{"id":"W2111112846","doi":"10.5539/jel.v1n1p77","title":"Reliability Evidence for Examination Cut Scores within a Medical School","year":2012,"lang":"en","type":"article","venue":"Journal of Education and Learning","topic":"Innovations in Medical Education","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Reliability (semiconductor); Competence (human resources); Item response theory; Psychology; Test (biology); Medical school; Statistics; Medical education; Psychometrics; Medicine; Clinical psychology; Mathematics; Social psychology","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":true,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.2182269,0.001398793,0.001813748,0.007587884,0.001531801,0.003834297,0.003427453,0.004006895,0.003406534],"category_scores_gemma":[0.4339868,0.002033962,0.004403126,0.005032099,0.006045671,0.002817089,0.004935957,0.002602349,0.001317665],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002200278,"about_ca_system_score_gemma":0.002420635,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003201254,"about_ca_topic_score_gemma":0.003180436,"domain_scores_codex":[0.7289889,0.1516925,0.02148113,0.01995218,0.07322592,0.004659398],"domain_scores_gemma":[0.2073103,0.6548072,0.03644417,0.04699144,0.05209892,0.002347976],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00223528,0.0004051903,0.9068298,0.002234254,0.003756035,0.0002150712,0.003761865,0.002627084,0.001039529,0.007475193,0.004267495,0.06515324],"study_design_scores_gemma":[0.0004028404,0.00243117,0.9435342,0.003309339,0.002616021,0.00101969,0.003026685,0.01746492,0.004603053,0.007508419,0.01388361,0.000199977],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8542172,0.02816916,0.0775492,0.0038485,0.001600108,0.0009234144,0.002670904,0.0004465231,0.03057493],"genre_scores_gemma":[0.9804449,0.001205403,0.01511063,0.0003795627,0.0002671698,0.0003278147,0.001603787,0.0001173571,0.0005433558],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2182269,"threshold_uncertainty_score":0.9640654,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.04873771077640658,"score_gpt":0.4205654949674104,"score_spread":0.3718277841910038,"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."}}