Choice-induced Misspecified Mental Models: A Follow-up Study

How to cite

Fan, Qiaofeng (Tony). 2025. "Choice-induced Misspecified Mental Models: A Follow-up Study." AEA RCT Registry. AEARCTR-0016509. https://doi.org/10.1257/rct.16509-1.0

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

Qiaofeng (Tony) Fan*

* Primary investigator

Registry ID
AEARCTR-0016509
DOI
https://doi.org/10.1257/rct.16509-1.0
First registered
Last updated
Status
In development
Trial dates
2025-08-04 – 2025-10-31
Countries
Private
Registry record
http://www.socialscienceregistry.org/trials/16509

Abstract

The current study is part of a project that proposes, tests, and seeks to understand a novel behavioral hypothesis: After agents control or manipulate a certain factor — their “choice variable” — to potentially influence a payoff-relevant outcome, they tend to form mental models
that overestimate the causal impact of their choice variable on the focal outcome. I term this hypothesis the Choice-induced Model Distortion hypothesis.

In an earlier study, I validated the Choice-induced Model Distortion hypothesis and provided some initial evidence for its underlying mechanisms. In particular, I ruled out selective learning about the choice variable as the main driver of the belief distortion. The current study extends the earlier study with a few additional treatments, to further disentangle several competing mechanisms that potentially generate the belief distortion effect, and to establish the robustness of the effect to certain variations in the design.

Intervention

Participants learn from a multivariate dataset before making choices involving one randomly selected variable and stating their beliefs about the data-generating process (i.e., relationships between variables). See the pre-analysis plan for more detail.

Experimental design

The current study consists of four different treatments:
(1) Baseline treatment: implements the study procedures (learning stage, choice task, belief elicitations) without change.
(2) RevealChoiceInconsequential treatment: In this treatment, right after the choice task, but before the belief elicitations, participants are informed that the belief elicitations (instead of the choice task) has been randomly selected to determine their final bonus.
(3) BeliefBeforeChoice treatment: This treatment essentially reverses the order of the choice task and the belief elicitations.
(4) FullDataset treatment: This treatment examines the robustness of the belief distortion effect to making the learning stage more transparent and informative to participants.

See the pre-analysis plan for more detail.

Design details

Randomization method
Randomization done in office by a computer
Randomization unit
Individual
Clusters
N/A
Observations
3200
Arms
800 in each treatment
Primary outcomes
Propensity to state different variables (attributes) as outcome-relevant, in multiple choice or elicited model (function) forms; probabilistic beliefs about the chances that different variables (attributes) are outcome-relevant.

Cited by

No citation data available for this registration yet.

Keywords