior to the work event, and they will never find out which worker was the winner and how much they would have earned without the intervention
● Pre-Work Info, Post-Work Intervention: workers will be informed of the redistribution and the cost of redistribution prior to the work event, but they will find out after the work event which worker was the winner and how much they would have earned without the intervention
● Post-Work Info, Pre-Work Intervention: workers will only be informed of the redistribution and the cost of redistribution after the work event, but they will never find out which worker was the winner and how much they would have earned without the intervention
● Post-Work Info, Post-Work Intervention: workers will only be informed of the redistribution and the cost of redistribution after the work event and they will find out after the work event which worker was the winner and how much they would have earned without the intervention

Spectators know that, in the case there is no redistribution, workers will never find out that redistribution was a possibility.

After the redistribution choice, the spectators will be asked a series of hypothetical questions about how workers would feel and behave under different redistribution costs and initial inequalities. Each spectator will be asked these questions for the conditions of their own treatment about either a high or low initial inequality (randomly assigned) and for three situations: no redistribution, redistribution at no cost, and redistribution at either a low or high cost (randomly assigned).

Spectators will be asked comprehension questions throughout the study. They will not be able to continue with the study until they correctly answer the comprehension questions. At the end of the study, we will ask the spectators a series of demographic questions as well as questions for what they believe the study is about and whether/how they thought about workers’ use of Gen AI (like ChatGPT) in the counting zeros task when making their redistribution choice.

In Stage 2, we will recruit subjects from mTurk or Prolific to act as our workers. They will find out about the work task and about the standard payment structure. Each worker pair will be assigned to one of the spectators, with each spectator having a 1-in-4 chance of being assigned to a worker pair within their treatment, which will determine whether redistribution occurs, at what cost, and the timing of the information. The matched spectators’ choices will be implemented, and the workers will do the work task. Workers will respond to a survey about their perceptions of the redistribution related to fairness, both in their own condition and in other hypothetical conditions similar to those used in the spectator survey. We will also implement comprehension questions and ask demographic questions after the work task.

Randomization for spectators will be done by a computer. Allocation of workers into treatments is based on which of the 4 identical survey invitations they receive/respond to – this is essentially random. Randomization for workers into pairs will be based on when they enter the study within a treatment. The random assignment of worker pairs to spectators will be done by a computer.

For both spectators and workers, we will constrain the sample’s gender balance to be half (self-reported) men and half (self-reported) not-men (i.e. women and any other gender).

For both spectators and workers, we will implement sample restrictions for respondent quality appropriate to the platform on which that stage is run.

In stage 1, the randomization unit is the spectator. In stage 2, the randomization unit is the worker pair.

No

For stage 1, we plan to have 1600 spectators. For stage 2, we plan to have 400 worker pairs.

For stage 1, we plan to have 1,600 spectators. For stage 2, we plan to have 800 workers.

We plan to have 400 spectators and 100 worker pairs for each treatment.

Based on Bartling et al. (2018), comparing the treatments WTA-No Choice with WTA-No Expectation, spectators on average do not distribute 48% vs 33% of the time. In order to detect a significant difference at 5% level of significance with 80% power we need a minimum sample size of 334 for subjects in treatments with pre and post-work info conditional on the work intervention treatment. Based on Charite et al. (2022), comparing the treatment with the control, spectators on average do not distribute 25% vs 12% of the time. In order to detect a significant difference at 5% level of significance with 80% power we need a minimum sample size of 278 for subjects in treatments with pre and post-work intervention conditional on the work info treatment. Given that our distribution hypothesis is conditional on willingness to pay, to account for that in the regression we have decided to have 400 per treatment, for a total of 1,600 spectators. For workers, each pair of workers will be matched with 4 spectators, and one of their decisions will be implemented randomly, therefore, we will need 800 workers (400 pairs). References Bartling, Björn and Cappelen, Alexander W. and Ekström, Mathias and Sorensen, Erik and Tungodden, Bertil, Fairness in Winner-Take-All Markets. (April 30, 2018). NHH Dept. of Economics Discussion Paper No. 08/2018 Charité, J., Fisman, R., Kuziemko, I., Zhang, K., 2022. Reference points and redistributive preferences: Experimental evidence. J. Public Econ. 216, 104761.