{"id":"W4404247756","doi":"10.1186/s12874-024-02387-z","title":"Evidence pointing toward invalidity of the SF-8 physical and mental scales: a fusion validity assessment","year":2024,"lang":"en","type":"article","venue":"BMC Medical Research Methodology","topic":"Geriatric Care and Nursing Homes","field":"Health Professions","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"York University; University of Alberta","funders":"University of Alberta","keywords":"Psychology; Mental health; Psychometrics; Test validity; Clinical psychology; Applied psychology; Psychiatry","routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":true,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.1712345,0.0007512251,0.0006134619,0.007491915,0.0027681,0.002786183,0.003469538,0.001796128,0.004195455],"category_scores_gemma":[0.4040176,0.0007622387,0.00287193,0.006699451,0.01278515,0.00331838,0.004961287,0.004277114,0.0005135885],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.005135624,"about_ca_system_score_gemma":0.006121029,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01474672,"about_ca_topic_score_gemma":0.0132707,"domain_scores_codex":[0.8414604,0.0672093,0.02548093,0.01004239,0.05259266,0.003214274],"domain_scores_gemma":[0.3240247,0.4898081,0.06779934,0.03328692,0.08311354,0.001967403],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0007094569,0.0001534336,0.8773329,0.002380624,0.00109637,0.0006074061,0.01221549,0.000536521,0.001099093,0.01927205,0.004965906,0.07963084],"study_design_scores_gemma":[0.0001965081,0.0006420834,0.8891688,0.008368427,0.001305205,0.003137256,0.0219984,0.005313234,0.005297442,0.02705025,0.03731961,0.0002027129],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8121977,0.0186564,0.06039247,0.04700039,0.002002714,0.000977189,0.003103027,0.0001530987,0.05551694],"genre_scores_gemma":[0.9850093,0.001565023,0.009360103,0.002463555,0.0003084679,0.000201756,0.0007308223,0.00004880243,0.0003122451],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.8287655,"threshold_uncertainty_score":0.9055853,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.7846350592512534,"score_gpt":0.663055238900969,"score_spread":0.1215798203502845,"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."}}