State-card readability under tricolor priming: an equivalence test

Participants take a brief (approximately 5-minute) survey administered via oTree on their own devices, in person, in a classroom setting at one of four Mexico City universities (CIDE, ITAM, Ibero, possibly Anahuac del Sur).

2026-05-06

2026-10-31

`state_memory_either` -- a binary indicator equal to 1 if the participant correctly identifies the donor state via either open-ended recall (Screen 6) or 6-alternative forced choice recognition (Screen 7), and 0 otherwise.

The OR-rule reflects the substantive question of interest: whether the participant has retained the donor-state identity from the exposure. A participant who recognizes the state from a list but cannot freely recall it has, in substance, retained the state's identity. Restricting the outcome to recall-only would treat such a participant as a memory failure, mismatching the substantive question. Free recall and recognition are reported as components alongside the primary for transparency.

Primary outcome components (reported alongside the primary)

- `recall_state_correct` -- open-ended free recall of the donor state (Screen 6), scored with a fuzzy canonical-name match (Levenshtein-tolerant against the canonical Spanish state name).
- `state_recognition_correct` -- 6-alternative forced choice recognition of the donor state from a fixed list of six Mexican states with per-participant shuffle (Screen 7).


Additional measured outcomes

The below outcome is part of the instrument by design. It is reported descriptively -- per-arm rates and a companion two-proportion test -- for transparency about the data collected. It is not pre-committed gates on H1: the verdict on the readability confound is not conditioned on any of these outcomes.

- `org_recall_correct` -- open-ended free recall of the donor organization shown on the card (Screen 5). Also serves as the buffer task between the card exposure and the state-recall question.

Regarding the primary outcome components: Both components are reported with the same TOST equivalence test as the primary (plus-or-minus 10 pp margin, Newcombe 90% CI, alpha = 0.05 per tail), plus companion NHST, so a reader can see how equivalence behaves on each component as well as on the OR combination.

Two-arm between-subject randomized controlled trial. Individual-level random assignment. No clustering. Fixed-N, no peeking.

Not available

At-individual level, deterministic per participant from the anonymous participation code: the code is hashed to a 64-bit integer and used to seed the choice of arm, the choice of beneficiary state from the 6-state pool, the bonus-question draw, and the per-participant shuffle of the 6-alternative forced choice recognition options. Reproducibility from the saved code is exact.

University affiliates (students and staff)

No

Not applicable

200 university affiliates (students and staff) total. Fixed-N, no peeking.

100 per arm

At the planned N (100 per arm), Monte Carlo TOST power on the OR primary is 80.3% (B = 20,000 simulation replications; 95% Monte Carlo CI [80.0%, 80.9%]), at alpha = 0.05 per tail, equivalence margin plus-or-minus 10 pp, anchored at p_neutral = 0.97 and p_flag = 0.95. The anchor reflects an expectation of high recall under the stimulus design: the cards are clear single-state illustrations shown for 10 seconds, intentionally well above the threshold at which differential reading speed would be the bottleneck. A small calibration pilot (N approximately 30, May 2026, CIDE undergraduate convenience sample) validated this anchor empirically and confirmed instrument functioning (exposure time tight at 10.15-10.43 s; zero bonus-computation mismatches; clean arm x state and arm x bonus-question balance). Pilot data are not included in the registered analysis and the pilot was conducted on a separate population from the main wave. Power was computed by direct Monte Carlo simulation of the TOST procedure on the OR primary, not by closed-form normal approximations. At the empirical-ceiling anchor (0.97 / 0.95), the power function is discrete-stepped in (k_flag, k_neutral) against the integer thresholds of the Newcombe-CI band edges, making asymptotic normal approximations unreliable. The exact simulation code, anchor sweeps, and per-N power values are in `analysis/power_sweep.R` and `analysis/power_fine.R`. MDES (minimum detectable effect size) -- reported under both interpretations, since the TOST design admits two: 1. As equivalence margin. Delta = 10 pp. This is the smallest true |p_flag - p_neutral| that the test is set up to exclude. Effects below this threshold are declared equivalent. 2. As conventional power-bounded MDES on the companion NHST. Under a standard two-sided two-proportion test at alpha = 0.05 with 80% power at p_neutral = 0.97, the smallest detectable absolute difference at N = 100/arm is approximately plus-or-minus 13 pp. This is the MDES for the companion NHST, not for the primary inference.