{"id":"W2118339612","doi":"10.1111/j.1524-4733.2007.00255.x","title":"Reliability of Two Instruments for Critical Assessment of Economic Evaluations","year":2007,"lang":"en","type":"article","venue":"Value in Health","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":39,"is_retracted":false,"has_abstract":false,"ca_institutions":"University of Calgary","funders":"","keywords":"Reliability (semiconductor); Intraclass correlation; Generalizability theory; Cronbach's alpha; Quality (philosophy); Inter-rater reliability; Variance (accounting); Statistics; Psychology; Scale (ratio); Rating scale; Mathematics; Psychometrics; Economics; Accounting; Geography","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false,"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":[0.1952437,0.001888634,0.002505064,0.01002448,0.001066112,0.004575664,0.001998902,0.003718685,0.002260925],"category_scores_gemma":[0.6823907,0.0007686818,0.002868513,0.006286803,0.004247869,0.00433128,0.003718358,0.003039147,0.0007524567],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002165393,"about_ca_system_score_gemma":0.002688671,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001248721,"about_ca_topic_score_gemma":0.0006258115,"domain_scores_codex":[0.757185,0.1596522,0.0269929,0.01222949,0.04145887,0.002481458],"domain_scores_gemma":[0.1161111,0.7795309,0.03289165,0.02898589,0.04114212,0.001338375],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.009473047,0.00118329,0.658531,0.001505628,0.004541744,0.0002707784,0.006018016,0.01839134,0.002005479,0.05132501,0.00732753,0.2394271],"study_design_scores_gemma":[0.001894271,0.006822984,0.6558572,0.001828099,0.003215743,0.001241186,0.002398864,0.1784449,0.01080967,0.1169747,0.01974702,0.0007654489],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.5733418,0.005480348,0.3886752,0.00137729,0.001097234,0.003746723,0.004384007,0.0005831622,0.02131422],"genre_scores_gemma":[0.9431546,0.0004325199,0.05149316,0.0002133339,0.0001602326,0.00247371,0.001548951,0.00008713357,0.0004363833],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.8047563,"threshold_uncertainty_score":0.9924078,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.4349332589155737,"score_gpt":0.5530560586814062,"score_spread":0.1181227997658325,"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."}}