{"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":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.2497883,0.001044944,0.001182305,0.004688742,0.001445849,0.003575536,0.002488806,0.001652626,0.002442722],"category_scores_gemma":[0.6003127,0.0008162565,0.001685484,0.002515049,0.003529236,0.007552114,0.003957634,0.003172155,0.0007916786],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00270336,"about_ca_system_score_gemma":0.006644797,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002463968,"about_ca_topic_score_gemma":0.003210558,"domain_scores_codex":[0.8343189,0.1156827,0.01631293,0.006021363,0.02562544,0.00203863],"domain_scores_gemma":[0.3136898,0.5323958,0.0374094,0.02939797,0.08418118,0.002925961],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"bench_or_experimental","study_design_scores_codex":[0.00200313,0.0005064146,0.309886,0.004581148,0.000897085,0.000373887,0.01948695,0.002587981,0.006295268,0.01278957,0.01902436,0.6215682],"study_design_scores_gemma":[0.0007525178,0.01183023,0.7124708,0.01133932,0.001691672,0.003156234,0.02494777,0.05658071,0.0469507,0.07362461,0.05593748,0.0007179851],"study_design_candidate":"bench_or_experimental","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.6248903,0.0067469,0.2971929,0.04989542,0.001688602,0.005710538,0.0007280737,0.001397899,0.01174927],"genre_scores_gemma":[0.8837368,0.0006420782,0.1090471,0.003251349,0.0003183667,0.001839794,0.0001979705,0.0001029902,0.000863703],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.7502117,"threshold_uncertainty_score":0.9251446,"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."}}