Reach the right people
Targeting
Targeting interventions identify a subset of a population with a different expected response — high-risk patients, eligible-but-not-enrolled families, frequent absentees — and direct an intervention there. The mechanism amplifies marginal effects by concentrating resources where they have the largest impact.
10
Experiments
2
Policy areas
1972–2019
Year span
6 / 10
Positive
When it works
When the population has heterogeneous response to the intervention and the targeting variable is observable. Predictive risk modeling has unlocked targeting in domains (child welfare, hospital readmissions) where it wasn't previously feasible.
Watch out for
Targeting can encode bias. If the model uses historical enforcement data, targeting follows historical enforcement patterns — not actual underlying risk. Validate the target population against ground truth, not just convenience proxies.
Targeting across policy areas
Public Safety· 8 experiments
- Positive
Minneapolis Foot Patrol Experiment
Minneapolis Police Department / George Mason University · Minneapolis, MN, USA · 2019
Effect: Violent crime at target segments: −22%; no statistically significant displacement detected within 2 blocks; non-violent crime: not significantly affected
- Null
Body-Worn Camera RCT — Washington DC
Metropolitan Police Department / George Mason University · Washington, DC, USA · 2016
Effect: No significant effect on use of force, complaints, arrests, or assaults; null across all pre-registered outcomes
- Positive
LEAD — Law Enforcement Assisted Diversion (Seattle)
Seattle Police Department / King County Prosecuting Attorney / Public Defender Association / Vera Institute / University of Washington evaluation · Seattle, Washington, USA · 2011
Effect: LEAD participants were 39% less likely than matched comparisons to be arrested for new felonies in the two-year follow-up window; 87% more likely to obtain housing; and significantly more likely to be receiving income from legitimate sources. System-cost savings (averted jail nights, prosecution, court costs) modestly exceeded program costs.
- Positive
Philadelphia Foot Patrol Experiment
Temple University / Philadelphia Police Department · Philadelphia, USA · 2009
Effect: −23% violent crime in foot patrol beats vs. control beats
- Positive
Hawaii's Opportunity Probation with Enforcement (HOPE)
First Circuit Court, State of Hawaii / RAND Corporation · Honolulu, HI, USA · 2004
Effect: HOPE vs. control: positive drug tests −72%; missed appointments −61%; new arrests −55%; days incarcerated −48% (fewer revocations meant less jail time overall despite more frequent brief sanctions)
- Positive
Philadelphia Hot Spots Policing
Philadelphia Police Department / Lawrence Sherman / Jerry Ratcliffe · Philadelphia, USA · 1995
Effect: −15% crime incidents at treated hot spots; no evidence of displacement to adjacent areas; 'diffusion of benefits' observed (crime fell in areas immediately adjacent to treated spots)
- Positive
Kansas City Gun Experiment — Directed Hot-Spots Patrol
University of Maryland / Kansas City Police Department · Kansas City, MO, USA · 1992
Effect: Gun crimes fell 49% in target beat during intervention; drive-by shootings fell from 9 to 1 during intervention period; gun seizures increased 65%; no significant increase in gun crime in surrounding beats (no displacement detected); comparison beat showed no significant change
- Null
Kansas City Preventive Patrol Experiment
Kansas City Police Department / Police Foundation · Kansas City, USA · 1972
Effect: No significant difference across conditions on any primary outcome
Public Health· 2 experiments
- Null
Camden Coalition Hotspotting RCT
MIT J-PAL North America / Camden Coalition of Healthcare Providers · Camden, New Jersey, USA · 2014
Effect: Readmission rate: 62.3% treatment vs. 61.7% control — a non-significant difference of 0.6 percentage points (95% CI: -7 to +8 pp). The intervention produced essentially no effect on the primary outcome. Total Medicaid spending and ED utilization were also unchanged.
- Mixed
Oregon Medicaid Lottery
Harvard University / MIT / NBER · Oregon, USA · 2008
Effect: Physical health: no significant effects on blood pressure, cholesterol, or diabetes management at 2 years. Mental health: significantly lower rate of positive depression screening (−9.2 pp). Financial security: significantly lower catastrophic medical expenses (−25%), lower medical debt, lower probability of borrowing money for medical care. Healthcare use: significantly higher (primary care, hospitalizations, prescriptions).