{"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":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.009226554,0.0009591193,0.000808911,0.001608114,0.001560697,0.001218094,0.001274795,0.000798494,0.001127127],"category_scores_gemma":[0.03241068,0.0004558857,0.001499526,0.001205634,0.0009459849,0.0008798053,0.001439031,0.001272332,0.0007012258],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001287685,"about_ca_system_score_gemma":0.001913473,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01720099,"about_ca_topic_score_gemma":0.01660092,"domain_scores_codex":[0.9959336,0.001500147,0.000460659,0.0005692292,0.001070691,0.0004657603],"domain_scores_gemma":[0.9766491,0.00305887,0.001712447,0.00826813,0.009028341,0.001283083],"domain_codex":null,"domain_gemma":"reproducibility","domain_candidate":"reproducibility","domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.002148264,0.004081462,0.8997283,0.0001244472,0.0004095348,0.0007087006,0.02175501,0.0004293546,0.0141186,0.0003564533,0.00124448,0.05489529],"study_design_scores_gemma":[0.000196586,0.005083371,0.9767035,0.00003135064,0.000151408,0.0005656211,0.008160716,0.0009838291,0.005599527,0.000344681,0.002114555,0.0000650209],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9981441,0.00002475589,0.0006066244,0.00003480648,0.00001850829,0.0002654103,0.0001575782,0.00002556549,0.000722535],"genre_scores_gemma":[0.9975048,0.00001929043,0.0009679022,0.00003936189,0.00000794648,0.0003330121,0.0003341186,0.00001523157,0.0007784191],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9907734,"threshold_uncertainty_score":0.04879528,"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."}}