{"id":"W2520223558","doi":"10.3138/cjpe.261","title":"Data Quality Evaluation for Program Evaluators","year":2016,"lang":"en","type":"article","venue":"Canadian Journal of Program Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":16,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Quality (philosophy); Psychology; Computer science; Philosophy; Epistemology","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":true,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.6795222,0.0009339792,0.002921546,0.009125547,0.004275409,0.01544404,0.004632197,0.003588237,0.005775798],"category_scores_gemma":[0.7974698,0.0007566596,0.002605198,0.008960254,0.009793296,0.008244547,0.008508016,0.008663053,0.001259327],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.01893633,"about_ca_system_score_gemma":0.0617254,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.007094366,"about_ca_topic_score_gemma":0.005277685,"domain_scores_codex":[0.196534,0.6873311,0.05487468,0.004523768,0.05486862,0.001867829],"domain_scores_gemma":[0.06681586,0.630734,0.0431126,0.05556329,0.2005682,0.003206094],"domain_codex":"methods","domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.001325415,0.0004344393,0.02831305,0.01415756,0.001479989,0.0001296372,0.01045431,0.003428326,0.000956168,0.1946225,0.04936895,0.6953297],"study_design_scores_gemma":[0.001867632,0.003011792,0.039499,0.0896488,0.002757988,0.0006423272,0.01342699,0.02620366,0.01528404,0.3671069,0.4398507,0.0007001094],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.02695448,0.02873369,0.6636242,0.1970764,0.005218356,0.01778427,0.001840223,0.0009831153,0.05778528],"genre_scores_gemma":[0.3844564,0.01022684,0.5555944,0.02128855,0.001861102,0.0202911,0.001034833,0.0004729253,0.004773924],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.6795222,"threshold_uncertainty_score":0.3952062,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.8261497658997451,"score_gpt":0.6791874459859031,"score_spread":0.146962319913842,"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."}}