under reviewAI Generatedfintech The Cash Transfer Design Trial (CTDT): One Factorial RCT, Same Total Dollars, Randomized Structure
Fund a single, pre-registered, multi-site randomized controlled trial that holds total transfer value and target population constant while randomizing only the two variables the field keeps conflating: payment structure (lump sum vs. equal-value monthly installments) and duration (a 2-year horizon vs. a 10-year-plus horizon). Run it as a 2x2 factorial design across at least three sites chosen for economic diversity — a low-income rural setting, a middle-income urban setting, and a high-income welfare-state setting — so results aren't just an artifact of one economy's labor market. Every arm receives the identical total nominal transfer value (adjusted for local purchasing power) by the point of the first cross-arm comparison, which is the design flaw that made Kenya's own three-arm study only partially conclusive: real-world implementers have never combined this kind of structural randomization with genuine multi-site replication.
The mechanism is straightforward, not novel research methodology — it's a coordination problem, not a science problem. A consortium of 2-3 existing cash-transfer implementers (GiveDirectly-type organizations, a national social ministry, a philanthropic guaranteed-income funder) would need to agree to a shared pre-registered protocol, shared outcome battery (employment status, enterprise formation, business revenue, financial stability, subjective wellbeing, measured at 12, 24, and 60 months), and independent third-party evaluation with a single analysis plan locked before data collection. The single hardest design element — informed consent for randomized payment structure — gets handled by recruiting only from populations who consent to structure randomization specifically in exchange for a modest structure-randomization stipend, which addresses the ethical concern that people might strongly prefer one structure over another.
What this buys policymakers that no current study can: a direct, apples-to-apples answer to "if we have $X to spend per recipient, should we pay it out as a lump sum or spread it over 24 months or spread it over 10 years?" — the actual question every legislature designing a guaranteed income program has to answer and currently cannot, because no existing dataset varies structure while holding total value and population fixed.
