{"id":"W2028787956","doi":"10.1111/medu.12621","title":"A systematic review of validity evidence for checklists versus global rating scales in simulation‐based assessment","year":2015,"lang":"en","type":"review","venue":"Medical Education","topic":"Simulation-Based Education in Healthcare","field":"Medicine","cited_by":328,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of British Columbia; University of Calgary","funders":"","keywords":"Checklist; Context (archaeology); Confidence interval; Medicine; Reliability (semiconductor); MEDLINE; Systematic review; Validity; Psychometrics; Psychology; Clinical psychology; Internal 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","metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.005417682,0.0003874013,0.002879255,0.0002570637,0.00005734006,0.00001869939,0.0003130361,0.0004719738,0.0001488694],"category_scores_gemma":[0.06753957,0.0003221561,0.0004596268,0.001186373,0.00008987558,0.0001112786,0.00003081452,0.0004145085,0.00001433016],"about_ca_system_candidate":true,"about_ca_system_consensus":true,"about_ca_system_score_codex":0.003863647,"about_ca_system_score_gemma":0.04533872,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001027581,"about_ca_topic_score_gemma":0.0000234919,"domain_scores_codex":[0.9935123,0.001090313,0.002854622,0.0005431589,0.00170419,0.0002954011],"domain_scores_gemma":[0.9825313,0.01105945,0.002164104,0.001066308,0.002661212,0.0005176008],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"systematic_review","study_design_gemma":"systematic_review","study_design_scores_codex":[0.000029467,0.0003807468,0.0005545624,0.9679756,0.00004407854,6.523666e-7,0.00004692188,0.0000421148,4.04307e-9,0.00007529717,0.0011077,0.02974288],"study_design_scores_gemma":[0.0007758173,0.000172082,0.0001753637,0.9815278,0.002053453,0.000005558303,0.000103143,0.007699332,6.783728e-8,0.00003514534,0.007226312,0.000225935],"study_design_candidate":"systematic_review","study_design_consensus":"systematic_review","genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.00002825728,0.9855518,0.001265722,0.001877008,0.001960678,0.009167932,0.00003786068,0.00004252218,0.00006820891],"genre_scores_gemma":[0.002593593,0.9867468,0.003020385,0.001308089,0.0006226932,0.003975002,0.001660581,0.00004785033,0.00002495565],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.06212189,"threshold_uncertainty_score":0.9999604,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.3337116927577552,"score_gpt":0.610282715440884,"score_spread":0.2765710226831288,"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."}}