{"id":"W2990367555","doi":"10.1016/j.cptl.2019.10.001","title":"Guidance for high-stakes testing within pharmacy educational assessment","year":2019,"lang":"en","type":"article","venue":"Currents in Pharmacy Teaching and Learning","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":15,"is_retracted":false,"has_abstract":false,"ca_institutions":"University of Alberta","funders":"","keywords":"Generalizability theory; Pharmacy; Reliability (semiconductor); Item response theory; Psychology; Standardized test; Medical education; Engineering ethics; Medicine; Psychometrics; Mathematics education; Clinical psychology; Engineering; 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":[0.01971838,0.0002206046,0.0003699174,0.0005291524,0.0006496189,0.0003472416,0.0005669466,0.00002880742,0.0001991036],"category_scores_gemma":[0.09219567,0.0001833856,0.00006091471,0.0007838117,0.00004907444,0.0004797764,0.0001656249,0.001529908,0.00002304002],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00007467904,"about_ca_system_score_gemma":0.0001477784,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000218291,"about_ca_topic_score_gemma":0.00000139669,"domain_scores_codex":[0.9954864,0.001750604,0.0008592739,0.0007651596,0.0007069104,0.0004316736],"domain_scores_gemma":[0.9561568,0.04268391,0.0006102383,0.0002461916,0.0001770207,0.0001258079],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.00001641673,0.00005805624,0.6353139,0.00002634821,0.000008117803,4.78087e-7,0.0006478714,0.003508255,0.0007422139,0.0007543283,0.0004202321,0.3585038],"study_design_scores_gemma":[0.005966288,0.0001576182,0.3979483,0.0005586366,0.00003428774,0.00002921621,0.003369301,0.4868816,0.0001727516,0.01709053,0.0869825,0.000808948],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9801601,0.0007307559,0.01053147,0.0006395346,0.0029067,0.0003539324,0.000006453015,0.00007348544,0.004597595],"genre_scores_gemma":[0.8856263,0.00001311117,0.1125864,0.0001319016,0.0002970964,0.00003960042,0.000008926114,0.0000191187,0.001277542],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.4833734,"threshold_uncertainty_score":0.9154512,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.4303950496053853,"score_gpt":0.5504070374077461,"score_spread":0.1200119878023608,"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."}}