{"id":"W3155215713","doi":"10.5539/elt.v14n5p23","title":"Identifying Guessing in English Language Tests via Rasch Fit Statistics: An Exploratory Study","year":2021,"lang":"en","type":"article","venue":"English Language Teaching","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Rasch model; Psychology; Polytomous Rasch model; Test (biology); Statistics; Item response theory; Social psychology; Psychometrics; Mathematics education; Developmental psychology; Mathematics","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":true,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.04793762,0.0005475354,0.0008153571,0.002382186,0.001550244,0.002780577,0.001456678,0.001137738,0.001561672],"category_scores_gemma":[0.1711856,0.0004891099,0.0007766988,0.002378887,0.002034431,0.002617455,0.001901857,0.001762424,0.0006196765],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001092977,"about_ca_system_score_gemma":0.001163309,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002155827,"about_ca_topic_score_gemma":0.003402964,"domain_scores_codex":[0.9658831,0.02500588,0.001855144,0.00159049,0.004635352,0.001029952],"domain_scores_gemma":[0.7719437,0.2024634,0.008005942,0.007309107,0.009453028,0.0008247865],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0008439702,0.00408298,0.6109318,0.000568073,0.0002008484,0.00172752,0.2215353,0.006549042,0.005823855,0.007312286,0.001802741,0.1386215],"study_design_scores_gemma":[0.0001276997,0.008071388,0.5509968,0.0007095316,0.0001387023,0.002349823,0.3358211,0.06188211,0.01825845,0.009521536,0.01182733,0.0002956359],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9910909,0.0000497487,0.007359063,0.0001021284,0.000003645895,0.0002848308,0.00008733966,0.00002139932,0.001001008],"genre_scores_gemma":[0.9831946,0.00008147626,0.01534285,0.00006325651,0.000007455716,0.0003794164,0.0001753063,0.00002269317,0.0007328184],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.04793762,"threshold_uncertainty_score":0.2535214,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2478259227889233,"score_gpt":0.467829404316962,"score_spread":0.2200034815280387,"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."}}