{"id":"W4407931412","doi":"10.1177/02655322251319284","title":"The relationship between English language proficiency test scores and academic achievement: A longitudinal study of two tests","year":2025,"lang":"en","type":"article","venue":"Language Testing","topic":"Higher Education Learning Practices","field":"Social Sciences","cited_by":11,"is_retracted":false,"has_abstract":true,"ca_institutions":"York University","funders":"","keywords":"Psychology; Language proficiency; Test (biology); Language assessment; Achievement test; Academic achievement; Mathematics education; English language; Test of English as a Foreign Language; Standardized test","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":true,"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":[0.003607478,0.0003226394,0.0003787478,0.001176498,0.001590806,0.00120793,0.0006249201,0.0006640399,0.0009654352],"category_scores_gemma":[0.00618503,0.0003496382,0.0006463816,0.0008280507,0.0005909179,0.001064584,0.001323449,0.001910691,0.0004568044],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0009516902,"about_ca_system_score_gemma":0.001494855,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0320573,"about_ca_topic_score_gemma":0.0433431,"domain_scores_codex":[0.9988966,0.0003235363,0.00008000054,0.0001901206,0.0002492935,0.0002603764],"domain_scores_gemma":[0.9942625,0.0007995369,0.00156428,0.0004889593,0.001219451,0.001665159],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00009697174,0.0003605229,0.9965993,0.000002926676,0.00004129541,0.00005618694,0.0008394063,0.00002212848,0.0001843897,0.00002998337,0.00007802354,0.001688904],"study_design_scores_gemma":[0.000005235174,0.0004228163,0.9977024,0.0000072039,0.00002344721,0.00008078521,0.001175869,0.0001274218,0.0001486885,0.00002794593,0.0002691392,0.000008972283],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9995515,0.00005118058,0.0000523905,0.00003801438,0.000004100916,0.000007143208,0.00009026697,0.000001938664,0.0002035743],"genre_scores_gemma":[0.9990771,0.0000403565,0.00008264254,0.000021738,0.000003211619,0.00001725897,0.0002352027,0.000002029534,0.0005205034],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.0320573,"threshold_uncertainty_score":0.06374145,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1073036299005186,"score_gpt":0.4529902981757598,"score_spread":0.3456866682752412,"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."}}