{"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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.08646503,0.0008325783,0.0009616695,0.003290237,0.001251057,0.002049503,0.001737504,0.001183836,0.000788663],"category_scores_gemma":[0.2503866,0.0005312273,0.00122869,0.001510979,0.002346675,0.002072304,0.003495042,0.001459379,0.0003009542],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0008876019,"about_ca_system_score_gemma":0.001538762,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.006735221,"about_ca_topic_score_gemma":0.009549269,"domain_scores_codex":[0.9458189,0.03592457,0.005032162,0.00386706,0.008062251,0.001295123],"domain_scores_gemma":[0.7556139,0.1619147,0.03839745,0.02170486,0.01903655,0.003332581],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0001852737,0.00008472854,0.9724224,0.00009090106,0.0002597569,0.0002845727,0.00335284,0.0003747154,0.0003555712,0.0003237021,0.0003588511,0.02190671],"study_design_scores_gemma":[0.0000321352,0.0008080449,0.9800633,0.0002611279,0.0002451501,0.001421879,0.00363471,0.006455245,0.00373592,0.001660814,0.001614576,0.00006710594],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9837459,0.0007762527,0.01059229,0.0004261133,0.00008277447,0.000246175,0.0001399698,0.00007402764,0.003916554],"genre_scores_gemma":[0.9947089,0.0001045043,0.004492241,0.0001236683,0.00002583755,0.0001671686,0.00009096554,0.00001765202,0.0002690274],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.08646503,"threshold_uncertainty_score":0.4572763,"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."}}