AfD Project

Project Purpose

This repository contains the thesis workflow on AfD polling support, media salience, framing/content indicators, and a supplementary media tone layer.

Minimal Navigation

• Primary project guide: README.md (this file)
• Data map and canonical files: data/README.md
• Notebook map and outputs: notebooks/README.md

Key thesis-facing notes:

• docs/final_empirical_results_summary.md
• docs/supplementary_media_tone_results_note.md
• docs/pre_writing_empirical_synthesis_note.md

Main Structure

• Corpora/: source corpus layer (monthly main corpus + shock corpora)
• archive_pipeline/: archived post-retrieval technical pipeline stages
• data/: canonical analysis-ready tables and panel outputs
• notebooks/: analysis/reporting notebooks
• figures/: notebook-generated figure outputs
• docs/: interpretation and methods notes
• scripts/: rebuild helpers

Canonical Final Outputs

• Main panel:
  • data/panel/monthly_full_analysis_panel_2013_2025.parquet
  • data/panel/monthly_full_analysis_panel_with_tone_2013_2025.parquet
• Synthesis tables:
  • data/panel/thesis_master_shock_results_table.csv
  • data/panel/thesis_dataset_counts_summary.csv
• Synthesis note:
  • docs/pre_writing_empirical_synthesis_note.md

Rebuild Commands

Clean layer refresh:

• python scripts/build_postretrieval_dataset.py
• python scripts/build_postretrieval_dataset.py --run-archived-pipeline

Salience rebuild:

• python scripts/build_monthly_media_salience_indicator.py

Main corpus processing (only when corpus batches change):

• python "Corpora/Main Corpus/run_main_corpus.py"
• full reprocess: python "Corpora/Main Corpus/run_main_corpus.py" --rerun --reextract
• target month example: python "Corpora/Main Corpus/run_main_corpus.py" --only 2021-09 --rerun --reextract

Note:

• Corpus scripts read source PDFs from raw_unarchive folders.
• Shock-specific corpora (for example Cologne) also read only from raw_unarchive; run their local audit_*.py then build_clean_*.py scripts after archive updates.

Polling Provenance Boundary

• DAWUM is the primary source from 2017-01 onward.
• Historical manual supplement (Wahlrecht-based) covers 2013-09 to 2016-12.

Project Rules

• Do not manually edit canonical generated datasets in data/.
• Regenerate via scripts/notebooks.
• Tone indicators are supplementary; core analysis uses salience + framing/content indicators.