{"id":"W4210362394","doi":"10.1111/jep.13657","title":"High quality (certainty) evidence changes less often than low‐quality evidence, but the magnitude of effect size does not systematically differ between studies with low versus high‐quality evidence","year":2022,"lang":"en","type":"article","venue":"Journal of Evaluation in Clinical Practice","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":24,"is_retracted":false,"has_abstract":true,"ca_institutions":"McMaster University","funders":"Agency for Healthcare Research and Quality","keywords":"Medicine; Grading (engineering); Evidence-based medicine; Statistics; Systematic review; Publication bias; Meta-analysis; Standard deviation; Econometrics; MEDLINE; Mathematics; Internal medicine; Alternative medicine","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":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.287252,0.002378779,0.01139874,0.02828316,0.001664849,0.01238038,0.004232265,0.006049623,0.004403572],"category_scores_gemma":[0.712868,0.003520995,0.02040854,0.01944792,0.005027984,0.01105096,0.007335226,0.005267861,0.0008224989],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.006823748,"about_ca_system_score_gemma":0.00835263,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003590282,"about_ca_topic_score_gemma":0.00828472,"domain_scores_codex":[0.4840605,0.2113071,0.2239997,0.0208873,0.05766754,0.002078057],"domain_scores_gemma":[0.1565471,0.6670787,0.1267917,0.01781941,0.03051425,0.001248803],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"systematic_review","study_design_gemma":"observational","study_design_scores_codex":[0.00282656,0.0001004337,0.0409839,0.6550891,0.1792884,0.0005693083,0.002508607,0.0007011042,0.001347044,0.005202376,0.005577894,0.1058053],"study_design_scores_gemma":[0.004141509,0.001006517,0.07114558,0.530021,0.332211,0.001396224,0.001906622,0.001112506,0.002730281,0.02083145,0.03303242,0.0004649842],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"review","genre_gemma":"empirical","genre_scores_codex":[0.02722243,0.9109869,0.02420583,0.01128505,0.00423188,0.008382901,0.005767915,0.0002588237,0.007658313],"genre_scores_gemma":[0.5604828,0.3089997,0.08395576,0.01616925,0.003355726,0.02146442,0.004034729,0.0002975047,0.001240154],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.712748,"threshold_uncertainty_score":0.8789452,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.8952655282853392,"score_gpt":0.6881634937659316,"score_spread":0.2071020345194077,"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."}}