{"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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.07651779,0.002358333,0.002828114,0.006725502,0.003150334,0.006944834,0.007009426,0.02061984,0.02830298],"category_scores_gemma":[0.2792679,0.002366286,0.003975214,0.002541165,0.003520883,0.00745585,0.006648197,0.01313871,0.02334842],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.003874559,"about_ca_system_score_gemma":0.02816106,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01075773,"about_ca_topic_score_gemma":0.03454231,"domain_scores_codex":[0.9124352,0.05845216,0.01061546,0.001655542,0.01429127,0.002550371],"domain_scores_gemma":[0.6378967,0.2776324,0.008695264,0.008774429,0.05531483,0.01168652],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.0002661805,0.001615411,0.004306097,0.002015932,0.00006808798,0.00201778,0.002675316,0.002556743,0.001778914,0.0113769,0.4017235,0.569599],"study_design_scores_gemma":[0.001516878,0.001232278,0.03055819,0.04048677,0.0003111385,0.009462742,0.006007435,0.01454344,0.006030217,0.1160368,0.7731078,0.0007062979],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.00931341,0.01420985,0.5205374,0.2673746,0.004728708,0.01254921,0.001941129,0.01790425,0.1514414],"genre_scores_gemma":[0.02107747,0.003961006,0.9225193,0.02700634,0.0008757435,0.006822955,0.0005919463,0.0008857222,0.0162595],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.07651779,"threshold_uncertainty_score":0.4046696,"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."}}