{"id":"W2142200036","doi":"10.1002/sim.3110","title":"Reliability analysis for continuous measurements: Equivalence test for agreement","year":2007,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":34,"is_retracted":false,"has_abstract":true,"ca_institutions":"Toronto Western Hospital; University Health Network; Memorial University of Newfoundland; Canadian Blood Services; University of Toronto","funders":"","keywords":"Repeatability; Equivalence (formal languages); Reliability (semiconductor); Inter-rater reliability; Computer science; Reliability engineering; Intra-rater reliability; Statistics; Test (biology); Consistency (knowledge bases); Mathematics; Rating scale; Artificial intelligence; Engineering","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":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.2180608,0.001763618,0.004732963,0.00894103,0.001709374,0.00326608,0.003075668,0.002330831,0.005267062],"category_scores_gemma":[0.5577832,0.0009568849,0.004923812,0.008624847,0.005386297,0.005923394,0.005558687,0.00462281,0.001551976],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002057025,"about_ca_system_score_gemma":0.00441909,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001116363,"about_ca_topic_score_gemma":0.0007126562,"domain_scores_codex":[0.6621769,0.2416153,0.0218199,0.02046098,0.05212159,0.001805277],"domain_scores_gemma":[0.386167,0.5346315,0.01635979,0.02859927,0.03315164,0.001090761],"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":[0.004907642,0.0008974779,0.08637467,0.01016012,0.00964405,0.0007209774,0.01545015,0.02454078,0.004751404,0.1881354,0.01696744,0.6374499],"study_design_scores_gemma":[0.001800259,0.01033305,0.143543,0.004874429,0.00464783,0.003145656,0.007238552,0.2409981,0.008948206,0.4989619,0.07434412,0.001164898],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.02266398,0.001605032,0.9627304,0.0004554449,0.000474817,0.003451759,0.0007384124,0.0005028035,0.007377364],"genre_scores_gemma":[0.4004761,0.001060589,0.5779858,0.000394271,0.0003669595,0.0167083,0.001197499,0.0005557671,0.001254606],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.2180608,"threshold_uncertainty_score":0.9642702,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2076967185162903,"score_gpt":0.4449493606354994,"score_spread":0.237252642119209,"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."}}