{"id":"W1969608664","doi":"10.1207/s15324818ame1703_4","title":"Comparability of Bilingual Versions of Assessments: Sources of Incomparability of English and French Versions of Canada's National Achievement Tests","year":2004,"lang":"en","type":"article","venue":"Applied Measurement in Education","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":62,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of British Columbia","funders":"","keywords":"Comparability; Differential item functioning; Psychology; Reading (process); Item response theory; Psychometrics; Developmental psychology; Linguistics; Mathematics","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":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.008387373,0.0001096597,0.00049297,0.0004315183,0.00004459875,0.000004626017,0.0003663378,0.00005234744,0.00002657251],"category_scores_gemma":[0.02685922,0.00009530023,0.00005264134,0.001388067,0.000232924,0.00006646567,0.0000929767,0.0001000504,6.989185e-8],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000342068,"about_ca_system_score_gemma":0.003462436,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.03255771,"about_ca_topic_score_gemma":0.01511523,"domain_scores_codex":[0.9956943,0.0002184989,0.001385108,0.0003144009,0.002249623,0.0001380937],"domain_scores_gemma":[0.9913903,0.003827856,0.001246788,0.0004199934,0.003048027,0.00006706948],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00006717769,0.002282373,0.9710136,0.0001585683,0.00003808071,1.666875e-8,0.002194308,0.004583712,0.01188098,0.002343472,0.0001626945,0.005274998],"study_design_scores_gemma":[0.0007111569,0.0001160835,0.9460227,0.0001031654,0.00001943878,1.014703e-7,0.009303373,0.00003704233,0.02926517,0.01426952,0.00007830784,0.00007388488],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9961377,0.0001505411,0.0004005036,0.000064529,0.0003507019,0.0003676402,0.00003405909,0.000003186382,0.002491179],"genre_scores_gemma":[0.9884601,0.000006330346,0.01148844,0.000008167554,0.00001623357,0.00001060651,0.000005461011,0.000002825405,0.000001867343],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.02499086,"threshold_uncertainty_score":0.981338,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2715701847682983,"score_gpt":0.4269062826454618,"score_spread":0.1553360978771635,"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."}}