{"id":"W3011227050","doi":"10.1186/s40536-020-00080-3","title":"Are large surveys of adult literacy skills as comparable over time as we think?","year":2020,"lang":"en","type":"article","venue":"Large-scale Assessments in Education","topic":"Education Systems and Policy","field":"Social Sciences","cited_by":4,"is_retracted":false,"has_abstract":true,"ca_institutions":"Institut National de la Recherche Scientifique","funders":"Canadian Institutes of Health Research; Social Sciences and Humanities Research Council of Canada; Fonds de Recherche du Québec-Société et Culture","keywords":"Ceteris paribus; Literacy; Cohort; Adult literacy; Cohort effect; Survey data collection; Psychology; Demographic economics; Demography; Medicine; Pedagogy; Sociology; Economics; Statistics","routes":{"ca_aff":true,"ca_fund":true,"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":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.001309349,0.0001651225,0.000339218,0.0001323952,0.0002458285,0.0001379319,0.0003594802,0.0001555897,0.002385242],"category_scores_gemma":[0.0004287607,0.0001748054,0.00008273478,0.0008497009,0.00005409803,0.0006948002,0.00005194241,0.0002122532,0.0006744289],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002096038,"about_ca_system_score_gemma":0.001365283,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00464516,"about_ca_topic_score_gemma":0.0008774623,"domain_scores_codex":[0.997077,0.001050437,0.0005379693,0.0003542646,0.0005463241,0.0004339334],"domain_scores_gemma":[0.9982982,0.0002094209,0.000556651,0.0002987313,0.0003843767,0.0002526045],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00001119141,0.002644965,0.747181,0.0001189508,0.00003434353,9.334114e-7,0.1806115,0.000008426026,0.0001383273,0.02239483,0.0448529,0.002002661],"study_design_scores_gemma":[0.0009574575,0.00007850242,0.5208185,0.0004150042,0.00002615075,9.764053e-7,0.08744726,0.0002669477,0.0002607111,0.003954035,0.3853517,0.0004227372],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9500549,0.0001894498,0.0001344871,0.006366874,0.001249998,0.0005611796,0.00008240377,0.00005418037,0.04130647],"genre_scores_gemma":[0.9828734,0.0001558496,0.0006428299,0.003175296,0.0007096202,0.00005894277,0.0001125606,0.00002232453,0.01224919],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.3404988,"threshold_uncertainty_score":0.9985267,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02048207184716733,"score_gpt":0.4094220301978794,"score_spread":0.388939958350712,"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."}}