{"id":"W2069905881","doi":"10.1097/00001888-200110001-00038","title":"Relationships among Item Characteristics, Examine Characteristics, and Response Times on USMLE Step 1","year":2001,"lang":"en","type":"article","venue":"Academic Medicine","topic":"Innovations in Medical Education","field":"Medicine","cited_by":26,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Context (archaeology); Test (biology); Computerized adaptive testing; United States Medical Licensing Examination; Disadvantaged; Psychology; Educational measurement; Computer science; Medical education; Medicine; Developmental psychology; Psychometrics; Medical school; Pedagogy","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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.008856405,0.0005311904,0.0005115667,0.001598947,0.0003674972,0.0007109535,0.0007762397,0.0007612613,0.004775207],"category_scores_gemma":[0.06818934,0.0003867604,0.0005150216,0.001242429,0.0004011688,0.00106226,0.001065785,0.001409229,0.0008963396],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0006856521,"about_ca_system_score_gemma":0.0005794745,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002069836,"about_ca_topic_score_gemma":0.002930299,"domain_scores_codex":[0.9949876,0.002512289,0.0005475579,0.0005547171,0.001162674,0.0002352858],"domain_scores_gemma":[0.8840771,0.08054727,0.02336933,0.0034556,0.006307142,0.002243521],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.001012926,0.0002925973,0.9893934,0.00003652139,0.00009617167,0.00005910337,0.0006127566,0.0003293577,0.0005761379,0.000053182,0.0004090387,0.007128966],"study_design_scores_gemma":[0.0000142735,0.0005734673,0.9976082,0.00001038434,0.00001843528,0.00007345167,0.0002582914,0.0006970473,0.0004093982,0.00004232409,0.0002856406,0.000009041506],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.998593,0.0000840178,0.0003004062,0.00007418358,0.000006424068,0.00006404379,0.0003481395,0.00002742969,0.0005023678],"genre_scores_gemma":[0.9963875,0.000071437,0.001205889,0.00004970739,0.000009787003,0.0001863541,0.0007694977,0.00003037871,0.001289489],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.008856405,"threshold_uncertainty_score":0.04683769,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.05064341965010042,"score_gpt":0.3302135982187392,"score_spread":0.2795701785686387,"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."}}