{"id":"W2081701430","doi":"10.3138/cjpe.29.1.62","title":"Assessing the Quality of Aboriginal Program Evaluations","year":2014,"lang":"en","type":"article","venue":"Canadian Journal of Program Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":7,"is_retracted":false,"has_abstract":true,"ca_institutions":"Université Laval","funders":"","keywords":"Treasury; Popularity; Quality (philosophy); Context (archaeology); Government (linguistics); Program evaluation; Strengths and weaknesses; Corporate governance; Replicate; Business; Political science; Public administration; Psychology; Finance; Geography; Statistics","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":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.5048721,0.001476171,0.00413732,0.01716251,0.002040611,0.006047614,0.003208497,0.001255977,0.002045672],"category_scores_gemma":[0.7597043,0.001208183,0.00766255,0.02121109,0.0043309,0.003078907,0.004474984,0.001610303,0.0001007094],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.01609104,"about_ca_system_score_gemma":0.02956002,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0548638,"about_ca_topic_score_gemma":0.07341382,"domain_scores_codex":[0.4396013,0.4115202,0.07419215,0.008228734,0.06446396,0.001993697],"domain_scores_gemma":[0.1504117,0.6232551,0.07561405,0.03555488,0.1130961,0.002068256],"domain_codex":null,"domain_gemma":"evaluation","domain_candidate":"evaluation","domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.005460544,0.0002303734,0.4355335,0.09497305,0.1655967,0.0002084902,0.01040699,0.01133885,0.0006903281,0.01142467,0.007618411,0.2565181],"study_design_scores_gemma":[0.002560167,0.004591452,0.4858947,0.1099735,0.2720604,0.0006113741,0.01032659,0.02513041,0.01072872,0.02488847,0.05249285,0.0007413338],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.5100587,0.2615255,0.1467985,0.01704636,0.002019162,0.0161768,0.01752905,0.0006176547,0.02822839],"genre_scores_gemma":[0.9473459,0.008280066,0.03860728,0.0005942407,0.0001534006,0.002871971,0.001640901,0.00006315855,0.0004429921],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.983909,"threshold_uncertainty_score":0.6105809,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.5441703005540053,"score_gpt":0.6713453947850856,"score_spread":0.1271750942310803,"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."}}