Financial Distress and the Cost of Learning: Evidence from a Field Experiment

Workers are individually randomly assigned to one of two treatment arms. In the cash-relief arm (Arm 1, n = 200), each worker receives a cash advance of $38, approximately five days of earnings and 2–3 weeks of typical lean-season salary delivered privately at their home two days before the work session. The advance is framed as an early installment of their expected compensation, consistent with occasional company practice. Workers in this arm receive the remaining expected earnings plus their actual piece-rate earnings at the end of the session; total compensation is identical across arms, with only the timing of the first installment varying. In the control arm (Arm 2, n = 200), workers receive no advance and no unusual contact before the session; all compensation is paid at the end of the session, so they arrive in their natural lean-season financially stressed state. On arrival, all workers are offered a binary technology choice: (A) manual corn shelling by hand, which requires no preparation and can begin immediately; or (B) a mechanical corn sheller, which yields higher piece-rate output but requires self-assembly after watching a five-minute instructional video on a tablet. The machine is provided at zero monetary cost. Workers may switch back to manual shelling at any time; pay depends solely on total output regardless of method. This design holds liquidity, information, and compensation constant across arms, isolating the cognitive margin of technology adoption.

2027-04-01

2027-05-31

Primary outcome 1: Initial entry into the learning-intensive technology. This is an indicator equal to one if the participant chooses the mechanical sheller rather than manual shelling at the initial task-choice point in the second work session.

Primary outcome 2: Post-assignment machine use. This is an indicator equal to one if the participant uses or chooses to continue using the mechanical sheller at the pre-specified post-assignment choice point after randomized initial production-method assignment. This outcome will also be decomposed by randomized starting method: switching from manual start to machine use, and retention among participants randomly assigned to start with the machine.

Initial entry into the learning-intensive technology is measured after standardized instruction on the mechanical sheller and before the randomized initial production-method assignment is implemented. The variable equals one if the participant chooses the mechanical sheller and zero if the participant chooses manual shelling. For the intent-to-treat analysis over all randomized participants, participants who do not attend the second session will be coded as not entering the mechanical task, with attrition and bounding analyses reported separately.

Post-assignment machine use is measured after the participant has worked under the randomly assigned initial production method and reaches the pre-specified opportunity to continue or switch. The variable equals one if the participant is using or chooses the mechanical sheller at that point and zero otherwise. Because initial production method is randomized, this outcome allows separate estimation of two margins: escape from the familiar manual method among participants randomly assigned to start manually, and retention on the unfamiliar technology among participants randomly assigned to start mechanically.

Secondary outcomes include: attendance at the second work session; completion of mechanical-sheller assembly; time required to complete assembly; abandonment during assembly; total output during the work session; productivity measured as output per unit of time; piece-rate or total earnings during the work session where applicable; time until first switch; self-reported perceived difficulty of the mechanical sheller; comprehension of standardized instructions; stated preference between manual and mechanical shelling after experience; and measures of short-run financial pressure, including debt repayment, urgent expenditures, and liquidity constraints measured before the second work session.

This study is a randomized field experiment conducted with workers in a maize-processing firm in rural Haiti. The study examines whether earlier access to previously earned wages changes willingness to adopt a more productive but unfamiliar production technology. The experiment has two randomized components. First, after all participants complete an initial paid manual shelling task, participants are individually randomized to receive the wages earned in that task either immediately or after the next work session. Total compensation is held constant. Second, during the next work session, after standardized instruction on the mechanical sheller and measurement of initial task choice, participants are individually randomized to an initial production method, manual shelling or mechanical shelling, subject to the firm’s real production need for both types of output.

The design therefore creates a 2 by 2 structure: immediate payment plus manual start; immediate payment plus machine start; deferred payment plus manual start; deferred payment plus machine start. The first randomization identifies the effect of earned-wage payment timing on technology entry and post-assignment machine use. The second randomization identifies path dependence in production method by separating the effect of starting with a familiar low-productivity method from starting with the unfamiliar higher-throughput method. The primary sample consists of eligible workers recruited from the firm’s labor pool who complete the first paid work session and are randomized to payment timing. The planned sample is 300 workers. Enrollment will occur through work sessions organized with the firm. Randomization will be implemented by the research team using a computer-generated assignment list, stratified by work session.

Not available

Randomization will be conducted by computer by the research team before each work session. The payment-timing assignment will be randomized at the individual-worker level using a reproducible randomization seed and stratification by work session. The initial production-method assignment in the second session will also be randomized at the individual-worker level, stratified by work session and subject to the firm’s operational requirement that both manual- and machine-shelled corn be produced. Assignment lists will be prepared before treatment revelation. Field staff will reveal assignments only at the relevant implementation point.

Individual worker. Payment timing is randomized at the individual-worker level. Initial production method in the second work session is also randomized at the individual-worker level. Work session fixed effects will be included in the main analysis to account for stratification and session-level conditions.

No

Not applicable. Randomization is at the individual level; there is no clustering. 400 individual workers total (200 per arm).

400 workers (individuals). An observation is considered complete only if the worker participated in the full four-hour production task and consented to the post-session survey. Data collection continues until 400 completed post-session surveys are reached or at the end of the scheduled lean-season field period (April-May 2027).

Planned assignment is 150 workers to immediate payment and 150 workers to deferred payment. Within the second work session, planned assignment is 200 workers to manual start and 200 workers to machine start, yielding approximately 100 workers in each of four cells: immediate payment plus manual start; immediate payment plus machine start; deferred payment plus manual start; deferred payment plus machine start. Actual cell sizes may differ slightly because of attendance, operational production constraints, or incomplete participation, all of which will be documented.

The planned sample is 400 completed behavioral observations. Participants are individually randomized to the earned-wage payment-timing treatment, with approximately 200 workers assigned to immediate payment of previously earned Day 1 wages and 200 workers assigned to deferred payment of the same earned wages after the Day 2 work session. In the second work session, participants are also individually randomized to the starting production method, with approximately 200 workers assigned to begin with manual shelling and 200 workers assigned to begin with the mechanical sheller. Under balanced assignment, this yields approximately 100 workers in each payment-timing-by-starting-method cell. The study has two co-primary binary outcomes: initial machine preference before starting-method randomization, and post-assignment machine use after the revision opportunity. The confirmatory family contains two one-sided tests in the pre-specified positive direction. We will control the familywise error rate using the Holm step-down procedure. For conservative planning, power calculations use a one-sided alpha of 0.025 for each primary outcome. For the first primary outcome, initial machine preference, the benchmark assumes a control-group machine-preference rate of 0.30 and a treatment-group rate of 0.45. This corresponds to a 15 percentage-point increase. With 200 workers per payment-timing arm and a one-sided alpha of 0.025, the design has approximately 88 percent power to detect this effect. The minimum detectable effect for 80 percent power is approximately 13.5 percentage points when the control-group mean is 0.30. For the second primary outcome, post-assignment machine use, the benchmark assumes that immediate access to previously earned wages increases starting-method-standardized machine use by 15 percentage points. One illustrative case is a 15 percentage-point increase in manual-to-machine switching among participants randomly assigned to begin manually and a 15 percentage-point increase in machine retention among participants randomly assigned to begin with the machine. With approximately 100 workers in each payment-timing-by-starting-method cell, corresponding to 400 completed behavioral observations overall, the design has approximately 89 percent power to detect this starting-method-standardized 15 percentage-point effect using a conservative one-sided alpha of 0.025. The minimum detectable starting-method-standardized effect for 80 percent power is approximately 13.3 to 13.5 percentage points. The starting-method-specific decomposition estimates are not powered as separate confirmatory hypotheses. With approximately 100 treated and 100 control workers within a given starting-method subgroup, a 15 percentage-point subgroup effect has only about 59 to 62 percent power under a one-sided alpha of 0.025, depending on the baseline rate. The minimum detectable subgroup effect for 80 percent power is approximately 18 to 19 percentage points. For this reason, the manual-starter and machine-starter estimates will be reported as pre-specified decomposition analyses used to interpret the source of the pooled post-assignment effect, not as separate confirmatory tests. If the scheduled lean-season field period ends before 400 completed behavioral observations are reached, the final analysis will report the realized sample size, the assignment-based study flow, treatment-arm attrition, and updated minimum detectable effects. The study will not report post-hoc power calculations.