Why Don't Farmers Insure? Sunk-Cost Perception and the Demand for Agricultural Insurance

The intervention is the offer of a risk-management tool, with five conditions tested within-subject. Each participant — a smallholder grape grower in the district of Cascas, region La Libertad, Northern Peru — faces a sequence of simulated two-season crop production decisions under weather risk, organized in five blocks of four rounds. In each round, the participant is shown the probability of a good harvest year. In the first four conditions, the participant is offered a single tool, which they may accept or decline:

Insurance (I). Premium α is deducted from harvest revenue; an indemnity is paid if the year is bad.
Emergency Loan (EL). A loan is disbursed if the year is bad; it is repaid with interest δ in the following season (which is deterministically good).
Cheap Insurance (CI). Identical to Insurance, but the premium is reduced to Jα with J = 1/2, framed to the participant as a 50% subsidy on the premium. Indemnity unchanged.
Cheap Emergency Loan (CEL). Identical to Emergency Loan, but the interest is reduced to Jδ with J = 1/2, framed as a 50% subsidy on the interest. Loan amount unchanged.
Head-to-Head (HH). The participant is shown Insurance and Emergency Loan side by side, both at full (unsubsidized) cost, and chooses which of the two tools to use for that round; there is no opt-out. This block yields a direct revealed-preference ranking between the two tools, complementing the accept/decline margin of conditions 1–4.
The proportional cost reduction (J = 1/2) is identical across CI and CEL by design, so the expected-cost reductions induced by "cheapening" each tool are matched. This proportional symmetry is what suggests the wasted-premium mechanism: if the adoption gap between Insurance and Emergency Loans is driven by the premium being perceived as "wasted" money when no loss occurs, the gap should not be the same when both tools are made cheaper by the same proportion.

Participants receive a fixed payment plus a performance-based payment determined by the profit earned in a single round drawn at random — by a physical draw the participant observes — from all rounds played. The performance payment increases linearly at a pre-specified rate per sol of profit above a threshold, with the threshold set below the lowest attainable profit so the variable payment is always positive and every outcome remains payoff-relevant. (Amended 2026-08-11: the original rule, linear in total profit with no threshold, left too little payment variation across choices; see amendment history.) Working capital is provided at the start of each round, so liquidity is never a binding constraint. The instrument is a custom Spanish-language HTML survey that runs fully offline on Android tablets.

2026-08-12

2026-09-02

A binary indicator, recorded by the tablet at the moment of decision, equal to 1 if the participant accepts the risk-management tool offered in that round and 0 otherwise. Each participant contributes 20 such observations: 4 rounds in each of the 5 within-subject blocks.

Derived condition-level statistic. For each of the four conditions (Insurance, Emergency Loan, Cheap Insurance, Cheap Emergency Loan), the participant-level adoption rate is the share of rounds within that condition's block in which the participant accepted the tool. Population-level adoption rates for each condition are estimated by the pooled regression described below.

Two pre-registered hypothesis tests:

H1 — Adoption gap. The difference in population mean adoption rates between Emergency Loan and Insurance. Tests whether the emergency loan is adopted at a higher rate than insurance.
H2 — Sunk-cost contrast. The change in the adoption gap when both tools are made cheaper by the same proportion (50% reduction in the priced dimension of each tool). Defined as the gap between Cheap Emergency Loan and Cheap Insurance, minus the gap between Emergency Loan and Insurance. A negative contrast is consistent with sunk-cost perception of the insurance premium driving the adoption gap.
Both quantities are estimated from a pooled linear-probability regression of the binary accept indicator on condition dummies (Insurance is the omitted reference), with individual and round fixed effects, and standard errors clustered at the participant level. Inference is two-sided. Directional alternatives are pre-committed for interpretation but not used to relax the rejection criterion. No family-wise error correction is applied to the two primary tests (the hypotheses target substantively distinct claims — existence of a gap, and a specific mechanism for it); a Bonferroni-adjusted p-value (multiplied by 2) is reported as a supplementary column for transparency.

Within-subject crossover lab-in-the-field experiment. The five conditions described under Intervention are the five levels of a within-subject factor; each participant is exposed to all five.

Block structure. Each participant plays R = 20 rounds, organized into K = 5 blocks of r = 4 consecutive rounds. Within a block the offered condition is held fixed; between blocks it changes. In four of the blocks the participant accepts or declines a single tool (I, EL, CI, CEL); in the fifth (Head-to-Head), the participant chooses which of the two full-cost tools — Insurance or Emergency Loan — to use in each round. The order of the five blocks varies across participants (see Randomization Method below).

Within-block variation. Within each block, the four rounds vary the probability of a good harvest year over the grid θ^H ∈ {0.3, 0.4, 0.6, 0.9}, presented in randomized order. Probability is shown to the participant before each decision. Probability is a source of within-subject, within-block variation; it is not a treatment dimension.

Round structure. Each round has a two-season horizon. Season-1 yield is high (Y^H) with probability θ^H and low (Y^L) with probability 1 − θ^H. Season-2 yield is high deterministically. The Season-1 outcome is realized by the participant drawing blindly from an opaque bag of 10 colored balls whose color mix is set, round by round, to match the displayed probability; the round selected for the performance-based payoff is drawn from an opaque bag at the end of the session. Both draws are physical and observable by the participant — to support trust in the randomness of payoff-relevant events.

Identification. The within-subject design absorbs all time-invariant individual heterogeneity through individual fixed effects. Round fixed effects absorb fatigue/learning across the global round counter. The Williams-balanced block-order assignment (see Randomization Method) ensures that block-position effects and first-order carryover are orthogonal in expectation to the treatment contrasts. The primary analysis is a pooled linear-probability regression of the binary accept indicator on condition dummies (Insurance is the omitted reference) over the four accept/decline conditions, with individual and round fixed effects, and standard errors clustered at the participant level. The Head-to-Head block's outcome — a binary indicator for choosing Insurance over the loan — does not enter this regression; it is analyzed separately as a direct revealed-preference measure between the two tools. The confirmatory sample comprises participants who played under the revised performance-payment rule (from 2026-08-12); data collected in the first fieldwork weekend under the original payment rule are classified as pilot and reported separately, with a pooled robustness specification including a payment-rule indicator.

Not available

All randomization is performed in advance by a computer, in the office, using a fixed seed; the seed and the generating script are committed to the project repository so the full assignment is reproducible. No randomization is performed in the field. Two levels of randomization are pre-generated and keyed to each participant's anonymous, pre-printed card_id:

Block (treatment) order. Each participant's sequence of the five conditions (I, EL, CI, CEL, HH) is assigned from a Williams-balanced design for K = 5 conditions. Because K is odd, the balanced construction uses 2K = 10 sequences; it balances both (a) condition assignment across block positions and (b) every ordered adjacent pair of conditions, so first-order carryover is differenced out across participants. Sequences are assigned to participants in equal proportions (15 per sequence at the target N = 150), with any overflow allocated uniformly at random.
Within-block probability order. For each (participant, block) cell, the four rounds use an independent uniform random permutation of the good-year probability grid {0.3, 0.4, 0.6, 0.9}.
The pre-generated assignments are loaded onto tablets before deployment; each tablet carries its own disjoint set of card codes. When the enumerator enters a participant's card code at the start of a session, the tablet retrieves that card_id's block sequence and within-block probability permutations.

Two payoff-relevant events that occur during the session — the Season-1 yield realization in each round and the single round drawn at the end of the session for the performance-based payment — are realized using physical instruments visible to the participant: the yield by the participant drawing blindly from an opaque bag of 10 colored balls whose mix matches that round's probabilities, and the paid round by drawing a numbered ticket from an opaque bag. These are randomization events but they are not assignment to treatment; they are recorded on the tablet by the enumerator after they occur.

The individual participant is the unit of randomization for treatment assignment. Each participant receives an independently drawn block order from the Williams-balanced Latin square, and independently drawn within-block probability orders for each of the four blocks.

There is no clustering of treatment assignment above the participant: communal centers and session time slots are used as logistical groupings only, with each participant within a session receiving their own randomly assigned block sequence. No higher-level (community, session, household) randomization is used.

Yes

150 participants. The design is not clustered: the individual participant is the unit of randomization (see Randomization Unit). Two sets of pilot data are excluded from the confirmatory sample and not pooled with the main analysis: a pre-fieldwork pilot of n = 10 (May 2026), and the 29 participants from the first fieldwork weekend (Aug 8–9, 2026), reclassified as pilot after a revision of the performance-payment rule (see amendment of 2026-08-11).

3,000 decision-level observations (150 participants × 20 rounds each): 2,400 accept/decline decisions in the four single-tool conditions (the sample for the primary regression) and 600 head-to-head choices in the fifth condition.

All 150 individuals are exposed to all five within-subject conditions (crossover design). Each participant contributes 4 decisions per condition, so each arm has 150 participants and 600 round-level decisions.

Outcome unit: probability of accepting the offered tool (percentage points). With N = 150, 4 decisions per participant per condition (600 decisions per condition), and the Williams-balanced crossover absorbing individual and round effects, the variance of a two-condition contrast of adoption rates is 2σ²/600, where σ is the residual standard deviation of the binary outcome. Using the conservative bound σ = 0.5 (binary outcome, no variance reduction from fixed effects): MDE at 80% power, 5% two-sided ≈ 2.80 × √(2·0.25/600) ≈ 8.1 percentage points for H1 (Emergency Loan − Insurance adoption gap), and ≈ 2.80 × √(4·0.25/600) ≈ 11.4 percentage points for H2 (the difference-in-differences contrast, which combines four condition means). Because individual and round fixed effects absorb substantial variance in practice, these are upper bounds on the true MDEs.