{"id":"W2932922973","doi":"10.3138/jvme.0917-128r1","title":"Inter-Rater Reliability of Grading Undergraduate Portfolios in Veterinary Medical Education","year":2019,"lang":"en","type":"article","venue":"Journal of Veterinary Medical Education","topic":"Innovations in Medical Education","field":"Medicine","cited_by":10,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Grading (engineering); Intraclass correlation; Inter-rater reliability; Medicine; Cohort; Psychology; Veterinary medicine; Medical education; Statistics; Mathematics; Pathology; Clinical psychology; Biology; Rating scale; Psychometrics","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":true,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.1111653,0.0005387107,0.001161247,0.002403998,0.0006083377,0.001685584,0.001014354,0.0008611468,0.0007978962],"category_scores_gemma":[0.1860626,0.000387401,0.001059305,0.001641727,0.001516416,0.001336357,0.002717668,0.0007461829,0.0006414384],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0007696335,"about_ca_system_score_gemma":0.0006594864,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0007504505,"about_ca_topic_score_gemma":0.001030198,"domain_scores_codex":[0.8669106,0.08360379,0.01540683,0.009534702,0.02264907,0.00189499],"domain_scores_gemma":[0.7849936,0.1305229,0.02071417,0.01792375,0.04381131,0.002034201],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.002656044,0.0005222149,0.826434,0.0007969448,0.001891238,0.0002328643,0.01552625,0.003052127,0.005420904,0.0009601211,0.001330938,0.1411762],"study_design_scores_gemma":[0.0001094384,0.001622948,0.9753714,0.0002368594,0.0003556026,0.0004282761,0.002448859,0.008990813,0.005918894,0.001192309,0.003226464,0.00009803062],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.964963,0.001611072,0.0248375,0.0001483458,0.0002742508,0.0009389037,0.0003859229,0.0001249307,0.006716083],"genre_scores_gemma":[0.992991,0.0001914896,0.005437095,0.00003308286,0.00005269953,0.0003530585,0.0002660896,0.00003778663,0.0006378377],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1111653,"threshold_uncertainty_score":0.5879052,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.03548445773637397,"score_gpt":0.3961437057285639,"score_spread":0.36065924799219,"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."}}