{"id":"W1506179204","doi":"10.1002/hec.3100","title":"Test–Retest Reliability of Capability Measurement in the UK General Population","year":2014,"lang":"en","type":"article","venue":"Health Economics","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":44,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of British Columbia","funders":"Medical Research Council; National Institute for Health and Care Research","keywords":"Reliability (semiconductor); Test (biology); Reliability engineering; Population; Statistics; Medicine; Psychology; Environmental health; Mathematics; Engineering; Biology","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":[],"category_scores_codex":[0.06232097,0.0002175906,0.001177188,0.0002206077,0.0001795096,0.00004358341,0.0004242823,0.0001505389,0.0001144245],"category_scores_gemma":[0.008321796,0.0002382391,0.0001428864,0.0001811743,0.00009216144,0.0002692802,0.0000429561,0.0002507728,0.0001883461],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001828239,"about_ca_system_score_gemma":0.0003105688,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01480672,"about_ca_topic_score_gemma":0.006313752,"domain_scores_codex":[0.9917333,0.0008726825,0.006070425,0.0006790232,0.0001093077,0.0005352879],"domain_scores_gemma":[0.993894,0.001639864,0.003047052,0.001151809,0.00009426101,0.0001729637],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.000011233,0.0001910335,0.9246212,0.0005167922,0.000007404435,3.705056e-8,0.00130437,0.002249125,5.276718e-7,0.0676591,0.002773928,0.0006653032],"study_design_scores_gemma":[0.0007156366,0.0001853131,0.9106058,0.0000476633,0.000002564364,0.000002029002,0.0002209426,0.02512406,0.000002478472,0.04581481,0.01702845,0.0002502886],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9539498,0.0006627944,0.0006970746,0.04043763,0.0005529504,0.001362186,0.0002000564,0.00002800437,0.002109511],"genre_scores_gemma":[0.9848445,0.0001671446,0.002323105,0.01216546,0.0002801412,0.0001212083,0.00004842848,0.00002463562,0.00002535788],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.05399917,"threshold_uncertainty_score":0.9962568,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.3379082102048664,"score_gpt":0.4085599441313604,"score_spread":0.07065173392649399,"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."}}