{"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":"codex-gemma-dda1882f352a","candidate_categories":["metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.0005514061,0.0003678057,0.0005955912,0.0001375985,0.00004897804,0.00009217174,0.0006596113,0.0004027183,0.0002474318],"category_scores_gemma":[0.003538382,0.0003370127,0.0001018527,0.0001907092,0.00003437563,0.000236297,0.001111894,0.001119045,0.000101075],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0004416819,"about_ca_system_score_gemma":0.0002206989,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0003808636,"about_ca_topic_score_gemma":0.002927713,"domain_scores_codex":[0.9971143,0.000192677,0.0008079679,0.0009587102,0.0006747128,0.0002516284],"domain_scores_gemma":[0.9963958,0.0003594518,0.001188496,0.001619363,0.000330173,0.000106686],"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":[0.0001612669,0.002776858,0.005082688,0.001870124,0.0005616518,0.0001199114,0.01174506,0.00005736065,0.0184531,0.9186165,0.03759785,0.002957658],"study_design_scores_gemma":[0.0001755056,0.00002409266,0.00112049,0.0005006532,0.00006328438,6.476709e-7,0.0004309531,0.001146091,0.01233321,0.9763867,0.007419082,0.0003992931],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.02905861,0.001847575,0.8695836,0.008177755,0.0003343728,0.003779342,0.0001279928,0.003213516,0.08387721],"genre_scores_gemma":[0.5044718,0.0003167122,0.4929967,0.0006162162,0.0001727399,0.001015543,0.00008304515,0.00007760173,0.0002497158],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.4754131,"threshold_uncertainty_score":0.9999082,"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."}}