{"id":"W4405995310","doi":"10.1016/j.ijresmar.2024.12.004","title":"On the persistent mischaracterization of Google and Facebook A/B tests: How to conduct and report online platform studies","year":2025,"lang":"en","type":"article","venue":"International Journal of Research in Marketing","topic":"Social Media in Health Education","field":"Social Sciences","cited_by":22,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of British Columbia","funders":"Social Sciences and Humanities Research Council; École Supérieure des Sciences Economiques et Commerciales; Social Sciences and Humanities Research Council of Canada; Universiteit van Amsterdam","keywords":"Internet privacy; Advertising; Business; Social media; World Wide Web; Computer science","routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.01232835,0.00004343713,0.0001210171,0.0004164086,0.0001535304,0.00007403178,0.0002053549,0.00003579817,0.00001158025],"category_scores_gemma":[0.07536548,0.00003283022,0.00002479147,0.0002805079,0.0002692881,0.0001447,0.00008126017,0.0002778597,2.393761e-7],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0003641601,"about_ca_system_score_gemma":0.0004152948,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000171289,"about_ca_topic_score_gemma":0.0002709826,"domain_scores_codex":[0.9980216,0.0005995133,0.0003248705,0.00009718713,0.0008149197,0.0001419062],"domain_scores_gemma":[0.9851599,0.01297425,0.0002448583,0.0000547829,0.001502134,0.00006407807],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.002137742,0.000480511,0.550465,0.0004736019,0.0006319482,0.0001712673,0.1136151,0.00003507583,0.006257463,0.02304258,0.008773808,0.2939159],"study_design_scores_gemma":[0.0009936627,0.000457502,0.5900989,0.006502485,0.00002319363,0.00005296155,0.3262557,0.0001658921,0.0004816141,0.009679829,0.06511512,0.0001730793],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8787447,0.000331654,0.000003680073,0.1185735,0.0009428109,0.000433995,0.000002529604,0.000001947364,0.0009652376],"genre_scores_gemma":[0.9976676,0.001261802,0.0003280181,0.0001470261,0.0003209146,0.00003391357,0.000001055836,0.000003038227,0.0002366719],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2937428,"threshold_uncertainty_score":0.9324231,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.408085861121674,"score_gpt":0.5628094697192918,"score_spread":0.1547236085976178,"validation_status":"score_only:v0-immature-baseline","note":"Baseline scores from an immature model (maturity gate not passed). Scores rank; they never assert a category."}}