Replication package: Institutional Trust and the Intention to Migrate: A Meta-Analysis of the Trust-Emigration Relationship, with Attention to Youth

Abstract

Effect-size files, estimation code, artefacts, a self-test and a manifest for the article "Institutional Trust and the Intention to Migrate: A Meta-Analysis of the Trust-Emigration Relationship, with Attention to Youth". It contains: the locked effect-size file of record for the individual-level evidence, seven single-context survey studies from six countries, each carrying the statistic located in its source, the conversion applied to it, the number of variables it is conditioned on, the sign decision, the construct it measures, the behavioural margin of its outcome and the pool it enters; the published statistics from which the correlation of three studies is recovered, and those of two further studies reported only as model-adjusted estimates; the cross-national file, three multi-country surveys covering 27, 33 and 135 countries, held on their native logit scale with the design-based or clustered standard error each source publishes; the country-level file, thirteen governance and corruption studies coded for direction only; the record of a study eligible for the review and inadmissible for synthesis, with the reason; and the chain in R, four scripts, covering the recovery of correlations with their conversion uncertainty, the principal pool of correlations, the model-adjusted estimates study by study, the wider trust-family pool and its age split, the conversion sensitivities, leave-one-out over the pool and the cross-national and driver evidence study by study; one artefact per reported result and the two figures the article carries. Every synthesis estimate in the article is produced by a script in 01_code and written to 04_estimates or 05_exhibits, none by hand. Effects are harmonised on Fisher's z and pooled by restricted maximum likelihood with the Knapp-Hartung adjustment; the prediction interval, not the pooled mean, is the article's headline. The self-test carries fifteen assertions in both directions: the estimator must recover a planted -0.30 and must not reject a planted zero; every core effect must be a correlation conditioned on no substantive covariate, and no model-adjusted estimate may enter a pool; the recovered correlations must match the conversion script and reproduce the published R-squared; a coefficient on the raw scale of its variables must not pass for a standardised one; the inadmissible study must enter no pool; and every row must record its source statistic. Raw inputs are the primary studies, published sources identified in the article's reference list and not redistributed. 02_data carries what they become after extraction, and each row records the statistic and table it came from. 00_README_run_order.md gives the run order and MANIFEST.csv lists every file with its checksum. No number here is produced outside R. Software: R 4.3.3 with the metafor package.