{"id":"W4392846070","doi":"10.1086/730455","title":"Do Preregistration and Preanalysis Plans Reduce <i>p</i>-Hacking and Publication Bias? Evidence from 15,992 Test Statistics and Suggestions for Improvement","year":2024,"lang":"en","type":"article","venue":"Journal of Political Economy Microeconomics","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":28,"is_retracted":false,"has_abstract":true,"ca_institutions":"Wilfrid Laurier University; University of Ottawa","funders":"Leibniz-Gemeinschaft","keywords":"Statistics; Test (biology); Hacker; Publication bias; Psychology; Computer science; Data science; Computer security; Mathematics; Biology","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":["metaresearch"],"category_scores_codex":[0.7120254,0.001958279,0.004507374,0.01029321,0.002349242,0.009491947,0.007450616,0.007482585,0.005735291],"category_scores_gemma":[0.9036897,0.003096031,0.009682631,0.01637452,0.008420786,0.01199059,0.004439618,0.007827282,0.001479077],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.003303056,"about_ca_system_score_gemma":0.01503279,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001551054,"about_ca_topic_score_gemma":0.001926727,"domain_scores_codex":[0.2780248,0.5811096,0.07703213,0.02111247,0.03918279,0.00353824],"domain_scores_gemma":[0.01873268,0.8265535,0.07920977,0.06400357,0.009999532,0.001500984],"domain_codex":"methods","domain_gemma":"methods","domain_candidate":"methods","domain_consensus":"methods","study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.01703815,0.0009931072,0.2010919,0.02861311,0.02688656,0.0007080854,0.007037697,0.007179063,0.002770332,0.02240671,0.03002997,0.6552453],"study_design_scores_gemma":[0.01325434,0.01984278,0.5854515,0.02207304,0.0468825,0.002355514,0.004181599,0.03481067,0.03128932,0.1404276,0.09815943,0.001271658],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.2482569,0.08534741,0.5187605,0.09024213,0.008177889,0.01570139,0.007192724,0.00756611,0.01875502],"genre_scores_gemma":[0.6502411,0.004148886,0.322873,0.01124092,0.001737307,0.006832256,0.001190817,0.0007594038,0.0009762712],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.2879746,"threshold_uncertainty_score":0.355124,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1212162986094238,"score_gpt":0.3940866936875985,"score_spread":0.2728703950781747,"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."}}