Upload folder using huggingface_hub

README.md

@@ -0,0 +1,139 @@
+---
+license: cc-by-4.0
+task_categories:
+  - tabular-classification
+  - tabular-regression
+language:
+  - en
+tags:
+  - health-financing
+  - out-of-pocket
+  - catastrophic-expenditure
+  - financial-protection
+  - synthetic
+  - healthcare
+  - sub-saharan-africa
+pretty_name: Out-of-Pocket Health Expenditure
+size_categories:
+  - 10K<n<100K
+configs:
+  - config_name: low_oop_insured
+    data_files: data/oop_low_oop_insured.csv
+  - config_name: moderate_oop_mixed
+    data_files: data/oop_moderate_oop_mixed.csv
+    default: true
+  - config_name: high_oop_unprotected
+    data_files: data/oop_high_oop_unprotected.csv
+---
+
+# Out-of-Pocket Health Expenditure
+
+## Abstract
+
+This synthetic dataset models household-level out-of-pocket (OOP) health spending across three sub-Saharan African scenarios: low OOP with strong insurance (Rwanda-like), moderate OOP with partial insurance (Ghana/Kenya-like), and high OOP with minimal financial protection (Nigeria/Chad-like). Each scenario contains 10,000 household records capturing OOP spending by category, catastrophic health expenditure, coping strategies, care-seeking behaviour, and impoverishment. Parameters are grounded in WHO AFRO financial protection reports, the Global Health Expenditure Database, and peer-reviewed literature.
+
+## Parameterization Evidence Table
+
+| Parameter | Value | Source | Year |
+|-----------|-------|--------|------|
+| SSA OOP as % CHE (median) | ~36% | WHO GHED | 2022 |
+| Rwanda OOP as % CHE | ~12% | WHO GHED / Exemplars | 2022 |
+| Nigeria OOP as % CHE | ~77% | WHO GHED | 2022 |
+| OOP >25% of health spending | 31 SSA countries | WHO AFRO UHC Report | 2023 |
+| OOP >50% of health spending | 11 SSA countries | WHO AFRO UHC Report | 2023 |
+| Catastrophic spending (>10%) | 95M people in Africa | WHO AFRO | 2019 |
+| Impoverishment by OOP | 150M+ in Africa | WHO AFRO | 2019 |
+| Average OOP per person (Africa) | ~$35/year | WHO AFRO | 2019 |
+| Medicines as % of OOP | 40-60% | Akazili et al. systematic review | 2022 |
+| CHE incidence range (SSA) | 1-48% (10% threshold) | PMC9047424 | 2022 |
+| SSA govt health budget share | <7% median (Abuja target 15%) | ODI / GHED | 2023 |
+
+## Scenario Design
+
+| Scenario | Exemplar | Insurance | OOP % CHE | Catastrophic | Mean OOP |
+|----------|----------|-----------|-----------|-------------|----------|
+| low_oop_insured | Rwanda/SA | 84% | 12% | 6.1% | $20 |
+| moderate_oop_mixed | Ghana/Kenya | 29% | 36% | 22.5% | $116 |
+| high_oop_unprotected | Nigeria/Chad | 5% | 72% | 36.3% | $271 |
+
+## Dataset Schema
+
+| Column | Type | Description |
+|--------|------|-------------|
+| id | int | Record identifier |
+| household_size | int | Household members |
+| ses_quintile | int | Wealth quintile (1=poorest) |
+| residence | cat | Urban/rural |
+| head_age | int | Household head age |
+| head_sex | cat | Household head sex |
+| has_insurance | binary | Health insurance coverage |
+| chronic_conditions | int | Number of chronic conditions |
+| household_income_usd | int | Annual income (USD) |
+| health_events_12m | int | Health events in past year |
+| sought_care | binary | Sought formal healthcare |
+| primary_service | cat | Main service type used |
+| facility_type | cat | Type of health facility |
+| oop_total_usd | float | Total annual OOP (USD) |
+| oop_medicines_usd | float | OOP on medicines |
+| oop_outpatient_usd | float | OOP on outpatient care |
+| oop_inpatient_usd | float | OOP on inpatient care |
+| oop_transport_usd | float | OOP on transport |
+| oop_diagnostic_usd | float | OOP on diagnostics |
+| catastrophic_expenditure | binary | OOP >10% of income |
+| impoverished_by_oop | binary | Pushed below poverty line |
+| oop_pct_income | float | OOP as % of income |
+| coping_strategy | cat | Primary coping mechanism |
+| forgone_care | binary | Needed but did not seek care |
+| year | int | Survey year (2020-2024) |
+
+## Validation
+
+![Validation Report](validation_report.png)
+
+## Usage
+
+```python
+from datasets import load_dataset
+ds = load_dataset("electricsheepafrica/oop-health-expenditure",
+                  name="moderate_oop_mixed")
+df = ds['train'].to_pandas()
+```
+
+```bash
+pip install -r requirements.txt
+python generate_dataset.py --all-scenarios --n 10000 --seed 42
+python validate_dataset.py
+```
+
+## Limitations
+
+- **Synthetic data**: All records are computationally generated and must not be used for clinical or policy decisions.
+- **Self-medication OOP**: Estimated for non-care-seekers using pharmacy/self-medication patterns; actual spending may vary.
+- **Income proxy**: Based on SES quintile assignment, not measured household survey data.
+- **Temporal**: Does not model year-on-year policy changes or COVID-19 spending shocks.
+
+## References
+
+1. WHO AFRO (2023). *Towards UHC: Tracking Financial Protection in the WHO African Region*. https://www.afro.who.int/publications/towards-universal-health-coverage
+2. WHO AFRO (2023). UHC Day Report. https://www.afro.who.int/news/uhc-day-high-health-care-costs
+3. WHO GHED (2022). https://apps.who.int/nha/database/
+4. Akazili J et al. (2022). Catastrophic health expenditure in SSA: systematic review. *BMJ Open* 12(4). PMC9047424
+5. ODI (2023). Health spending in SSA. https://odi.org/en/insights/what-do-we-know-about-health-spending
+6. Exemplars in Global Health (2023). Rwanda CBHI. https://www.exemplars.health/topics/primary-health-care/rwanda
+7. World Bank (2022). WDI. https://data.worldbank.org
+
+## Citation
+
+```bibtex
+@dataset{electricsheepafrica_oop_health_2025,
+  title   = {Out-of-Pocket Health Expenditure Dataset},
+  author  = {Electric Sheep Africa},
+  year    = {2025},
+  url     = {https://huggingface.co/datasets/electricsheepafrica/oop-health-expenditure},
+  license = {CC-BY-4.0}
+}
+```
+
+## License
+
+CC-BY-4.0

generate_dataset.py

@@ -0,0 +1,331 @@
+#!/usr/bin/env python3
+"""
+Literature-Informed Out-of-Pocket Health Expenditure Dataset
+=============================================================
+
+Each record = ONE household's annual out-of-pocket health spending profile.
+
+PHASE 1 — LITERATURE RESEARCH & EVIDENCE SYNTHESIS
+=====================================================
+Sources:
+[1] WHO AFRO (2023). Towards UHC: Tracking Financial Protection in
+    the WHO African Region. 150M+ pushed into poverty by OOP; 95M
+    making catastrophic payments (>10% of budget) in 2019.
+    URL: https://www.afro.who.int/publications/towards-universal-health-coverage
+
+[2] WHO AFRO (2023). UHC Day Report. OOP >25% of health spending in
+    31 SSA countries; >50% in 11; >70% in 3. Average $35/person/yr.
+    Medicines & outpatient care are main drivers. Rural, elderly HH
+    heads, multigenerational households most affected.
+    URL: https://www.afro.who.int/news/uhc-day-high-health-care-costs
+
+[3] WHO GHED (2022). OOP as % current health expenditure (CHE):
+    SSA median ~36%. Range: Botswana 5%, South Africa 8%, Rwanda 12%,
+    Ghana 37%, Kenya 24%, Nigeria 77%, Chad 66%.
+    URL: https://apps.who.int/nha/database/
+
+[4] Akazili et al. (2022). Catastrophic health expenditure in SSA:
+    systematic review. CHE incidence 1-48% across studies (10% threshold).
+    Impoverishment 1-28%. Medicines 40-60% of OOP. DOI: PMC9047424
+
+[5] ODI (2023). Health spending in SSA — GHED analysis. Govt health
+    expenditure rose 10% in 2020 (pandemic). Most countries <7% of
+    budget to health, far below 15% Abuja target.
+    URL: https://odi.org/en/insights/what-do-we-know-about-health-spending
+
+[6] Exemplars in Global Health (2023). Rwanda CBHI co-payment 10%
+    cap, OOP ~12% of CHE. $0.29 flat fee per visit.
+    URL: https://www.exemplars.health/topics/primary-health-care/rwanda
+
+[7] World Bank (2022). Poverty line: $2.15/day ($785/yr).
+    SSA average per-capita health expenditure ~$78.87.
+    URL: https://data.worldbank.org
+
+PHASE 2 — CAUSAL STRUCTURE (DAG)
+===================================
+ROOT: household_size, ses_quintile, residence, head_age, head_sex,
+      insurance_status, chronic_illness_count
+INTERMEDIATE: health_events_12m, care_sought, service_type,
+              facility_type, total_cost
+LEAF: oop_spending, oop_category_breakdown, catastrophic_expenditure,
+      impoverishment, coping_strategy
+
+PHASE 4 — SCENARIOS
+=====================
+1. low_oop_insured: Strong insurance + low OOP (Rwanda-like)
+2. moderate_oop_mixed: Partial insurance (Ghana/Kenya-like)
+3. high_oop_unprotected: Minimal insurance, high OOP (Nigeria/Chad-like)
+"""
+
+import numpy as np
+import pandas as pd
+import argparse
+import os
+
+SCENARIOS = {
+    'low_oop_insured': {
+        'description': 'Strong insurance, low OOP (Rwanda-like)',
+        'exemplar': 'Rwanda/South Africa',
+        'oop_pct_che': 0.12,           # [3] Rwanda 12%
+        'mean_oop_usd': 18,            # [2] well below $35 avg
+        'sd_oop_usd': 15,
+        'insurance_rate': 0.85,         # [6] Rwanda >85%
+        'catastrophic_10pct_rate': 0.04,# [1] low with insurance
+        'impoverishment_rate': 0.03,
+        'medicines_share': 0.35,        # [4] lower with covered meds
+        'outpatient_share': 0.30,
+        'inpatient_share': 0.15,
+        'transport_share': 0.10,
+        'diagnostic_share': 0.10,
+        'chronic_prevalence': 0.18,
+        'care_seeking_rate': 0.80,      # [6] insured 2x utilisation
+    },
+    'moderate_oop_mixed': {
+        'description': 'Partial insurance, moderate OOP (Ghana/Kenya-like)',
+        'exemplar': 'Ghana/Kenya',
+        'oop_pct_che': 0.36,           # [3] SSA median
+        'mean_oop_usd': 45,            # near SSA average
+        'sd_oop_usd': 40,
+        'insurance_rate': 0.30,
+        'catastrophic_10pct_rate': 0.14,# [4] mid-range
+        'impoverishment_rate': 0.10,
+        'medicines_share': 0.45,        # [4] medicines dominant
+        'outpatient_share': 0.25,
+        'inpatient_share': 0.12,
+        'transport_share': 0.10,
+        'diagnostic_share': 0.08,
+        'chronic_prevalence': 0.22,
+        'care_seeking_rate': 0.55,
+    },
+    'high_oop_unprotected': {
+        'description': 'Minimal insurance, very high OOP (Nigeria/Chad-like)',
+        'exemplar': 'Nigeria/Chad',
+        'oop_pct_che': 0.72,           # [3] Nigeria 77%
+        'mean_oop_usd': 80,
+        'sd_oop_usd': 70,
+        'insurance_rate': 0.05,
+        'catastrophic_10pct_rate': 0.28,# [4] high end
+        'impoverishment_rate': 0.20,
+        'medicines_share': 0.55,        # [4] 40-60% medicines
+        'outpatient_share': 0.20,
+        'inpatient_share': 0.08,
+        'transport_share': 0.12,
+        'diagnostic_share': 0.05,
+        'chronic_prevalence': 0.25,
+        'care_seeking_rate': 0.35,
+    },
+}
+
+SERVICE_TYPES = ['outpatient_consultation', 'inpatient_admission', 'medicines_only',
+                 'diagnostic_test', 'maternal_care', 'dental_care',
+                 'chronic_disease_management', 'emergency_care']
+
+FACILITY_TYPES = ['public_hospital', 'public_health_centre', 'private_hospital',
+                  'private_clinic', 'pharmacy', 'traditional_healer', 'none']
+
+COPING_STRATEGIES = ['savings', 'borrowing', 'selling_assets', 'reducing_food',
+                     'community_support', 'delayed_treatment', 'no_coping_needed',
+                     'forgoing_care']
+
+
+def generate_dataset(n=10000, seed=42, scenario='moderate_oop_mixed'):
+    rng = np.random.default_rng(seed)
+    sc = SCENARIOS[scenario]
+    records = []
+
+    for idx in range(n):
+        rec = {'id': idx + 1}
+
+        # ── ROOT NODES ──────────────────────────────────────────────
+        rec['household_size'] = max(1, int(rng.poisson(4.8)))
+        rec['ses_quintile'] = int(rng.choice([1, 2, 3, 4, 5],
+            p=[0.25, 0.22, 0.20, 0.18, 0.15]))
+        rec['residence'] = rng.choice(['urban', 'rural'], p=[0.38, 0.62])
+        rec['head_age'] = int(np.clip(rng.normal(40, 14), 18, 85))
+        rec['head_sex'] = rng.choice(['male', 'female'], p=[0.62, 0.38])
+
+        # Insurance status — conditional on scenario & SES
+        ins_prob = sc['insurance_rate'] + (rec['ses_quintile'] - 3) * 0.04
+        ins_prob = np.clip(ins_prob, 0.01, 0.98)
+        rec['has_insurance'] = 1 if rng.random() < ins_prob else 0
+
+        # Chronic conditions
+        rec['chronic_conditions'] = int(rng.poisson(
+            sc['chronic_prevalence'] * (1 + (rec['head_age'] - 30) / 80)))
+        rec['chronic_conditions'] = min(rec['chronic_conditions'], 5)
+
+        # Annual household income (USD)
+        income_base = {1: 280, 2: 550, 3: 1100, 4: 2400, 5: 5500}
+        rec['household_income_usd'] = max(50, int(rng.normal(
+            income_base[rec['ses_quintile']],
+            income_base[rec['ses_quintile']] * 0.30)))
+
+        # ── INTERMEDIATE NODES ──────────────────────────────────────
+
+        # Number of health events in 12 months
+        event_lambda = 1.5 + rec['chronic_conditions'] * 0.8
+        rec['health_events_12m'] = max(0, int(rng.poisson(event_lambda)))
+
+        # Care seeking — conditional on insurance, SES
+        seek_prob = sc['care_seeking_rate']
+        if rec['has_insurance']:
+            seek_prob = min(seek_prob + 0.15, 0.95)
+        if rec['ses_quintile'] <= 2:
+            seek_prob = max(seek_prob - 0.10, 0.10)
+        rec['sought_care'] = 1 if (rec['health_events_12m'] > 0 and
+                                   rng.random() < seek_prob) else 0
+
+        # Service type (most recent episode)
+        if rec['sought_care']:
+            if rec['chronic_conditions'] > 0 and rng.random() < 0.4:
+                rec['primary_service'] = 'chronic_disease_management'
+            else:
+                svc_p = np.array([0.30, 0.08, 0.25, 0.10, 0.10, 0.02, 0.05, 0.10])
+                svc_p = svc_p / svc_p.sum()
+                rec['primary_service'] = rng.choice(SERVICE_TYPES, p=svc_p)
+
+            # Facility type
+            if rec['has_insurance']:
+                fac_p = np.array([0.30, 0.35, 0.15, 0.10, 0.08, 0.02, 0.0])
+            elif rec['ses_quintile'] >= 4:
+                fac_p = np.array([0.15, 0.10, 0.30, 0.25, 0.15, 0.05, 0.0])
+            else:
+                fac_p = np.array([0.15, 0.25, 0.05, 0.10, 0.30, 0.10, 0.05])
+            fac_p = fac_p / fac_p.sum()
+            rec['facility_type'] = rng.choice(FACILITY_TYPES, p=fac_p)
+        else:
+            rec['primary_service'] = 'none'
+            rec['facility_type'] = 'none'
+
+        # ── OOP SPENDING CALCULATION ────────────────────────────────
+
+        rec['oop_total_usd'] = 0.0
+        rec['oop_medicines_usd'] = 0.0
+        rec['oop_outpatient_usd'] = 0.0
+        rec['oop_inpatient_usd'] = 0.0
+        rec['oop_transport_usd'] = 0.0
+        rec['oop_diagnostic_usd'] = 0.0
+
+        if rec['sought_care']:
+            # Base cost depends on service and facility
+            svc_cost_mult = {
+                'outpatient_consultation': 1.0, 'inpatient_admission': 5.0,
+                'medicines_only': 0.6, 'diagnostic_test': 1.2,
+                'maternal_care': 3.0, 'dental_care': 2.0,
+                'chronic_disease_management': 1.8, 'emergency_care': 4.0, 'none': 0}
+            fac_cost_mult = {
+                'public_hospital': 1.0, 'public_health_centre': 0.5,
+                'private_hospital': 3.0, 'private_clinic': 2.0,
+                'pharmacy': 0.4, 'traditional_healer': 0.3, 'none': 0}
+
+            base = max(2, rng.lognormal(np.log(sc['mean_oop_usd']), 0.8))
+            base *= svc_cost_mult.get(rec['primary_service'], 1.0)
+            base *= fac_cost_mult.get(rec['facility_type'], 1.0)
+
+            # Insurance reduces OOP
+            if rec['has_insurance']:
+                base *= 0.20  # insurance covers ~80%
+
+            # Multiple events scale (diminishing)
+            base *= (1 + 0.3 * min(rec['health_events_12m'] - 1, 5))
+
+            total = max(0, base)
+            rec['oop_total_usd'] = round(total, 2)
+
+            # Breakdown by category [4]
+            rec['oop_medicines_usd'] = round(total * sc['medicines_share'], 2)
+            rec['oop_outpatient_usd'] = round(total * sc['outpatient_share'], 2)
+            rec['oop_inpatient_usd'] = round(total * sc['inpatient_share'], 2)
+            rec['oop_transport_usd'] = round(total * sc['transport_share'], 2)
+            rec['oop_diagnostic_usd'] = round(total * sc['diagnostic_share'], 2)
+
+        elif rec['health_events_12m'] > 0:
+            # Self-medication / pharmacy OOP for those who didn't seek formal care
+            # [2] Medicines are primary OOP driver even without formal care
+            self_med = max(0, rng.lognormal(
+                np.log(sc['mean_oop_usd'] * 0.4), 0.9))
+            self_med *= rec['health_events_12m']
+            rec['oop_total_usd'] = round(self_med, 2)
+            rec['oop_medicines_usd'] = round(self_med * 0.85, 2)
+            rec['oop_transport_usd'] = round(self_med * 0.15, 2)
+
+        # ── LEAF NODES ──────────────────────────────────────────────
+
+        # Catastrophic health expenditure (>10% of income) [1]
+        rec['catastrophic_expenditure'] = 1 if (
+            rec['household_income_usd'] > 0 and
+            rec['oop_total_usd'] / rec['household_income_usd'] > 0.10) else 0
+
+        # Impoverishment (pushed below $2.15/day poverty line)
+        poverty_line = 785  # $2.15/day * 365
+        rec['impoverished_by_oop'] = 1 if (
+            rec['household_income_usd'] >= poverty_line and
+            (rec['household_income_usd'] - rec['oop_total_usd']) < poverty_line) else 0
+
+        # OOP as % of income
+        rec['oop_pct_income'] = round(
+            rec['oop_total_usd'] / max(1, rec['household_income_usd']) * 100, 1)
+
+        # Coping strategy
+        if rec['oop_total_usd'] <= 0:
+            rec['coping_strategy'] = 'no_coping_needed'
+        elif rec['oop_pct_income'] < 5:
+            rec['coping_strategy'] = rng.choice(
+                ['savings', 'no_coping_needed'], p=[0.4, 0.6])
+        elif rec['oop_pct_income'] < 15:
+            rec['coping_strategy'] = rng.choice(
+                ['savings', 'borrowing', 'community_support', 'reducing_food'],
+                p=[0.35, 0.30, 0.15, 0.20])
+        else:
+            rec['coping_strategy'] = rng.choice(COPING_STRATEGIES,
+                p=[0.10, 0.25, 0.15, 0.20, 0.05, 0.15, 0.0, 0.10])
+
+        # Forgone care (didn't seek despite needing)
+        rec['forgone_care'] = 1 if (
+            rec['health_events_12m'] > 0 and not rec['sought_care']) else 0
+
+        rec['year'] = rng.choice([2020, 2021, 2022, 2023, 2024],
+                                 p=[0.10, 0.15, 0.20, 0.25, 0.30])
+
+        records.append(rec)
+
+    df = pd.DataFrame(records)
+
+    # ── Summary ─────────────────────────────────────────────────────
+    print(f"\n{'='*65}")
+    print(f"OOP Health Expenditure — {scenario}")
+    print(f"  Exemplar: {sc['exemplar']} | n={n} | seed={seed}")
+    print(f"{'='*65}")
+    seekers = df[df['sought_care'] == 1]
+    print(f"  Insurance rate:        {df['has_insurance'].mean()*100:.1f}%")
+    print(f"  Care seeking:          {df['sought_care'].mean()*100:.1f}%")
+    print(f"  Mean OOP (seekers):    ${seekers['oop_total_usd'].mean():.1f}")
+    print(f"  Median OOP (seekers):  ${seekers['oop_total_usd'].median():.1f}")
+    print(f"  Catastrophic (>10%):   {df['catastrophic_expenditure'].mean()*100:.1f}%"
+          f"  (target ~{sc['catastrophic_10pct_rate']*100:.0f}%)")
+    print(f"  Impoverished:          {df['impoverished_by_oop'].mean()*100:.1f}%"
+          f"  (target ~{sc['impoverishment_rate']*100:.0f}%)")
+    print(f"  Forgone care:          {df['forgone_care'].mean()*100:.1f}%")
+    return df
+
+
+if __name__ == '__main__':
+    parser = argparse.ArgumentParser(
+        description='Generate OOP Health Expenditure Dataset')
+    parser.add_argument('--all-scenarios', action='store_true')
+    parser.add_argument('--n', type=int, default=10000)
+    parser.add_argument('--seed', type=int, default=42)
+    args = parser.parse_args()
+
+    os.makedirs('data', exist_ok=True)
+
+    if args.all_scenarios:
+        for sc_name in SCENARIOS:
+            df = generate_dataset(n=args.n, seed=args.seed, scenario=sc_name)
+            fname = os.path.join('data', f'oop_{sc_name}.csv')
+            df.to_csv(fname, index=False)
+            print(f"  -> Saved to {fname}\n")
+    else:
+        df = generate_dataset(n=args.n, seed=args.seed)
+        df.to_csv(os.path.join('data', 'oop_moderate_oop_mixed.csv'), index=False)
+        print("  -> Saved\n")

requirements.txt

@@ -0,0 +1,3 @@
+numpy>=1.24
+pandas>=2.0
+matplotlib>=3.7

validate_dataset.py

@@ -0,0 +1,178 @@
+#!/usr/bin/env python3
+"""Validation & Visualization for OOP Health Expenditure Dataset."""
+
+import pandas as pd
+import numpy as np
+import matplotlib.pyplot as plt
+import os
+import glob
+
+def load_scenarios(data_dir='data'):
+    dfs = {}
+    for f in sorted(glob.glob(os.path.join(data_dir, 'oop_*.csv'))):
+        basename = os.path.basename(f).replace('.csv', '')
+        name = basename[4:]  # strip leading 'oop_' prefix only
+        dfs[name] = pd.read_csv(f)
+    return dfs
+
+def main():
+    dfs = load_scenarios()
+    if not dfs:
+        print("No CSV files found in data/. Run generate_dataset.py first.")
+        return
+
+    all_df = pd.concat(
+        [df.assign(scenario=name) for name, df in dfs.items()], ignore_index=True)
+
+    fig, axes = plt.subplots(4, 2, figsize=(16, 20))
+    fig.suptitle('Out-of-Pocket Health Expenditure — Validation Report',
+                 fontsize=14, fontweight='bold', y=0.98)
+    colors = {'low_oop_insured': '#2ecc71',
+              'moderate_oop_mixed': '#f39c12',
+              'high_oop_unprotected': '#e74c3c'}
+    labels = {'low_oop_insured': 'Low OOP (Rwanda-like)',
+              'moderate_oop_mixed': 'Moderate OOP (Ghana/Kenya)',
+              'high_oop_unprotected': 'High OOP (Nigeria/Chad)'}
+    scenarios = list(dfs.keys())
+
+    # Panel 1: OOP spending distribution (truncated)
+    ax = axes[0, 0]
+    for s in scenarios:
+        vals = dfs[s].loc[dfs[s]['oop_total_usd'] > 0, 'oop_total_usd'].clip(upper=300)
+        if len(vals) > 0:
+            ax.hist(vals, bins=50, alpha=0.5, label=labels.get(s, s),
+                    color=colors[s], density=True)
+    ax.set_xlabel('OOP Spending (USD, capped at $300)')
+    ax.set_ylabel('Density')
+    ax.set_title('Panel 1: OOP Spending Distribution')
+    ax.legend(fontsize=7)
+
+    # Panel 2: Catastrophic expenditure by scenario
+    ax = axes[0, 1]
+    cat_rates = [dfs[s]['catastrophic_expenditure'].mean() * 100 for s in scenarios]
+    targets = [4, 14, 28]
+    x = np.arange(len(scenarios))
+    ax.bar(x - 0.15, cat_rates, 0.3, label='Observed',
+           color=[colors[s] for s in scenarios], alpha=0.8)
+    ax.bar(x + 0.15, targets, 0.3, label='Target', color='grey', alpha=0.5)
+    ax.set_xticks(x)
+    ax.set_xticklabels([labels.get(s, s) for s in scenarios], fontsize=7, rotation=15)
+    ax.set_ylabel('Catastrophic Expenditure (%)')
+    ax.set_title('Panel 2: Catastrophic Expenditure Rate')
+    ax.legend(fontsize=8)
+    for i, (o, t) in enumerate(zip(cat_rates, targets)):
+        ax.text(i - 0.15, o + 0.5, f'{o:.1f}%', ha='center', fontsize=7)
+
+    # Panel 3: OOP breakdown by category
+    ax = axes[1, 0]
+    cats = ['medicines', 'outpatient', 'inpatient', 'transport', 'diagnostic']
+    width = 0.25
+    for i, s in enumerate(scenarios):
+        d = dfs[s]
+        seekers = d[d['oop_total_usd'] > 0]
+        if len(seekers) == 0:
+            continue
+        total = seekers['oop_total_usd'].sum()
+        shares = [seekers[f'oop_{c}_usd'].sum() / total * 100 for c in cats]
+        ax.bar(np.arange(len(cats)) + i * width, shares, width,
+               label=labels.get(s, s), color=colors[s], alpha=0.8)
+    ax.set_xticks(np.arange(len(cats)) + width)
+    ax.set_xticklabels([c.capitalize() for c in cats], fontsize=8)
+    ax.set_ylabel('Share of Total OOP (%)')
+    ax.set_title('Panel 3: OOP Breakdown by Category')
+    ax.legend(fontsize=7)
+
+    # Panel 4: OOP by SES quintile
+    ax = axes[1, 1]
+    for s in scenarios:
+        means = dfs[s].groupby('ses_quintile')['oop_total_usd'].mean()
+        ax.plot(means.index, means.values, 'o-', label=labels.get(s, s),
+                color=colors[s], linewidth=2)
+    ax.set_xlabel('SES Quintile (1=poorest)')
+    ax.set_ylabel('Mean OOP (USD)')
+    ax.set_title('Panel 4: Mean OOP by Wealth Quintile')
+    ax.legend(fontsize=8)
+
+    # Panel 5: Coping strategies
+    ax = axes[2, 0]
+    strats = ['savings', 'borrowing', 'selling_assets', 'reducing_food',
+              'community_support', 'delayed_treatment', 'forgoing_care']
+    for i, s in enumerate(scenarios):
+        d = dfs[s]
+        counts = [d['coping_strategy'].value_counts().get(st, 0) / len(d) * 100
+                  for st in strats]
+        ax.bar(np.arange(len(strats)) + i * 0.25, counts, 0.25,
+               label=labels.get(s, s), color=colors[s], alpha=0.8)
+    ax.set_xticks(np.arange(len(strats)) + 0.25)
+    ax.set_xticklabels([s.replace('_', '\n') for s in strats], fontsize=6)
+    ax.set_ylabel('Percentage (%)')
+    ax.set_title('Panel 5: Coping Strategies')
+    ax.legend(fontsize=7)
+
+    # Panel 6: Cross-scenario key metrics
+    ax = axes[2, 1]
+    metrics = ['Insurance %', 'Care Seeking %', 'Catastrophic %',
+               'Forgone Care %']
+    for i, s in enumerate(scenarios):
+        d = dfs[s]
+        vals = [d['has_insurance'].mean() * 100,
+                d['sought_care'].mean() * 100,
+                d['catastrophic_expenditure'].mean() * 100,
+                d['forgone_care'].mean() * 100]
+        ax.bar(np.arange(len(metrics)) + i * 0.25, vals, 0.25,
+               label=labels.get(s, s), color=colors[s], alpha=0.8)
+    ax.set_xticks(np.arange(len(metrics)) + 0.25)
+    ax.set_xticklabels(metrics, fontsize=8)
+    ax.set_ylabel('Percentage (%)')
+    ax.set_title('Panel 6: Cross-Scenario Key Metrics')
+    ax.legend(fontsize=7)
+
+    # Panel 7: OOP as % of income distribution
+    ax = axes[3, 0]
+    for s in scenarios:
+        vals = dfs[s].loc[dfs[s]['oop_pct_income'] > 0, 'oop_pct_income'].clip(upper=50)
+        if len(vals) > 0:
+            ax.hist(vals, bins=50, alpha=0.5, label=labels.get(s, s),
+                    color=colors[s], density=True)
+    ax.axvline(10, color='red', linestyle='--', alpha=0.7, label='10% threshold')
+    ax.set_xlabel('OOP as % of Income')
+    ax.set_ylabel('Density')
+    ax.set_title('Panel 7: OOP as % of Household Income')
+    ax.legend(fontsize=7)
+
+    # Panel 8: Correlation heatmap
+    ax = axes[3, 1]
+    num_cols = ['household_size', 'ses_quintile', 'has_insurance',
+                'chronic_conditions', 'oop_total_usd', 'household_income_usd',
+                'catastrophic_expenditure', 'forgone_care']
+    corr = all_df[num_cols].corr()
+    im = ax.imshow(corr, cmap='RdBu_r', vmin=-1, vmax=1, aspect='auto')
+    ax.set_xticks(range(len(num_cols)))
+    ax.set_yticks(range(len(num_cols)))
+    ax.set_xticklabels([c.replace('_', '\n') for c in num_cols], fontsize=5,
+                       rotation=45, ha='right')
+    ax.set_yticklabels([c.replace('_', '\n') for c in num_cols], fontsize=5)
+    ax.set_title('Panel 8: Correlation Heatmap')
+    fig.colorbar(im, ax=ax, fraction=0.046)
+    for i in range(len(num_cols)):
+        for j in range(len(num_cols)):
+            ax.text(j, i, f'{corr.iloc[i, j]:.2f}', ha='center', va='center',
+                    fontsize=4.5, color='white' if abs(corr.iloc[i, j]) > 0.5 else 'black')
+
+    plt.tight_layout(rect=[0, 0, 1, 0.96])
+    plt.savefig('validation_report.png', dpi=150, bbox_inches='tight')
+    plt.close()
+    print("Saved validation_report.png")
+
+    print("\n=== VALIDATION SUMMARY ===")
+    for s in scenarios:
+        d = dfs[s]
+        seekers = d[d['sought_care'] == 1]
+        print(f"\n{labels.get(s, s)}:")
+        print(f"  Insurance: {d['has_insurance'].mean()*100:.1f}%")
+        print(f"  Mean OOP (seekers): ${seekers['oop_total_usd'].mean():.1f}")
+        print(f"  Catastrophic: {d['catastrophic_expenditure'].mean()*100:.1f}%")
+        print(f"  Forgone care: {d['forgone_care'].mean()*100:.1f}%")
+
+if __name__ == '__main__':
+    main()