{"id":"W4246957039","doi":"10.1177/016146811511700107","title":"Cautions about Inferences from International Assessments: The Case of PISA 2009","year":2015,"lang":"en","type":"article","venue":"Teachers College Record The Voice of Scholarship in Education","topic":"Educational Assessment and Improvement","field":"Decision Sciences","cited_by":14,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of British Columbia","funders":"","keywords":"Context (archaeology); Graduation (instrument); Higher education; Ranking (information retrieval); Jurisdiction; International comparisons; Reading (process); Literacy; Scale (ratio); International education; Political science; Economic growth; Public relations; Geography; Economics; 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":[0.006455437,0.0001493364,0.0002259949,0.0003436765,0.0001544746,0.0001761106,0.001336932,0.00008522714,0.0003030695],"category_scores_gemma":[0.004002853,0.00008992755,0.00008994449,0.001163828,0.000208659,0.001006616,0.0001320454,0.0004026578,0.00004498822],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0003282454,"about_ca_system_score_gemma":0.002013483,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.005002232,"about_ca_topic_score_gemma":0.003863167,"domain_scores_codex":[0.9967199,0.0007932642,0.0009708391,0.0003203235,0.001023556,0.00017213],"domain_scores_gemma":[0.9954652,0.0018794,0.0008157225,0.0006584876,0.001084255,0.00009693984],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0001139364,0.002037934,0.764295,0.000007077315,0.0001397982,0.000002801064,0.01418649,0.0001454448,0.0005057736,0.01878934,0.03773594,0.1620405],"study_design_scores_gemma":[0.0007045114,0.0001692179,0.6167843,0.0001317592,0.00006165869,0.00002183668,0.2147194,0.0008498316,0.0003179862,0.1272615,0.03874739,0.0002306956],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9791136,0.0004336849,0.0001238059,0.0108862,0.003059048,0.0004432663,0.0001009968,0.000009090036,0.005830253],"genre_scores_gemma":[0.9941311,0.00003763114,0.002310946,0.0002373263,0.0002346131,0.0001124717,0.00002916433,0.000008566447,0.00289816],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2005329,"threshold_uncertainty_score":0.7561912,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1616855812454026,"score_gpt":0.4616570141959139,"score_spread":0.2999714329505113,"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."}}