{"id":"W4391790392","doi":"10.1177/20531680241233439","title":"Promoting Reproducibility and Replicability in Political Science","year":2024,"lang":"en","type":"article","venue":"Research & Politics","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":22,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Ottawa","funders":"Laura and John Arnold Foundation","keywords":"Reproducibility; Politics; Political science; Data science; Psychology; Computer science; Statistics; Mathematics; Law","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":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.1492238,0.00009134643,0.0001799655,0.0008301322,0.0003560678,0.001022927,0.0006336092,0.00005503581,0.0002383856],"category_scores_gemma":[0.08185679,0.00006637618,0.00003737287,0.002471084,0.001997282,0.000658947,0.0006131658,0.000632834,0.0001983457],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0007914108,"about_ca_system_score_gemma":0.003041169,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0004283606,"about_ca_topic_score_gemma":0.00007676868,"domain_scores_codex":[0.9904854,0.0008822721,0.0007500185,0.002075656,0.004382824,0.001423841],"domain_scores_gemma":[0.9915569,0.003998198,0.00002851477,0.002555364,0.001252509,0.0006085059],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.000005638049,0.0001070934,0.1797951,0.00007710597,0.000002057597,0.00001382303,0.002750179,0.000002530529,0.001395801,0.7848968,0.0003171401,0.03063678],"study_design_scores_gemma":[0.00009511466,0.0001097782,0.4067348,0.0000616375,0.000001369988,0.00001688789,0.001381741,0.05699906,0.003690712,0.5231546,0.007654347,0.00009993646],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9572753,0.0001485951,0.00003198716,0.02315113,0.0001351882,0.000362901,0.000007792864,0.00004294491,0.01884413],"genre_scores_gemma":[0.9983147,0.000009672979,0.0003253799,0.0001410278,0.0001458133,0.00002392035,7.667667e-7,0.000006749362,0.001031908],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2617422,"threshold_uncertainty_score":0.9864113,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.5427626271480739,"score_gpt":0.6652534810633781,"score_spread":0.1224908539153042,"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."}}