{"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":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.1198193,0.0002241139,0.0004295542,0.0008411665,0.0003592836,0.0005976603,0.001735116,0.0001551483,0.004151989],"category_scores_gemma":[0.03614303,0.000138128,0.0002207526,0.0009224303,0.000188008,0.002083612,0.00005046154,0.0001535728,0.0001334366],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0009769184,"about_ca_system_score_gemma":0.0158549,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0002005005,"about_ca_topic_score_gemma":0.02048876,"domain_scores_codex":[0.9880323,0.002183068,0.002190892,0.0006101159,0.006436285,0.000547372],"domain_scores_gemma":[0.9813163,0.001315263,0.00180972,0.001387477,0.0134303,0.000740936],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.0000463233,0.00008894667,0.01420033,0.000005934851,0.00004606166,5.248787e-7,0.000198097,0.0001756776,0.00008467055,0.0004989415,0.01385101,0.9708035],"study_design_scores_gemma":[0.006468705,0.001899504,0.1511644,0.0001561122,0.0005924234,0.00003147067,0.0009698402,0.2121554,0.0002008224,0.05728446,0.5685998,0.0004771123],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.898655,0.00245797,0.03073147,0.02914176,0.007745378,0.02452703,0.0007696941,0.0001118354,0.005859864],"genre_scores_gemma":[0.9789172,0.00001572125,0.01894486,0.0001911933,0.0006103023,0.0008631262,0.0001633669,0.00002493939,0.000269304],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9703264,"threshold_uncertainty_score":0.9973848,"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."}}