{"id":"W2007345088","doi":"10.1371/journal.pone.0073990","title":"The Case for Using the Repeatability Coefficient When Calculating Test–Retest Reliability","year":2013,"lang":"en","type":"article","venue":"PLoS ONE","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":474,"is_retracted":false,"has_abstract":true,"ca_institutions":"Dalhousie University","funders":"Curtin University of Technology","keywords":"Intraclass correlation; Repeatability; Reliability (semiconductor); Strengths and weaknesses; Computer science; Pearson product-moment correlation coefficient; Statistics; Test (biology); Measure (data warehouse); Reliability engineering; Correlation coefficient; Data mining; Medical physics; Machine learning; Mathematics; Reproducibility; Medicine; Psychology; Engineering","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":["metaresearch"],"category_scores_codex":[0.4940352,0.002530921,0.005727397,0.009974007,0.003664764,0.01062555,0.006213953,0.008707317,0.001440989],"category_scores_gemma":[0.75715,0.001976619,0.004891769,0.01124025,0.01495066,0.0162407,0.007405201,0.01882054,0.001546703],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.004378862,"about_ca_system_score_gemma":0.009408763,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.008443343,"about_ca_topic_score_gemma":0.008540183,"domain_scores_codex":[0.3533055,0.489402,0.05007901,0.03239944,0.07223766,0.00257642],"domain_scores_gemma":[0.1461179,0.7178635,0.03199954,0.05380201,0.0482715,0.001945594],"domain_codex":"methods","domain_gemma":"methods","domain_candidate":"methods","domain_consensus":"methods","study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.001156886,0.000185463,0.1146,0.009080076,0.006611285,0.001379752,0.02302811,0.004791981,0.003284951,0.1871354,0.05450526,0.5942408],"study_design_scores_gemma":[0.0004619239,0.003649607,0.08527841,0.03127594,0.004109183,0.008933462,0.01298787,0.03452655,0.01126062,0.5341878,0.2714023,0.001926418],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.02263666,0.06789897,0.7834129,0.08140676,0.0192135,0.0008153017,0.0004081923,0.001189588,0.02301807],"genre_scores_gemma":[0.3229043,0.01125865,0.6215421,0.02839028,0.009178137,0.001747906,0.0002232873,0.001584668,0.003170605],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.5059648,"threshold_uncertainty_score":0.6239446,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2965950402626327,"score_gpt":0.3528709709396792,"score_spread":0.0562759306770465,"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."}}