Misguided Inference and Asymmetric Correction

How to cite

Drobner, Christoph; Orhun, Yesim. 2026. "Misguided Inference and Asymmetric Correction." AEA RCT Registry. AEARCTR-0019173. https://doi.org/10.1257/rct.19173-1.0

Download: BibTeX · RIS — for reference managers without a browser translator. Zotero and most others pick up this page's citation metadata directly.

Christoph Drobner* , Yesim Orhun

* Primary investigator

Registry ID
AEARCTR-0019173
DOI
https://doi.org/10.1257/rct.19173-1.0
First registered
Last updated
Status
In development
Trial dates
2026-08-01 – 2027-12-31
Countries
United States of America ()
JEL codes
C91, D83, D84
Registry record
http://www.socialscienceregistry.org/trials/19173

Abstract

This study investigates how overconfidence and underconfidence about own ability shape inferences about the economic environment when outcomes depend on both ability and the environment. We focus on a setting where individuals can switch between environments. We test the hypothesis that overconfident individuals are more likely than underconfident individuals to select environments they perceive as more ability-contingent, and that this selection generates asymmetric belief correction (i.e., overconfident participants correct their biases more than underconfident participants do).

Experimental design

Participants complete a two-stage online experiment. In Stage 1, they take a 12-question logic quiz of randomly assigned difficulty and report an incentivized belief about how their performance on six shared questions compares to a reference pool of 100 earlier participants. In Stage 2, over two rounds, one of two managers generates 50 paid tokens per round. One manager pays a green token when the participant beats a randomly drawn member of the reference pool; the other pays green at a fixed rate that is disclosed to participants as a numerical probability. By design this rate equals the participant's own performance rate, so the two managers produce statistically identical token streams. After each round, participants report incentivized beliefs about their own performance and about which manager they face. A manager is randomly assigned before round 1. After round 1, all participants state whether they want to stay or switch; for half, randomly assigned, this choice is implemented (active-choice treatment), and for the other half the round-2 manager is randomly assigned (control). Further details are provided in the pre-analysis plan.

Design details

Randomization method
Randomization is conducted by computer at the individual-participant level. Quiz difficulty (Easy vs. Difficult) is block randomized within strata defined by the participant's score on the six shared quiz questions, which are answered before difficulty is assigned, crossed with gender. The manager-assignment regime (active-choice vs. control) is block randomized after the round-1 belief reports, within strata defined by round-1 performance-belief bias, whether the participant updated their performance belief in round 1, gender, and errors on the repeat-comprehension check. Order and placement features are independently randomized. Further details are provided in the pre-analysis plan.
Randomization unit
Individual participant.
Clusters
2,000 individual participants (no clustering).
Observations
2,000 participants who complete the entire experiment.
Arms
2,000 participants total. Quiz difficulty: approximately 1,000 Easy, approximately 1,000 Difficult. Manager-assignment regime: 1,000 active-choice, 1,000 control.
Primary outcomes
1. Whether the participant chooses the manager they believe is more likely to be the Performance Manager. 2. Reduction in the absolute bias of the participant's belief about their own performance between rounds 1 and 2.

Cited by

No citation data available for this registration yet.

Keywords