{"id":"W2018691396","doi":"10.1111/j.1745-3992.2010.00173.x","title":"Application of Think Aloud Protocols for Examining and Confirming Sources of Differential Item Functioning Identified by Expert Reviews","year":2010,"lang":"en","type":"article","venue":"Educational Measurement Issues and Practice","topic":"Educational and Psychological Assessments","field":"Psychology","cited_by":96,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of New Brunswick; University of British Columbia","funders":"","keywords":"Differential item functioning; Think aloud protocol; Psychology; Empirical evidence; Cognitive psychology; Expert opinion; Applied psychology; Social psychology; Item response theory; Computer science; Developmental psychology; Psychometrics; Epistemology; Medicine; Human–computer interaction","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":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.2149297,0.002266272,0.001834006,0.009078635,0.002895811,0.002945993,0.002780356,0.001528558,0.002329929],"category_scores_gemma":[0.3916547,0.00140646,0.00128525,0.004398292,0.002474734,0.002871722,0.004249005,0.002393657,0.002093779],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00187649,"about_ca_system_score_gemma":0.007874333,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0008753639,"about_ca_topic_score_gemma":0.002535016,"domain_scores_codex":[0.6542087,0.2567131,0.04287118,0.0111398,0.03349718,0.00157003],"domain_scores_gemma":[0.427949,0.3650914,0.04525364,0.04969629,0.1102313,0.001778442],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"qualitative","study_design_scores_codex":[0.001991789,0.001022342,0.011671,0.007714304,0.0006141101,0.0009419167,0.1177458,0.002327097,0.07361138,0.008690528,0.009687918,0.7639818],"study_design_scores_gemma":[0.002536988,0.01374987,0.07995953,0.01048903,0.001718504,0.004535196,0.0980249,0.04438656,0.3638791,0.09725705,0.2807628,0.002700473],"study_design_candidate":"qualitative","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.157169,0.001558288,0.7670733,0.001473126,0.0009516002,0.0556288,0.001137828,0.002285451,0.01272266],"genre_scores_gemma":[0.08723245,0.0007140122,0.8424261,0.0004201975,0.0001558065,0.06655532,0.0003914263,0.0002678598,0.001836753],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.2149297,"threshold_uncertainty_score":0.9681315,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2023495907107836,"score_gpt":0.4728441178860088,"score_spread":0.2704945271752253,"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."}}