{"id":"W2058922083","doi":"10.1080/15305058.2014.891223","title":"Uncovering Substantive Patterns in Student Responses in International Large-Scale Assessments—Comparing a Latent Class to a Manifest DIF Approach","year":2014,"lang":"en","type":"article","venue":"International Journal of Testing","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":19,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of British Columbia","funders":"","keywords":"Generalizability theory; Differential item functioning; Psychology; Item response theory; Latent class model; Homogeneous; Contrast (vision); Homogeneity (statistics); Reading comprehension; Cognitive psychology; Scale (ratio); Comprehension; Social psychology; Psychometrics; Developmental psychology; Reading (process); Statistics; Mathematics","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.01271289,0.0001915751,0.0004777049,0.002268738,0.00005397756,0.0004898458,0.002060759,0.0000604468,0.00005155797],"category_scores_gemma":[0.04956838,0.0001566582,0.0001239468,0.001241185,0.00002201252,0.0005337393,0.0006443435,0.0005094996,0.000009028298],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0004680891,"about_ca_system_score_gemma":0.00007811494,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001236245,"about_ca_topic_score_gemma":0.0001802323,"domain_scores_codex":[0.9943575,0.0006221823,0.001743144,0.0004423351,0.002468567,0.0003663188],"domain_scores_gemma":[0.9848703,0.01264912,0.001107773,0.0002317816,0.001005467,0.0001355358],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0001725475,0.0002701917,0.9451922,0.000002883766,0.00003857865,0.0001312257,0.0009504888,0.03054462,0.0005526301,0.0001358239,0.00002790995,0.02198095],"study_design_scores_gemma":[0.001320004,0.0001434564,0.9347823,0.0003514331,0.000004739696,0.0001692207,0.00272909,0.05844578,0.00006666108,0.001492769,0.0003381016,0.0001564736],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9067892,0.00002264433,0.08631536,0.0004242969,0.001872557,0.0001116952,0.000008516719,0.00001328271,0.004442409],"genre_scores_gemma":[0.9277487,0.000005808057,0.07160754,0.0001918723,0.0003485806,0.000007057259,0.000001502481,0.00001353982,0.00007538901],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.03685549,"threshold_uncertainty_score":0.9584375,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.3759875606329405,"score_gpt":0.4999896769691466,"score_spread":0.1240021163362061,"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."}}