{"id":"W1967929637","doi":"10.1016/j.jclinepi.2004.07.010","title":"The standard gamble demonstrated lower reliability than the feeling thermometer","year":2005,"lang":"en","type":"article","venue":"Journal of Clinical Epidemiology","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":27,"is_retracted":false,"has_abstract":false,"ca_institutions":"University of Toronto; McMaster University","funders":"Medical Research Council Canada","keywords":"Intraclass correlation; Reliability (semiconductor); Statistics; Psychology; Variance (accounting); Feeling; Audiology; Physical therapy; Medicine; Mathematics; Psychometrics; Social psychology","routes":{"ca_aff":true,"ca_fund":true,"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":[0.03664403,0.001143734,0.001276955,0.001670098,0.0006570268,0.001848405,0.0008795547,0.002038338,0.004381045],"category_scores_gemma":[0.1575528,0.0006727157,0.002670141,0.001306019,0.001265607,0.002074229,0.001485104,0.001565575,0.002787185],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000514027,"about_ca_system_score_gemma":0.0007267497,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00212238,"about_ca_topic_score_gemma":0.00270924,"domain_scores_codex":[0.9827271,0.009802241,0.001871677,0.002200237,0.002975398,0.0004233377],"domain_scores_gemma":[0.7349074,0.2061832,0.0105105,0.01586165,0.03124931,0.001287894],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.006862452,0.0009440994,0.7520632,0.001718772,0.002347062,0.0002014407,0.006531483,0.007604972,0.005302313,0.005750106,0.007683581,0.2029905],"study_design_scores_gemma":[0.000290576,0.004197003,0.9409669,0.0008423842,0.001465412,0.0005508177,0.001621938,0.02347152,0.005073345,0.008823726,0.01240013,0.0002962545],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9362065,0.002338918,0.03984012,0.0005960526,0.001593625,0.0007212047,0.001485316,0.0005403699,0.0166778],"genre_scores_gemma":[0.982104,0.0004180452,0.01154591,0.0003552197,0.0002288797,0.0004177896,0.0006726914,0.0002161582,0.004041196],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.963356,"threshold_uncertainty_score":0.1937944,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.6268485298520915,"score_gpt":0.5692764832966586,"score_spread":0.05757204655543291,"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."}}