{"id":"W4313892487","doi":"10.1007/978-3-031-21796-8_8","title":"None of the Above: Integrity Concerns of Standardized English Proficiency Tests","year":2023,"lang":"en","type":"book-chapter","venue":"Ethics and integrity in educational contexts","topic":"Multilingual Education and Policy","field":"Social Sciences","cited_by":3,"is_retracted":false,"has_abstract":false,"ca_institutions":"University of British Columbia; University of Calgary","funders":"","keywords":"Equity (law); English language; Coronavirus disease 2019 (COVID-19); Work (physics); Language proficiency; Test (biology); Standardized test; Political science; Psychology; Engineering ethics; Sociology; Mathematics education; Engineering; Medicine; Law","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":["research_integrity"],"consensus_categories":[],"category_scores_codex":[0.00754045,0.0002985812,0.0004551124,0.0007020483,0.002067324,0.004458474,0.001398069,0.005725258,0.008437685],"category_scores_gemma":[0.03062329,0.0003611649,0.0003588409,0.0006295224,0.007320533,0.006760738,0.002450034,0.00731427,0.003537668],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001676199,"about_ca_system_score_gemma":0.004964087,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.009802816,"about_ca_topic_score_gemma":0.01763633,"domain_scores_codex":[0.9935684,0.00244752,0.0005393184,0.0006533695,0.002419799,0.0003715653],"domain_scores_gemma":[0.9825689,0.01098224,0.0006813409,0.001231326,0.004037241,0.0004989512],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"qualitative","study_design_scores_codex":[0.00002865151,0.00006164375,0.003390281,0.0001506752,0.00001024379,0.0003378735,0.004757232,0.0001741325,0.0004556012,0.7674794,0.1210348,0.1021194],"study_design_scores_gemma":[0.00001236892,0.00006820064,0.007376985,0.0007913568,0.00002757911,0.001748845,0.003570262,0.0004717678,0.002965066,0.343382,0.6395016,0.00008390255],"study_design_candidate":"qualitative","study_design_consensus":null,"genre_codex":"other","genre_gemma":"empirical","genre_scores_codex":[0.01650922,0.007541795,0.01626635,0.1323939,0.003122922,0.00009233339,0.0002907099,0.0001641836,0.8236186],"genre_scores_gemma":[0.313107,0.003229087,0.01332274,0.06699356,0.002286542,0.0002257391,0.0002031316,0.0002192494,0.600413],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.9942747,"threshold_uncertainty_score":0.03987819,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1820764381213399,"score_gpt":0.4948918957819943,"score_spread":0.3128154576606543,"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."}}