{"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":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.07430344,0.0008927176,0.0009164136,0.00344865,0.001548484,0.002632087,0.001353192,0.0006525397,0.002076098],"category_scores_gemma":[0.1899419,0.0005357808,0.001304346,0.003386508,0.004274826,0.002559716,0.004015861,0.001468049,0.0003674475],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001607059,"about_ca_system_score_gemma":0.001883784,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00332186,"about_ca_topic_score_gemma":0.004363356,"domain_scores_codex":[0.9338647,0.0418472,0.008290452,0.0039256,0.0103185,0.00175353],"domain_scores_gemma":[0.8127816,0.1338195,0.01832343,0.01555776,0.01791389,0.001603727],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0003563903,0.00015318,0.9005221,0.0003126958,0.0004076783,0.0001789392,0.02625144,0.000616929,0.00239105,0.004543534,0.0005585836,0.06370756],"study_design_scores_gemma":[0.00008410593,0.0006473677,0.9522521,0.000475354,0.0002181879,0.0005914571,0.01975676,0.006445002,0.006370705,0.009871946,0.003173488,0.0001134588],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9703879,0.0003377303,0.02038265,0.0007743238,0.0001195347,0.0002496778,0.000160821,0.00005324746,0.007534118],"genre_scores_gemma":[0.9943156,0.00005483731,0.004823813,0.0001541135,0.00001434892,0.0002261264,0.0001199331,0.00002585573,0.0002653367],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9256966,"threshold_uncertainty_score":0.3929589,"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."}}