## Citation and Research Question Doe, J., and Roe, A. (2021). Does text-message outreach raise benefit take-up? Evidence from a state pilot. *Journal of Policy Analysis and Management, 40*(3), 611 to 640. The authors ask whether low-cost reminder messages increase enrollment in a means-tested nutrition program among households already determined eligible (p. 612). The question matters because non-take-up among eligible households is large and persistent, and reminders are cheap relative to outreach workers (p. 613). ## Background and Contribution Earlier work documented that a third of eligible households never enroll, and attributed the gap to information costs, stigma, and administrative burden (p. 614). Field experiments on reminders in other domains (tax credits, college aid) found modest but reliable effects. No prior study randomized reminders at the household level inside an operating state system (p. 615). The authors claim two contributions: a clean household-level randomization, and administrative outcome data rather than self-report (p. 615). ## Data and Methods Administrative records from one state agency, 2018 to 2019, covering 41,220 households flagged eligible but unenrolled (p. 617). Unit of analysis is the household. Design is a randomized controlled trial: households were assigned to a control group, a single-message arm, or a three-message arm, with assignment stratified by county (p. 618). The outcome is enrollment within 90 days. Treatment is message arm. Controls include household size, prior enrollment history, and county fixed effects (p. 619). The authors pre-registered the analysis and report an intent-to-treat estimate; they handle undeliverable numbers by keeping those households in their assigned arm (p. 620). ## Key Findings The single-message arm raised 90-day enrollment by 2.4 percentage points over a control mean of 11.1 percent (95 percent CI 1.6 to 3.2; p. 624). The three-message arm raised it by 4.1 points (CI 3.3 to 4.9; p. 624). Effects were larger for households with a prior enrollment spell (5.8 points) than for first-time eligibles (2.9 points; p. 626). Effects did not differ by household size or urban versus rural county (p. 627). Cost per additional enrollment was 14 dollars in the three-message arm (p. 628). ## Policy Implications The authors recommend states adopt a three-message sequence as a default step before any in-person outreach (p. 631). My inference: the stronger effect among prior enrollees suggests the messages work mainly by reducing hassle for people who already understand the program, so the harder first-time population likely needs something beyond reminders. State benefit agencies and county offices are the actors who would implement this. ## Limitations and Critique Points The authors acknowledge a single state, a single program, and a 90-day window that may miss delayed enrollment (p. 633). Internal validity: assignment was clean, but message delivery was not verified, so the estimate blends the effect of receiving a message with the effect of being assigned one. The ITT framing is honest but understates the per-message effect. External validity: the pilot ran before a 2020 policy change that simplified the application. Reminder effects may be smaller when the form itself is easier. Measurement: enrollment is observed, but retention is not. A reminder that pulls in households who then churn out at recertification would look successful here and fail in practice. Alternative explanation: the three-message arm could be working through a second channel if household members forwarded messages; the paper cannot rule out spillover within households. Extension: a follow-up should track twelve-month retention and test a message that addresses the first-time population directly, for example by naming the single document most applicants fail to submit.