{"id":"W1831219925","doi":"10.1037/spq0000126","title":"Generalizability theory reliability of written expression curriculum-based measurement in universal screening.","year":2015,"lang":"en","type":"article","venue":"School Psychology Quarterly","topic":"Reading and Literacy Development","field":"Psychology","cited_by":19,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of British Columbia","funders":"University of Houston","keywords":"Generalizability theory; PsycINFO; Psychology; Curriculum; Reliability (semiconductor); Curriculum-based measurement; Context (archaeology); Mathematics education; Sample (material); Sample size determination; Clinical psychology; Developmental psychology; Statistics; Pedagogy; Curriculum development; MEDLINE; Mathematics; Chemistry; Curriculum mapping","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.004949654,0.0003044016,0.0004892347,0.0003492707,0.00005048163,0.00001781969,0.00050423,0.0003238781,0.000721625],"category_scores_gemma":[0.0002614169,0.0002708124,0.0001412222,0.00043671,0.0002523655,0.0001703039,0.00002360094,0.0004882683,0.0001939932],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002352135,"about_ca_system_score_gemma":0.0001828712,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0003580231,"about_ca_topic_score_gemma":0.00003430058,"domain_scores_codex":[0.9950181,0.002111098,0.0008997025,0.0008881529,0.000536827,0.0005461566],"domain_scores_gemma":[0.9975936,0.0001105529,0.0002803326,0.001244738,0.0004195404,0.0003512478],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.002149135,0.002323611,0.9245481,0.00004235054,0.00006903905,0.00006246538,0.007908235,0.0001019922,0.00190731,0.001072129,0.03770564,0.02210999],"study_design_scores_gemma":[0.009443151,0.00193792,0.9600521,0.0002010539,0.00003993421,0.00001483323,0.00605659,0.0001331942,0.0003279655,0.01119823,0.009941981,0.0006530855],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9727283,0.0003347551,0.01575179,0.0009805646,0.001382056,0.0005128823,0.00001322388,0.0001108082,0.008185631],"genre_scores_gemma":[0.992682,0.000001023584,0.006322692,0.0005574198,0.00008797799,0.0000886221,0.00002209391,0.00002725514,0.000210918],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.03550396,"threshold_uncertainty_score":0.9999744,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.04176173805231315,"score_gpt":0.3287780740287186,"score_spread":0.2870163359764054,"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."}}