{"id":"W2184853928","doi":"10.1037/abn0000069","title":"Method matters: Understanding diagnostic reliability in DSM-IV and DSM-5.","year":2015,"lang":"en","type":"article","venue":"Journal of Abnormal Psychology","topic":"Mental Health Research Topics","field":"Psychology","cited_by":162,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Toronto","funders":"National Institute of Mental Health","keywords":"Reliability (semiconductor); Psychology; Medical diagnosis; DSM-5; Clinical psychology; Test (biology); Clinical trial; Psychiatry; Medicine; Pathology","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":[],"consensus_categories":[],"category_scores_codex":[0.005466392,0.0001711979,0.0004739663,0.0004682158,0.00005018876,0.00002601293,0.0003217581,0.0002370327,0.0005219974],"category_scores_gemma":[0.0006521947,0.0001496578,0.00007103269,0.0002922282,0.0002093791,0.0002630894,0.00008601433,0.001013338,0.00008480642],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0004103603,"about_ca_system_score_gemma":0.0001066733,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001436879,"about_ca_topic_score_gemma":0.0000529751,"domain_scores_codex":[0.9964767,0.001264774,0.0009602593,0.0003052335,0.0003733333,0.0006196938],"domain_scores_gemma":[0.9972919,0.001343891,0.0003852046,0.0003315588,0.0001216828,0.000525795],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.006070479,0.002077429,0.7176399,0.0003092827,0.00021776,0.004153816,0.01485738,0.0001144243,0.0002036723,0.01687757,0.1887813,0.04869698],"study_design_scores_gemma":[0.01635977,0.006487811,0.8065045,0.0002887729,0.00006779897,0.009040867,0.01101176,0.0001358757,0.00003907642,0.08911767,0.06036041,0.0005856797],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8979822,0.002384065,0.02390891,0.04430829,0.004449777,0.0006211343,0.00001089115,0.00002706117,0.02630767],"genre_scores_gemma":[0.9827591,0.0002179199,0.01037624,0.00581964,0.0004633928,0.00001243207,0.000001742265,0.00003107664,0.0003184718],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1284209,"threshold_uncertainty_score":0.6102867,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2257624653789143,"score_gpt":0.5066243302847697,"score_spread":0.2808618649058554,"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."}}