{"id":"W2885300578","doi":"10.1186/s12909-018-1294-z","title":"Development, administration, and validity evidence of a subspecialty preparatory test toward licensure: a pilot study","year":2018,"lang":"en","type":"article","venue":"BMC Medical Education","topic":"Innovations in Medical Education","field":"Medicine","cited_by":6,"is_retracted":false,"has_abstract":true,"ca_institutions":"London Health Sciences Centre; Western University","funders":"","keywords":"Licensure; Summative assessment; Test (biology); Medical education; CLARITY; Content validity; Generalizability theory; United States Medical Licensing Examination; Medicine; Psychology; Objective structured clinical examination; Construct validity; Educational measurement; Subspecialty; External validity; Test validity; Staffing; Curriculum; Formative assessment; Family medicine; Psychometrics; Clinical psychology; Nursing; Social psychology; Mathematics education; 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":[0.09905322,0.0007262525,0.0005324036,0.001598902,0.001195927,0.001598278,0.001782352,0.001208557,0.001991392],"category_scores_gemma":[0.2262267,0.0007501663,0.0009232659,0.001162862,0.002311805,0.002001231,0.002274459,0.001930047,0.0009237428],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002341978,"about_ca_system_score_gemma":0.00739756,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002698651,"about_ca_topic_score_gemma":0.002944818,"domain_scores_codex":[0.942572,0.03894747,0.004631012,0.002427139,0.01008277,0.001339523],"domain_scores_gemma":[0.7098576,0.2069811,0.01438103,0.01692683,0.04713934,0.004714055],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.004516674,0.02910413,0.7261979,0.0009270835,0.0002127238,0.0006655001,0.02555602,0.001016437,0.007326934,0.0008803077,0.00136303,0.2022333],"study_design_scores_gemma":[0.002279627,0.09393465,0.8543685,0.0008149148,0.0002957983,0.001208638,0.01209248,0.006924253,0.0184758,0.0005739463,0.008898661,0.0001327858],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9883348,0.0001017798,0.004491718,0.0002280703,0.00004713145,0.004450608,0.00009620639,0.00004384487,0.002205762],"genre_scores_gemma":[0.977002,0.0002165433,0.01780464,0.0002631152,0.00003770192,0.003723011,0.0002952007,0.00004046199,0.0006174408],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.09905322,"threshold_uncertainty_score":0.5238497,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1453263486835266,"score_gpt":0.4330176939737351,"score_spread":0.2876913452902086,"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."}}