{"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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.008047015,0.0005560822,0.0005488629,0.00225042,0.0005196131,0.001491965,0.000550086,0.0004026708,0.002588206],"category_scores_gemma":[0.03276218,0.0001434223,0.000866248,0.001462196,0.0003455665,0.001072045,0.001100218,0.0004945186,0.0009499522],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0005822137,"about_ca_system_score_gemma":0.001896563,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001592962,"about_ca_topic_score_gemma":0.00544742,"domain_scores_codex":[0.9947692,0.002548964,0.0005619912,0.0004294362,0.001511907,0.0001785773],"domain_scores_gemma":[0.973202,0.01257388,0.004448663,0.002184039,0.005529121,0.002062261],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0008034665,0.004169297,0.6813898,0.0001701249,0.0002637879,0.000127648,0.001133183,0.002964169,0.006888864,0.0008182258,0.001757441,0.2995139],"study_design_scores_gemma":[0.0001054049,0.008751153,0.9640856,0.0001318616,0.0001528504,0.0001754176,0.001146785,0.01189067,0.009630787,0.0008076951,0.003023189,0.00009860956],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9834791,0.00009841601,0.009044452,0.00009675632,0.00005508594,0.0004369696,0.0002378824,0.0002222268,0.006329056],"genre_scores_gemma":[0.9859596,0.0001040795,0.01224496,0.00003510983,0.00001769256,0.0001809534,0.0002859544,0.00001896591,0.001152496],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.008047015,"threshold_uncertainty_score":0.04255718,"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."}}