{"id":"W2003736935","doi":"10.1186/1472-6920-14-30","title":"Multiple tutorial-based assessments: a generalizability study","year":2014,"lang":"en","type":"article","venue":"BMC Medical Education","topic":"Innovations in Medical Education","field":"Medicine","cited_by":3,"is_retracted":false,"has_abstract":true,"ca_institutions":"Medical Council of Canada; Université Laval; Université de Sherbrooke","funders":"","keywords":"Generalizability theory; Medical education; Educational measurement; MEDLINE; Psychology; Computer science; Medical physics; Medicine; Curriculum; Pedagogy","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":[0.1626672,0.0008692734,0.00103129,0.003514518,0.0009592459,0.00184703,0.001355923,0.00162519,0.003090836],"category_scores_gemma":[0.2846596,0.000616722,0.003367929,0.002024337,0.00282549,0.003494657,0.003296615,0.001389424,0.0007498912],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001031777,"about_ca_system_score_gemma":0.001057245,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001021077,"about_ca_topic_score_gemma":0.001225296,"domain_scores_codex":[0.8867405,0.07639525,0.009652317,0.0116151,0.01441124,0.001185656],"domain_scores_gemma":[0.480333,0.4013785,0.02368221,0.06468266,0.02815098,0.001772567],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00184125,0.001122613,0.9357013,0.0005292344,0.00220298,0.000206872,0.008711262,0.0009069653,0.002086103,0.001077872,0.0007529659,0.04486059],"study_design_scores_gemma":[0.0005149539,0.01597804,0.9545151,0.0004034781,0.002311113,0.0008705126,0.006478021,0.007125453,0.004601053,0.002504462,0.004589357,0.000108465],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9563508,0.0008615935,0.03121012,0.0002927982,0.0001953094,0.002927943,0.0004519543,0.0001531433,0.007556415],"genre_scores_gemma":[0.9944398,0.00007351078,0.003877367,0.000111093,0.0000390465,0.0008958308,0.0002228169,0.00004460917,0.0002959602],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1626672,"threshold_uncertainty_score":0.8602765,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02907386443334496,"score_gpt":0.4148771174322712,"score_spread":0.3858032529989263,"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."}}