{"id":"W3096723063","doi":"10.1037/pas0000958","title":"Development of embedded performance validity indicators in the NIH Toolbox Cognitive Battery.","year":2020,"lang":"en","type":"article","venue":"Psychological Assessment","topic":"Traumatic Brain Injury Research","field":"Medicine","cited_by":19,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of British Columbia; University of Windsor","funders":"Vancouver Coastal Health Research Institute; Michael Smith Health Research BC; WorkSafeBC","keywords":"Psychology; Cutoff; Univariate; PsycINFO; Multivariate analysis; Receiver operating characteristic; Psychometrics; Cognition; Test validity; Battery (electricity); Multivariate statistics; Clinical psychology; Audiology; Psychiatry; Medicine; MEDLINE; Machine learning; Computer science","routes":{"ca_aff":true,"ca_fund":true,"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":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.001089072,0.0001516766,0.0003408301,0.0001168777,0.00005987538,0.00002070637,0.00029724,0.00009562358,0.000944239],"category_scores_gemma":[0.0002298874,0.00008882504,0.00006519578,0.0007346782,0.0001536645,0.00006205186,0.00006709963,0.0005840564,0.00007619463],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00006108103,"about_ca_system_score_gemma":0.0001144491,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":5.715118e-7,"about_ca_topic_score_gemma":4.559883e-7,"domain_scores_codex":[0.9978097,0.0002402021,0.000528694,0.0003510665,0.0007606445,0.000309691],"domain_scores_gemma":[0.9990335,0.0004321671,0.0001133732,0.0002214977,0.00005409344,0.0001453579],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.001432531,0.00524803,0.6422966,0.0006394066,0.0002160049,0.0001646059,0.01899977,0.000002833005,0.005045545,0.0007033505,0.003121061,0.3221303],"study_design_scores_gemma":[0.001611777,0.001026465,0.9920751,0.00009469697,0.00001922773,0.00001390496,0.002392951,0.0001113352,0.001490614,0.0000216407,0.001034491,0.0001077398],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9737127,0.00001098813,0.001127909,0.004321805,0.00004705399,0.0007807552,0.000007575107,0.0000297316,0.01996145],"genre_scores_gemma":[0.986504,0.00000965664,0.009807399,0.003417423,0.00005281844,0.0001528538,0.00002752094,0.000008999685,0.00001939453],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.3497786,"threshold_uncertainty_score":0.999969,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2969758375088455,"score_gpt":0.4643634012355106,"score_spread":0.1673875637266651,"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."}}