Data and code for “Not all weak links are equal: Domain-asymmetric compensation in consumer evaluations of airline service”
Description
This dataset supports an observational study of domain-asymmetric compensation in consumer evaluations of airline service. It contains processed numerical records for 375,525 TripAdvisor airline reviews written in 2016–2025, including the 209,974 complete seven-attribute profiles used for the primary analysis across 49 Asian airlines. Value-for-money ratings are available for 208,650 complete profiles. It also contains 1,200 paired independent narrative-coding records and final adjudicated labels, 490 carrier-year classifications, held-out predictions, analysis scripts, numerical outputs, bootstrap draws and verification records. The scripts estimate categorical additive and polynomial service-evaluation models, conditional-average contrasts, carrier-held-out prediction, narrative prevalence, business-model sensitivity, full nonlinear bootstrap inference, and a value-for-money alternative outcome. The release preserves the distinction between conditional-average and common-score contrasts and reports the non-replication of the primary conditional-average finding for value for money. Public numerical records use replacement review and author-cluster identifiers. Original review text, profile links, exact dates, source identifiers and private coding notes are excluded. The workflow starts from validated numeric observations and frozen coding decisions. Reproducibility instructions, a data dictionary, per-file descriptions and checksums are included. See documentation/PROVENANCE.md for implementation provenance and verification scope.