{"id":"W4393029196","doi":"10.1016/j.jneumeth.2024.110111","title":"Measuring the replicability of our own research","year":2024,"lang":"en","type":"article","venue":"Journal of Neuroscience Methods","topic":"Health, Environment, Cognitive Aging","field":"Environmental Science","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"Dalhousie University","funders":"Natural Sciences and Engineering Research Council of Canada; Fondation Brain Canada; Simons Foundation","keywords":"Replicate; Psychology; Anxiety; Cognitive psychology; Reliability (semiconductor); Cognition; Developmental psychology; Neuroscience; Psychiatry","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":[0.4085333,0.001116585,0.001818848,0.006620748,0.002258751,0.005356133,0.00331585,0.00295485,0.002397739],"category_scores_gemma":[0.669853,0.001152051,0.001853591,0.003763844,0.01156914,0.006593275,0.007986064,0.003032332,0.001061665],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002469036,"about_ca_system_score_gemma":0.002375472,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001614023,"about_ca_topic_score_gemma":0.001209436,"domain_scores_codex":[0.6248796,0.2280926,0.04002218,0.03402708,0.0706469,0.002331713],"domain_scores_gemma":[0.1745587,0.4275435,0.05783982,0.2456742,0.09098748,0.003396417],"domain_codex":null,"domain_gemma":"reproducibility","domain_candidate":"reproducibility","domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.002173684,0.0008069648,0.5971806,0.003479584,0.006828242,0.001266624,0.06827933,0.004195832,0.01666554,0.06032302,0.008369588,0.230431],"study_design_scores_gemma":[0.0005542708,0.006704116,0.6065908,0.003831971,0.003698144,0.004858102,0.0238725,0.01982398,0.0290249,0.152238,0.1478323,0.000970995],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.5399666,0.01267222,0.3382007,0.01389019,0.005573476,0.005038659,0.001768058,0.001048487,0.08184156],"genre_scores_gemma":[0.909963,0.0009872755,0.07818009,0.002066677,0.001050327,0.003279,0.0006169998,0.0005911545,0.003265381],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.5914667,"threshold_uncertainty_score":0.7293837,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.3218217066765193,"score_gpt":0.5147701931513586,"score_spread":0.1929484864748394,"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."}}