{"id":"W4401988558","doi":"10.1371/journal.pone.0307999","title":"Accumulating evidence across studies: Consistent methods protect against false findings produced by p-hacking","year":2024,"lang":"en","type":"article","venue":"PLoS ONE","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":8,"is_retracted":false,"has_abstract":true,"ca_institutions":"Queen's University","funders":"","keywords":"Hacker; Biology; Bioinformatics; Medicine; Computational biology; Computer science; Computer security","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":[0.6935701,0.005015357,0.009900494,0.01359963,0.004922165,0.01373907,0.01237833,0.01578663,0.01061341],"category_scores_gemma":[0.9026824,0.005259373,0.01515528,0.01236346,0.02682792,0.01580699,0.01898719,0.02108195,0.002012837],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.006597443,"about_ca_system_score_gemma":0.0151014,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002888477,"about_ca_topic_score_gemma":0.002246089,"domain_scores_codex":[0.2238702,0.6693136,0.04115872,0.03612816,0.02703786,0.002491367],"domain_scores_gemma":[0.04182518,0.8486556,0.02375614,0.07544372,0.009211067,0.001108145],"domain_codex":"methods","domain_gemma":"methods","domain_candidate":"methods","domain_consensus":"methods","study_design_codex":"theoretical_or_conceptual","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.008162999,0.0007960518,0.04793089,0.01980958,0.0371242,0.004998078,0.02297624,0.06194227,0.003241885,0.393986,0.03095204,0.3680799],"study_design_scores_gemma":[0.003803213,0.001667417,0.004381288,0.006229811,0.004722541,0.001205276,0.001278861,0.0842719,0.005592585,0.8537097,0.03254483,0.0005925908],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.01503996,0.002551924,0.9490984,0.01157208,0.001828406,0.01195815,0.001006526,0.002556771,0.00438782],"genre_scores_gemma":[0.1675755,0.0005238538,0.8008264,0.005960695,0.0004464755,0.02316714,0.0003254662,0.0004741608,0.0007003951],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.3064299,"threshold_uncertainty_score":0.3778826,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.9447732269234966,"score_gpt":0.6460625260506303,"score_spread":0.2987107008728663,"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."}}