← Back to T&S Working Group Projects
Participatory AI Lab • Computational Policy Simulation

Stakeholder Policy Analyst: N=500 Double-Blind Simulation

Research Question
How do different narrative and causal framing strategies (drawn from Deborah Stone’s Policy Paradox taxonomy) shift stakeholder alignment scores, Key Veto-Player neutrality (Weight 10 ≥ 40), and the weighted Political Power Index (PPI) in a simulated suburban municipal council hearing on terminating an Automated License Plate Reader (ALPR) contract?
500
Total Simulation Runs (50 Replications × 10 Treatments)
4,000
Stakeholder Alignment Shift Records (500 Runs × 8 Personas)
46.26%
Control Baseline Weighted PPI (Σ Weight = 49)
41.77% – 65.29%
Observed Mean Final PPI Range Across 10 Treatments

1. Methodology

Experimental design separating stakeholder construction (Sabatier's Advocacy Coalition Framework) from experimental treatments (Deborah Stone's Policy Paradox).

Step 1

Control Baseline

Loads a fixed municipal scenario (anonymized suburban city, Pop. ~125,000, MHI > $115,000; proposal to terminate a $275,000/yr ALPR contract and reallocate funds to radar speed trailers, crosswalk lighting, and evening transit) and 8 stakeholders (S1–S8) defined via ACF belief tiers (Deep Core, Policy Core, Secondary Instrumental Aspect). Baseline Weighted PPI = 46.26%.

Step 2

Experimental Treatments

Selects one of 10 narrative/causal treatments from Deborah Stone’s Policy Paradox across a balanced randomized schedule (50 replications per treatment = 500 runs). Provides only the unlabeled 100-word public-comment text, omitting theoretical labels.

Step 3

Double-Blind Evaluation

Evaluates the 8 frozen ACF stakeholder profiles against the unlabeled public-comment text to record each stakeholder’s post-treatment alignment score (0–100) and engaged ACF belief layer without access to the treatment ID or theoretical category.

Step 4

Outcome Calculation

Computes weighted PPI in Python: PPI = Σ(Alignment_i × Weight_i) / 49. Evaluates passage criteria: (1) Final PPI ≥ 60.0%, (2) all Weight-10 stakeholders (S1, S2, S3, S4) have Alignment ≥ 40, and (3) Council vote tally (S1, S2, S4, S5 ≥ 50) is at least 3–1.

2. Findings

Summary statistics by treatment strategy and mean post-treatment alignment scores across the 8 stakeholder personas. Click any row to view the exact unlabeled stimulus text.

Symbolic Device: Polysemic Ambiguity (Ch. 7) N = 50 • Base PPI: 46.26% • Mean Final PPI: 65.29% (+19.03%) • SD: 2.49 • Veto-Safe: 100.0% • Pass Rate: 98.0%

STONE_SYMBOL_AMBIGUITY — Strategic Ambiguity ('Visible Safety Reinvestment')

"We all share the same goal: keeping our city the safest, best-governed community in the region. By transitioning our $275,000 annual expenditure away from an expiring third-party cloud lease and putting 100% of those dollars into city-owned radar speed feedback trailers, illuminated school crosswalks, and corridor patrols, fiscal conservatives get tangible city-owned assets, homeowners get immediate neighborhood traffic calming, the City Attorney eliminates third-party data liability, and residents' privacy is fully protected."
Treatment Strategy ID & Label N Mean PPI (Δ) SD(PPI) VetoSafe%
Key ≥ 40
Hostile%
Key ≤ 20
Pass% S1: Mayor
W:10 | B:34
S2: Sr CM
W:10 | B:28
S3: Atty
W:10 | B:38
S4: Sw CM
W:10 | B:56
S5: Eq CM
W:5 | B:68
S6: HOA
W:5 | B:25
S7: CivLib
W:5 | B:92
S8: OpenSp
W:2 | B:76
Python Pipeline

simulate_500_runs.py

Python script executing the 500 simulation runs and logging results to SQLite.

Download Python Script (.py)
Experimental Treatments

rhetorical_library_stone.json

JSON specification of the 10 Deborah Stone treatments and unlabeled podium speech stimuli.

Download Stone Library (.json)
ACF Control Baseline

Baseline Control Personas (.json)

JSON specification of the 8 anonymized ACF stakeholder personas, weights, and initial alignments.

Download Control Baseline (.json)
Run-Level Dataset (N=500)

simulation_runs_500.csv

500 simulation rows recording initial PPI, final PPI, delta, minimum key alignment, council vote tally, and pass status.

Download 500 Runs (.csv)
Shift-Level Dataset (N=4,000)

stakeholder_shifts_4000.csv

4,000 individual stakeholder shift rows (500 runs × 8 personas) with initial alignment, final alignment, delta, and stance.

Download 4,000 Shifts (.csv)
Relational Database

policy_experiment.db

SQLite database containing the runs (N=500) and stakeholder_shifts (N=4,000) tables.

Download SQLite DB (.db)