{"id":"W2078744383","doi":"10.1037/a0028195","title":"Estimating the accuracy of neurocognitive effort measures in the absence of a “gold standard”.","year":2012,"lang":"en","type":"article","venue":"Psychological Assessment","topic":"Traumatic Brain Injury Research","field":"Medicine","cited_by":3,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Alberta","funders":"","keywords":"Malingering; Bayesian probability; Computer science; Context (archaeology); False positive paradox; Gold standard (test); Econometrics; Receiver operating characteristic; Neurocognitive; Test (biology); Statistics; Machine learning; Markov chain Monte Carlo; Artificial intelligence; Psychology; Mathematics; Clinical psychology; Cognition; Biology","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":[0.00294333,0.0001139333,0.0002961959,0.00005330234,0.00003837587,0.00001168361,0.0003190346,0.00005612772,0.0001606085],"category_scores_gemma":[0.001214564,0.00005025731,0.00008635976,0.000340154,0.0002836002,0.00006666335,0.00006389883,0.0004816208,0.000005484804],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00003280662,"about_ca_system_score_gemma":0.00004148222,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00001611536,"about_ca_topic_score_gemma":0.000001624658,"domain_scores_codex":[0.9978367,0.0003222236,0.0004500851,0.0001705775,0.0009031131,0.0003172657],"domain_scores_gemma":[0.997519,0.001695899,0.0001769309,0.0004058414,0.0001374849,0.00006490314],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.001085871,0.0043268,0.7830441,0.0003878824,0.0001679619,0.00004175057,0.005750443,0.00003607239,0.01164927,0.007514866,0.002544533,0.1834504],"study_design_scores_gemma":[0.0008391355,0.0007907714,0.9954583,0.0001415249,0.00002213128,0.00004793459,0.0009959321,0.0002278022,0.0004287785,0.000820028,0.000172701,0.00005494551],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9749978,0.00009200249,0.002532017,0.002447294,0.000100293,0.001007632,0.00001105534,0.00001320712,0.01879865],"genre_scores_gemma":[0.9914683,0.00001084087,0.007742625,0.0005602479,0.00006974652,0.0001249771,0.000002441196,0.000007271713,0.00001354997],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2124142,"threshold_uncertainty_score":0.209243,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1816961584107314,"score_gpt":0.487004782171133,"score_spread":0.3053086237604017,"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."}}