{"id":"W1984249886","doi":"10.1016/j.asw.2008.10.002","title":"How accurate are ESL students’ holistic writing scores on large-scale assessments?—A generalizability theory approach","year":2008,"lang":"en","type":"article","venue":"Assessing Writing","topic":"EFL/ESL Teaching and Learning","field":"Arts and Humanities","cited_by":66,"is_retracted":false,"has_abstract":false,"ca_institutions":"","funders":"University of Alberta; Francis Crick Institute","keywords":"Generalizability theory; Reliability (semiconductor); Variance components; Psychology; Variance (accounting); Rating scale; Scale (ratio); Mathematics education; Variation (astronomy); Statistics; Mathematics; Developmental psychology; Accounting; Geography","routes":{"ca_aff":false,"ca_fund":true,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaepi_narrow","sts","scholarly_communication"],"consensus_categories":[],"category_scores_codex":[0.002858428,0.0004504026,0.000536965,0.0001892047,0.004088159,0.003202459,0.0005260287,0.0001214696,0.0001392467],"category_scores_gemma":[0.0005540972,0.0004173171,0.0002061956,0.0001121434,0.000352728,0.001116696,0.0002149347,0.001126587,0.00002342085],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001515316,"about_ca_system_score_gemma":0.00007573103,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00007664643,"about_ca_topic_score_gemma":0.00002852087,"domain_scores_codex":[0.9957989,0.00109023,0.0005706124,0.0008054897,0.0008523421,0.0008823837],"domain_scores_gemma":[0.9978193,0.0007585853,0.0005263423,0.0005282813,0.0002142943,0.0001531985],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"qualitative","study_design_scores_codex":[0.00009026797,0.002774438,0.7152587,0.00101,0.0003885967,0.0003358118,0.07983955,0.001801999,0.0006580344,0.1775016,0.002639662,0.01770136],"study_design_scores_gemma":[0.005564654,0.0005505587,0.3071328,0.004143525,0.0004749919,0.0002382511,0.5994053,0.03637354,0.0003661023,0.005635583,0.03546541,0.004649227],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9447283,0.0002840909,0.00989984,0.0002113836,0.0004276708,0.0002288632,0.0000267156,0.0005181329,0.04367502],"genre_scores_gemma":[0.9898009,0.00001314849,0.003061542,0.0006727139,0.001895853,0.00004424292,0.00008053925,0.0001028448,0.004328187],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.5195658,"threshold_uncertainty_score":0.9998279,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1330202938737769,"score_gpt":0.3594221099872191,"score_spread":0.2264018161134423,"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."}}