Demand for Advance Payments: Financial Literacy, Wage Theft and Credit Constraints

The study is an incentivised hiring experiment embedded in a follow-up survey of households in high-migration villages of Western Odisha, India, a source region for seasonal migrant labour to India's brick kilns. During the survey, enumerators offer each respondent the chance to apply for a real job vacancy at a partner kiln for the upcoming season. If hired, respondents choose between two contracts that trade off the size of an upfront advance against the piece rate paid for per brick. The high advance contract pays a larger sum on recruitment but a lower rate per brick. The low advance contract pays a smaller advance but a higher rate per brick, and is worth more in present value terms at typical productivity levels and local interest rates.

The intervention is an information treatment that corrects respondents' beliefs about how much they would take home under each contract. Half of respondents are randomly assigned to treatment and half to an active control group. In the control group, respondents estimate for themselves what their group would take home over the season under each contract, net of repaying the advance, at each of their reported output scenarios. In the treatment group, respondents are instead told these figures, computed from their own reported output and group size. The same divergence is repeated in a later hypothetical multiple price list, where treatment respondents are told the take-home pay implied by each contract before choosing while control respondents are not. The intervention tests whether limited demand for low advance, high piece rate contracts is partly driven by underestimation of the earnings those contracts offer.

2026-07-04

2026-08-30

Switch point in the hypothetical multiple price list (MPL): the lowest low advance piece rate at which the respondent prefers the low advance contract to the high advance contract, expressed as an index from 1 to 6. This is the single primary outcome of the randomised experiment and the confirmatory test of the primary hypothesis (H1), that accurate information on earnings under the two contracts raises demand for the low advance, high piece rate contract. The hypothesised treatment effect is negative, meaning treated respondents switch to the low advance contract at a less generous piece rate.

The MPL presents the respondent with five rows, each a choice between a fixed high advance contract and a low advance contract whose piece rate rises from row 1 to row 5. The switch point is the index of the lowest row at which the respondent prefers the low advance contract. It is constructed from the respondent's choices reordered by piece rate, so it does not depend on the presentation order, which is randomised to ascending or descending at the respondent level. Respondents who prefer the high advance contract at all five rows are assigned an index of 6. Respondents whose reordered choices switch more than once are assigned the first row at which they prefer the low advance contract and are flagged, with a sensitivity analysis that excludes them. A lower switch point indicates stronger demand for the low advance contract, so the hypothesised negative treatment effect corresponds to respondents abandoning the high advance contract at a less generous piece rate.

Application for the low advance job (secondary outcome of the primary hypothesis H1). An indicator equal to 1 if the respondent registers as an applicant for the real job and either accepts only the low advance contract or accepts both contracts and states a preference for the low advance one, and 0 otherwise. This is an incentivised, real-stakes measure, treated as an underpowered validation of the primary result. Hypothesised direction: positive under the information treatment.

Belief gap in earnings differences (confirmatory outcome of secondary hypothesis H2, the mechanism). The amount by which a control respondent's perceived earnings difference between the low and high advance contracts falls short of the objective difference. Hypothesised direction: negative, meaning the perceived gap understates the true one.

Change in contract demand when wage protection is removed (confirmatory outcome of secondary hypothesis H3). The cross-contract difference in how a respondent's acceptance of each contract changes when told that wage-dispute protection would be removed. Hypothesised direction: negative, meaning removal reduces acceptance of the low advance contract by more.

Cross-contract difference in expected earnings loss to wage theft (confirmatory outcome of secondary hypothesis H4). Hypothesised direction: positive, meaning the expected loss is larger under the low advance contract.

Change in contract demand when an outside loan is offered (confirmatory outcome of secondary hypothesis H5). The cross-contract difference in how acceptance of each contract changes when the respondent is offered a hypothetical outside loan. Hypothesised direction: positive, meaning loan access shifts demand toward the low advance contract.

Outcome 1 (application for the low advance job): built from the interest registration module. It equals 1 for respondents who register as applicants and either accept only the low advance contract or accept both and name the low advance one as their preference. It equals 0 for all other surveyed households, including those who register but accept or prefer the high advance contract, those who do not register, and those who are ineligible and so are not offered registration. The outcome is incentivised because a random subset of applicants will be invited to recruitment events with the employer. Because vacancies are limited it is underpowered, so it is read as a validation of the primary outcome rather than a stand-alone confirmatory test.

Outcome 2 (belief gap, H2): defined for the control group, since treated respondents are told the correct earnings. The perceived difference is the respondent's own reported take-home pay under the low advance contract minus that under the high advance contract, at their normal output scenario and net of repaying the advance. The objective difference is the same quantity computed from the respondent's reported normal output and group size. The belief gap is the perceived difference minus the objective difference. All open-ended rupee amounts are winsorised at the 1st and 99th percentiles.

Outcome 3 (wage protection removal, H3): all eligible respondents, whether or not they registered, are asked whether and under which contracts they would apply if wage-dispute protection were removed. For each contract, the first difference is acceptance under removal minus status-quo acceptance. The confirmatory object is the low advance first difference minus the high advance first difference, tested as a paired comparison.

Outcome 4 (expected loss to wage theft, H4): for each contract, the respondent reports the probability of being paid fairly, underpaid, or overpaid relative to the fair amount at their normal season, by allocating ten stones, and the take-home amounts they would receive if underpaid or overpaid. The expected loss is the probability-weighted rupee shortfall of pay below the fair amount, netting out any expected overpayment. The confirmatory object is the expected loss under the low advance contract minus that under the high advance contract, pooled across arms. As a robustness check the loss is also divided by group size.

Outcome 5 (outside loan, H5): eligible respondents are asked whether they would take a hypothetical loan disbursed when the advances are paid and repaid after the season, and, if so, which contracts they would accept and prefer with the loan in place. Status-quo acceptance of a contract equals 1 if the respondent accepts only that contract, or accepts both and prefers it. Acceptance with the loan is defined the same way, with respondents who decline the loan held at their status-quo value so their change is zero. For each contract the first difference is acceptance with the loan minus status-quo acceptance, and the confirmatory object is the low advance first difference minus the high advance first difference. The share who would take the loan is reported alongside.

The study is a randomised controlled trial embedded in a follow-up survey of roughly 1,000 households in high-migration villages of Western Odisha, India. It has a single randomised treatment, an information intervention that gives respondents accurate figures for their expected take-home pay under brick-kiln contracts that differ in the size of the advance and the piece rate. Households are randomly assigned to the information treatment or to an active control group. Treated respondents receive these earnings figures for two sets of contracts: the two contracts attached to a real, advertised job vacancy, and the series of hypothetical contracts at a wider range of piece rates. Control respondents instead estimate the same earnings for themselves in both cases. The arms are otherwise handled identically.

The primary hypothesis is tested by comparing the two arms on the primary outcome, the switch point in the hypothetical multiple price list. The design also supports an incentivised secondary test on a real application decision, since respondents can register for a genuine vacancy and a random subset of applicants will be invited to recruitment events with the employer.

The remaining, secondary hypotheses are tested mainly through within-respondent comparisons rather than the randomised treatment. The first concerns the mechanism behind the information effect: whether respondents' perceived earnings difference between the two contracts understates the objective difference. This belief gap is measured in the control group, since treated respondents are told the correct earnings. The others weigh two alternative drivers of contract demand, wage theft and credit constraints. For these, respondents evaluate both contracts and their choices are re-elicited under a set of common hypothetical scenarios, namely removal of wage-dispute protection and access to an outside loan.

Randomisation is stratified by village and by a baseline measure of demand for advance payment, and was carried out ahead of fieldwork with a pre-specified seed applied to all 1,494 baseline households. Both stratification variables are controlled for in analysis through randomisation-block fixed effects, and standard errors are clustered at the village level.

Not available

Randomisation was carried out by computer, using R, in the office ahead of fieldwork, with a pre-specified seed. It was applied to all 1,494 baseline households and stratified by village and by a baseline measure of demand for advance payments. Within each randomisation block (a village by baseline-stratum cell), half of the households were assigned to each arm at random. The leftover household from each odd-sized block was pooled with other leftover households in the same village, and these were split between the two arms as evenly as possible, randomly deciding the extra assignment where their number was odd. By construction, the number of treated and control households differs by at most one in every village.

The individual household. Randomisation is stratified within village, and standard errors are clustered at the village level for inference.

No

Approximately 71 villages. Treatment is not assigned at the cluster level, but standard errors are clustered at the village level, so villages are the relevant clusters for inference. The follow-up targets 1,000 households, and the final number of villages will depend on within-village attrition, since villages are visited in turn until the target is reached.

1,000 households (the target for the follow-up survey), drawn from a baseline of 1,494 households across 71 villages.

Approximately 500 households information treatment, 500 households control. Assignment was made in a 50/50 split within each randomisation block across all 1,494 baseline households (about 747 treatment and 747 control). Given that attrition is orthogonal to treatment and balanced within village, the realised follow-up sample of 1,000 households will be close to a 500/500 split, with the exact numbers depending on which households are successfully resurveyed.

Primary outcome, the MPL switch point (index 1 to 6): the minimum detectable effect at 80% power is 0.40 index steps, or 0.24 standard deviations (outcome SD 1.66), or 10% of the control mean of 4.05. Secondary outcome, application for the low advance contract (binary): 8 to 10 percentage points at a 32% base rate, or 0.17 to 0.21 standard deviations. Both are simulated at alpha 0.05 with treatment re-randomised within village-by-baseline-MPL blocks and standard errors clustered by village.