{"id":"W3134552356","doi":"10.24926/iip.v12i1.2235","title":"Providing Validation Evidence for a Clinical-Science Module: Improving Testing Reliability with Quizzes","year":2021,"lang":"en","type":"article","venue":"INNOVATIONS in pharmacy","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":2,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Alberta","funders":"","keywords":"Reliability (semiconductor); Syllabus; Generalizability theory; Coursework; Computer science; Mathematics education; Psychology; Medical education; Medical physics; 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","bibliometrics"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.03557128,0.0001511761,0.0003209511,0.000852417,0.0005163065,0.0005463248,0.0009093013,0.00004051771,0.00002997372],"category_scores_gemma":[0.6842359,0.0001168634,0.00005473219,0.02237763,0.0003216029,0.001797193,0.0003108077,0.0003300244,0.000004894969],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001477024,"about_ca_system_score_gemma":0.001344262,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00009018964,"about_ca_topic_score_gemma":0.000006501179,"domain_scores_codex":[0.9952639,0.0004453441,0.001645165,0.001131765,0.001083186,0.0004306012],"domain_scores_gemma":[0.7844868,0.1973052,0.002156101,0.002327668,0.01351012,0.0002141413],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00003496507,0.00008556999,0.6706815,0.0000206504,0.000003280257,0.000004848771,0.00008834593,0.00119372,0.0129158,0.0007442317,0.0001774494,0.3140496],"study_design_scores_gemma":[0.003184,0.000291753,0.562407,0.0005183358,0.00004297444,0.00008640495,0.001287151,0.3188831,0.06009376,0.04751461,0.004908085,0.0007828033],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.788799,0.0001162647,0.2077932,0.001395714,0.0007114215,0.0004726602,0.000005749097,0.00007528291,0.0006307393],"genre_scores_gemma":[0.6681767,0.00001014305,0.331318,0.000218067,0.0001158977,0.00008197511,0.000001669004,0.000008055337,0.00006948989],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.6486647,"threshold_uncertainty_score":0.9984022,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.8471072521476786,"score_gpt":0.6358239551542438,"score_spread":0.2112832969934347,"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."}}