{"id":"W2607007324","doi":"10.1080/08957347.2017.1316276","title":"For Which Boys and Which Girls Are Reading Assessment Items Biased Against? Detection of Differential Item Functioning in Heterogeneous Gender Populations","year":2017,"lang":"en","type":"article","venue":"Applied Measurement in Education","topic":"School Choice and Performance","field":"Social Sciences","cited_by":14,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of British Columbia","funders":"Institute of Education Sciences","keywords":"Differential item functioning; Psychology; Disadvantage; Developmental psychology; Socioeconomic status; Gender bias; Item response theory; Demography; Psychometrics; Social psychology; Population","routes":{"ca_aff":true,"ca_fund":false,"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":[],"consensus_categories":[],"category_scores_codex":[0.0133013,0.0003145933,0.000761644,0.001756271,0.0005972525,0.001558517,0.0005744162,0.0007291082,0.001555315],"category_scores_gemma":[0.0424449,0.0002222016,0.0007379734,0.001287418,0.001299051,0.001416907,0.0009649447,0.0005413471,0.000522071],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0007135457,"about_ca_system_score_gemma":0.0006652625,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00849327,"about_ca_topic_score_gemma":0.01166703,"domain_scores_codex":[0.9915791,0.004188235,0.000660038,0.001341269,0.001565711,0.0006656476],"domain_scores_gemma":[0.9844714,0.006947741,0.004468632,0.002306802,0.00149288,0.0003125237],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0001318703,0.00002009119,0.9649494,0.00004566904,0.0001517375,0.0001223298,0.004935246,0.00008604203,0.001162302,0.0006192533,0.0003829623,0.02739306],"study_design_scores_gemma":[0.000006922896,0.00007816102,0.9901148,0.00009120972,0.0000543998,0.0003694034,0.006009462,0.0004412839,0.0006585334,0.001267754,0.0008886815,0.00001939108],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9948577,0.0005628876,0.001863734,0.0003622836,0.00002320113,0.00003081892,0.0002279033,0.00001147822,0.002059865],"genre_scores_gemma":[0.9988844,0.00009274847,0.0006024002,0.00008453861,0.000007218902,0.00001696846,0.00009628361,0.000006534653,0.0002088077],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.0133013,"threshold_uncertainty_score":0.07034481,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1191182427796675,"score_gpt":0.3618901915618662,"score_spread":0.2427719487821987,"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."}}