{"id":"W4283261567","doi":"10.1080/09645292.2022.2091113","title":"Low-stakes standardized tests in British Columbia, Canada: system accountability and/or individual feedback?","year":2022,"lang":"en","type":"article","venue":"Education Economics","topic":"School Choice and Performance","field":"Social Sciences","cited_by":4,"is_retracted":false,"has_abstract":true,"ca_institutions":"McMaster University","funders":"","keywords":"Accountability; Regression discontinuity design; Scale (ratio); Numeracy; Psychology; Standardized test; Academic achievement; Reading (process); Point system; Literacy; Mathematics education; Statistics; Political science; Geography; Economics; Pedagogy; Mathematics; Law","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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.003935845,0.000568484,0.0008293319,0.001582057,0.004518954,0.002324121,0.00283464,0.0008365015,0.002644981],"category_scores_gemma":[0.01266676,0.0005648072,0.0004236999,0.004809173,0.002413724,0.0008362114,0.001493151,0.002356194,0.0003367558],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.08153081,"about_ca_system_score_gemma":0.1076369,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.999092,"about_ca_topic_score_gemma":0.9996623,"domain_scores_codex":[0.9947028,0.0008529675,0.0002339718,0.0006680255,0.001897893,0.001644464],"domain_scores_gemma":[0.9826517,0.002351216,0.001789232,0.0005979323,0.009056487,0.003553354],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0004587694,0.0004538836,0.9389074,0.000149996,0.0001054038,0.0002450696,0.004183654,0.0007981206,0.0004581066,0.001187866,0.008251217,0.04480055],"study_design_scores_gemma":[0.00002817401,0.00007202351,0.993666,0.00006323827,0.00003696343,0.00002851307,0.003016644,0.0005698364,0.0002844343,0.0001152292,0.002089825,0.00002919069],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9892516,0.0007478425,0.0002801776,0.001146312,0.00002329466,0.00008800229,0.001988267,0.00002571669,0.00644877],"genre_scores_gemma":[0.9939121,0.0002959761,0.0003645192,0.0002067111,0.000004239754,0.0000400878,0.00113117,0.00001294837,0.004032194],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.08153081,"threshold_uncertainty_score":0.5915502,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.01928093764362444,"score_gpt":0.2679370934014547,"score_spread":0.2486561557578302,"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."}}