{"id":"W4294991525","doi":"10.3352/jeehp.2022.19.26","title":"Acceptability of the 8-case objective structured clinical examination of medical students in Korea using generalizability theory: a reliability study","year":2022,"lang":"en","type":"article","venue":"Journal of Educational Evaluation for Health Professions","topic":"Innovations in Medical Education","field":"Medicine","cited_by":4,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Generalizability theory; Reliability (semiconductor); Objective structured clinical examination; Statistics; Psychology; Variance components; Medicine; Mathematics; Medical education","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":false,"about_ca":true,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.07940491,0.0001254693,0.0005791098,0.0003712911,0.0004582029,0.000005960639,0.0004799665,0.0001282229,0.002030569],"category_scores_gemma":[0.04658343,0.00008993773,0.0002073848,0.001275094,0.0002692171,0.0001375258,0.0002099016,0.001015569,1.162092e-7],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001937008,"about_ca_system_score_gemma":0.01527935,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001353809,"about_ca_topic_score_gemma":0.00007924203,"domain_scores_codex":[0.9834635,0.008775816,0.003374238,0.0003049609,0.003880585,0.0002009478],"domain_scores_gemma":[0.9900115,0.002343776,0.003033063,0.0005608895,0.003909737,0.0001410887],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0006879034,0.01527098,0.934068,0.0003538197,0.0001124125,8.119763e-7,0.015557,0.0008856417,0.00002030776,0.001726318,0.000883876,0.03043297],"study_design_scores_gemma":[0.003436212,0.0008601471,0.9464089,0.0001125956,0.0001730654,0.000101412,0.03439641,0.005715279,0.00001256365,0.008693021,0.00003303382,0.00005736243],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9739179,0.00008462194,0.0003931664,0.01855519,0.002636263,0.004358293,0.00003405782,0.000002260638,0.00001824434],"genre_scores_gemma":[0.995617,0.00000446749,0.003111651,0.0006675909,0.0002378398,0.0002918145,0.00003542092,0.00001281537,0.00002138812],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.03282148,"threshold_uncertainty_score":0.9988817,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1615378778704149,"score_gpt":0.599957094874539,"score_spread":0.4384192170041241,"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."}}