{"id":"W4405148704","doi":"10.1080/15434303.2024.2438142","title":"The Academic Achievement of Undergraduate Students with Different TOEFL iBT Score Profiles: A Replication Study","year":2024,"lang":"en","type":"article","venue":"Language Assessment Quarterly","topic":"Online Learning and Analytics","field":"Computer Science","cited_by":2,"is_retracted":false,"has_abstract":true,"ca_institutions":"York University","funders":"York University","keywords":"Test of English as a Foreign Language; Replication (statistics); Psychology; Mathematics education; Academic achievement; Language assessment; Statistics; Mathematics","routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":true,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0006087375,0.0001795869,0.000184372,0.0000829928,0.0001384244,0.0003819167,0.001024307,0.00003848883,0.000001599297],"category_scores_gemma":[0.000008750815,0.0001005638,0.00005236595,0.0004311984,0.0000371502,0.0002062565,0.0001001279,0.0004261221,0.000007986389],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00007472916,"about_ca_system_score_gemma":0.00009430547,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00005202865,"about_ca_topic_score_gemma":0.00005835441,"domain_scores_codex":[0.9978987,0.0001915785,0.0003501787,0.0005156952,0.0007971546,0.0002467186],"domain_scores_gemma":[0.9985532,0.0001471957,0.0001489745,0.001026401,0.0000648014,0.00005941532],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.0000558437,0.003123515,0.351705,0.0003883801,0.001308943,0.0002647144,0.06044709,0.0001842119,0.005221935,0.08809556,0.000703849,0.488501],"study_design_scores_gemma":[0.003173164,0.01934534,0.7479639,0.001436942,0.0005008889,0.00003749141,0.0498673,0.1651597,0.001135501,0.00935493,0.0007807754,0.001244067],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9491369,0.0002421828,0.0436085,0.006020373,0.0001382099,0.0005741208,0.000003515794,0.000195934,0.00008030186],"genre_scores_gemma":[0.9977668,0.00001660896,0.0009135941,0.00004804061,0.00007220569,0.00008784634,0.00001154122,0.00001580267,0.001067565],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.4872569,"threshold_uncertainty_score":0.4100872,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.01246481434015625,"score_gpt":0.349279442253663,"score_spread":0.3368146279135067,"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."}}