{"id":"W3096173211","doi":"10.37213/cjal.2020.30436","title":"Examining Rater Performance on the CELBAN Speaking: A Many-Facets Rasch Measurement Analysis","year":2020,"lang":"en","type":"article","venue":"Canadian Journal of Applied Linguistics","topic":"Interpreting and Communication in Healthcare","field":"Health Professions","cited_by":9,"is_retracted":false,"has_abstract":true,"ca_institutions":"Queen's University","funders":"Queen's University","keywords":"Rasch model; Psychology; Licensure; Language proficiency; Language assessment; Grammar; Rating scale; Inter-rater reliability; Reliability (semiconductor); Test (biology); Context (archaeology); Item response theory; Psychometrics; Applied psychology; Scale (ratio); Judgement; Medical education; Clinical psychology; Mathematics education; Medicine; Developmental psychology; Linguistics","routes":{"ca_aff":true,"ca_fund":true,"ca_venue":true,"about_ca":true,"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":[0.04330741,0.0008822973,0.001115751,0.002621937,0.00118855,0.001925761,0.0008515132,0.0006586378,0.001480205],"category_scores_gemma":[0.09299967,0.000438915,0.00237897,0.002472017,0.001345204,0.001145231,0.002064058,0.001189393,0.0007279228],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001015355,"about_ca_system_score_gemma":0.001687,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.005367324,"about_ca_topic_score_gemma":0.009698868,"domain_scores_codex":[0.9621688,0.01632781,0.004292942,0.003859648,0.01248268,0.0008681447],"domain_scores_gemma":[0.891386,0.05633153,0.01393988,0.01036381,0.02707097,0.0009077502],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0006969003,0.0003561631,0.8591685,0.0005621993,0.001653038,0.0001222252,0.01783266,0.002648215,0.003301384,0.002187431,0.003009804,0.1084615],"study_design_scores_gemma":[0.00009138586,0.001813845,0.9514599,0.0003200677,0.0005502193,0.0002611893,0.006898668,0.02330045,0.005592442,0.002504473,0.007020112,0.0001871909],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9281992,0.0004710948,0.06171585,0.0001955197,0.0001339717,0.001816166,0.000988378,0.0002346784,0.006245103],"genre_scores_gemma":[0.9706139,0.0001342196,0.02510856,0.00006808298,0.00003431684,0.001955243,0.0008209572,0.00006279669,0.001201974],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.04330741,"threshold_uncertainty_score":0.2290342,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1965768547620296,"score_gpt":0.3578816742305013,"score_spread":0.1613048194684717,"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."}}