Polarization and Belief Entrenchment Through Memory

How to cite

Tang, Li; Kuang, Pei; Weber, Michael. 2026. "Polarization and Belief Entrenchment Through Memory." AEA RCT Registry. AEARCTR-0018416. https://doi.org/10.1257/rct.18416-1.0

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

Li Tang* , Pei Kuang , Michael Weber

* Primary investigator

Registry ID
AEARCTR-0018416
DOI
https://doi.org/10.1257/rct.18416-1.0
First registered
Last updated
Status
In development
Trial dates
2026-04-21 – 2027-10-31
Countries
Private
Registry record
http://www.socialscienceregistry.org/trials/18416

Abstract

This study examines how memory processes, specifically rehearsal and cue-triggered retrieval, shape belief formation about a salient U.S. policy. Participants complete a two-part online survey conducted approximately 24 hours apart. In Part 1, they read and evaluate a set of opinion headlines about the policy. Across randomly assigned conditions, participants either (a) complete a brief writing task that directs rehearsal of previously viewed headlines, or (b) are exposed to a brief visual cue at the beginning of Part 2. Part 2 elicits recall of the Part 1 content and participants' current views on the policy. The design holds the information environment constant across conditions and varies only the retrieval environment, allowing the study to isolate the role of memory in belief formation. The study contributes to research on belief formation and political disagreement.

Intervention

The study implements two sets of interventions across separate experiments:

(1) Rehearsal. Between the two survey waves, participants are randomly assigned to one of three short writing tasks. Two direct retrieval toward different subsets of the Part 1 content; the third is unrelated to the policy (neutral control).

(2) Cue. At the start of Part 2, participants are randomly assigned to one of three conditions. Two display a brief visual stimulus linked to the Part 1 content; the third proceeds without any additional display (no-cue control).

The information environment is identical across conditions; only the retrieval environment is manipulated.

Experimental design

The study uses a two-part online experimental design implemented across two separate experiments. In Part 1, participants view a balanced set of 20 opinion-article headlines about a salient U.S. policy, with each headline accompanied by a shape indicator. Across all conditions, participants observe the same 20 headlines in randomized order. After each headline, participants classify it according to its stance on the policy, which produces an individual-level record of encoding accuracy. Participants also report an overall emotional reaction to the set of headlines and complete an instructed-response attention check. The information environment is held constant across all conditions within each experiment.

In Experiment 1 (Rehearsal), participants are randomly assigned to one of three groups that differ in a short writing task administered at the end of Part 1. In one group, participants are asked to write about as many headlines as they can remember from Part 1 that are most consistent with their own view on the policy. In a second group, participants are asked to write about headlines that are most inconsistent with their own view. A third, neutral-control group is asked to write about an unrelated topic. All three groups face similar cognitive effort, and writing-task responses are quality-incentivized.

In Experiment 2 (Cue), there is no treatment in Part 1. Random assignment occurs at the start of Part 2, immediately before elicitation. Two groups see a brief visual display associated with the Part 1 shape indicators; a third no-cue-control group proceeds directly to elicitation. The cue contains no new information about the policy itself.

Part 2 of both experiments is conducted approximately 24 hours after Part 1. Participants are recontacted, receive a brief neutral reminder that the Part 1 headlines are relevant for the follow-up, and complete a common battery of questions eliciting recall of the Part 1 content, policy sentiment, policy support, and subjective uncertainty about the Part 1 information environment. Participants also complete a recognition test consisting of the Part 1 headlines presented alongside additional decoys. Accuracy bonuses are attached to the recall and recognition tasks.

The experimental design allows for comparisons of recall, beliefs, and uncertainty across participants exposed to the same information environment but assigned to different retrieval conditions, as well as comparisons within self-reported political-identity groups measured at the end of Part 1.

Design details

Randomization method
Done within qualtrics using javascript.
Randomization unit
Individual participant.
Clusters
N/A
Observations
Around 4200 individuals.
Arms
Around 700 per treatment arm.
Primary outcomes
Primary outcomes are measured at Part 2, 24 hours after Part 1: (1) Recall. Self-reported recall of Part 1 content. (2) Policy sentiment. Self-reported sentiment toward the policy. (3) Policy support. Self-reported support for the policy. (4) Posterior uncertainty. Self-reported uncertainty about the Part 1 information.
Secondary outcomes
(1) Recognition performance. Accuracy on a Part 2 recognition test. (2) Encoding accuracy. Accuracy on the Part 1 per-headline classification task.

Cited by

No citation data available for this registration yet.

Keywords