Preferences for Redistribution and Social Mobility: The Role of Motivated Beliefs.

We study two types of motivated beliefs about social mobility that relate to preferences for redistribution. For one, people might be overconfident with respect to the degree of upward mobility that they will experience, which leads them to prefer less redistribution. Apart from that, people may hold a specific redistribution preference in the first place and then form their beliefs about social mobility in order to justify their redistribution preferences. We test for (i) the existence of overconfidence and/or "justifying beliefs" with regards to social mobility as well as (ii) their relative importance for the redistribution preferences in an online experiment on Prolific.

In our experiment, subjects perform an effort task and receive a payoff which may or may not depend on their performance in the effort task. We interpret the degree to which subjects can actually influence their own payoff via effort provision as the amount of social mobility in our setting. After the effort task subjects enter a redistribution stage, where they vote on a preferred redistributive tax to determine the final payoffs for everyone. In our control treatment we ask subjects without them knowing of the redistribution stage about their belief that effort rather than luck was deterministic for their pre-tax payoff payments. In our justifying beliefs treatment we ask subjects the same question, but inform them about the existence of a redistribution stage prior to the question. With this information they can potentially form their beliefs in order to justify their preferred redistribution demand. Our overconfidence treatment is also similar to our control treatment, the difference here being that we do not inform subjects about their pre-tax payoff in the overconfidence treatment. Thus, overconfidence matters for the decision on the preferred tax rate in our overconfidence treatment, but not in our control treatment, since subjects already know their own payoffs in the control treatment. Across all our treatments we additionally vary the probability with which effort rather than luck determines payoffs. We do this in order to gain an understanding of how motivated beliefs are formed under different levels of social mobility.

2022-08-19

2022-08-26

We have two types of primary outcomes: Beliefs and the preferred tax rate.

1.) Beliefs:
1.1: Effort Performance Beliefs: We measure the beliefs about subjects' own rank in the effort task. This allows us to estimate the degree of overconfidence for each subject.
1.2: Payoff Determination Belief: We also measure the beliefs that subjects state regarding the probability that their payoff was determined by effort rather than luck. We compare these beliefs between the justifying beliefs treatment and the control treatment conditional on the payoff of each subject.

The measurement of beliefs allows us to assess, whether or not / to what degree overconfidence and justifying beliefs exist in our setting. While this is of interest for us, we further want to understand whether and to which extent the existence of such motivational beliefs influences preferences for redistribution. Therefore, we measure the preferred tax rates.

2.) Preferred tax rate: We measure the tax rate (in %) that subjects indicate as their preferred tax rate in the experiment. We compare this tax rate between our treatments.

Regarding the justifying beliefs treatment, we compare the preferred tax rate to the preferred tax rates in the control treatment conditional on the pre-tax payoff. According to our null hypothesis we expect subjects with a high (low) pre-tax payoff to vote for lower (higher) taxes in the justifying beliefs treatment relative to the control treatment due to the possibility to justify their preferred redistribution preference. We also compare preferred tax rates for different effort/luck combinations in determining the payoffs between the justifying beliefs and the control treatment. We hypothesize similar differences in preferred taxes between the justifying beliefs and the control treatment (in contrast to the comparison of the overconfidence and control treatment, see below).

Regarding the overconfidence treatment we compare the preferred tax rates between our overconfidence treatment with the preferred tax rates in our control treatment. We expect the preferred tax rates to be relatively lower on average in the overconfidence treatment. We also compare preferred tax rates for different effort/luck combinations in determining the payoffs between the overconfidence and the control treatment. We hypothesize that a possible gap in preferred taxes between the overconfidence and the control treatment reduces with stronger luck dependence.

As secondary outcomes we measure risk preferences of subjects. Besides, standard data on subjects' demographic background will be provided to us by Prolific. The measurement of risk preferences is important, because in our overconfidence treatment subjects make their tax choice without knowing their own payoffs, which makes risk preferences relevant for their decision. By controlling for risk preferences we can eliminate the effect of risk preferences in our analyses.

At the start of our experiment subjects perform an effort task, more specifically the slider task of Gill and Prowse (2012). We ask subjects to indicate their belief regarding their own performance in the effort task relative to others. Afterwards, subjects receive a payoff which is either determined on the basis of their performance in the effort task or determined by sheer luck. Whether a subject's payoff is determined by effort or luck is random: With a specific probability p, effort is deterministic for the payoff. We vary this probability p across all our treatments using three different values for p, which are 0.8, 0.5 and 0.2. After subjects have learnt their payoff, we ask them to state how likely it is that their payoff was determined by effort rather than luck. Finally, subjects vote upon a redistributive tax with efficiency losses, which influences their own and other subjects' payoffs. We also record subjects' risk preferences and other demographic variables in a questionnaire after the experiment.

The above mentioned design is exactly the design in our control group.
In our justifying beliefs treatment we tell subjects about the existence of a redistribution stage before they indicate their beliefs regarding whether it was effort or luck which determined their pre-tax payoffs. Thus, subjects can possibly alter their belief report in order to justify a specific redistribution preference, a behavior which is impossible in our control treatment due to the unawareness of a redistribution stage being played later.
In our overconfidence treatment we also stick to the above mentioned design of the control group, except for the fact that we do not tell subjects about their pre-tax payoffs. Thus, overconfidence can potentially affect their preferred tax rates.

Below we describe the experimental design for our control group. Our treatment groups share the same experimental design with only some slight alterations, which we also describe below. Our experiment consists of six stages:

Stage 1: Subjects perform an effort task. In our case the effort task is the slider task of Gill and Prowse (2012). In this task subjects have to adjust sliders that range from a minimum to a maximum number, such that each slider shows a specific number. In our case, we have sliders that range from 0 until 100 which subjects need to adjust to exactly 50. Subjects have one minute to adjust as many sliders as possible at the number 50. Subjects' performance is measured as the number of correctly adjusted sliders. Before the effort task, we inform subjects that their performance in the effort task may positively affect their chances to earn money in the experiment but we do not give further details about the experiment.

Stage 2: We measure subjects' belief about their effort performance by asking them to rank themselves among 7 others, who have previously done the slider task. Since subjects on Prolific do not perform the experiment at the same time (this would be feasible on Prolific in principle, but it is too costly for our experimental budget and also rather complicated to conduct in practice), we conducted a pre-study on Prolific, where participating subjects only did the slider task. We use the data on performance of those participating subjects in the pre-study as comparison for the subjects in our main experiment i.e. the "7 others" are taken from this pre-study. We tell our subjects that they are ranked against subjects from the pre-study. We incentivize this belief elicitation by paying a bonus after the experiment only if a subject ranks herself exactly correctly among the 7 others.

Stage 3: We determine a payoff for every subject. This payoff is either determined by the performance in the effort task or by sheer luck. We have eight different payoffs, ranging from 1 EC (experimental currency, which we later convert to real money) to 8 EC in steps of 1EC. If a subject's payoff is determined according to effort, we simply take her rank in the effort task among the 7 others and assign the payoff according to this rank: If she ranks first, she gets 8 EC, rank 2 gets 7 EC etc. until rank 8, who gets 1 EC. If a subject's payoff is determined by luck, we simply assign each payoff with a 12.5% probability. Whether a subject's payoff is determined by effort or luck is random: With a specific probability p, effort is deterministic for the payoff. We vary this probability p across all our treatments: We use three different values for p (0.8; 0.5 and 0.2). We explain the above mentioned mechanism which determines payoffs to our subjects. Subjects see their assigned payoff, but not whether their payoff was determined according to effort or luck.

Stage 4: We ask subjects about how likely they think it is that their payoff was determined by effort rather than luck. Note that this is a non-trivial question, because subjects can combine the information they have from knowing their assigned payoff with their belief about their effort task performance to form a "posterior" belief about the probability that effort rather than luck was deterministic for their payoff. For instance, a subject who thinks of herself as ranking well in the effort task and who received a high payoff can plausibly infer that it is more likely that her payoff was being determined by effort rather than luck. We do not incentivize this belief elicitation because we want to give room for the formation of motivated beliefs.

Stage 5: Subjects decide upon a redistributive tax between 0% and 100%. The tax is redistributive, as it brings high and low payoffs closer together. The tax comes at a per subject efficiency cost, which increases with the tax rate. We randomly pick one subject's tax choice as being pivotal i.e. as the tax which applies to a group of 8 subjects, the pivotal subject included, of course. We use this simple voting mechanism to deter strategic voting. Subjects are aware of this voting mechanism.

Stage 6: Subjects choose from a menu of risky and riskless hypothetical payoffs. We include this stage in order to measure subjects' risk preferences. All choices in stage 6 are unincentivized.

Our treatments follow mainly the same experimental design with slight alterations:
In our justifying beliefs treatment everything is the same as in our control group, except that we tell subjects about stage 5 before stage 4 i.e. we inform subjects that they will make a choice about a redistributive tax before they state their beliefs regarding whether effort or luck determined their pre-tax payoffs. Thus, they can potentially report higher (lower) beliefs about the importance of effort in determining payoffs in order to justify a low (high) redistribution preference.

In our overconfidence treatment, everything is the same as in our control group, except that we do not inform subjects about their payoffs in stage 3, which is why we also omit stage 4 in this treatment: Asking about the probability that effort determined a subject's payoff is relatively nonsensical, if the subject has no information other than the baseline probability.

Across all our treatments we additionally vary the degree to which a subject’s payoff is determined by effort rather than luck. In the luck condition, subjects’ payoff is determined to 80% by luck and only to 20% by effort. In the effort condition, it is vice versa. In the neutral condition, payoffs are determined according to a 50/50 effort/luck mix. We expect overconfidence to have less of an effect on redistribution preferences in more luck-based treatment conditions. We also study the impact of the effort/luck mix on the effect of justifying beliefs on redistribution preferences.

We run the experiment online via Prolific. We embed our experiment in Prolific such that an individual is randomly allocated to a treatment group when participating in our experiment. The exact randomization method will be chosen according to Prolific's standard method for RCTs.

We randomize on the individual level i.e. every individual is randomly allocated to exactly one treatment. No individual can be in two treatments.

No

No clustering.

About 450 subjects from the online Platform Prolific. The exact feasible sample size will depend on the completion time that subjects need for the experiment, as subjects are paid per hour on Prolific. We can only estimate a rough completion time of our experiment at this point, which is why the estimated 450 participants should be considered a rough estimate.

50 per treatment.
Explanation: We have one control group, the justifying belief treatment and the overconfidence treatment. Across these three treatment groups we vary the amount of effort versus luck which determines payoffs. We use three different effort versus luck parameter constellations: High effort importance, medium effort importance, and low effort importance. Thus, we have a 3x3 treatment design yielding 9 different treatments.