{"id":"W4240041394","doi":"10.31234/osf.io/kdurz","title":"Measurement Practices in Large-Scale Replications: Insights from Many Labs 2","year":2020,"lang":"en","type":"preprint","venue":"","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":6,"is_retracted":false,"has_abstract":true,"ca_institutions":"McGill University","funders":"McGill University","keywords":"Replication (statistics); Interpretability; Reliability (semiconductor); Construct validity; Scale (ratio); Psychology; Applied psychology; Construct (python library); Sample (material); Data science; Psychometrics; Computer science; Clinical psychology; Statistics; Artificial intelligence; Mathematics; Geography; Cartography","routes":{"ca_aff":true,"ca_fund":true,"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":[0.5696647,0.001883233,0.002763455,0.006368489,0.004794655,0.01133398,0.005801305,0.003397407,0.003663555],"category_scores_gemma":[0.8474924,0.002248744,0.003661479,0.009697077,0.01012512,0.01296487,0.008193937,0.004906304,0.0009022043],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.005582747,"about_ca_system_score_gemma":0.01413818,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.007487115,"about_ca_topic_score_gemma":0.01224151,"domain_scores_codex":[0.3093969,0.6317971,0.02052958,0.01396763,0.02252649,0.001782353],"domain_scores_gemma":[0.05948314,0.7704561,0.02831817,0.09617253,0.04424807,0.001321893],"domain_codex":"methods","domain_gemma":"reproducibility","domain_candidate":"reproducibility","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.001844814,0.001073578,0.2113171,0.01590906,0.008105352,0.002227153,0.2178828,0.00808927,0.003400739,0.1632585,0.01699045,0.3499012],"study_design_scores_gemma":[0.001745485,0.002841785,0.1702233,0.020376,0.006405496,0.002315929,0.06192477,0.04799255,0.00791521,0.5700716,0.1070745,0.001113347],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.2522281,0.0254949,0.659483,0.02774127,0.001174955,0.007543831,0.00245034,0.001494553,0.02238908],"genre_scores_gemma":[0.6664603,0.002078004,0.3197554,0.003277328,0.0003611704,0.00597507,0.0008115713,0.0005207882,0.0007604143],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.4303353,"threshold_uncertainty_score":0.5306801,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.3281582637090656,"score_gpt":0.4435800413767325,"score_spread":0.1154217776676668,"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."}}