{"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":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.003793014,0.00006048264,0.0001489368,0.0003395367,0.0003706666,0.00005582604,0.0002058274,0.00006251009,0.0002627821],"category_scores_gemma":[0.0008652661,0.00005624609,0.00005290844,0.0005775901,0.0001191186,0.0002324103,0.000003736913,0.0001255597,0.000004472576],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0004073486,"about_ca_system_score_gemma":0.003671048,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.03582073,"about_ca_topic_score_gemma":0.2876629,"domain_scores_codex":[0.9983095,0.0004436616,0.000433758,0.00007534263,0.0004913763,0.0002463683],"domain_scores_gemma":[0.9983751,0.000105875,0.0003170375,0.00008210356,0.0008648365,0.000255032],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.000001070262,0.00006130531,0.7745739,0.000005937395,0.000008926842,0.000001872273,0.1892006,0.0000887719,0.00003724549,0.0002911048,0.000261564,0.03546782],"study_design_scores_gemma":[0.0003775456,0.00009623917,0.8560091,0.0001468472,0.00001950133,0.00001256398,0.1362626,0.0001358089,0.0000104285,0.0004361274,0.006411742,0.00008160767],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9968402,0.0002169679,0.00003470197,0.0005735154,0.00034438,0.0004561156,0.000006451059,0.000005692739,0.001521913],"genre_scores_gemma":[0.9980586,0.00001615923,0.00136924,0.00003953774,0.0004051264,0.00002609526,0.000003623934,0.000006360498,0.00007530721],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2518421,"threshold_uncertainty_score":0.9705998,"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."}}