{"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":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.1345872,0.0005874601,0.001641483,0.003480117,0.001354173,0.006354971,0.003059705,0.002354123,0.003414818],"category_scores_gemma":[0.4719467,0.0006588154,0.001382306,0.007134188,0.005217103,0.006132192,0.002972542,0.002347586,0.0007944495],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.003754258,"about_ca_system_score_gemma":0.005754689,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.1227949,"about_ca_topic_score_gemma":0.07014492,"domain_scores_codex":[0.8754196,0.08143558,0.007281539,0.01130082,0.02204553,0.002516797],"domain_scores_gemma":[0.5529637,0.2062105,0.08692878,0.08253399,0.06647388,0.004889194],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0009030809,0.0001548521,0.7359579,0.002340063,0.004821843,0.0002363295,0.01061797,0.002851372,0.0008175873,0.04327318,0.0520443,0.1459816],"study_design_scores_gemma":[0.0002182718,0.0004109711,0.8781281,0.00485424,0.001337891,0.0001729398,0.007866797,0.00334662,0.001889626,0.02841744,0.07315791,0.0001992804],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.569065,0.05034678,0.0921889,0.1972982,0.01120663,0.001192943,0.02177083,0.0008175336,0.05611314],"genre_scores_gemma":[0.9551619,0.003763508,0.01619509,0.01724384,0.001600623,0.0005897747,0.003329048,0.000144868,0.001971293],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.8654128,"threshold_uncertainty_score":0.7117738,"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."}}