{"id":"W2113879845","doi":"10.1080/15305058.2012.690140","title":"Investigating Sources of Differential Item Functioning in International Large-Scale Assessments Using a Confirmatory Approach","year":2013,"lang":"en","type":"article","venue":"International Journal of Testing","topic":"Educational and Psychological Assessments","field":"Psychology","cited_by":28,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of British Columbia","funders":"","keywords":"Differential item functioning; Psychology; Equivalence (formal languages); Measurement invariance; Scale (ratio); Item response theory; Pairwise comparison; Cognition; Psychometrics; Differential (mechanical device); Confirmatory factor analysis; Clinical psychology; Developmental psychology; Social psychology; Statistics; Structural equation modeling","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":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0005426865,0.0001592879,0.0002679794,0.0004102665,0.00005731134,0.000101194,0.0006016135,0.00009568985,0.001483085],"category_scores_gemma":[0.0004099965,0.0001402225,0.0001130669,0.0002157831,0.00007873706,0.0004341546,0.0001097158,0.0004281606,0.00001552508],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001329327,"about_ca_system_score_gemma":0.00009565746,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0002448435,"about_ca_topic_score_gemma":0.00000240246,"domain_scores_codex":[0.9976323,0.0001644189,0.001050061,0.000218086,0.0006995623,0.0002355973],"domain_scores_gemma":[0.997197,0.0005211098,0.001192564,0.0001012531,0.0008863432,0.0001016838],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00003079099,0.0008229513,0.9822505,0.00001242672,0.0002678721,0.00001275772,0.00165101,0.0008305908,0.008220622,0.0006127974,0.0002408252,0.005046816],"study_design_scores_gemma":[0.001775918,0.0001239491,0.9684172,0.0003553685,0.0000270134,0.0004090928,0.006223802,0.02041273,0.000142482,0.00174366,0.000178378,0.0001904236],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9762044,0.000061361,0.007825714,0.0002183147,0.003229065,0.0001115335,0.00001109994,0.00001223175,0.01232627],"genre_scores_gemma":[0.9676636,0.000001545708,0.03116799,0.0002112895,0.0007186123,0.00001325388,0.00001355278,0.00001537609,0.0001947451],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.02334227,"threshold_uncertainty_score":0.9994297,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1081283061996754,"score_gpt":0.4028085569126335,"score_spread":0.2946802507129581,"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."}}