{"id":"W4387445954","doi":"10.3928/01484834-20230815-05","title":"Scoring Methods in Script Concordance Tests: An Exploratory Psychometric Study","year":2023,"lang":"en","type":"article","venue":"Journal of Nursing Education","topic":"Clinical Reasoning and Diagnostic Skills","field":"Medicine","cited_by":2,"is_retracted":false,"has_abstract":true,"ca_institutions":"Université du Québec","funders":"","keywords":"Concordance; Spearman's rank correlation coefficient; Test (biology); Psychology; Clinical psychology; Medicine; Statistics; Mathematics; Internal medicine","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.09622407,0.0007114199,0.0006246355,0.00302898,0.0008221662,0.001860399,0.0009886732,0.0006622128,0.001383617],"category_scores_gemma":[0.2037798,0.0005278966,0.001152339,0.002930065,0.001485644,0.002238678,0.002293228,0.00111223,0.0002962165],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001417708,"about_ca_system_score_gemma":0.002055443,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0008837437,"about_ca_topic_score_gemma":0.001246305,"domain_scores_codex":[0.9135437,0.06150327,0.006388361,0.001713648,0.0159435,0.0009074787],"domain_scores_gemma":[0.6481162,0.3024228,0.01723284,0.008496841,0.02206322,0.00166811],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.001954188,0.004012028,0.6790555,0.0008496554,0.0003144928,0.0002740316,0.01717332,0.002139338,0.00143449,0.001190755,0.0006996838,0.2909025],"study_design_scores_gemma":[0.001208859,0.041826,0.8515627,0.002403842,0.0008023265,0.003005535,0.0276777,0.04689762,0.0116334,0.004099446,0.008534309,0.000348174],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9807779,0.0004639288,0.01405272,0.00009357744,0.0000339922,0.001985412,0.0001059436,0.00004505095,0.002441492],"genre_scores_gemma":[0.9638895,0.0002631376,0.03323376,0.00006716683,0.00002409859,0.002126765,0.0001529379,0.00001977191,0.0002228552],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.09622407,"threshold_uncertainty_score":0.5088876,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1869778306894005,"score_gpt":0.549873423595076,"score_spread":0.3628955929056755,"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."}}