{"id":"W2789355765","doi":"10.1080/09500782.2018.1430825","title":"Initial assessment for K-12 English language support in six countries: revisiting the validity–reliability paradox","year":2018,"lang":"en","type":"article","venue":"Language and Education","topic":"Multilingual Education and Policy","field":"Social Sciences","cited_by":7,"is_retracted":false,"has_abstract":true,"ca_institutions":"Institute for Christian Studies; University of Toronto","funders":"","keywords":"Construct validity; Construct (python library); Reliability (semiconductor); Psychology; Writing assessment; Validity; Language assessment; Argument (complex analysis); Variance (accounting); Equity (law); Predictive validity; Computer science; Mathematics education; Psychometrics; Political science; Developmental psychology; Economics","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":[],"consensus_categories":[],"category_scores_codex":[0.001562155,0.0000754894,0.00009861156,0.00004853036,0.0003776762,0.0001291014,0.0001227782,0.00006561731,0.0005342978],"category_scores_gemma":[0.00143602,0.00006181691,0.00002983716,0.0001295784,0.0001728224,0.0001610809,0.00001590297,0.00009371294,0.00001114919],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000101567,"about_ca_system_score_gemma":0.0008988917,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.005913733,"about_ca_topic_score_gemma":0.002716662,"domain_scores_codex":[0.9990661,0.0001963982,0.0001932477,0.0001747874,0.0001464124,0.0002230873],"domain_scores_gemma":[0.9990991,0.0004100563,0.000084238,0.0001724,0.0001585175,0.00007574208],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"qualitative","study_design_scores_codex":[0.00001336166,0.00007414096,0.01630482,0.00005377238,0.000003500822,3.238097e-7,0.8794711,2.884158e-7,0.00002553133,0.006640363,0.002962709,0.09445009],"study_design_scores_gemma":[0.0003030791,0.00005610177,0.0232629,0.00004458024,0.00002114845,8.493087e-7,0.5162464,0.00005439486,0.0002823407,0.0005811165,0.4589698,0.0001772435],"study_design_candidate":"qualitative","study_design_consensus":"qualitative","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9549515,0.00009797986,0.00004473055,0.003493765,0.001179838,0.0004880157,0.00001615452,0.00003521209,0.03969285],"genre_scores_gemma":[0.9890103,0.00004010516,0.00041967,0.00156445,0.00678514,0.00008996322,0.00005266284,0.000007116496,0.002030583],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.4560071,"threshold_uncertainty_score":0.8939834,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.04921096980836722,"score_gpt":0.4818473775180079,"score_spread":0.4326364077096407,"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."}}