{"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":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.08440655,0.003306514,0.01289451,0.01468502,0.001677286,0.01400916,0.006889733,0.01176646,0.004662692],"category_scores_gemma":[0.2399666,0.001902039,0.003642855,0.01617365,0.03032742,0.02654573,0.007876617,0.01430784,0.003336005],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.003495447,"about_ca_system_score_gemma":0.01107837,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001778671,"about_ca_topic_score_gemma":0.002015358,"domain_scores_codex":[0.8563383,0.06882838,0.02596654,0.006863461,0.04044553,0.001557832],"domain_scores_gemma":[0.610328,0.3351598,0.02501561,0.009708659,0.01755492,0.002233133],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.0003798386,0.0001282789,0.002162852,0.04512649,0.001834178,0.0001744826,0.001925149,0.001050344,0.0004769255,0.2417904,0.04069464,0.6642565],"study_design_scores_gemma":[0.0004336022,0.000448141,0.005033592,0.06269553,0.003618574,0.003334086,0.002380859,0.001780316,0.001142711,0.547768,0.3708443,0.0005203561],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.0008315027,0.8730987,0.05673955,0.05375074,0.009579377,0.0001012941,0.0002120639,0.0001218141,0.005564966],"genre_scores_gemma":[0.07494909,0.6804658,0.1148896,0.07850616,0.04336712,0.001682391,0.0003660358,0.0002913956,0.005482451],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.9155934,"threshold_uncertainty_score":0.4463899,"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."}}