{"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":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch","research_integrity"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.338635,0.001950502,0.005169361,0.009132072,0.002032887,0.009437256,0.007195896,0.00755438,0.00841357],"category_scores_gemma":[0.7761388,0.001328577,0.00732342,0.009835204,0.02028263,0.01529107,0.006365193,0.008282808,0.001361411],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00224756,"about_ca_system_score_gemma":0.002290662,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0009856833,"about_ca_topic_score_gemma":0.0004523107,"domain_scores_codex":[0.5633448,0.3134839,0.02643677,0.04531513,0.04834903,0.003070412],"domain_scores_gemma":[0.04271248,0.9022298,0.02495198,0.02243602,0.006645303,0.00102448],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.007580128,0.000649517,0.4621421,0.01479922,0.04741804,0.001983978,0.003975195,0.0206492,0.002757011,0.1361844,0.01549244,0.2863688],"study_design_scores_gemma":[0.002339367,0.005296096,0.2013003,0.008020394,0.01623178,0.004850641,0.003608386,0.0729649,0.01096807,0.6438463,0.02988619,0.0006876258],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.361317,0.04831207,0.4888328,0.03218252,0.004075743,0.001994136,0.005618663,0.001473217,0.0561939],"genre_scores_gemma":[0.9681336,0.001702083,0.02445599,0.002602477,0.0008434081,0.0008358683,0.0006396293,0.0002316883,0.0005552007],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.9924456,"threshold_uncertainty_score":0.8155808,"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."}}