{"id":"W2316658219","doi":"10.1119/perc.2013.pr.047","title":"Development of a General Undergraduate Estimation Skills Survey (GUESS)","year":2014,"lang":"en","type":"article","venue":"","topic":"Statistics Education and Methodologies","field":"Mathematics","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of British Columbia","funders":"","keywords":"Test (biology); Class (philosophy); Set (abstract data type); Computer science; Mathematics education; Estimation; Baseline (sea); Machine learning; Multiple choice; Artificial intelligence; Statistics; Psychology; Mathematics; Engineering; Significant difference","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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.00864167,0.0004782189,0.0007396667,0.00284807,0.0003291392,0.0007993055,0.0007964472,0.0004726052,0.008094393],"category_scores_gemma":[0.01779151,0.000497263,0.000502633,0.001146073,0.000275894,0.0009715058,0.001254379,0.001088761,0.005352197],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0006981123,"about_ca_system_score_gemma":0.001846133,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001263094,"about_ca_topic_score_gemma":0.001831046,"domain_scores_codex":[0.9969638,0.0009513982,0.0006302915,0.0002403802,0.001022662,0.0001915075],"domain_scores_gemma":[0.9869673,0.003422224,0.0009470028,0.001379432,0.006310574,0.0009734905],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.0005053204,0.003920892,0.2816458,0.0005323817,0.00006367981,0.0001670961,0.002606128,0.002510294,0.009159138,0.001794187,0.01403019,0.6830649],"study_design_scores_gemma":[0.0003896727,0.01069538,0.8267412,0.0003649892,0.00009929307,0.0005039999,0.005328432,0.01929664,0.01850888,0.003466928,0.1144276,0.0001769942],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"methods","genre_scores_codex":[0.7871758,0.0002012629,0.122964,0.001012895,0.0002683635,0.04627764,0.01669897,0.003137551,0.02226339],"genre_scores_gemma":[0.5369214,0.0006401062,0.3692778,0.0007881021,0.0001330345,0.04664608,0.02645441,0.0003455611,0.01879342],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.00864167,"threshold_uncertainty_score":0.0457021,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.200871181720234,"score_gpt":0.4406332209790296,"score_spread":0.2397620392587956,"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."}}