{"id":"W2038816208","doi":"10.5539/hes.v1n2p107","title":"Fairness of IELTS Test Scores in University Admission","year":2011,"lang":"en","type":"article","venue":"Higher Education Studies","topic":"Student Assessment and Feedback","field":"Social Sciences","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Test (biology); Context (archaeology); Language proficiency; Psychology; Language assessment; Interpretation (philosophy); Empirical research; Medical education; Mathematics education; Applied psychology; Medicine; Computer science; Statistics","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":true,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.1347471,0.000261038,0.0007328728,0.003503164,0.002186707,0.004319474,0.001477027,0.0009704322,0.002934508],"category_scores_gemma":[0.4027041,0.0002234406,0.0005175149,0.002790307,0.003518268,0.002703679,0.004517158,0.001869722,0.0005302361],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.004917426,"about_ca_system_score_gemma":0.00440748,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.005653388,"about_ca_topic_score_gemma":0.004867621,"domain_scores_codex":[0.8803708,0.07100375,0.009837171,0.004667598,0.03039426,0.003726363],"domain_scores_gemma":[0.6546758,0.1784073,0.06327245,0.02423263,0.06828411,0.01112769],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.001253176,0.0004247863,0.8370126,0.00009655071,0.00008974319,0.0001715821,0.01643189,0.0007005756,0.000587415,0.004516552,0.002636499,0.1360786],"study_design_scores_gemma":[0.0000387816,0.0007721302,0.9665496,0.0002270994,0.00002759783,0.0001772416,0.01403624,0.003399379,0.002452657,0.007759092,0.004449198,0.0001109449],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9654029,0.000666131,0.006040947,0.004464765,0.0003661232,0.0002128681,0.0001689062,0.0001034113,0.02257397],"genre_scores_gemma":[0.9982666,0.00004221197,0.0006563761,0.0001219482,0.00004784206,0.0000337568,0.0000304698,0.000009275058,0.000791441],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1347471,"threshold_uncertainty_score":0.7126195,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1157244094817193,"score_gpt":0.3958591156896718,"score_spread":0.2801347062079525,"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."}}