{"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":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0009098735,0.00004729127,0.00013837,0.0000279274,0.0006784594,0.0003682187,0.0001961283,0.00003341503,0.0006730481],"category_scores_gemma":[0.00009847481,0.00008439695,0.00001476641,0.0001598887,0.00005804435,0.0003036445,0.0000439081,0.0001385054,0.000002349005],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001940159,"about_ca_system_score_gemma":0.01093635,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.9750393,"about_ca_topic_score_gemma":0.9996,"domain_scores_codex":[0.9991189,0.0001127872,0.0002340615,0.000204196,0.0001329613,0.0001970906],"domain_scores_gemma":[0.9995141,0.0001216136,0.0001037379,0.0001214104,0.00004743452,0.00009167121],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0000118222,0.00006663926,0.9571826,0.00003107499,0.000006821698,5.841015e-7,0.005276078,0.00004959697,2.982929e-7,0.00006540376,0.01989675,0.01741237],"study_design_scores_gemma":[0.000271451,0.00001400244,0.7381039,0.00001622121,0.000004881763,0.0000051446,0.112819,0.00001846496,0.000001189045,0.00005265718,0.1485644,0.0001286561],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9952403,0.0001289313,2.31775e-7,0.0005546645,0.001355369,0.0002602183,0.0002455589,0.00001958044,0.002195124],"genre_scores_gemma":[0.997618,0.00007203116,0.00004156374,0.0004258509,0.0003076985,0.000122669,0.00003397128,0.000007064702,0.001371112],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2190787,"threshold_uncertainty_score":0.9946707,"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."}}