{"id":"W3033267246","doi":"10.3389/fpsyg.2020.01088","title":"Using Corpus Analyses to Help Address the DIF Interpretation: Gender Differences in Standardized Writing Assessment","year":2020,"lang":"en","type":"article","venue":"Frontiers in Psychology","topic":"Gender Studies in Language","field":"Social Sciences","cited_by":7,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Saskatchewan","funders":"","keywords":"Psychology; Differential item functioning; Test (biology); Interpretation (philosophy); Spelling; Boston Naming Test; Cognitive psychology; Interpretability; Linguistics; Cognition; Developmental psychology; Psychometrics; Item response theory","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":[],"consensus_categories":[],"category_scores_codex":[0.0007654434,0.0001374124,0.0003651905,0.0001475185,0.0002247488,0.00006169883,0.0005222018,0.00009408096,0.0001072791],"category_scores_gemma":[0.000338429,0.0001126452,0.00006012513,0.0008024619,0.000316901,0.00008481824,0.0001288985,0.0002891703,0.00000310836],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001881636,"about_ca_system_score_gemma":0.00008404407,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0002963815,"about_ca_topic_score_gemma":0.0007796161,"domain_scores_codex":[0.9976231,0.0008451862,0.0003635666,0.0004082584,0.0003375009,0.0004224124],"domain_scores_gemma":[0.9994127,0.0001451167,0.0001062551,0.0001938576,0.00005129741,0.00009082555],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"qualitative","study_design_scores_codex":[0.0001852909,0.00006790153,0.7678468,0.00001838404,0.0001426895,0.00006841345,0.1867623,0.0002175509,0.0003531142,0.0007898447,0.01141232,0.03213548],"study_design_scores_gemma":[0.002539195,0.0002083138,0.3911778,0.0001062635,0.00008932716,0.000004296961,0.5808117,0.00526132,0.00003520855,0.007938968,0.01111528,0.000712248],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.3447977,0.00388378,0.5730327,0.01465341,0.005413484,0.001061052,0.0000295456,0.00009050427,0.05703788],"genre_scores_gemma":[0.9706467,0.0001273625,0.0259028,0.003069,0.0001979923,0.00003319229,0.000001820146,0.00001070847,0.00001044508],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.625849,"threshold_uncertainty_score":0.459354,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1870382146697567,"score_gpt":0.4757906005330269,"score_spread":0.2887523858632702,"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."}}