Replication repository for "Measuring What Matters for the Viability of Inclusive Insurance: Behavioural Diagnostics from 169 Community-Based Health Insurance Schemes Across 38 Countries" (programme version 3.18)

This repository provides the data, code, and documentation needed to reproduce every table and figure in the manuscript "Measuring What Matters for the Viability of Inclusive Insurance: Behavioural Diagnostics from 169 Community-Based Health Insurance Schemes Across 38 Countries", under peer review at The Geneva Papers on Risk and Insurance — Issues and Practice. The manuscript tests three behavioural pillars — Trust (T), Consensus (C), and Dual Benefit (DB) — combined multiplicatively, against the standard supply-side specification of government-health, external-aid, and out-of-pocket shares of health expenditure. The empirical test uses a registry of 169 documented community-based health insurance (CHI) schemes across 38 low- and middle-income countries, with an audit-clean analytical subset of 149 schemes in 31 countries. The repository contains: The 169-scheme registry with country, year, source documentation, and binary sustainability outcome Raw external data (Alesina et al. 2003 fractionalization indices, WHO Global Health Expenditure database, WHO Service Coverage Index) Three audit-traceable pillar datasets at country level: Trust (from primary survey sources documented per country in TRUST_EXTRACTION_PROTOCOL.md), Consensus (operationalised from Alesina fractionalization), and Dual Benefit (constructed from WHO SCI and out-of-pocket share of current health expenditure, WHO official 2021 series) Derived analytical datasets at country and scheme level Eleven R scripts, driven by a one-click entry point (replicate.R), that rebuild the analytical datasets and reproduce all six tables, three figures, the multiplicative-axiom tests, and the robustness battery from the audit-clean raw data A calibration-gated robustness script (Script 10) covering temporal bracketing, single-scheme-country exclusion, and scheme-count weighting; the gate must reproduce the audited baseline estimates from the deposited data before any check runs A numerical-claims audit script that verifies every statistic reported in the manuscript against the recomputed values, with PASS/FAIL flags Pre-computed outputs for verification The replication pipeline runs end-to-end in well under a minute via source("replicate.R"). The required R packages are listed in the README and are installed automatically on first run. The pipeline has been executed on Windows (R 4.6.1) and independently re-executed on Linux (R 4.3.3) with content-identical data outputs. Every numerical claim in the manuscript can be traced to a specific row, column, and computation in this repository.