# Design codebook You are coding the design of trials for a review of randomized trials of **unconditional cash in high-income countries** and its effects on recipients' **mental health** and **employment**. For each trial you get the trial's id and the local text files of its reports (reports, working papers, articles, protocols, registrations). Read what you need from them; search within them freely. Work alone. Don't look anything up elsewhere, and don't read numbers of outcome results except as this codebook asks (sample sizes and timing are design facts; effect estimates are not needed here). ## 1. Is the trial eligible? All of these must hold: 1. **Randomized**: people or households were assigned to arms at random, including a lottery among eligible applicants. A comparison group formed any other way (matched, waitlisted by date, chosen by staff, people who applied too late) is not randomized. 2. **Unconditional cash**: at least one arm got money without conditions on what recipients do (no work, training, school, health, or spending requirement), by cash, bank transfer, or a card usable like cash (not one restricted to food or to some kinds of purchase). 3. **Flat**: the amount didn't fall as the recipient's earnings or income rose during the trial. A negative income tax, a top-up to a threshold, or a benefit withdrawn as earnings rise is not flat. (Eligibility by low income at enrolment is fine.) 4. **A control arm** that got no cash, a token amount (for example a payment for completing surveys, or $50 a month), or the usual benefits. 5. **A high-income country** (World Bank classification at the time of the trial). 6. **Mental health or employment** of the people paid (or the adult the payment was meant for) is among the outcomes the trial's reports present, or plan to present. If any fails, code `eligible: false` with `reason` one of: `not_randomized`, `not_unconditional_cash`, `not_flat`, `no_control_arm`, `not_high_income`, `no_eligible_outcome`, `not_a_trial` (the files describe something other than a trial), `other` (say what). ## 2. For an eligible trial, code - `country`, `sites` (cities or regions), `population` (a few words: who could enrol), `years` (when payments ran, "YYYY-MM to YYYY-MM"). - `arms`: every arm, each `{"label": the report's name for it, "kind": "recurring" | "lump_sum" | "control", "amount": number, "currency": "USD" | "EUR" | ..., "per": "month" | "week" | "two_weeks" | "once" | "year", "months_paid": number of months over which payments were made (a lump sum: 0), "n_randomized": number or null}`. A recurring arm that started with a lump sum and then paid monthly is `recurring`; give its monthly amount and note the lump sum in `notes`. A control arm's `amount` is its token payment per `per` (0 if none). - `comparisons`: each eligible comparison of a cash arm against a control arm, `{"id": "-", "treated": arm label, "control": arm label, "kind": "recurring" | "lump_sum"}`. A trial with two cash arms and one control has two comparisons. A comparison between two cash arms is not one of these. - `randomization`: `"individual"`, `"household"`, or `"cluster"`, and `stratified_by` if the report says. - `payment_start`: when the first payment was made, "YYYY-MM" (or months after enrolment if only that is given), and `payment_months`: how many months payments lasted for recurring arms. - `waves`: every follow-up survey or data collection after payments began, each `{"label": the report's name for it (for example "12-month"), "months_since_first_payment": number, "during_payments": true if payments were still being made at that time or had ended less than one month before, otherwise false, "measures": ["mental_health", "employment", "stress", "life_satisfaction", "hours", "earnings"] — those the reports present results for at that wave}`. Use the months the report states; if it gives only dates, compute months from the first payment. - `mental_health_instruments`: the instruments the reports use for mental health, named as in the report (for example "Kessler 6 (K6)", "Kessler 10", "PHQ-9", "PHQ-8", "CES-D 10", "GAD-7", "MHI-5", "SF-12 mental component", "WHO-5", "an index of …"). - `employment_measures`: how the reports measure work (for example "employed (any paid work)", "employed full-time", "hours per week", "earnings", "days employed from registers"). - `evaluator`: the organization(s) that ran the evaluation, as the report names them. - `result_reports`: the ids of the files that hold the trial's outcome results for mental health and employment (the most complete first). - `notes`: anything else that matters for the comparison (cash added to services both arms got; payments to only some household members; arms that changed during the trial). ## Output A JSON array, one object per trial in the order given: `{"trial": "", "eligible": true|false, "reason": "", "why": "", ... the fields above if eligible}`. Use `null` where the reports don't say.