{"id":"W2025888974","doi":"10.1371/journal.pone.0114663","title":"Assessing the Assessment in Emergency Care Training","year":2014,"lang":"en","type":"article","venue":"PLoS ONE","topic":"Innovations in Medical Education","field":"Medicine","cited_by":23,"is_retracted":false,"has_abstract":true,"ca_institutions":"McMaster University","funders":"Erasmus Universitair Medisch Centrum Rotterdam; Stichting Beroepsopleiding Huisarts","keywords":"Checklist; Generalizability theory; Scale (ratio); Reliability (semiconductor); Construct validity; Psychology; Inter-rater reliability; Medicine; Clinical psychology; Validity; Emergency department; Psychometrics; Rating scale; Nursing","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":[],"category_scores_codex":[0.0158611,0.0004464975,0.0004102671,0.001168908,0.0005484128,0.00109916,0.0009199722,0.0005901271,0.002225616],"category_scores_gemma":[0.08175173,0.0002001375,0.0003620676,0.0009051299,0.001127044,0.001340986,0.002291634,0.0009779903,0.0007329045],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001197178,"about_ca_system_score_gemma":0.002459534,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001263334,"about_ca_topic_score_gemma":0.001961693,"domain_scores_codex":[0.9850761,0.009068353,0.001280695,0.0005590578,0.00350729,0.0005084132],"domain_scores_gemma":[0.964502,0.01600306,0.007189346,0.001414106,0.009781912,0.001109576],"domain_codex":null,"domain_gemma":"evaluation","domain_candidate":"evaluation","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.0004574074,0.0007796814,0.4149121,0.002022364,0.0001715883,0.0002697664,0.005823866,0.00185922,0.003067173,0.001287302,0.009233406,0.5601162],"study_design_scores_gemma":[0.0001145991,0.002190532,0.9487782,0.002631751,0.00009715193,0.002211252,0.007246922,0.004958053,0.00371075,0.002773111,0.02519274,0.00009489517],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9211128,0.005406681,0.03895622,0.002970476,0.0006226518,0.002197383,0.0007205525,0.000235609,0.02777758],"genre_scores_gemma":[0.9561747,0.002021179,0.03708655,0.0003743239,0.0001331265,0.001504785,0.0003293254,0.00002891016,0.002347112],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9841389,"threshold_uncertainty_score":0.08388257,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1074143091989156,"score_gpt":0.3999828747110414,"score_spread":0.2925685655121258,"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."}}