{"id":"W4376106412","doi":"10.1016/j.ajpe.2023.100081","title":"Using Automatic Item Generation to Create Multiple-Choice Questions for Pharmacy Assessment","year":2023,"lang":"en","type":"article","venue":"American Journal of Pharmaceutical Education","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":12,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Alberta","funders":"","keywords":"Item bank; Pharmacy; Summative assessment; Multiple choice; Quality (philosophy); Computer science; Test (biology); Sample (material); Medical education; Scope (computer science); Worksheet; Set (abstract data type); Item response theory; Medicine; Formative assessment; Psychometrics; Psychology; Family medicine; Mathematics education; Clinical psychology","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":[],"consensus_categories":[],"category_scores_codex":[0.0594645,0.001670199,0.0008220134,0.004704203,0.0009408753,0.001907797,0.001809131,0.001053624,0.005086143],"category_scores_gemma":[0.2176794,0.000745313,0.001081301,0.003062387,0.001003918,0.002143672,0.003303382,0.001537106,0.002118723],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001546916,"about_ca_system_score_gemma":0.002695907,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001219947,"about_ca_topic_score_gemma":0.001936615,"domain_scores_codex":[0.9492677,0.04076735,0.002859279,0.002188162,0.004503139,0.0004143088],"domain_scores_gemma":[0.7074109,0.2345359,0.01152097,0.01919334,0.02646597,0.0008728982],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"bench_or_experimental","study_design_scores_codex":[0.0008616938,0.00153568,0.03885741,0.0009599332,0.000167883,0.0002361636,0.00558807,0.01228535,0.01208609,0.006400092,0.007159565,0.9138619],"study_design_scores_gemma":[0.002577653,0.006957383,0.1252959,0.00220468,0.0004560179,0.00180992,0.006693183,0.6278227,0.102618,0.05713946,0.06558184,0.000843305],"study_design_candidate":"bench_or_experimental","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.1048783,0.0001557759,0.8742625,0.0004620565,0.0001118512,0.01141931,0.0005843343,0.004116656,0.00400924],"genre_scores_gemma":[0.124396,0.00006538948,0.8675748,0.0001223915,0.00003009016,0.006424805,0.0005849791,0.0002021963,0.0005994723],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.0594645,"threshold_uncertainty_score":0.3144821,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.7467467061115505,"score_gpt":0.6691299934978945,"score_spread":0.07761671261365599,"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."}}