{"id":"W2060998495","doi":"10.1002/cjs.10103","title":"Planning and analysis of measurement reliability studies","year":2011,"lang":"en","type":"article","venue":"Canadian Journal of Statistics","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":6,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Guelph; University of Waterloo","funders":"","keywords":"Reliability (semiconductor); Plan (archaeology); Measure (data warehouse); Mathematics; Statistics; Humanities; Computer science; Geography; Philosophy; Data mining; Physics","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":true,"about_ca":true,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.009263743,0.00009682337,0.0005170935,0.0006847206,0.0001090329,0.00004088013,0.0003365724,0.00003381897,0.0002532391],"category_scores_gemma":[0.01098147,0.00006795349,0.00009855678,0.0006800096,0.0003209941,0.0001161367,0.000019502,0.0001253408,0.000002069674],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001610423,"about_ca_system_score_gemma":0.0005331883,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0009710394,"about_ca_topic_score_gemma":0.008186801,"domain_scores_codex":[0.996953,0.0002084384,0.001147096,0.0001633289,0.001345645,0.0001824849],"domain_scores_gemma":[0.9949303,0.0005321247,0.0006714135,0.0002771204,0.003213568,0.0003755346],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00003912107,0.00004172382,0.9619161,0.00003255727,0.00121203,0.00006699218,0.01161898,0.001004233,0.00005344729,0.002767774,0.009338873,0.01190816],"study_design_scores_gemma":[0.0002330402,0.0003125824,0.93508,0.00008079864,0.0009489494,0.000005335329,0.005781964,0.0004673072,0.0001761233,0.05436087,0.002428413,0.0001246488],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8692992,0.00648817,0.1202747,0.0004170615,0.001023171,0.0002085028,0.000405343,0.000003305309,0.001880527],"genre_scores_gemma":[0.9811581,0.00004378837,0.0186987,0.00005555309,0.00001710346,5.510883e-7,5.221548e-7,0.000002981083,0.00002268114],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1118589,"threshold_uncertainty_score":0.9973494,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.4922490454234743,"score_gpt":0.3848306343614787,"score_spread":0.1074184110619956,"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."}}