{"id":"W1970971885","doi":"10.1016/j.jclinepi.2011.10.019","title":"A confidence interval approach to sample size estimation for interobserver agreement studies with multiple raters and outcomes","year":2012,"lang":"en","type":"article","venue":"Journal of Clinical Epidemiology","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":131,"is_retracted":false,"has_abstract":false,"ca_institutions":"Western University; York University","funders":"","keywords":"Confidence interval; Triage; Sample size determination; Reliability (semiconductor); Medicine; Estimation; Medical physics; Sample (material); Interval estimation; Statistics; Scale (ratio); Mathematics; Psychiatry; Power (physics)","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"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.09069584,0.0001803115,0.002053035,0.00009117781,0.0000845443,0.00003203431,0.0004777616,0.0001088144,0.00002711038],"category_scores_gemma":[0.5352219,0.00008927724,0.0003826142,0.0001218673,0.0003215185,0.0003679059,0.0001625698,0.0002908761,0.00001083593],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00005151686,"about_ca_system_score_gemma":0.0000494451,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00001758943,"about_ca_topic_score_gemma":0.0000239838,"domain_scores_codex":[0.9912939,0.002782632,0.004568639,0.0003588049,0.0006178538,0.0003781493],"domain_scores_gemma":[0.6780325,0.3153101,0.003940808,0.0005730192,0.001509238,0.0006343322],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.001103802,0.0003167372,0.9745274,0.00003082355,0.0002538756,3.571976e-7,0.0006800868,0.001985873,0.000007718636,0.001437612,0.005383465,0.01427224],"study_design_scores_gemma":[0.002237365,0.00287811,0.9346225,0.0001292804,0.00009984955,0.00001553144,0.002355095,0.004781458,0.00001742879,0.04831425,0.004381427,0.0001676564],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.5510898,0.0002165327,0.4377364,0.009640635,0.0009378351,0.0003496362,0.000005271525,0.000003412427,0.00002046452],"genre_scores_gemma":[0.7412383,0.00005265777,0.2546178,0.003840247,0.0001733146,0.00001950056,3.934305e-7,0.000004288615,0.0000535332],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.4445261,"threshold_uncertainty_score":0.93632,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.7278829789470725,"score_gpt":0.5892656106725227,"score_spread":0.1386173682745498,"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."}}