{"id":"W1771332781","doi":"10.1371/journal.pone.0135162","title":"Evaluation of Nine Consensus Indices in Delphi Foresight Research and Their Dependency on Delphi Survey Characteristics: A Simulation Study and Debate on Delphi Design and Interpretation","year":2015,"lang":"en","type":"article","venue":"PLoS ONE","topic":"Delphi Technique in Research","field":"Social Sciences","cited_by":124,"is_retracted":false,"has_abstract":true,"ca_institutions":"McGill University","funders":"Wellcome Trust","keywords":"Delphi method; Delphi; Stakeholder; Statistics; Descriptive statistics; Pairwise comparison; Computer science; Mathematics; Economics; Management","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":[0.03401948,0.0001640533,0.0003370265,0.0007420665,0.0002290356,0.0001336266,0.0001991788,0.0001636413,0.00001350875],"category_scores_gemma":[0.01400952,0.0001479422,0.00001125397,0.0006037886,0.0004512724,0.0002095103,0.0001173747,0.0003599684,0.000003244886],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002583174,"about_ca_system_score_gemma":0.0003574855,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002758313,"about_ca_topic_score_gemma":0.008598048,"domain_scores_codex":[0.9903202,0.005969575,0.0004840847,0.0004633522,0.002404695,0.0003580934],"domain_scores_gemma":[0.9917858,0.005421283,0.0002011321,0.0002388199,0.002171559,0.0001813901],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.008736445,0.01110502,0.5824661,0.0002534151,0.0005551822,0.00004769557,0.2631313,0.001558288,0.003961695,0.0006425203,0.000107968,0.1274344],"study_design_scores_gemma":[0.003479113,0.00664193,0.4680203,0.0008026544,0.0001240599,0.000001480164,0.02625261,0.473132,0.002305349,0.01879942,0.000003627292,0.0004375127],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9957432,0.0002279426,0.0003413121,0.0002837045,0.00001967379,0.003002026,0.00002083898,0.00003506421,0.0003262612],"genre_scores_gemma":[0.9992133,0.0001320032,0.0004191004,0.00001262624,0.00003498216,0.0001326455,0.00001408904,0.0000214307,0.00001986803],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.4715737,"threshold_uncertainty_score":0.9946802,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.677004819242767,"score_gpt":0.5118827177054857,"score_spread":0.1651221015372814,"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."}}