{"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":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.001154729,0.0001035847,0.0001530637,0.00018786,0.0006814459,0.0001405372,0.0001199432,0.000103453,0.0000064717],"category_scores_gemma":[0.0003532072,0.000115786,0.00002300619,0.0002033912,0.00002656432,0.0002652766,0.0000158742,0.0001211044,9.379851e-7],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00044154,"about_ca_system_score_gemma":0.0003030027,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001212185,"about_ca_topic_score_gemma":0.04451766,"domain_scores_codex":[0.9987578,0.00006286146,0.0003140557,0.0002479444,0.000402651,0.0002147248],"domain_scores_gemma":[0.9990309,0.00004499794,0.00034063,0.0002171321,0.0003097595,0.00005660948],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00007642937,0.0003425983,0.8873059,0.0001416556,0.00002362272,3.894926e-8,0.004390021,0.0009449901,0.07174831,0.001141685,0.00008596254,0.03379886],"study_design_scores_gemma":[0.0006331923,0.00002042232,0.9895396,0.0001173794,0.00002486791,1.571572e-7,0.003411435,0.001065597,0.003839413,0.0006933319,0.0004971424,0.0001574859],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9932019,0.00006212814,0.001510347,0.0002503477,0.0009722851,0.0007675533,0.000002399528,0.00001631747,0.003216722],"genre_scores_gemma":[0.9987636,0.0000645998,0.0003062924,0.00003016684,0.0003640564,0.0004199852,0.00002083211,0.00001083198,0.0000196769],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1022338,"threshold_uncertainty_score":0.9729174,"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."}}