Targeted Welfare Contrast Evaluations
RCTClinical Trial
Standard experimental evaluations measure aggregate average treatment effects, failing to quantify the welfare gains specifically attributable to targeting policies or adaptive treatment assignment relative to traditional uniform rollouts.
Picture this
Imagine evaluating a hospital network by comparing three policies: giving every patient the same average medicine, giving everyone no medicine, or assigning each patient group the specific medicine that works best for them. Welfare contrasts mathematically calculate the exact extra recovery percentage achieved by matching the right medicine to the right group compared to giving everyone a uniform prescription.
What the evidence says
The optimal targeted policy (Delta_2) generated a 1.7 percentage point employment gain over control (a 35% relative increase, 95% credible set [0.001, 0.034]), whereas the optimal non-targeted policy (Delta_3) produced only a 0.6 percentage point gain (95% credible set [-0.015, 0.027]).
- Who was studied
- N = 3,770 jobseekers (1,663 Syrian refugees and 2,107 Jordanians) in urban Amman, Irbid, and Mafraq, Jordan.
- How
- Comparison of three distinct welfare contrasts (Delta_1: adaptive vs random assignment; Delta_2: optimal targeted vs control; Delta_3: optimal non-targeted vs control) using Bayesian posterior expectations.
What to do
Compute stratum-weighted welfare contrasts comparing optimal conditional treatment assignments against non-targeted counterfactuals to isolate the empirical value of policy targeting.
From the source
"If we compare the optimal targeted policy to a counterfactual where no intervention is given (welfare contrast Delta_2), we estimate a gain in employment of 1.7 percentage points... The optimal non-targeted policy, on the other hand, delivers a gain in employment of about half of a percentage point."
An_Adaptive_Targeted_Field_Experiment_Job_Search_Assistance_for.pdf