{"id":"W2055830213","doi":"10.1186/1471-2288-13-134","title":"How confidence intervals become confusion intervals","year":2013,"lang":"en","type":"article","venue":"BMC Medical Research Methodology","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":52,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Alberta; University of British Columbia","funders":"","keywords":"Confidence interval; Harm; Confusion; MEDLINE; Medicine; Point estimation; Statistical significance; Psychology; Statistics; Social psychology; Mathematics","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","scholarly_communication","open_science","insufficient_payload"],"consensus_categories":["metaresearch","insufficient_payload"],"category_scores_codex":[0.717023,0.000380291,0.005582958,0.001247337,0.0002651778,0.001891003,0.007432532,0.0005313557,0.2608238],"category_scores_gemma":[0.9161035,0.0001751917,0.001836277,0.002146498,0.001361908,0.0004795353,0.001963127,0.001406083,0.02783901],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00007336694,"about_ca_system_score_gemma":0.0007742319,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0004314333,"about_ca_topic_score_gemma":0.0006230352,"domain_scores_codex":[0.535297,0.4216807,0.01107571,0.002493015,0.02790735,0.001546209],"domain_scores_gemma":[0.1218879,0.842664,0.005284731,0.01208706,0.0144243,0.00365198],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","study_design_scores_codex":[0.00003837542,0.0001216194,0.01081347,0.0002480993,0.000202949,0.00005424163,0.0009746597,0.000001650678,0.003020752,0.04195027,0.5410313,0.4015426],"study_design_scores_gemma":[0.0009875626,0.0006976826,0.02801173,0.0005038665,0.00009051008,0.000238088,0.009674029,0.02505388,0.001196299,0.3545539,0.5784,0.0005924797],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.1270233,0.001563918,0.8276038,0.0331499,0.0009645898,0.001677968,0.000005608925,0.00001793265,0.007993005],"genre_scores_gemma":[0.6240155,0.0004583626,0.2258106,0.003085554,0.0008260259,0.0008188924,0.00001298875,0.0000502482,0.1449219],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.6017932,"threshold_uncertainty_score":0.9991452,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.9715599684711017,"score_gpt":0.7146335475274728,"score_spread":0.2569264209436289,"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."}}