{"id":"W3043201942","doi":"10.5539/ijel.v10n6p1","title":"Investigating the Validity of a University-Level ESL Speaking Placement Test via Mixed Methods Research","year":2020,"lang":"en","type":"article","venue":"International Journal of English Linguistics","topic":"EFL/ESL Teaching and Learning","field":"Arts and Humanities","cited_by":2,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Rubric; CLARITY; Psychology; Test (biology); Mathematics education; Construct validity; Test validity; Medical education; Clinical psychology; Psychometrics; Medicine","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":true,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.1408551,0.000662579,0.001308039,0.004290302,0.002403234,0.004373763,0.002922273,0.001436405,0.001182341],"category_scores_gemma":[0.2350847,0.0007681363,0.00140278,0.003063535,0.002908047,0.003318079,0.003386326,0.001099388,0.0003496792],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.003341319,"about_ca_system_score_gemma":0.006032436,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0028901,"about_ca_topic_score_gemma":0.005373636,"domain_scores_codex":[0.8587845,0.1050543,0.008808569,0.007161489,0.01834498,0.001846186],"domain_scores_gemma":[0.7115374,0.2062852,0.03271746,0.01550562,0.03170119,0.002253074],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0009558978,0.004696155,0.7153527,0.001987178,0.001091884,0.000439367,0.05850856,0.001015257,0.003479213,0.003903942,0.0004598804,0.2081099],"study_design_scores_gemma":[0.001012793,0.02594662,0.8005118,0.003794457,0.001430643,0.0008183694,0.1091069,0.01606765,0.01773515,0.008067556,0.01511417,0.0003939403],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9793937,0.0005075876,0.01316705,0.0002249122,0.00005822393,0.003812866,0.0001474057,0.000026924,0.002661313],"genre_scores_gemma":[0.9506183,0.0003152427,0.03961993,0.0003677534,0.00006028366,0.008283587,0.0002299758,0.00002267778,0.0004821946],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1408551,"threshold_uncertainty_score":0.7449217,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2390646065262667,"score_gpt":0.3937657712575501,"score_spread":0.1547011647312835,"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."}}