{"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":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","research_integrity"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.04220158,0.0001368416,0.0005274945,0.0001804129,0.0006055174,0.00001954172,0.0004889115,0.0003273145,0.0006425267],"category_scores_gemma":[0.0622824,0.00008203363,0.0001659795,0.000749942,0.001026455,0.00009694816,0.001378057,0.002326006,0.00003316991],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002390427,"about_ca_system_score_gemma":0.002801596,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001205474,"about_ca_topic_score_gemma":0.0005113246,"domain_scores_codex":[0.9603918,0.03585654,0.0005403745,0.0005023211,0.001990233,0.0007187343],"domain_scores_gemma":[0.9020259,0.09688958,0.00009473456,0.0003912416,0.0002530108,0.0003454975],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.001105164,0.0007078313,0.2281148,0.02199838,0.0001884682,0.0001186118,0.0756682,0.000001736159,0.04265341,0.02752271,0.1959186,0.4060021],"study_design_scores_gemma":[0.003645066,0.002936078,0.677917,0.02591752,0.0002790998,0.0001765073,0.08901288,0.02009063,0.007616736,0.1203407,0.05109366,0.0009740772],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9626284,0.002023634,0.01357415,0.01764342,0.002550151,0.0008146708,0.00001506301,0.00006139382,0.0006891452],"genre_scores_gemma":[0.9868348,0.001628881,0.009988306,0.0001514631,0.0009280141,0.0000962258,0.00000534462,0.00001985449,0.0003471562],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.4498022,"threshold_uncertainty_score":0.9999757,"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."}}