{"id":"W4416319867","doi":"10.2139/ssrn.5763589","title":"Limited consensus in expert opinions on studies evaluating the design, conduct, analysis, or reporting of health research: a survey study","year":2025,"lang":"en","type":"preprint","venue":"SSRN Electronic Journal","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":false,"ca_institutions":"Impact; University of Ottawa; Bruyère; Children’s Health Research Institute; McMaster University; University of Toronto","funders":"","keywords":"Terminology; Flexibility (engineering); Qualitative research; Survey research; Survey data collection; Process (computing); MEDLINE; Survey methodology","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[{"model":"gemma","categories":["metaresearch"],"domain":"methods","study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low","status":"direct model label, unvalidated"},{"model":"gpt","categories":["metaresearch"],"domain":"methods","study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"high","status":"direct model label, unvalidated"}],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.6070454,0.001053754,0.00468632,0.01679315,0.00350368,0.008424032,0.004032286,0.006543916,0.00460805],"category_scores_gemma":[0.8322502,0.00280545,0.006009099,0.01439569,0.006883048,0.01234924,0.01247122,0.005614566,0.0007289978],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.006069376,"about_ca_system_score_gemma":0.01327128,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002913663,"about_ca_topic_score_gemma":0.003269242,"domain_scores_codex":[0.2008189,0.5362232,0.1708315,0.02468771,0.06298431,0.004454326],"domain_scores_gemma":[0.03356687,0.8448634,0.05059974,0.02518976,0.04344834,0.002331934],"domain_codex":"methods","domain_gemma":"methods","domain_candidate":"methods","domain_consensus":"methods","study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.005996296,0.001106984,0.4462708,0.05608545,0.02296447,0.001037208,0.1643227,0.001078865,0.002615025,0.00854692,0.01561584,0.2743595],"study_design_scores_gemma":[0.004574145,0.005685037,0.5195475,0.1149551,0.03638729,0.006635828,0.1585962,0.009909199,0.007230502,0.03778766,0.09687015,0.001821461],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.7475796,0.08749673,0.09681191,0.0372714,0.001721168,0.007020654,0.003325428,0.0003142757,0.01845893],"genre_scores_gemma":[0.9528874,0.007545554,0.02179525,0.01201319,0.0005916783,0.003565326,0.001007975,0.0001758043,0.0004178091],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.3929546,"threshold_uncertainty_score":0.4845831,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.9830111413737759,"score_gpt":0.7377449015228527,"score_spread":0.2452662398509232,"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."}}