{"id":"W4392819524","doi":"","title":"Performance-based accountability systems: types, instrumental logics and effects on effectiveness and equity in educational systems in Europe and Canada. A comparative study using PISA 2012.","year":2017,"lang":"fr","type":"preprint","venue":"HAL (Le Centre pour la Communication Scientifique Directe)","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":false,"ca_institutions":"","funders":"","keywords":"Accountability; Equity (law); Political science; Mathematics education; Accounting; Economics; Psychology; Law","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":false,"about_ca":true,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01208399,0.0003310466,0.0006256918,0.004126815,0.004590723,0.006130885,0.001720428,0.0007893589,0.003441237],"category_scores_gemma":[0.05639199,0.0002522382,0.000452514,0.008757818,0.005611822,0.002729989,0.005016609,0.001407737,0.0001876201],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.08474918,"about_ca_system_score_gemma":0.1006043,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.9728737,"about_ca_topic_score_gemma":0.9763174,"domain_scores_codex":[0.9863312,0.003073913,0.0005790475,0.0009009414,0.005753487,0.003361288],"domain_scores_gemma":[0.9424756,0.01766666,0.005795657,0.001889881,0.02506899,0.007103243],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0008176923,0.0004121907,0.829009,0.0002137678,0.0002496212,0.0001279953,0.01824388,0.004432867,0.0004679313,0.05258015,0.005687749,0.08775716],"study_design_scores_gemma":[0.00003683202,0.00006700906,0.9785542,0.0001073865,0.0001138867,0.00001981984,0.01022374,0.00190246,0.0006295302,0.002560613,0.005742649,0.00004174367],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9749268,0.001182343,0.0008242721,0.001890185,0.0000274629,0.00007009094,0.001060091,0.00003086265,0.0199878],"genre_scores_gemma":[0.998555,0.0001100201,0.0001444018,0.00002542077,0.000002963712,0.000008712303,0.0001432495,0.000005652587,0.001004681],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9152508,"threshold_uncertainty_score":0.6149012,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1170480752426027,"score_gpt":0.4023130205782668,"score_spread":0.2852649453356642,"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."}}