Incentives, Expectations, and Political Identity

We manipulate various features of a survey used to elicit respondent beliefs.

We investigate how survey design influences economic expectations reported by individuals with differing political affiliations, focusing on whether the effect of incentives varies around a major shift in political power. Specifically, we conduct the study in multiple waves both before and after the 2024 presidential election to assess whether changes in political leadership alter how participants incorporate information and respond to monetary rewards for accurate forecasts. More concretely, we elicit expectations use flat-fee payments and various forms of marginal incentives. Additionally, we implement a standard information-provision RCT to study how incentive structure influences learning rates across self-reported political identity and whether/how this changes whenever political power changes.

2025-03-27

2025-05-01

We are primarily interested in (1) how different incentives influence point expectations of inflation (and other macroeconomic variables) across self-reported political identity (2) Whether/how this changes after a shift in political power (3) Similar to 1,2 but regarding the link between expectations and perceptions of inflation across political parties (4) Similar to 1,2 but for learning rates in an information-provision RCT.

We randomly assign subjects recruited via Prolific to variations of a survey meant to elicit the same distribution of beliefs. Our interest is in how different survey designs influence the underlying distribution of beliefs we collect.

We implement a multi-wave survey experiment, with data collection occurring just prior to the 2024 U.S. presidential election and then again shortly afterward. During each wave, participants answer questions about their demographic profile, political affiliation, and economic expectations—focusing on outcomes such as inflation, gas prices, or general economic conditions. In both the pre- and post-election surveys, we will randomly assign participants to different versions of the questionnaire. Specifically, these versions will vary in the incentives we use to elicit expectations. More concretely, we vary whether subjects face marginal monetary incentives tied to forecast accuracy. We do this in an RCT framework, sometimes incentivizing priors and posteriors, sometimes neither, and sometimes only one of the two. We will invite the same participants in both waves so that we can observe within-subject changes over time; however, we will also supplement the second wave with additional, newly recruited subjects for cross-sectional comparisons to mitigate concerns about survey learning effects confounding our within-subject results.

The experiment’s core objective is to identify how incentivization influences reported beliefs across self-reported political identies. In particular, our interest is in whether imposing marginal incentives can mitigate or resolve long-standing empirical puzzles related to inflation expectations and political identity. Because the pre- and post-election timing captures a substantial real-world event, we can also measure how shifts in political power or in participants’ perceptions of that power affect their reliance on certain types of information or incentives. For instance, we will compare how participants respond to expert forecasts or to performance-based bonuses before and after the election results are clear. This setup enables us to track whether heightened or diminished trust in government outcomes, stemming from election results, manifests in systematic changes to economic expectations or willingness to exert effort in making accurate predictions.

Participants will receive a baseline show-up payment for each wave, but the precise bonus scheme will vary by treatment. In the performance-pay conditions, if participants’ forecasts are sufficiently close to realized metrics (for instance, the actual inflation rate published months later), they receive an additional bonus. If participants are in a control condition with no performance-based incentives, they simply receive a fixed compensation for completing the survey. Each wave’s questionnaire will last approximately 15 to 25 minutes and include attention checks to ensure data quality. At the conclusion of both waves, participants will be debriefed about the nature of the study and provided with details on their final compensation, including any performance-based bonus that can only be computed once relevant economic statistics become available. Through this design, we aim to shed light on how political events and various survey design elements together shape individuals’ stated economic beliefs.

Randomization done by software coded in oTree and deployed on Prolific.

Individual

No

We do not cluster.

1000 participants in our original wave. Our goal is to call back all of those 1000 participants in our second wave and to include roughly 500 new participants in our second wave.

250 per treatment in the original wave. Hopefully, 250 recalls per treatment in the second wave. Roughly125 - 250 fresh recruits per treatment in the follow wave.

We conducted a power analysis to determine the sample size required to detect various effect magnitudes (measured by Cohen’s d) with a significance level of 0.05 and a power of 0.80. Cohen’s d is the standardized mean difference between the treatment and control groups, calculated as (M1 – M2) / SD_pooled. Conventional interpretations label d=0.2 as a small effect, d=0.5 as medium, and d=0.8 as large. Based on these guidelines, we decided on a sample size of 250 subjects per treatment arm, which allows sufficient power to detect small differences (or precisely estimate a null effect) at a one-percent level of significance (with power equal to 0.8) for typical between-group comparisons.