{"id":"W7093303755","doi":"10.17605/osf.io/skb3j","title":"Are Risk Assessment Tools More Accurate than Unstructured Judgments in Predicting Violent, Any, and Sexual Offending? A Meta-Analysis of Direct Comparison Studies","year":2025,"lang":"en","type":"dataset","venue":"OSF Preprints (OSF Preprints)","topic":"Psychopathy, Forensic Psychiatry, Sexual Offending","field":"Psychology","cited_by":0,"is_retracted":false,"has_abstract":false,"ca_institutions":"","funders":"Social Sciences and Humanities Research Council","keywords":"Risk assessment; Identification (biology); Risk management tools; Interpretation (philosophy)","routes":{"ca_aff":false,"ca_fund":true,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01839548,0.001692031,0.007092563,0.002425169,0.0008047803,0.002313726,0.003615448,0.002077771,0.0174729],"category_scores_gemma":[0.06926472,0.001087299,0.01209503,0.003185819,0.0007320044,0.001535251,0.002274193,0.002238557,0.001949151],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001160296,"about_ca_system_score_gemma":0.001766278,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.007735528,"about_ca_topic_score_gemma":0.01731253,"domain_scores_codex":[0.9880871,0.006744461,0.002408746,0.001572054,0.0009087916,0.0002789038],"domain_scores_gemma":[0.9557561,0.03338423,0.005164489,0.003480831,0.001710848,0.0005035637],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"meta_analysis","study_design_gemma":"meta_analysis","study_design_scores_codex":[0.02831396,0.0002846328,0.05146218,0.2102712,0.6105703,0.0001982716,0.000409019,0.001088543,0.0007770868,0.00113788,0.06611481,0.0293722],"study_design_scores_gemma":[0.04326035,0.0005539696,0.08414946,0.01932294,0.7990211,0.0004081757,0.0002449389,0.0009531793,0.0006836491,0.00358459,0.04760851,0.0002091464],"study_design_candidate":"meta_analysis","study_design_consensus":"meta_analysis","genre_codex":"dataset","genre_gemma":"dataset","genre_scores_codex":[0.06823141,0.2920775,0.009227381,0.003900094,0.0008232633,0.002953236,0.6173445,0.0009057943,0.004536839],"genre_scores_gemma":[0.6267576,0.04248809,0.02559327,0.005125668,0.0004975704,0.02191132,0.2705706,0.0008568115,0.006199035],"genre_candidate":"dataset","genre_consensus":"dataset","teacher_disagreement_score":0.01839548,"threshold_uncertainty_score":0.09728581,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.09409489194662782,"score_gpt":0.4064889892503991,"score_spread":0.3123940973037713,"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."}}