{"id":"W4294349341","doi":"10.1080/87565641.2022.2105847","title":"Cross-Cultural Performance Validity Testing: Managing False Positives in Examinees with Limited English Proficiency","year":2022,"lang":"en","type":"article","venue":"Developmental Neuropsychology","topic":"Traumatic Brain Injury Research","field":"Medicine","cited_by":18,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Windsor","funders":"","keywords":"Psychology; False positive paradox; Limited English proficiency; Mediation; Artificial intelligence; Computer science","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":[0.0004835937,0.0002164521,0.0002795494,0.0003366063,0.0003084085,0.00004703214,0.0002849088,0.00004210334,0.0002585166],"category_scores_gemma":[0.0003839938,0.0001856851,0.00002661243,0.001200931,0.0003375673,0.0002134662,0.0002577381,0.0007214952,0.00002315352],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002054961,"about_ca_system_score_gemma":0.000125998,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00001304992,"about_ca_topic_score_gemma":0.000002759137,"domain_scores_codex":[0.9977456,0.0002265427,0.000377717,0.0005868044,0.0004940907,0.0005692727],"domain_scores_gemma":[0.9991854,0.0002125648,0.00008524749,0.0002259093,0.0001762295,0.000114655],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.000730412,0.0004421366,0.9780974,0.00007139547,0.00002035963,0.0006831307,0.002664903,0.00003589505,0.008167514,0.0000449995,0.0001066048,0.008935291],"study_design_scores_gemma":[0.0021435,0.001603393,0.9909971,0.00004309015,0.000006820547,0.001854807,0.001428013,0.0002265692,0.000899572,0.000008965953,0.000575915,0.0002122271],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.980779,0.00001753943,0.000007645445,0.0003644143,0.0002589289,0.0006431781,0.000009034343,0.0001454187,0.01777488],"genre_scores_gemma":[0.9951146,0.000005459992,0.003252458,0.0006481713,0.0000434278,0.0002034043,0.00005136527,0.00003792882,0.0006432169],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.01713166,"threshold_uncertainty_score":0.7572019,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1208384285849731,"score_gpt":0.3556954365946144,"score_spread":0.2348570080096413,"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."}}