{"id":"W2995932983","doi":"10.3138/cjpe.56989","title":"Comparison of Canadian and American Graduate Evaluation Education Programs","year":2019,"lang":"en","type":"article","venue":"Canadian Journal of Program Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":6,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Victoria","funders":"","keywords":"Professionalization; Graduate education; Graduate students; Population; Political science; Medical education; Program evaluation; Higher education; Library science; Sociology; Public administration; Medicine; Demography; Computer science","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":true,"about_ca":true,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.01694237,0.0001553093,0.0004243647,0.002656962,0.0001565219,0.0003346057,0.0003886101,0.00006462626,0.0009349904],"category_scores_gemma":[0.001737669,0.0001303522,0.0001048131,0.002218883,0.0001831463,0.0006522115,0.00001009769,0.0002299877,0.00006905767],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0006368313,"about_ca_system_score_gemma":0.02039799,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.08616996,"about_ca_topic_score_gemma":0.6335765,"domain_scores_codex":[0.9942293,0.0007922028,0.001260696,0.00027178,0.003112454,0.0003336212],"domain_scores_gemma":[0.9901083,0.0001321496,0.001507627,0.0003645919,0.007075273,0.000812122],"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":[0.000007503712,0.00004601507,0.3015782,0.0000054795,0.00001693021,1.593056e-7,0.001436804,0.0005673218,0.0000243357,0.0001776186,0.001020863,0.6951188],"study_design_scores_gemma":[0.0009514865,0.001577129,0.7378334,0.0001238033,0.0002171648,0.00002619802,0.0120984,0.1978437,0.0001243758,0.004259171,0.04470013,0.0002450168],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9884154,0.0004518343,0.00004371235,0.002432052,0.001060681,0.002292809,0.000004724064,0.000005103905,0.005293696],"genre_scores_gemma":[0.997087,0.000016308,0.002420412,0.0001017479,0.0001000462,0.00009077742,0.00004840834,0.00001226046,0.0001230565],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.6948738,"threshold_uncertainty_score":0.9999783,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.5628449880548297,"score_gpt":0.5884740088916572,"score_spread":0.0256290208368275,"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."}}