{"id":"W4225106913","doi":"10.1002/jaba.921","title":"Agreement between visual inspection and objective analysis methods: A replication and extension","year":2022,"lang":"en","type":"article","venue":"Journal of Applied Behavior Analysis","topic":"Behavioral and Psychological Studies","field":"Psychology","cited_by":7,"is_retracted":false,"has_abstract":true,"ca_institutions":"Université de Montréal","funders":"Fonds de Recherche du Québec - Santé","keywords":"Visual inspection; Inter-rater reliability; Replicate; Classifier (UML); Support vector machine; Replication (statistics); Artificial intelligence; Psychology; Machine learning; Pattern recognition (psychology); Extension (predicate logic); Computer science; Statistics; Mathematics; Developmental psychology; Rating scale","routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[{"model":"gemma","categories":[],"domain":null,"study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"high","status":"direct model label, unvalidated"},{"model":"gpt","categories":[],"domain":null,"study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low","status":"direct model label, unvalidated"}],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.2770073,0.002069394,0.001429845,0.002470333,0.002037181,0.002783106,0.003174644,0.00192211,0.003680166],"category_scores_gemma":[0.40179,0.001611386,0.003388439,0.001806439,0.004780836,0.004229559,0.005026081,0.002684057,0.001086694],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001843013,"about_ca_system_score_gemma":0.004398371,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003416575,"about_ca_topic_score_gemma":0.004215426,"domain_scores_codex":[0.7955887,0.1421765,0.01565123,0.01443154,0.03033159,0.001820499],"domain_scores_gemma":[0.4336626,0.2978264,0.0354352,0.1278729,0.10315,0.002052858],"domain_codex":null,"domain_gemma":"reproducibility","domain_candidate":"reproducibility","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.01537185,0.01905938,0.3578642,0.01280634,0.007218658,0.0007614969,0.08223128,0.003825813,0.04520471,0.01137169,0.01173163,0.432553],"study_design_scores_gemma":[0.007664002,0.04130545,0.7798827,0.004577471,0.005694424,0.001370323,0.01708965,0.02795826,0.05434171,0.02305744,0.03588758,0.001171115],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.6657631,0.002213597,0.2686517,0.001007422,0.001715204,0.04116111,0.001388547,0.00106616,0.01703315],"genre_scores_gemma":[0.7577264,0.0003531954,0.2024404,0.0006000998,0.0002961273,0.03581657,0.0005618309,0.0003957701,0.001809619],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.7229927,"threshold_uncertainty_score":0.8915787,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.136103726726318,"score_gpt":0.4345176438425861,"score_spread":0.298413917116268,"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."}}