{"id":"W4417516558","doi":"10.1162/rest.a.1688","title":"The Power of Tests for Detecting <i>p</i> -Hacking","year":2025,"lang":"en","type":"article","venue":"The Review of Economics and Statistics","topic":"Information and Cyber Security","field":"Computer Science","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"Queen's University","funders":"","keywords":"Power (physics); Distribution (mathematics); Monotonic function; Flourishing; Statistical power; Empirical distribution function","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":[0.0006862635,0.0000384575,0.0001183595,0.00001132205,0.0001322829,0.00003742118,0.0002522878,0.000008908455,7.211259e-7],"category_scores_gemma":[0.0001730882,0.00002286194,0.00002644362,0.00005401568,0.00004010351,0.0000488954,0.00007759131,0.00003010345,3.00094e-7],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000005523559,"about_ca_system_score_gemma":0.00004245687,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000003267826,"about_ca_topic_score_gemma":0.00001122588,"domain_scores_codex":[0.999567,0.00001663617,0.000291224,0.0000461876,0.00001691017,0.00006198562],"domain_scores_gemma":[0.9989274,0.0005787811,0.0001964328,0.0002118015,0.00007672313,0.000008901076],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"not_applicable","study_design_scores_codex":[7.543226e-7,0.000001777309,0.000004313088,0.0004563928,0.000007469935,6.912841e-9,0.0001137888,0.00000466274,4.666508e-7,0.8729908,0.0007645228,0.125655],"study_design_scores_gemma":[0.0004651556,0.0001116446,0.0005663438,0.002258642,0.00006603572,0.000005724445,0.0001909765,0.1652179,0.0006542958,0.3484744,0.4817806,0.0002083595],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"review","genre_scores_codex":[0.001745012,0.03840226,0.9478405,0.002254008,0.000411349,0.0005932625,0.00009140166,0.00000916774,0.008653051],"genre_scores_gemma":[0.3291979,0.4743351,0.1884632,0.007503736,0.00003694621,0.00005580127,0.00000802799,0.00001238873,0.0003868245],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.7593772,"threshold_uncertainty_score":0.1017426,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.01644721000412936,"score_gpt":0.2744561221497097,"score_spread":0.2580089121455804,"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."}}