{"id":"W2954642664","doi":"10.1177/2515245919858072","title":"Failing Grade: 89% of Introduction-to-Psychology Textbooks That Define or Explain Statistical Significance Do So Incorrectly","year":2019,"lang":"en","type":"article","venue":"Advances in Methods and Practices in Psychological Science","topic":"Statistics Education and Methodologies","field":"Mathematics","cited_by":39,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Guelph","funders":"","keywords":"Fallacy; Psychology; Odds; Significance testing; Experimental psychology; Statistical significance; Epistemology; Logistic regression; Cognition; Statistics; Mathematics","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":true,"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":[0.009850525,0.0009725254,0.001187003,0.002469391,0.0007306725,0.003097877,0.00149059,0.0014905,0.01130624],"category_scores_gemma":[0.1168996,0.0006650682,0.0007009516,0.002027391,0.00135041,0.00179478,0.001929738,0.001906399,0.005904925],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002280735,"about_ca_system_score_gemma":0.003039478,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0032716,"about_ca_topic_score_gemma":0.006780125,"domain_scores_codex":[0.9891071,0.002063355,0.001665717,0.0009001001,0.005721676,0.0005421624],"domain_scores_gemma":[0.8771878,0.06015879,0.02467437,0.004480544,0.03005613,0.003442287],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0009760108,0.00184277,0.4561343,0.004756195,0.0003480319,0.002026651,0.02052757,0.001665531,0.01468739,0.006210695,0.1382188,0.3526061],"study_design_scores_gemma":[0.0002443793,0.001268657,0.6900061,0.004681637,0.0004878047,0.002834678,0.01070939,0.005125596,0.02202052,0.01210032,0.2502786,0.0002423809],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.91496,0.007610311,0.01922658,0.007661636,0.002081968,0.0008395779,0.005548197,0.002482274,0.03958948],"genre_scores_gemma":[0.9394822,0.003506287,0.01760491,0.007173838,0.0002621068,0.0004601293,0.003529612,0.0007378183,0.02724316],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9901495,"threshold_uncertainty_score":0.05209523,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2283246623473483,"score_gpt":0.6279325644064597,"score_spread":0.3996079020591115,"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."}}