Belief updating from explicit, implicit, and silent signals across multiple settings

How to cite

He, Simin; Gui, Shan; Hu, Gaoji. 2025. "Belief updating from explicit, implicit, and silent signals across multiple settings." AEA RCT Registry. AEARCTR-0017408. https://doi.org/10.1257/rct.17408-1.1

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

Simin He* , Shan Gui , Gaoji Hu

* Primary investigator

Registry ID
AEARCTR-0017408
DOI
https://doi.org/10.1257/rct.17408-1.1
First registered
Last updated
Status
On going
Trial dates
2025-12-01 – 2026-12-31
Countries
Private
JEL codes
C72, C78, C91, C92
Registry record
http://www.socialscienceregistry.org/trials/17408

Abstract

We conduct a series of experiments to examine how people update their beliefs in response to non-explicit or silent signals/activities across multiple settings. Experiment 1 studies an individual belief-formation environment, testing whether participants can infer an underlying state optimally after receiving imperfect exogenous signals, which may be explicit, implicit, or silent. Experiment 2 studies a game-theoretic environment using a “colored hats” problem, examining whether individuals correctly update their beliefs about the state of the world by observing others’ actions, which again vary in how explicitly they convey information. Experiment 3 studies a matching environment with incomplete information, investigating whether people learn about the underlying state of the matching market from others’ matching behavior or outcomes. The information structure varies across conditions, with participants observing either only matching outcomes or all market activities, including individual proposals, acceptances, and rejections.

Intervention

Experiment 1 (individual belief-formation).
In an individual belief-formation setting, participants are asked to infer the underlying state by observing signals. The experiment involves two dimensions of intervention: First, whether one type of signals is explicitly presented or replaced by the absence of a signal, while maintaining theoretical equivalence. Second, in the case of no signal, whether participants are explicitly reminded that there is no signal.

Experiment 2 (game-theoretic: “colored hats” problem).
In a game-theoretic setting, represented by a colored hats problem, participants must guess the underlying state (the color of their hats) by observing the hat color and actions of others over multiple periods. The experimental intervention has two dimensions: First, whether others’ actions are explicit (guessing one specific color), or silent (no action taken). Second, whether the action space explicitly contains the option of “no action”.

Experiment 3 (matching with incomplete information).
In a matching market with incomplete information setting, participants must infer the state of the market to make optimal matching decisions based on observed market activities. The intervention involves two dimensions: First, whether others' matching behavior is active (e.g., moving from unmatched to matched, or dissolving an existing match) or silent (e.g., remaining unmatched). Second, whether detailed matching activities (including proposals, acceptances, and rejections) are made public, allowing participants to observe the actions behind silent matching behavior.

Experimental design

Experiment 1.
There are two possible states of the world: Blue or Red, each occurring with a known probability. Participants receive imperfect exogenous signals about the true state. In Treatment Explicit, signals are shown as either blue or red; and each signal is more likely to occur under the corresponding true state (i.e., a blue signal is more likely to be drawn if the state is Blue). In treatment Silent, the blue signal is replaced with the absence of a signal, meaning it is not displayed to participants, while the red signal remains explicit. In Treatment Implicit, the signal structure mirrors that of Silent, but participants are given explicit information about the probability of receiving no signal and are informed whenever the signal is absent. In our experimental design, we use a between-subject design for these three treatments.

Experiment 2.
We consider a “colored hats” problem with two players, each wearing a colored hat: either Red or Blue. Both players are informed that at least one of the two hats is Red (or Blue, depending on the realized state of hat colors), and each player observes the color of the other player’s hat. The game is played for multiple periods (open-ended), with the following rules: in each period, players first observe the other player’s action (if any) from the previous period, and then simultaneously decide whether to guess the color of their own hat. Once a player makes a guess, it becomes final and cannot be changed in later periods. The game ends when both players have made a guess, or when both choose to proceed to the next round. Correct guesses made in later periods receive a discounted payoff, ensuring that players with sufficient information have no incentive to delay their guess.
We employ a 2x2 design that varies along two dimensions: game type and action space.
The game type is determined by the true state and consists of two possibilities:
A. The two players wear different hat colors.
B. The two players wear the same hat color.
The action space also has two versions:
a. (“Red”, “Blue”, no action). No action means not making a choice between “Red” and “Blue”. This is denoted as the Silent treatment.
b. (“Red”, “Blue”, “Cannot tell”). This is denoted as the Explicit treatment.

We consider a two-game-type by two-action-space design. The two game types (different hat colors or same hat color) are implemented within participants, while the two action spaces (whether “Cannot tell” is a displayed action) are assigned between participants.

Experiment 3.
We adopt a matching with incomplete information environment here (see pre-registration at https://www.socialscienceregistry.org/trials/16825/edit for background information).
We employ a 2x2 design that varies along two dimensions: game type and information structure.
Game types:
A. At the start of the market, all participants are unmatched. Participants can update their beliefs of the state by observing others either form a match or dissolve existing matches.
B. At the start of the markets, all participants are unmatched. Participants can update their beliefs of the state by observing others maintain the unmatched status quo.

Information structures:
a. Only dynamic matching outcomes are observable to everyone in the market. This is denoted as the Silent treatment.
b. Both dynamic matching outcomes and additional detailed activities, including individual acceptances and rejections, are observable to everyone in the market. This is denoted as the Explicit treatment.

We consider a two-game-type by two-market-information-structure design. The two game types (active matching behavior or maintaining the status quo) are implemented within participants, while the two market information structures (whether the acceptances and rejections activities are publicly displayed) are assigned between participants.

Design details

Randomization method
Experiment 1. Within each experiment session, multiple treatment orders are implemented by randomization; the randomization is pre-determined. Subjects who sign up for the experiment receive a seat number randomly before entering the laboratory; the seat number determines the treatment and the parameter order. Experiment 2. Within each experiment session, both treatment (Silent and Explicit) are implemented by randomization; the randomization is pre-determined. Subjects who sign up for the experiment receive a seat number randomly before entering the laboratory; the seat number determines the treatment and the parameter order. Experiment 3. Within each experiment session, both treatment (Silent and Explicit) are implemented by randomization; the randomization is pre-determined. Subjects who sign up for the experiment receive a seat number randomly before entering the laboratory; the seat number determines the treatment and the parameter order.
Randomization unit
Individual-level randomization
Clusters
Experiment 1. We aim to collect 40-50 subjects for each of the 3 treatments; within each treatment, they are roughly equally split among the two different parameter orders. Experiment 2. We aim to collect 50-60 subjects (25-30 independent pairs) for each of the two between treatments (Silent or Explicit). Experiment 3. We aim to collect 60-72 subjects (15-18 independent groups) for each of the two between treatments (Silent or Explicit).
Observations
Experiment 1. About 120-150 individuals, recruited via the subject pool of the Economic Lab of the Shanghai University of Finance and Economics. We only recruit first-year undergraduate students for this experiment, as they haven’t taken a Probability Theory or Statistical course yet (during university). Experiment 2. About 100-120 individuals, recruited via the subject pool of the Economic Lab of the Shanghai University of Finance and Economics; who have not participated Experiment 1. Experiment 3. About 120-144 individuals, recruited via the subject pool of the Economic Lab of the Shanghai Jiao Tong University; who have not participated Experiment 1, 2, or a related experiment we conducted before, all of which are conducted at the Shanghai University of Finance and Economics (see https://www.socialscienceregistry.org/trials/16825/edit).
Arms
Experiment 1. 40-50 subjects per treatment arm. Experiment 2. 50-60 subjects per treatment arm. Experiment 3. 60-72 subjects per treatment arm.
Primary outcomes
Experiment 1. Whether the choices are Bayesian optimal. Experiment 2. Whether the choices are correct/optimal given the information received. Experiment 3. At individual level, whether the individual reaches the Bayesian correct matching.

Cited by

No citation data available for this registration yet.

Keywords