{"id":"W2073056556","doi":"10.1016/j.jclinepi.2005.09.005","title":"“Precision” and “Accuracy”: Two Terms That Are Neither","year":2006,"lang":"en","type":"review","venue":"Journal of Clinical Epidemiology","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":194,"is_retracted":false,"has_abstract":false,"ca_institutions":"Baycrest Hospital; McMaster University; University of Toronto","funders":"","keywords":"Reliability (semiconductor); Conceptualization; Inter-rater reliability; Scale (ratio); Test (biology); Computer science; Synonym (taxonomy); Validity; Criterion validity; Psychology; Process (computing); Cognitive psychology; Psychometrics; Artificial intelligence; Statistics; Mathematics; Rating scale; Construct validity; Geography","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","metaepi_narrow","metaepi_broad","research_integrity"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.2454556,0.0005666703,0.01731947,0.001125968,0.0001216012,0.0001068938,0.002515462,0.00126043,0.00029993],"category_scores_gemma":[0.9239976,0.0002734778,0.003453757,0.001174284,0.000592939,0.0002423109,0.0005891158,0.003184811,0.0001010976],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00003854253,"about_ca_system_score_gemma":0.0002926205,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00001788585,"about_ca_topic_score_gemma":0.000006142411,"domain_scores_codex":[0.9344106,0.04206487,0.02042772,0.001182595,0.001182417,0.0007318214],"domain_scores_gemma":[0.1495854,0.8122269,0.036054,0.001120459,0.0004868119,0.0005264995],"domain_codex":null,"domain_gemma":"methods","domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.00002584851,0.00005045498,0.06419694,0.0001452878,0.0001051304,0.00006961007,0.000002108507,0.000006875588,4.056935e-9,0.0001823184,0.04666371,0.8885517],"study_design_scores_gemma":[0.0004751779,0.0002590205,0.03342006,0.002617276,0.0002631297,0.0008479923,0.00001574924,0.00003177483,7.143739e-9,0.08696296,0.8748732,0.0002336323],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.0002084478,0.9806215,0.009252972,0.001239733,0.0060287,0.0002293562,0.00001965283,0.00001455929,0.002385144],"genre_scores_gemma":[0.00003590116,0.9553931,0.03956634,0.001164817,0.002940106,0.000003601595,0.000001644241,0.00003358386,0.0008609406],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.8883181,"threshold_uncertainty_score":0.9999717,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.9346514191644626,"score_gpt":0.7216304917470712,"score_spread":0.2130209274173914,"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."}}