{"id":"W2902135892","doi":"10.1027/1614-2241/a000156","title":"Examining Validity Evidence of Self-Report Measures Using Differential Item Functioning","year":2018,"lang":"en","type":"article","venue":"Methodology","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":9,"is_retracted":false,"has_abstract":true,"ca_institutions":"Centre for Advancing Health Outcomes; Learning Partnership; University of British Columbia; Providence Health Care Research Institute; Providence Health Care","funders":"","keywords":"Differential item functioning; Item response theory; Psychology; Nonparametric statistics; Ordinal data; Scale (ratio); Ordinal Scale; Multilevel model; Logistic regression; Latent variable; Statistics; Econometrics; Psychometrics; Clinical psychology; Mathematics","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":[0.05156149,0.0002279295,0.0009014422,0.0008195868,0.0003480022,0.00009529106,0.000900591,0.0002271698,0.0006947915],"category_scores_gemma":[0.5043761,0.000171873,0.00017548,0.002282435,0.0003997112,0.0002990212,0.0004542281,0.0002758148,0.00002395948],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0000569484,"about_ca_system_score_gemma":0.0001397633,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001858269,"about_ca_topic_score_gemma":0.00001392827,"domain_scores_codex":[0.9859024,0.009768643,0.001537847,0.0009564214,0.001321845,0.000512843],"domain_scores_gemma":[0.775876,0.2188537,0.001913433,0.001506146,0.001672162,0.0001786298],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0001911295,0.00005864094,0.7453701,0.00002386914,0.00009969637,0.00004224344,0.001017094,0.00009423086,0.1364369,0.0003007706,0.0003963793,0.115969],"study_design_scores_gemma":[0.0006944019,0.001380992,0.8828159,0.0001967093,0.0002653912,0.001274401,0.001993847,0.01093573,0.0599493,0.03718641,0.00266003,0.0006468682],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.5052119,0.0001647395,0.491093,0.00002890183,0.002678838,0.00007176885,7.065198e-7,0.00006462595,0.0006854945],"genre_scores_gemma":[0.5563092,0.000007816942,0.4430392,0.00003427243,0.0004841175,0.000003038115,2.830139e-7,0.0000102538,0.0001118861],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.4528145,"threshold_uncertainty_score":0.976617,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.9393714709572591,"score_gpt":0.5778324970126306,"score_spread":0.3615389739446285,"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."}}