{"id":"W4381800707","doi":"10.5430/ijhe.v12n3p116","title":"Evaluating a University-Wide Mathematics Assessment for Use in Introductory Statistics Course Performance","year":2023,"lang":"en","type":"article","venue":"International Journal of Higher Education","topic":"Statistics Education and Methodologies","field":"Mathematics","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Mathematics education; Course (navigation); Test (biology); Computer science; Mathematics; Engineering","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":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.001437805,0.0001008773,0.0001850251,0.0004345067,0.0000451706,0.00005924178,0.0002902785,0.00004555489,0.000120814],"category_scores_gemma":[0.002329608,0.0001001519,0.00005022024,0.0002038233,0.00003807963,0.0002600775,0.00003348705,0.0001589703,0.000009438032],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0003808869,"about_ca_system_score_gemma":0.0009218187,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000007754714,"about_ca_topic_score_gemma":0.000006812962,"domain_scores_codex":[0.99867,0.0001315674,0.0004988712,0.0001191313,0.0004421954,0.0001382466],"domain_scores_gemma":[0.9941585,0.003771851,0.0005882148,0.0001298572,0.001300443,0.00005113034],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"observational","study_design_scores_codex":[0.0002669932,0.001967004,0.05502558,0.000364886,0.0003816052,0.00002171563,0.00401538,0.00113833,0.00108879,0.3780457,0.4970749,0.0606091],"study_design_scores_gemma":[0.002722241,0.0004141911,0.5513658,0.0006206517,0.00028766,0.00006683068,0.005879185,0.009328328,0.0003954849,0.3536799,0.07477146,0.0004682711],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"methods","genre_scores_codex":[0.8895909,0.00004995646,0.09164684,0.003219552,0.01448157,0.0004401334,0.0001551524,0.00005117052,0.0003647192],"genre_scores_gemma":[0.123229,0.0001005453,0.8661912,0.0001286385,0.0005440731,0.00001043179,0.00006559053,0.00002291898,0.009707641],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.7745444,"threshold_uncertainty_score":0.4084078,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2529458654991861,"score_gpt":0.5232242556622919,"score_spread":0.2702783901631058,"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."}}