{"id":"W2086620040","doi":"10.1080/01443410.2014.946890","title":"Exploring plausible causes of differential item functioning in the PISA science assessment: language, curriculum or culture","year":2014,"lang":"en","type":"article","venue":"Educational Psychology","topic":"Student Assessment and Feedback","field":"Social Sciences","cited_by":45,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Differential item functioning; Curriculum; Psychology; Scientific literacy; Literacy; Mainland China; Cross-cultural studies; Cultural diversity; Scale (ratio); Measurement invariance; Cross-cultural; China; Item response theory; Item analysis; Language proficiency; Variance (accounting); Mathematics education; Psychometrics; Developmental psychology; Pedagogy; Science education; Social psychology; Geography; Sociology; Confirmatory factor analysis","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":[],"consensus_categories":[],"category_scores_codex":[0.0008524708,0.00007959997,0.0001108346,0.0001535987,0.0004181768,0.0000720477,0.000475559,0.00003628698,0.0008137596],"category_scores_gemma":[0.0003016015,0.00005422369,0.00003218311,0.000806785,0.0003892729,0.0003438984,0.00003209812,0.0001584791,0.00002074919],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00006453037,"about_ca_system_score_gemma":0.0002820369,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0004908598,"about_ca_topic_score_gemma":0.0007887768,"domain_scores_codex":[0.9986439,0.0001964017,0.0001857098,0.0002339101,0.0004876991,0.0002523729],"domain_scores_gemma":[0.9993122,0.0003010428,0.0001051114,0.0001395333,0.00009956089,0.00004250169],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00001529955,0.0005801186,0.7513995,0.000009206457,0.00001344788,4.688991e-7,0.04959991,0.000006891292,0.001868177,0.1904907,0.004601943,0.001414407],"study_design_scores_gemma":[0.0002485153,0.00005246237,0.9620193,0.00001791393,0.00000817509,0.000001613,0.03022033,0.00001611081,0.00001764942,0.0007042759,0.006613357,0.0000802785],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9371946,0.0000224575,0.0001391653,0.00298208,0.00179657,0.0001441115,0.000001226271,0.00001339118,0.05770643],"genre_scores_gemma":[0.9971644,0.00003899654,0.0005699116,0.0002921752,0.0008782755,0.00007275675,0.000007207786,0.000004392263,0.0009719134],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2106199,"threshold_uncertainty_score":0.8910097,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1094805351232798,"score_gpt":0.4612824068570056,"score_spread":0.3518018717337257,"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."}}