{"id":"W2234202768","doi":"","title":"Prevalence of Gender DIF in Mixed Format High School Exit Examinations.","year":2001,"lang":"en","type":"article","venue":"","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":9,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Polytomous Rasch model; Differential item functioning; Item response theory; Psychology; Statistics; Social psychology; Developmental psychology; Psychometrics; Mathematics","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":false,"about_ca":true,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.005502596,0.00008937193,0.0002188636,0.00122977,0.00004849799,0.00005721007,0.0007094138,0.00005079709,0.003259221],"category_scores_gemma":[0.06817585,0.00005983,0.00005235268,0.003654932,0.0000563878,0.0004057216,0.0001481372,0.0001033805,0.0001130741],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00002137771,"about_ca_system_score_gemma":0.0000264145,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00009177448,"about_ca_topic_score_gemma":0.00001004243,"domain_scores_codex":[0.9975604,0.0003210321,0.0007322408,0.0003071974,0.0008444142,0.0002347259],"domain_scores_gemma":[0.9865406,0.01240449,0.0002370588,0.0005190928,0.0002192754,0.00007944003],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00001087889,0.00006519441,0.8894633,0.00001598066,0.000003674335,0.000005342526,0.00007610345,0.00008381408,0.0002515425,0.002220779,0.005789152,0.1020142],"study_design_scores_gemma":[0.0002781806,0.00003082595,0.951727,0.00001078221,0.000002480043,0.00001062285,0.0004974419,0.001460634,0.0005034736,0.04460562,0.0007952703,0.00007767692],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8803766,0.0002301735,0.06671795,0.0001169291,0.0003034375,0.0001284803,0.000003422095,0.00003167688,0.05209137],"genre_scores_gemma":[0.9495864,0.0002407398,0.04661294,0.00007875594,0.00002286506,0.000008558752,4.026465e-7,0.000003776274,0.003445554],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1019366,"threshold_uncertainty_score":0.9976519,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.5182594646204142,"score_gpt":0.4606609800442816,"score_spread":0.05759848457613254,"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."}}