Behavioral Simulation and Discourse Analysis of Institutional Corruption Framing
Description
This dataset supports the manuscript “Framing Institutional Corruption: A Behavioral Simulation of Moral Judgment and Accountability Narratives.” It provides the complete corpus, frame dictionaries, Python scripts, and simulation outputs used to examine how discursive framing—reformist, punitive/responsibility-forward, and neutral—reweights moral judgment, blame attribution, and trust under controlled, theory-guided conditions. The package enables full replication of the Phase 1 discourse analysis and the Phase 2 behavioral simulation and is intended for use in behavioral ethics, organizational studies, and computational social science.
Files
Steps to reproduce
The Python simulation and analysis follow a sequential pipeline linking the text corpus (Phase 1) with the behavioral simulation (Phase 2). Phase 1 – Corpus, frame coding, and topic-model diagnostics Corpus/ contains public documents from the Airbus, FIFA, and Sarkozy cases in plain-text format. 7. source_map.csv provides metadata assigning each document to Corporate, Legal, or Media and (where applicable) a case label. Run 3. corruption_analysis.py to apply the four frame dictionaries—reformism, deferral, externalization, minimization—using compiled regex. This script writes auditable outputs to Out/: counts_by_doc_and_frame.csv, counts_by_source_and_frame.csv, doc_stats.csv, and match ledgers (instance_matches.csv, instance_samples.csv) for precision checks and length-normalized reporting. Run make_triangle_overlay.py to tag Fraud Triangle cues (Pressure, Opportunity, Rationalization) and generate overlay diagnostics in Out/: triangle_overlay_matches.csv and triangle_overlay_matches_summary.csv. Optionally, run 5. topic_modelling_revised.py (default k = 6) to produce descriptive lexical-validation outputs in Out/: 5.1_topics_top_terms.csv, 5.2_doc_topic_assignments.csv, 5.3_topic_exemplars.csv, plus the exemplar figure. Phase 2 – Vignettes, agent architectures, and Monte Carlo simulation The vignette stimuli consist of three base scenarios (corporate, sports/governance, political) rewritten into reformist, punitive/responsibility-forward, and neutral versions with facts held constant (stored in Vignettes/ if included, otherwise embedded/loaded by the simulation script). Run 0. vignette_simulation_revised.py to execute the behavioral simulation with two rule-transparent evaluators: a deterministic rule-based (justice-motive) agent and a narrative-sensitive heuristic/NSBA agent parameterized from behavioral-ethics theory. The Monte Carlo design iterates across documented random seeds and outputs 0. vignette_simulation_results_mc.csv. Phase 2 – Descriptive analysis and diagnostics Run 1. moral_simulation_analysis_revised.py to aggregate Monte Carlo results and regenerate the manuscript’s main descriptive outputs: 1.1. simulation_summary.csv and 1.2. frame_effects_by_agent.csv, plus associated figures/tables. Optionally, run 1. make_diagnostics.py to export illustrative variance-partition checks: 1.3. anova_tables.csv and 1.4. tukey_contrasts.csv. Optionally, run 2. agent_comparison_table.py to generate the compact cross-agent comparison and 2. Agent comparison table.png. Reproducibility notes All agents are hand-specified and rule-transparent; no pre-trained language models or external APIs are used. Randomness in the heuristic agent is controlled through documented seed handling. Expected directional pattern: reformist > neutral > punitive for trust and reformist < neutral < punitive for organizational blame, with larger frame sensitivity for the heuristic agent than for the rule-based agent.
Institutions
- Universitatea de Vest din Timisoara Facultatea de Economie si de Administrare a Afacerilor