{"id":"W3133889348","doi":"10.24926/iip.v12i1.2110","title":"Validation Evidence using Generalizability Theory for an Objective Structured Clinical Examination","year":2021,"lang":"en","type":"article","venue":"INNOVATIONS in pharmacy","topic":"Innovations in Medical Education","field":"Medicine","cited_by":10,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Alberta","funders":"","keywords":"Generalizability theory; Objective structured clinical examination; Pharmacy; Reliability (semiconductor); Test (biology); Psychology; Univariate; Medical education; Multivariate statistics; Medicine; Statistics; Mathematics; Family medicine","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":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.004212678,0.0001503873,0.0002630418,0.0003973363,0.0001712956,0.00004335436,0.0001182085,0.000122245,0.0002294427],"category_scores_gemma":[0.01154721,0.0001600082,0.00006299765,0.00265465,0.0001537461,0.0007005252,0.00003749189,0.000401422,0.00000324995],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0004746967,"about_ca_system_score_gemma":0.001055695,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00002449738,"about_ca_topic_score_gemma":0.000007809396,"domain_scores_codex":[0.997343,0.0005573682,0.001036325,0.0005035918,0.0003363801,0.0002233414],"domain_scores_gemma":[0.9960089,0.0004649177,0.000254947,0.000487278,0.002734658,0.00004932608],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.0004288696,0.001996674,0.2158387,0.0003103148,0.0001343573,0.0000106197,0.002037904,0.0006735062,0.1503029,0.05245503,0.00162368,0.5741875],"study_design_scores_gemma":[0.008139398,0.0003437784,0.614242,0.0005115602,0.0004111379,0.0001674903,0.002428205,0.1240736,0.1986735,0.04607229,0.004204513,0.000732487],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9048235,0.00007346569,0.09163346,0.001089069,0.001226678,0.0008792384,0.00001255303,0.00005905666,0.0002029652],"genre_scores_gemma":[0.92422,0.00002390009,0.07220051,0.00206094,0.0006430673,0.0001521078,0.0005731644,0.00002563631,0.0001007169],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.573455,"threshold_uncertainty_score":0.996779,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2373819096283104,"score_gpt":0.533615937288704,"score_spread":0.2962340276603936,"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."}}