{"id":"W2911785435","doi":"10.5430/jct.v8n1p1","title":"Graded Response Method: Does Question Type Influence the Assessment of Critical Thinking?","year":2019,"lang":"en","type":"article","venue":"Journal of Curriculum and Teaching","topic":"Education and Critical Thinking Development","field":"Social Sciences","cited_by":6,"is_retracted":false,"has_abstract":true,"ca_institutions":"College of Family Physicians of Canada; University of Toronto","funders":"University of Toronto Mississauga; University of Toronto","keywords":"Critical thinking; Grading (engineering); Operationalization; Multiple choice; Taxonomy (biology); Mathematics education; Inference; Psychology; Critical appraisal; Statistics; Mathematics; Computer science; Epistemology; Artificial intelligence; Biology","routes":{"ca_aff":true,"ca_fund":true,"ca_venue":true,"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":[0.1246961,0.001083488,0.001541933,0.00279413,0.0009661176,0.002533934,0.001890144,0.001450726,0.00556554],"category_scores_gemma":[0.3729227,0.0005162846,0.001958356,0.003013557,0.001538428,0.003267575,0.002125795,0.001393124,0.001394815],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001568608,"about_ca_system_score_gemma":0.001250247,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0009031585,"about_ca_topic_score_gemma":0.001694736,"domain_scores_codex":[0.8522557,0.1064291,0.01221059,0.008113262,0.01970001,0.001291341],"domain_scores_gemma":[0.5078669,0.3947903,0.04828367,0.01382167,0.03315914,0.002078316],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0110133,0.002603588,0.5425884,0.00353436,0.003654213,0.0002065106,0.01839239,0.002264173,0.006965683,0.004767707,0.008510118,0.3954996],"study_design_scores_gemma":[0.00234398,0.01455532,0.8676131,0.003261401,0.00252557,0.0009036937,0.009284617,0.03754377,0.01467017,0.02040956,0.02608969,0.0007990779],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8657389,0.002338151,0.1086557,0.002049726,0.001047272,0.007660254,0.001133786,0.0006009796,0.01077509],"genre_scores_gemma":[0.9312755,0.000588665,0.05058609,0.0009101037,0.0002175555,0.01337555,0.0006207688,0.0002106674,0.002214983],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1246961,"threshold_uncertainty_score":0.659464,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.01412744076797895,"score_gpt":0.4142738720014842,"score_spread":0.4001464312335053,"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."}}