{"id":"W2560836925","doi":"10.1177/0962280216681599","title":"Interval estimation for a proportion using a double-sampling scheme with two fallible classifiers","year":2016,"lang":"en","type":"article","venue":"Statistical Methods in Medical Research","topic":"Advanced Statistical Process Monitoring","field":"Decision Sciences","cited_by":8,"is_retracted":false,"has_abstract":true,"ca_institutions":"Robarts Clinical Trials; Western University","funders":"","keywords":"Confidence interval; Statistics; Credible interval; Robust confidence intervals; Percentile; Coverage probability; Mathematics; Tolerance interval; Binomial proportion confidence interval; Sample size determination; Classifier (UML); Confidence distribution; Computer science; Artificial intelligence; Negative binomial distribution; Poisson distribution","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":[0.03655063,0.0002201013,0.0005724445,0.00064145,0.0003060784,0.0002157233,0.0008638664,0.0001961802,0.0006727817],"category_scores_gemma":[0.2619973,0.000123148,0.00005208405,0.00165022,0.001564012,0.0004799658,0.0003317578,0.0009040622,0.00003688771],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0004840808,"about_ca_system_score_gemma":0.0008811625,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00006806826,"about_ca_topic_score_gemma":0.00005329591,"domain_scores_codex":[0.9885982,0.001858767,0.001333907,0.001133174,0.005900903,0.001175013],"domain_scores_gemma":[0.9244221,0.07254213,0.0002011169,0.0005289936,0.001479224,0.0008264656],"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":[0.001160489,0.00008104646,0.0008784959,0.00005845346,0.0000105157,0.00004726081,0.0000775466,0.0001384616,0.002220145,0.1354358,0.0000574752,0.8598343],"study_design_scores_gemma":[0.00247752,0.0003411376,0.0004490081,0.000470093,0.000006924723,0.00001638675,0.0002858776,0.3509978,0.001257824,0.6426612,0.0008417054,0.0001945749],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.004192549,0.00003222238,0.9927521,0.001342167,0.0002461135,0.0007926053,0.00004035094,0.00004256211,0.000559322],"genre_scores_gemma":[0.1327312,0.000005115608,0.8666396,0.00003066594,0.0001549762,0.0002393282,0.000003671248,0.00003433251,0.0001611552],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.8596397,"threshold_uncertainty_score":0.9920739,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.7224746746721421,"score_gpt":0.731382168623437,"score_spread":0.008907493951294865,"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."}}