{"id":"W4401316857","doi":"10.1111/caje.12728","title":"Reproduce to validate: A comprehensive study on the reproducibility of economics research","year":2024,"lang":"en","type":"article","venue":"Canadian Journal of Economics/Revue canadienne d économique","topic":"scientometrics and bibliometrics research","field":"Decision Sciences","cited_by":5,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Transparency (behavior); Citation; Computer science; Confidentiality; Reproducibility; Code (set theory); Data science; Statistics; Computer security; Library science; Mathematics","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":true,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.4366081,0.0009584305,0.002179183,0.01741955,0.004426589,0.01258356,0.003121394,0.002978626,0.005243809],"category_scores_gemma":[0.8331152,0.001019392,0.004172372,0.02497242,0.009538316,0.01353776,0.01027653,0.002605468,0.001213183],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.005049296,"about_ca_system_score_gemma":0.00868567,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002080112,"about_ca_topic_score_gemma":0.001896025,"domain_scores_codex":[0.5164973,0.265905,0.07546452,0.02423273,0.1139071,0.003993385],"domain_scores_gemma":[0.02795728,0.748964,0.08127726,0.1037679,0.03635368,0.001679964],"domain_codex":null,"domain_gemma":"reproducibility","domain_candidate":"reproducibility","domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.001555906,0.0005496977,0.7999921,0.00447539,0.005261525,0.0009577811,0.02711041,0.002669609,0.001598838,0.01785075,0.008704743,0.1292732],"study_design_scores_gemma":[0.0004831163,0.002038503,0.8503985,0.005803631,0.003661133,0.002566755,0.01528717,0.01111369,0.00829551,0.04054812,0.05915968,0.0006441916],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8889035,0.01154312,0.04690779,0.007453497,0.000638378,0.002364585,0.004694285,0.0007071465,0.03678775],"genre_scores_gemma":[0.9856089,0.0006676633,0.009931002,0.0006583843,0.0002435295,0.0006131768,0.001221344,0.0002598735,0.0007960607],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.5633919,"threshold_uncertainty_score":0.6947625,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.8716767194993417,"score_gpt":0.4702523222071099,"score_spread":0.4014243972922317,"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."}}