{"id":"W2782384206","doi":"10.1177/0265532217750692","title":"Examining sources of variability in repeaters’ L2 writing scores: The case of the PTE Academic writing section","year":2018,"lang":"en","type":"article","venue":"Language Testing","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":11,"is_retracted":false,"has_abstract":true,"ca_institutions":"York University","funders":"","keywords":"Test (biology); Psychology; Context (archaeology); Boston Naming Test; Language assessment; Language proficiency; Mathematics education; Cognition","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":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.04395105,0.000150941,0.0003540737,0.0002827357,0.0003838981,0.00007337797,0.0009004445,0.0001184916,0.00003911367],"category_scores_gemma":[0.4279945,0.00008396868,0.00007881052,0.003934137,0.0004176251,0.000187578,0.00050096,0.0005721146,0.000001565338],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0000366806,"about_ca_system_score_gemma":0.00004893656,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001882361,"about_ca_topic_score_gemma":0.0001243278,"domain_scores_codex":[0.9947868,0.00238961,0.00140078,0.0005058862,0.0005480374,0.0003688909],"domain_scores_gemma":[0.8812267,0.1158117,0.001495027,0.001013234,0.0004048634,0.000048406],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.000006502964,0.000009143778,0.779924,0.00002368205,0.000004543639,0.00003758929,0.008377585,0.0001139314,0.01290258,0.00004890832,0.000008312566,0.1985432],"study_design_scores_gemma":[0.0004777123,0.0001543054,0.8202249,0.0009758305,0.0000330756,0.001429487,0.1217552,0.03262131,0.01640105,0.005641297,0.00001124032,0.0002746601],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9916693,0.0002110218,0.002710139,0.00006900854,0.0003475023,0.0002003829,0.000004981368,0.00005400499,0.004733643],"genre_scores_gemma":[0.984133,0.000001063758,0.01533073,0.00007202217,0.000409187,0.000006122288,2.022241e-7,0.000012225,0.00003543585],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.3840435,"threshold_uncertainty_score":0.9844536,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.3740090240133829,"score_gpt":0.4424069973241175,"score_spread":0.0683979733107346,"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."}}