The Role of Topic Choice in Cross-Partisan Conversations: Dataset and Analysis

This repository contains the full dataset and analysis code for the paper:

"The Role of Topic Choice in Cross-Partisan Conversations" James Houghton, Gus Cooney, Duncan Watts (2026) https://osf.io/preprints/socarxiv/nygt3_v3/

---

Abstract

Affective polarization in the United States — animosity between Republicans and Democrats — has escalated for decades, threatening the health of American democracy. Research on intergroup contact suggests that talking across party lines can reduce polarization, yet studies disagree on whether confronting or avoiding political disagreement is the more effective strategy. We address this debate using a large-scale integrative experiment in which Republicans and Democrats engaged in face-to-face video conversations, with levels of disagreement and political relevance systematically varied across a diverse set of topics. While some topics reduced affective polarization more than others, how "political" a topic was did not predict which conversations went well. Moreover, topic assignment explained just 2% of the variance in individual outcomes, with people assigned to the same topic often having entirely different experiences. What did correlate with conversational success was how individuals experienced the interaction (i.e., whether their partner listened, took their perspective, and made them feel heard). We suggest a shift in focus from choosing the "right" topic to understanding the detailed interactional dynamics that make cross-partisan conversation succeed.

Repository Contents

Code Notebooks

• analyze_study_data.ipynb Reproduces all figures and tables used in the paper. Contains code for all study outcome analyses.

• analyze_topics_pretest.ipynb Documents the pre-survey process used to select the final set of 10 discussion topics from an initial pool of 174. Includes exploratory analyses and figures for the supplement.

Figures

All figures are exported as both PDF and PNG:

From analyze_study_data.ipynb

File	Description
fig_2_topic_effects	Figure 2: Estimated effects of each topic on both outcomes
fig_3_single_predictors	Figure 3: Single-predictor regression panels
fig_4_cross_predictor_comparison	Figure 4: Cross-predictor comparison of effect sizes
fig_S4_balanced_samples	Figure S4: Sample balance across experimental conditions
fig_S5_change_histograms_kde	Figure S5: Outcome distributions (KDE)
fig_S6_no_conversation_control	Figure S6: Test-retest reliability from no-conversation control
fig_S7_disagreement_by_topic	Figure S7: Effect of dyadic disagreement broken out by topic
fig_S8_manipulation_check	Figure S8: Partner party recall (manipulation check)

From analyze_topics_pretest.ipynb

File	Description
fig_S1_example_topics_polarization_contentiousness	Figure S1: Example topics from the full pool
fig_S3_selected_topics_partisanship_contentiousness	Figure S3: Selected topics plotted on partisanship vs contentiousness

Tables

All tables are rendered as pandas DataFrames within the notebook:

Table	Description
Table 1	Primary regressions with cluster bootstrap inference (resampling topics)
Table 2	Pre-treatment individual attributes (attitudes, personality, demographics)
Table 3	Political attitude predictors of attitude change
Table 4	Conversation dynamics predictors with topic fixed effects and dyad-clustered SEs
Table 5	Discussion topics and their positions on topic-level dimensions
Table S1	R-squared values by category of predictors
Table S2	Scaling ranges for cross-predictor comparison
Table S3	Bayesian ROPE analysis for single-predictor regressions
Table S4	Effect of primary predictors on secondary outcomes (single regressions, BH-corrected)
Table S5	Primary regressions excluding floor-effect participants
Table S6	Floor effects on political attitude predictors
Table S7	Balance check: pretest measures regressed on primary dimensions (per-IV BH correction)
Table S8	Primary regressions controlling for pre-discussion affect
Table S9	Comparison of attritters and non-attritters on individual-level predictors
Table S10	Primary regressions with attritters included (worst-case imputation)

Data Files

Topic Definitions

• topic_list.csv Contains full topic texts and short name identifiers used in code and figures.

• topic_moments.csv Summary statistics for topic-level dimensions.

Pretest Data

• topic_pretest_identity_threat_responses.csv Qualtrics survey data capturing participants' fear of judgment (identity threat) for each topic.

• topic_pretest_responses.csv Main pre-survey dataset used to estimate contentiousness and partisanship for topic selection.

Study Data

• topic_study_responses.csv Follow-up topic feature responses collected during the study to refine feature estimates.

• study_data.csv Final merged and cleaned dataset from the experiment, ready for analysis.

• no_conversation_control_data.csv Test-retest control data from participants who completed pre- and post-test measures without a conversation.

Citation

Please cite the study as:

Houghton, J., Cooney, G., Watts, D. "The Role of Topic Choice in Cross-Partisan Conversations." 2026.