{"id":"W4366452411","doi":"10.3138/cjpe.0022.004","title":"Walking in Their Shoes: Students’ Perceptions of Large-Scale High-Stakes Testing","year":2008,"lang":"en","type":"article","venue":"Canadian Journal of Program Evaluation","topic":"Education Systems and Policy","field":"Social Sciences","cited_by":11,"is_retracted":false,"has_abstract":true,"ca_institutions":"Queen's University","funders":"","keywords":"Graduation (instrument); Test (biology); Literacy; Psychology; Perception; Scale (ratio); Mathematics education; Focus group; Medical education; Standardized test; Pedagogy; Sociology; Medicine; Engineering","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":true,"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":[0.00457848,0.0002823182,0.0003896746,0.0008716052,0.003045832,0.003455951,0.0009090517,0.001351315,0.004773636],"category_scores_gemma":[0.01782357,0.0003736495,0.000535129,0.0006948652,0.002749352,0.001212873,0.001945687,0.001659536,0.0003925672],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002741978,"about_ca_system_score_gemma":0.004012646,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.06783517,"about_ca_topic_score_gemma":0.09718947,"domain_scores_codex":[0.9946368,0.002120743,0.0003438597,0.0002315548,0.001501958,0.001165035],"domain_scores_gemma":[0.9813018,0.006050784,0.003541311,0.0004443043,0.002515986,0.006145698],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"qualitative","study_design_scores_codex":[0.0003297173,0.002341891,0.6045492,0.0002237279,0.00006478558,0.001620423,0.3537078,0.0002107156,0.002665988,0.0005463967,0.002339346,0.03140005],"study_design_scores_gemma":[0.00002692754,0.0009979063,0.3834781,0.0001278041,0.00003447078,0.0003430725,0.6084767,0.0004822399,0.0006542301,0.0002010792,0.005111802,0.00006569294],"study_design_candidate":"qualitative","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9982733,0.00003625024,0.00005901123,0.0003160305,0.000007019272,0.00001887312,0.00001165183,0.000003629893,0.00127429],"genre_scores_gemma":[0.9994362,0.00004461041,0.000061787,0.00009829836,0.000002881079,0.00001170553,0.00001097311,0.000002314604,0.0003312647],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.06783517,"threshold_uncertainty_score":0.1348807,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1872952066227057,"score_gpt":0.4436968600455949,"score_spread":0.2564016534228892,"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."}}