{"id":"W2037599414","doi":"10.1111/capa.12103","title":"What metrics? On the utility of measuring the performance of policy research: An illustrative case and alternative from Employment and Social Development Canada","year":2015,"lang":"en","type":"article","venue":"Canadian Public Administration","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":4,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Ottawa; Institute on Governance","funders":"","keywords":"Government (linguistics); Task (project management); Object (grammar); Performance measurement; Focus (optics); Computer science; Research Object; State (computer science); Policy development; Public policy; Management science; Operations research; Public economics; Econometrics; Economics; Sociology; Engineering; Political science; Public administration; Artificial intelligence; Regional science; Economic growth; Management","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":true,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.03529845,0.0008112639,0.001013381,0.007089575,0.01757728,0.02100192,0.002631302,0.004055202,0.00226805],"category_scores_gemma":[0.05641726,0.0003293803,0.0007258533,0.01346224,0.02761089,0.004843828,0.007595375,0.002910204,0.0001644424],"about_ca_system_candidate":true,"about_ca_system_consensus":true,"about_ca_system_score_codex":0.1532482,"about_ca_system_score_gemma":0.1481083,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.9245374,"about_ca_topic_score_gemma":0.9412951,"domain_scores_codex":[0.9465572,0.03294101,0.001132914,0.001361362,0.01217388,0.00583361],"domain_scores_gemma":[0.9320785,0.04725436,0.001948567,0.00237874,0.01345073,0.002889078],"domain_codex":null,"domain_gemma":"evaluation","domain_candidate":"evaluation","domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"not_applicable","study_design_scores_codex":[0.00006619428,0.00005411008,0.007547047,0.0001709666,0.00002554795,0.0007373165,0.01713037,0.004690874,0.0002399311,0.9445279,0.004067292,0.0207425],"study_design_scores_gemma":[0.0001768101,0.0003101185,0.03284868,0.002227895,0.0003172294,0.000752266,0.1861019,0.05364199,0.003224994,0.4786435,0.241343,0.0004116721],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"other","genre_gemma":"empirical","genre_scores_codex":[0.259939,0.007015065,0.03002494,0.1299957,0.0002592837,0.0007125899,0.0005472007,0.0001247971,0.5713813],"genre_scores_gemma":[0.9841027,0.001039821,0.009399471,0.000899597,0.00002542271,0.00009948714,0.00003449994,0.00001942892,0.004379547],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.9647015,"threshold_uncertainty_score":0.9821121,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.6470003141615459,"score_gpt":0.4995877186147932,"score_spread":0.1474125955467527,"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."}}