{"id":"W2769374615","doi":"10.14507/epaa.25.3299","title":"Policy incentives in Canadian large-scale assessment: How policy levers influence teacher decisions about instructional change","year":2017,"lang":"en","type":"article","venue":"Education Policy Analysis Archives","topic":"Educational Assessment and Improvement","field":"Decision Sciences","cited_by":8,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Incentive; Promotion (chess); Scale (ratio); Curriculum; Work (physics); Test (biology); Psychology; Mathematics education; Pedagogy; Political science; Public relations; Economics","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":"codex-gemma-dda1882f352a","candidate_categories":["metaepi_narrow","sts","scholarly_communication"],"consensus_categories":[],"category_scores_codex":[0.001503319,0.0003548752,0.0005357274,0.009591753,0.001718895,0.001695595,0.002099773,0.0001131419,0.0003629141],"category_scores_gemma":[0.006716684,0.0003176639,0.0004050763,0.003921045,0.0004713113,0.002338945,0.0003548834,0.0003901528,0.00007299498],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0009525673,"about_ca_system_score_gemma":0.01406262,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.4395935,"about_ca_topic_score_gemma":0.4460453,"domain_scores_codex":[0.9954638,0.0004447298,0.0008358089,0.0009523768,0.001322078,0.0009812148],"domain_scores_gemma":[0.9955937,0.0006359593,0.0008704654,0.001681904,0.00034657,0.0008714239],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.000007291603,0.0003819312,0.8091785,0.000002947491,0.0001580205,5.025595e-7,0.005423355,0.0001009915,0.0000861159,0.1180664,0.0008475964,0.06574637],"study_design_scores_gemma":[0.0002768186,0.00001939818,0.8943524,0.00003072615,0.00005592044,0.000001568952,0.004269636,0.001178954,0.00001570873,0.05997305,0.03954251,0.0002833588],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8096865,0.00006031345,0.0004223772,0.1586656,0.0003742216,0.0004328275,0.0002631094,0.00002095382,0.0300741],"genre_scores_gemma":[0.9800922,0.0001871555,0.003275122,0.002541007,0.002034892,0.0002617486,0.0001709482,0.00002098362,0.01141592],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1704057,"threshold_uncertainty_score":0.9999275,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.07839698053366613,"score_gpt":0.4586197873152658,"score_spread":0.3802228067815996,"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."}}