{"id":"W3087609374","doi":"10.1177/1362168820958512","title":"Statistical significance tests in language teaching research","year":2020,"lang":"en","type":"article","venue":"Language Teaching Research","topic":"Educational Assessment and Pedagogy","field":"Social Sciences","cited_by":5,"is_retracted":false,"has_abstract":false,"ca_institutions":"University of Victoria","funders":"","keywords":"Psychology; Language assessment; Statistical analysis; Mathematics education; Language proficiency; Linguistics; Statistics","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":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.1887779,0.003122094,0.00722092,0.01276366,0.006610012,0.006882937,0.008391095,0.00754241,0.02317078],"category_scores_gemma":[0.5812181,0.002207698,0.007500511,0.02070184,0.01537622,0.009338144,0.006297301,0.01627709,0.002410377],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.004649365,"about_ca_system_score_gemma":0.01072257,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002975415,"about_ca_topic_score_gemma":0.002825923,"domain_scores_codex":[0.5564673,0.3358694,0.02346221,0.02549613,0.05334164,0.005363321],"domain_scores_gemma":[0.1373479,0.8131065,0.01298021,0.02570805,0.009004058,0.001853262],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.009819437,0.003142245,0.127386,0.0216616,0.0228661,0.004538422,0.01552956,0.01883185,0.006416654,0.3445215,0.05040237,0.3748844],"study_design_scores_gemma":[0.004064936,0.01062135,0.1341929,0.007256345,0.006679426,0.004781263,0.01101058,0.1209387,0.01875191,0.6107767,0.07009518,0.000830729],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.04757646,0.00565226,0.9168936,0.002964829,0.00300058,0.003767762,0.002441021,0.002654922,0.01504855],"genre_scores_gemma":[0.3470129,0.00125309,0.6215969,0.001925666,0.001255712,0.01961825,0.001160185,0.001715756,0.004461632],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.8112221,"threshold_uncertainty_score":0.9983649,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2858507297296443,"score_gpt":0.6059525332191947,"score_spread":0.3201018034895505,"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."}}