The Cost of Grouping: Experimental Evidence on Youth Demand for Individual and Group-Based Seed Capital

This is a survey-embedded, incentivized valuation experiment rather than a programmatic treatment. Every sampled youth completes a dedicated DBDC valuation section in the SurveyCTO CAPI instrument, immediately after the baseline socioeconomic module and a four-question comprehension check. Each respondent is randomly assigned to exactly one of three valuation modules:
• M1: Individual grant (10,000 ETB) vs. self-formed group (group total B) — premium for autonomy when partners are self-chosen.
• M2: Individual grant (10,000 ETB) vs. program-assigned group (group total B) — premium for autonomy and partner choice combined.
• M3: Self-formed group (50,000 ETB) vs. program-assigned group (group total B) — pure value of partner choice, group structure held constant.

Within the assigned module the respondent answers two linked yes/no choices (double-bounded): the follow-up bid moves up after a rejection and down after an acceptance, bracketing the latent willingness-to-accept (WTA). Group amounts are denominated as the total received by a group of five (G = 5, fixed). The parity floor is 50,000 ETB (= 5 × 10,000) and the ceiling is 70,000 ETB (= 5 × 14,000). The starting bid is drawn uniformly from five bid paths in 5,000 ETB steps {50,000; 55,000; 60,000; 65,000; 70,000}.
Incentive compatibility is enforced by a public lottery: after fieldwork closes, one youth per kebele (16 primary winners across 16 kebeles) is drawn; one of each winner’s incentivized DBDC answers is selected at random and the chosen modality is paid for real as seed capital.

Intervention (public) Youth members of agribusiness groups complete an incentivized choice experiment valuing whether they would prefer the same seed-capital grant as an individual, in a group of their own choosing, or in a group assigned by the program. A subset of respondents (one per kebele) has one of their real choices implemented and paid out.

2026-08-30

2026-12-30

1. pi_M1 — median compensating premium to switch from a 10,000 ETB individual grant to a self-formed group (Module M1). 2. pi_M2 — median compensating premium to switch from a 10,000 ETB individual grant to a program-assigned group (Module M2). 3. pi_M3 — median compensating premium to switch from a self-formed group to a program-assigned group (Module M3); the partner-choice premium / cost of assignment. 4. Acceptance shares at parity— share accepting the group alternative at the 50,000 ETB parity bid, for self-formed and assigned.

Each median premium is the group-total ETB at which exactly 50% of the population would accept the switch, denominated as the premium over parity (group total − 50,000 ETB). It is identified by interval (bracket) regression on the DBDC bracket from the relevant module, with kebele fixed effects and standard errors clustered at the YAIG level. Acceptance shares at parity are read directly from the fitted demand curves at B = 50,000.

Full demand curves at the five bid levels {50,000; 55,000; 60,000; 65,000; 70,000} for each modality; cost of assignment as the residual pi_M2 − pi_M1; heterogeneity of premiums by pre-specified covariates (gender, age, prior business experience, prior group experience, prior group conflict, trust-network size); right-censoring rates at the 70,000 ETB ceiling by modality; validity diagnostics (comprehension-check pass rates, belief elicitation on trust in the draw/payout and subjective win probability, bid-path balance, and module balance).

Demand curves and heterogeneity are estimated with a stacked choice-level logit (two rows per respondent for her assigned module), with the premium-over-parity slope interacted with module and covariate dummies, kebele fixed effects, and YAIG-clustered standard errors.

Between-subjects, three-arm valuation experiment embedded in a CAPI survey of the full youth sample. Every respondent completes the experiment; there is no community- or YAIG-level treatment. Randomization operates over (a) which one of the three valuation modules a respondent answers, (b) the starting bid path within that module, and (c) post-fieldwork implementation draws (which youth win and which of their answers is implemented). Each respondent provides two double-bounded yes/no choices that bracket her WTA for the non-preferred modality. Incentive compatibility is secured by a binding, publicly announced lottery.
Experimental Design Details Modules: M1 (individual vs self-formed), M2 (individual vs assigned), M3 (self-formed vs assigned). Group size G = 5 fixed across both group arms, isolating the formation mechanism from group size. Bid paths (group totals, ETB):
Path B1 (start) B_L (lower) B_H (upper) Notes
1 50,000 n/a 55,000 Floor: identifies youth who accept at parity
2 55,000 50,000 60,000
3 60,000 55,000 65,000
4 65,000 60,000 70,000
5 70,000 65,000 n/a Ceiling: rejection ⇒ right-censored
DBDC bracket mapping: yes–yes ⇒ WTA ≤ B_L; yes–no ⇒ (B_L, B1]; no–yes ⇒ (B1, B_H]; no–no ⇒ WTA > B_H (right-censored; at the ceiling path the right censor is 70,000 ETB). After fieldwork, 16 primary winners (one per kebele) each have one randomly drawn incentivized answer implemented and paid.

Not available

1) Bid path: drawn on the fly inside the SurveyCTO instrument at the start of each interview using SurveyCTO’s random() function — uniform at the respondent level, no stratification. (2) Module assignment: pre-generated over the sampling frame by a seeded stata script (seed archived for reproducibility) and preloaded into the instrument (3) Implementation draws (winners; which answer is implemented) are made post-fieldwork by computer

Module assignment is stratified by YAIG (and therefore by kebele, since each YAIG nests within exactly one kebele); bid-path assignment is unstratified at the individual level.

Yes

183 YAIGs

planned number of observations Approximately 2,500 youth (each contributes two DBDC choice rounds within her single assigned module).

Three modules, individual-level equal-proportion allocation within each YAIG: approximately 833 respondents per module (M1, M2, M3) at N = 2,600.

At N = 2,500, three-way split, ICC = 0.10 at the YAIG level (sigma of log-WTA = 0.6): - Median premium gap pi_M2 − pi_M1 (cost of assignment): ~6,000 ETB group-total ≈ ~1,200 ETB per person (G = 5) at 80% power (simulation, interval regression). - Acceptance share, single module at parity (D_m(0) ≠ 0.5): ~5.6 percentage points (analytical, one-sample clustered binary, n = 833). - Acceptance-share gap between two arms at parity: ~7.8 pp at ICC = 0.10 (~7.3 pp at ICC = 0.05). - Gender-heterogeneity gap in acceptance share: ~10.7 pp (33% female assumption).