{"id":"W2095752327","doi":"10.1002/(sici)1097-0258(20000315)19:5<723::aid-sim379>3.0.co;2-a","title":"Interval estimation for Cohen's kappa as a measure of agreement","year":2000,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":217,"is_retracted":false,"has_abstract":true,"ca_institutions":"Western University","funders":"","keywords":"Confidence interval; Statistics; Mathematics; Measure (data warehouse); Kappa; Statistic; Variance (accounting); Delta method; Coverage probability; Asymptotic analysis; Computation; Asymptotic distribution; Cohen's kappa; Applied mathematics; Computer science; Algorithm","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":["metaresearch","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.007252351,0.0001381291,0.0004369779,0.000180623,0.00005625889,0.0000204042,0.0004393494,0.00005284913,0.006227912],"category_scores_gemma":[0.008945916,0.00009497673,0.00004560084,0.0003774818,0.0002492155,0.00007386786,0.00002588437,0.0001163929,0.0001100698],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00007963402,"about_ca_system_score_gemma":0.00007118362,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001450716,"about_ca_topic_score_gemma":0.0001421121,"domain_scores_codex":[0.9958637,0.0001877172,0.001203565,0.0003503025,0.002177019,0.0002176944],"domain_scores_gemma":[0.9970811,0.001633237,0.0002395192,0.0004441712,0.0005190375,0.00008291486],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.0004506093,0.0003054597,0.003310242,0.000110228,0.0000390168,0.00000870051,0.003275936,0.004834862,0.0005035896,0.02728542,0.09017988,0.869696],"study_design_scores_gemma":[0.004283762,0.002271252,0.01531838,0.0006632088,0.00008327864,0.000004389221,0.00209458,0.08427042,0.0005615975,0.8536577,0.03650804,0.000283358],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.09164632,0.0003324673,0.8774809,0.005971984,0.0009288197,0.001759466,0.0002601117,0.00002090377,0.02159907],"genre_scores_gemma":[0.934467,0.00005486016,0.06239018,0.0004828379,0.0001126822,0.00006699333,0.00004351584,0.00000935096,0.002372566],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.8694127,"threshold_uncertainty_score":0.9994022,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1194851870437682,"score_gpt":0.4260948892710158,"score_spread":0.3066097022272476,"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."}}