{"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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.1233407,0.0009755833,0.001432437,0.009259951,0.004450987,0.009475849,0.004812463,0.003978774,0.004021112],"category_scores_gemma":[0.403795,0.001018728,0.001782013,0.01720419,0.00594128,0.01084112,0.008374658,0.01105944,0.001356546],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.004731757,"about_ca_system_score_gemma":0.004752294,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.06360277,"about_ca_topic_score_gemma":0.0570082,"domain_scores_codex":[0.8421896,0.1065123,0.01616179,0.009717532,0.02200963,0.003409152],"domain_scores_gemma":[0.6802248,0.1921244,0.0378195,0.04654929,0.04093936,0.002342706],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"observational","study_design_scores_codex":[0.0005133136,0.0001490351,0.1657452,0.0026112,0.001002489,0.003325951,0.06578771,0.002771713,0.0004106335,0.3749123,0.2557524,0.127018],"study_design_scores_gemma":[0.0001092909,0.0001776639,0.2120106,0.009412308,0.0005717024,0.001628946,0.05718089,0.008953509,0.002879542,0.1964421,0.5102206,0.0004128695],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"other","genre_gemma":"empirical","genre_scores_codex":[0.2559168,0.009529003,0.1204283,0.2477533,0.008768906,0.002467717,0.02657746,0.001060184,0.3274983],"genre_scores_gemma":[0.799589,0.003025886,0.1118933,0.05857868,0.002267066,0.005941547,0.006654029,0.0008091734,0.01124132],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.1233407,"threshold_uncertainty_score":0.6522956,"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."}}