{"id":"W4323652773","doi":"10.1002/bsl.2609","title":"From “below chance” to “a single error is one too many”: Evaluating various thresholds for invalid performance on two forced choice recognition tests","year":2023,"lang":"en","type":"article","venue":"Behavioral Sciences & the Law","topic":"Traumatic Brain Injury Research","field":"Medicine","cited_by":21,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Windsor","funders":"","keywords":"Malingering; Psychology; Binomial distribution; Statistics; Audiology; Two-alternative forced choice; Test (biology); Cognitive psychology; Clinical psychology; Mathematics; Medicine","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":true,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.0238573,0.0004954271,0.0006841237,0.001708279,0.0004039247,0.001312033,0.0006861475,0.0009346567,0.0009108059],"category_scores_gemma":[0.1172321,0.0002202732,0.0007122936,0.0007633148,0.002006055,0.001140522,0.001091276,0.0006916898,0.0003328736],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0006753033,"about_ca_system_score_gemma":0.0006394257,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002349591,"about_ca_topic_score_gemma":0.00349211,"domain_scores_codex":[0.9821324,0.008938336,0.002172072,0.001428293,0.004666091,0.0006627758],"domain_scores_gemma":[0.8754776,0.1005507,0.0129377,0.003814838,0.005503196,0.001715941],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.001660735,0.0001928199,0.9835682,0.00004365647,0.0001430102,0.00008349774,0.0006833142,0.0006824392,0.000770178,0.0002111487,0.0001355519,0.01182555],"study_design_scores_gemma":[0.00007482338,0.004351735,0.9747213,0.00009098717,0.0001522036,0.001201517,0.001216959,0.01147155,0.005084093,0.001016978,0.000542801,0.00007527192],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9974021,0.0001238822,0.001535722,0.00003812198,0.00000961174,0.00003604177,0.00007667091,0.00001482446,0.000763151],"genre_scores_gemma":[0.9987586,0.0000332892,0.0009907908,0.00002875908,0.00000407759,0.00002531492,0.00008754249,0.000005103384,0.00006657621],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9761427,"threshold_uncertainty_score":0.1261709,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.5522029002402319,"score_gpt":0.4986658389475107,"score_spread":0.05353706129272118,"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."}}