{"id":"W2125507422","doi":"10.5054/tq.2010.222215","title":"Do Language Proficiency Test Scores Differ by Gender?","year":2010,"lang":"en","type":"article","venue":"TESOL Quarterly","topic":"EFL/ESL Teaching and Learning","field":"Arts and Humanities","cited_by":7,"is_retracted":false,"has_abstract":true,"ca_institutions":"Thompson Rivers University","funders":"","keywords":"Psychology; Test (biology); Language assessment; Language proficiency; Linguistics; Mathematics education","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":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0002210642,0.0001698265,0.0001480414,0.00006655847,0.0004333947,0.0003711321,0.0002314625,0.00005602109,0.003358142],"category_scores_gemma":[0.0000613554,0.0001333626,0.00006597691,0.00002492835,0.0001628135,0.0001630028,0.00001145707,0.0005606842,0.0006018548],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000008514879,"about_ca_system_score_gemma":0.00001637293,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0002888286,"about_ca_topic_score_gemma":0.0002473077,"domain_scores_codex":[0.99899,0.00004245461,0.0001831186,0.0002586039,0.0001961807,0.0003296561],"domain_scores_gemma":[0.999364,0.0001431181,0.00007122157,0.0003033348,0.00003627077,0.00008201114],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"not_applicable","study_design_scores_codex":[0.00001103575,0.0005638483,0.01921259,0.00008042104,0.0000323401,0.00002879017,0.7142143,2.921057e-7,0.07888282,0.06494991,0.05780117,0.06422246],"study_design_scores_gemma":[0.002940155,0.004612419,0.03086089,0.0001917869,0.0001794565,0.00007194792,0.1790646,0.001373776,0.002825079,0.007287085,0.7674415,0.003151303],"study_design_candidate":"qualitative","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9306219,0.0002259622,0.00003531479,0.0001915037,0.0004531216,0.0001248187,0.00003730125,0.0003365678,0.06797355],"genre_scores_gemma":[0.9831495,5.676178e-7,0.00007082741,0.0002115711,0.0007721644,0.0000183989,0.0000258917,0.0000316846,0.01571942],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.7096403,"threshold_uncertainty_score":0.9975529,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.01560257194511426,"score_gpt":0.2436205360457356,"score_spread":0.2280179641006214,"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."}}