Florida Replication

Repository for replication files for "Deep canvassing with automated conversational agents: personalized messaging to change attitudes".

Replication Instructions

Step 0: Data Preprocessing

Ensure /data/df_for_analysis_final.rds exists, then run:

cd code
Rscript pre_processing_senderdb.R

Outputs:

• /data/df_for_analysis_processed.rds (5,115 observations, 31 variables)
• /tables/attrition_summary.tex - Summary of pre-survey, control, and treatment group sizes and completion rates

Step 1: Main Analysis

Run the main analysis script (automatically sources utils.R):

Rscript main_analysis.R

Outputs:

• /tables/main_results.tex - Main treatment effects (all respondents)
• /tables/dem_results.tex - Subgroup analysis (Democrats)
• /tables/rep_results.tex - Subgroup analysis (Republicans)
• /tables/ind_results.tex - Subgroup analysis (Independents)

Step 2: Robustness Checks

Run robustness analysis:

Rscript robustness.R

Outputs:

• /tables/balance_table_all.tex - Covariate balance (all respondents)
• /tables/balance_table_completers.tex - Covariate balance (completers)
• /tables/placebo_leave_one_out.tex - Leave-one-out covariate placebo test
• /tables/placebo_outcomes.tex - Placebo outcomes test

Step 3: Appendix Analysis (Manski Bounds)

Generate Manski bounds with bootstrapped confidence intervals:

Rscript bootstrapping_manski.R     # Bootstrap confidence intervals

Note: If boot_UB.rds and boot_LB.rds already exist in /data/, the script will use them by default to quickly generate the table. To force regeneration of bootstrap results (e.g., after changing parameters), set USE_EXISTING_BOOTSTRAP <- FALSE at the top of the script.

Outputs:

• /data/boot_UB.rds, /data/boot_LB.rds - Bootstrap results (if generated)
• /tables/manski_bounds_ci.tex - Manski bounds with 95% confidence intervals

Step 4: Generate Figures

Run the figures script to create visualizations:

mkdir -p ../figures  # Create figures directory if it doesn't exist
Rscript figures.R

Outputs:

• /figures/Figure1_treatment_effects.png/.tiff - Treatment effects plot for all respondents (3 outcomes × 3 models)
• /figures/FigureSI2_treatment_effects_by_subgroup.png/.tiff - Treatment effects by party affiliation (Overall, Democrats, Republicans, Independents)

Step 5: Generate Codebook (Optional)

Rscript datamaid_codebook.R

Output: /test-data-codebook.Rmd - Data dictionary for all variables