{"id":"W4399879966","doi":"10.55016/ojs/ajer.v49i3.54986","title":"Differential Validity and Utility of Successive and Simultaneous Approaches to the Development of Equivalent Achievement Tests in French and English","year":2003,"lang":"en","type":"article","venue":"Alberta Journal of Educational Research","topic":"Student Assessment and Feedback","field":"Social Sciences","cited_by":8,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Alberta","funders":"","keywords":"Differential (mechanical device); Psychology; Multiculturalism; Socioeconomic status; Mathematics education; Multilingualism; Pedagogy; Sociology","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":true,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.002169721,0.00006593714,0.0001586609,0.0001391126,0.0002229942,0.00004878868,0.000196351,0.00003707284,0.0003922964],"category_scores_gemma":[0.008130779,0.00004708696,0.00001939811,0.0002403429,0.0003137006,0.0000970589,0.0001006865,0.0001757441,4.679562e-7],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00005898942,"about_ca_system_score_gemma":0.0007510767,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0007835382,"about_ca_topic_score_gemma":0.009446729,"domain_scores_codex":[0.9981923,0.0003739987,0.000363692,0.0001294666,0.0007411181,0.0001994912],"domain_scores_gemma":[0.989754,0.009519937,0.0001405005,0.00006988693,0.0003832939,0.0001324118],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0000379818,0.0004188618,0.9364279,0.00003342224,0.00003166797,1.376707e-7,0.05629671,0.000002698599,0.00006552353,0.005248436,0.0001419103,0.00129476],"study_design_scores_gemma":[0.0003580344,0.0001247723,0.9811452,0.00006667062,0.00001295172,9.246214e-7,0.01536426,0.00002375007,0.0002844718,0.0007737788,0.001778389,0.0000668073],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9915522,0.0002375015,0.000006903287,0.002202498,0.00009600021,0.0002388091,0.000001330485,1.32083e-7,0.005664585],"genre_scores_gemma":[0.9987558,0.0001006139,0.0002935285,0.000005122244,0.0000760644,0.000007884905,8.787443e-7,0.000002752987,0.0007573719],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.04471729,"threshold_uncertainty_score":0.973389,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2521266780511994,"score_gpt":0.430524562775349,"score_spread":0.1783978847241496,"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."}}