{"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":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","sts","research_integrity","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0207235,0.0001322052,0.0002167939,0.0003759865,0.001416266,0.0003632904,0.0009097832,0.0001500313,0.001235217],"category_scores_gemma":[0.01363992,0.0001304921,0.00004006201,0.0007622283,0.0005219213,0.0002841132,0.0002509854,0.004635355,0.000432042],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0004296503,"about_ca_system_score_gemma":0.001043176,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.05670701,"about_ca_topic_score_gemma":0.01349845,"domain_scores_codex":[0.9877522,0.007822153,0.0002961158,0.0005756666,0.002315215,0.001238652],"domain_scores_gemma":[0.9921761,0.006921076,0.00003756486,0.0002969426,0.0001294092,0.000438908],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"qualitative","study_design_scores_codex":[0.00005392035,0.0002889497,0.04742068,0.00007186762,0.00001160996,0.0003139138,0.7087585,0.00000727141,0.007956663,0.1743955,0.01252313,0.04819801],"study_design_scores_gemma":[0.0006203044,0.0002856849,0.03709951,0.000146325,0.0000052348,0.000002322633,0.874051,0.0004943857,0.000163126,0.004251343,0.08242878,0.0004520336],"study_design_candidate":"qualitative","study_design_consensus":"qualitative","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8301547,0.0004333204,0.0001813761,0.01788124,0.00009473193,0.0005248244,0.00002450673,0.0001147713,0.1505906],"genre_scores_gemma":[0.9897756,0.00001843935,0.002617927,0.0003374873,0.001507509,0.00007439974,0.00004604546,0.0000306408,0.005591969],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1701442,"threshold_uncertainty_score":0.9998838,"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."}}