What the evidence says about health interventions.

Health insurance expands access without reliably improving measurable physical health outcomes.

The RAND Health Insurance Experiment — the most expensive social science experiment ever conducted — found that free care increased utilization 40% but produced no significant improvement in health outcomes for the average enrollee. Fifty years later, the Oregon Medicaid Lottery reached the same conclusion: gaining Medicaid coverage increased ED visits and reduced financial strain, but produced no statistically significant improvement in blood pressure, cholesterol, or blood sugar control after two years. Mental health was the exception — both studies found coverage significantly improved psychological wellbeing. The evidence suggests that what people use insurance for often doesn't produce the health improvements that motivated the coverage expansion.

Low-cost point-of-use interventions outperform coverage expansions on cost-effectiveness.

Chlorine dispensers placed at community water sources in Kenya increased water treatment from near-zero to 53% of households at a cost of roughly $3 per household per year — and reduced diarrheal illness by 4.5 percentage points. Vaccination reminders paired with small incentives in Haryana, India increased full immunization rates by 18 percentage points in a randomized trial. A 2024 meta-analysis of SMS vaccination reminders across low- and middle-income countries found consistent 5–10 percentage point improvements in childhood vaccination coverage. The pattern: interventions that reduce friction at the moment of health action (attending a vaccination clinic, treating drinking water) produce reliable improvements at low cost. Interventions that expand financial access to care produce more uncertain effects.

Defaults and framing shift health behavior at population scale without mandates or financial incentives.

Cross-national evidence on organ donation opt-out laws found that countries with presumed consent (opt-out) achieve donation rates 25–30 percentage points higher than opt-in countries — one of the largest default effects documented in any domain. In a 2022 RCT on COVID-19 booster uptake, messages framing the vaccine as 'your dose' (ownership frame) outperformed control and other message variants. By contrast, a 700,000-person SMS experiment in Rhode Island found that off-the-shelf text reminders with no personalization produced no increase in vaccination rates — the effect size was precisely estimated at near zero. The implication: behavioral interventions work when they target the specific friction at the decision point, not as generic nudges.

Community-delivered and team-based care models substantially outperform individual referral models.

The IMPACT Collaborative Care trial randomized 1,801 older primary care patients with depression to team-based care (care manager, primary care physician, psychiatrist consultation) vs. usual referral. At 12 months, 45% of IMPACT patients responded to treatment vs. 19% in usual care — a difference that persisted at 24 months and produced cost savings. The IPS Supported Employment model for people with serious mental illness achieves competitive employment rates of 55–60% vs. 20–25% for traditional vocational rehabilitation, replicated across 20+ RCTs in multiple countries. Community-level mask promotion across 600 villages in Bangladesh increased mask adherence from 13% to 59% and reduced symptomatic COVID-19. Scale and embeddedness in local context consistently amplifies effect sizes in public health.

COVID-19 Vaccination SMS Reminders

Rhode Island, USA · 2021

Generic SMS reminders at scale produced no measurable vaccination increase. The friction wasn't information — most eligible recipients already intended to get vaccinated. The limiting factor was appointment access and logistics, not awareness.

Clean Cookstove Adoption in Rural India

India (Orissa state) · 2016

Subsidized clean cookstoves were adopted initially but largely abandoned within 4 years. Survey-based demand estimates dramatically overstated actual usage. The lesson: stated preferences and revealed preferences diverge sharply for products requiring behavior change.

Workplace Wellness Programs — Illinois RCT

University of Illinois, USA · 2016

A 4,834-employee randomized trial found that workplace wellness programs produced no improvement in health behaviors, health outcomes, or medical costs over three years, despite high participation rates. Screened-in populations may differ from the intended population.