{"id":"W2122573908","doi":"10.1177/0013164415574086","title":"A Ratio Test of Interrater Agreement With High Specificity","year":2015,"lang":"en","type":"article","venue":"Educational and Psychological Measurement","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":6,"is_retracted":false,"has_abstract":true,"ca_institutions":"Université du Québec à Trois-Rivières; University of Ottawa","funders":"","keywords":"Inter-rater reliability; Agreement; Psychology; Validation test; Test validity; Test (biology); Psychometrics; Statistics; Clinical psychology; Developmental psychology; Rating scale; Mathematics; Geology","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":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.00528333,0.0001701539,0.0002696943,0.00009512017,0.00008314366,0.00009737622,0.000380836,0.00005455595,0.001703188],"category_scores_gemma":[0.001946345,0.00009212989,0.00005254454,0.0003157601,0.0002312704,0.0001226023,0.00005558969,0.0001258733,0.0001564558],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001023451,"about_ca_system_score_gemma":0.0001087666,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00002859096,"about_ca_topic_score_gemma":0.00003355092,"domain_scores_codex":[0.9951897,0.0002334922,0.0006947219,0.0005495645,0.003134748,0.0001977349],"domain_scores_gemma":[0.9972599,0.0003461392,0.0002418447,0.0004372252,0.001473638,0.0002412234],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00133138,0.01477403,0.6793236,0.00004478641,0.0001753233,0.000005648903,0.003288074,0.0003445926,0.01798966,0.03277455,0.2157815,0.03416684],"study_design_scores_gemma":[0.001200804,0.002027648,0.8835647,0.00006686495,0.00002113406,0.00001151831,0.001129347,0.0000234463,0.001793398,0.09393515,0.01597508,0.0002509104],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9590281,0.0002649256,0.002938334,0.02089463,0.0007964507,0.0004871495,0.00001388905,0.0000118836,0.01556458],"genre_scores_gemma":[0.996219,0.00000853102,0.002267763,0.0004943745,0.0001915504,0.00005213121,0.000004024907,0.000003744781,0.0007589278],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2042411,"threshold_uncertainty_score":0.9992094,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.4746679401283067,"score_gpt":0.4088811515647052,"score_spread":0.06578678856360154,"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."}}