{"id":"W2498525818","doi":"10.1186/s12909-016-0714-1","title":"Validity of a new assessment rubric for a short-answer test of clinical reasoning","year":2016,"lang":"en","type":"article","venue":"BMC Medical Education","topic":"Clinical Reasoning and Diagnostic Skills","field":"Medicine","cited_by":17,"is_retracted":false,"has_abstract":true,"ca_institutions":"The Wilson Centre; Memorial University of Newfoundland; University Health Network; University of Toronto","funders":"University of Toronto","keywords":"Rubric; Generalizability theory; Certification; Medical education; Test (biology); Psychology; Competence (human resources); Construct validity; Medicine; Applied psychology; Clinical psychology; Psychometrics; Social psychology; Mathematics education","routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":true,"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":[0.02495299,0.000584109,0.0008004988,0.003872428,0.0008262742,0.001603988,0.001471029,0.0008450884,0.002327763],"category_scores_gemma":[0.1248134,0.000326849,0.001679991,0.00134915,0.001093418,0.001904164,0.002659708,0.001169769,0.0008296813],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001568204,"about_ca_system_score_gemma":0.002737847,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003179842,"about_ca_topic_score_gemma":0.005814304,"domain_scores_codex":[0.9747745,0.008404098,0.003191795,0.001119518,0.01197407,0.0005360172],"domain_scores_gemma":[0.8978279,0.05367587,0.01467431,0.005210097,0.02670172,0.001910209],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.001799226,0.001153212,0.5383631,0.0006661531,0.0003802091,0.0002415446,0.003724206,0.002381116,0.005518425,0.001799657,0.007925908,0.4360472],"study_design_scores_gemma":[0.0003611782,0.004006753,0.9518772,0.000720482,0.0001448751,0.001128837,0.002142263,0.01888288,0.004440944,0.002319108,0.01378986,0.0001856543],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.924799,0.001905275,0.04976867,0.0006902341,0.0006705906,0.005314295,0.001449381,0.0005712866,0.01483127],"genre_scores_gemma":[0.9202051,0.0005512249,0.07089284,0.0002627134,0.0001145003,0.004145087,0.001679179,0.00008163923,0.002067692],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.02495299,"threshold_uncertainty_score":0.1319656,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1331455526702405,"score_gpt":0.500389192385086,"score_spread":0.3672436397148455,"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."}}