{"id":"W2905051589","doi":"10.1177/0734282918816849","title":"Examining the Reliability of Scores From a Performance Assessment of Practice-Based Competencies","year":2018,"lang":"en","type":"article","venue":"Journal of Psychoeducational Assessment","topic":"Innovations in Medical Education","field":"Medicine","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"Medical Council of Canada; University of Alberta","funders":"","keywords":"Generalizability theory; Psychology; Reliability (semiconductor); Variance (accounting); Variation (astronomy); Variance components; Explained variation; Clinical psychology; Statistics; Applied psychology; Developmental psychology; Mathematics","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":[0.03468217,0.0005032073,0.000606334,0.002494869,0.0004400436,0.00106075,0.0006562984,0.0004825355,0.0006082747],"category_scores_gemma":[0.1338219,0.0002794146,0.0006058819,0.001253979,0.001055025,0.0008966235,0.001953114,0.0006107236,0.0004288698],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000420091,"about_ca_system_score_gemma":0.0005063267,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001497162,"about_ca_topic_score_gemma":0.002346888,"domain_scores_codex":[0.9659209,0.0149504,0.004017843,0.003048107,0.01129907,0.0007636225],"domain_scores_gemma":[0.873224,0.07551988,0.01501749,0.0101455,0.02514382,0.0009492646],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0004586334,0.0001254609,0.9346442,0.0001343629,0.0005419391,0.00009115409,0.005346401,0.0007971697,0.00473187,0.0002526397,0.0003622965,0.05251382],"study_design_scores_gemma":[0.00002202201,0.0007550991,0.9876511,0.00006194688,0.0001179674,0.0004050353,0.00146466,0.003935992,0.004121471,0.0002862582,0.00114578,0.00003259257],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9881182,0.0002223724,0.008817816,0.00005232074,0.00005411492,0.0001289868,0.000137224,0.00004655722,0.002422466],"genre_scores_gemma":[0.996452,0.00007599903,0.002815879,0.00001577664,0.00002028227,0.00007084234,0.0002352021,0.00002115418,0.0002929062],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9653178,"threshold_uncertainty_score":0.183419,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.04365185737418878,"score_gpt":0.4278329816498999,"score_spread":0.3841811242757111,"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."}}