Strategic Communication and the Underuse of Incentive-Based Credibility: Experimental Evidence

2026-07-12

2026-08-31

The previously specified hypotheses for senders and receivers remain part of the broader project.
For the additional CRED+ treatment, we test the following additional hypothesis:
- H5: Receivers are more likely to follow the sender’s message in CRED+ than in the existing CRED treatment. Consequently, the difference between type-II errors and type-I errors is lower in CRED+ than in CRED.
As in the previous pre-registration, type-I and type-II errors are defined relative to the receiver’s subjectively optimal decision. This decision is derived from the receiver’s initial belief that their answer is correct and the empirical truth frequency of the relevant message type. A type-I error occurs when a receiver follows a message that should not be followed according to this benchmark. A type-II error occurs when a receiver does not follow a message that should be followed according to this benchmark.
We test H5 using both non-parametric comparisons and regression analysis. The main comparison is between CRED+ and the existing CRED treatment. As in the previous pre-registration, the analyses focus on observations for which the sender’s message is not aligned with the receiver’s initial answer.

The previously specified secondary outcomes remain unchanged.
The only addition concerns the open-ended explanations and structured follow-up questions elicited in CRED+. These analyses are exploratory and are intended to support the interpretation of the behavioral mechanism. Specifically, we examine whether participants’ stated reasons for following or not following the sender’s message point to limited comprehension, persistent reliance on their initial answer, perceived credibility of the message, distrust of the sender, or other reasons.
For the open-ended responses, we will classify participants’ explanations into broad categories that capture the main stated reasons for their decisions. The structured follow-up questions will be analyzed descriptively and, where useful, related to observed follow/not-follow behavior. These analyses are not intended as confirmatory tests of additional hypotheses but as exploratory evidence on the mechanisms underlying receivers’ responses to credible messages.

This pre-registration extends one of our existing studies by adding one treatment, CRED+. The general design remains unchanged. Participants in the role of receivers know that there are two main stages of which only one will be randomly chosen for determining their bonus.

In stage 1, participants answer factual questions comparing outcomes under the Obama and Trump administrations. Specifically, they are asked whether the unemployment rate in a particular U.S. state, and the crime rate in another U.S. state, was lower during the Obama or the Trump administration. The two questions are presented in random order. Participants know that, if stage 1 is selected for payment, one of the two questions will be randomly chosen and they will receive a bonus if and only if their answer to that question is correct. For each question, we also ask participants how certain they are that their belief is correct.

In stage 2, receivers can revise their previous answers based on a message from a sender in a previous study. In that study, senders knew the correct answer. For each question, they could send one of two possible messages, suggesting that the correct answer was either Obama or Trump. Senders in treatment TRUMP knew that they would receive a bonus if and only if the receiver answered Trump. Senders in treatment OBAMA knew that they would receive a bonus if and only if the receiver answered Obama.

Receivers are informed that senders knew the correct answers but did not need to report truthfully. Receivers are also informed about the senders' incentives. After receiving the sender's message, they again choose whether they believe that Obama or Trump is the correct answer. Receivers know that, if stage 2 is selected for payment, they will receive a bonus if and only if their answer is correct.

To elicit receivers' responses to the senders' messages, we use the strategy method. For each of the two questions, receivers choose between Obama and Trump for each of the two possible sender messages. Payoffs are calculated based on the message the sender actually sent.
The existing treatment manipulation varies the senders' incentives:
• In treatment OBAMA, receivers know that senders got a bonus if and only if the receiver they were matched with answered Obama.
• In treatment TRUMP, receivers know that senders got a bonus if and only if the receiver they were matched with answered Trump.

The additional treatment introduced in this extension is CRED+:
• CRED+ builds on the existing credibility-information treatment. Before making their final decisions, receivers receive information about the empirical truth frequency of the relevant message type.
• After this credibility information is presented, receivers answer an additional comprehension/attention check. The purpose is to ensure that the information has been made salient and to measure whether the participant processed the information.
• After their final decisions in cases where the receiver's initial answer differs from the sender's message, receivers are asked to explain their decision. We first elicit an open-ended explanation and then ask structured questions about possible reasons for their behavior.

In the post-experimental questionnaire, we elicit basic demographic information and personal attitudes, such as political affiliation.
The experiment will be conducted online using the survey software Qualtrics and participants will be recruited using Prolific. We will require participants to reside in the US and show an approval rate of at least 95%. Participants who answered the majority of our first three comprehension questions incorrectly will not be allowed to continue and will hence not be included our sample. Moreover, in the post-experimental questionnaire, participants are asked whether they looked up the answers to the estimation questions on the internet. Those who respond “Yes” will be excluded from the statistical analysis.

Not available

All randomization procedures during the study are implemented by the survey software. This includes assignment to sender-incentive conditions, the order of questions, and the order of answer options.

Individual-level randomization

No

We cluster on the participant level and plan to obtain 300 participants/clusters.

This extension adds 300 participants in CRED+. Because we elicit each participant's decisions for both possible messages in two question domains, we obtain four observations per participant. Thus, CRED+ adds 1,200 observations in total. Since the main analyses focus on observations in which the receiver's initial answer differs from the sender's message, approximately half of these observations, i.e., 600 observations, are expected to be usable for the main statistical analysis.

For CRED+, we plan to recruit 300 participants. Participants will be split evenly across the two sender-incentive conditions, with 150 participants in the condition in which senders were incentivized to induce Trump answers and 150 participants in the condition in which senders were incentivized to induce Obama answers.