{"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":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.06133523,0.0001421073,0.001112097,0.0004787694,0.00009399807,0.00001097394,0.0002305278,0.0001039532,0.0001514785],"category_scores_gemma":[0.005262034,0.0001917476,0.0001411838,0.0001682382,0.0001288691,0.0002713705,0.00003876455,0.0001483927,0.00004176686],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001949524,"about_ca_system_score_gemma":0.001072231,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.008833193,"about_ca_topic_score_gemma":0.0003731543,"domain_scores_codex":[0.9909812,0.0004660563,0.007427881,0.0004995378,0.0001173823,0.0005079364],"domain_scores_gemma":[0.9927536,0.004069351,0.002324839,0.0005521892,0.0001090983,0.0001909168],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"observational","study_design_scores_codex":[0.00001962648,0.0002805237,0.2658592,0.001489366,0.00001644961,5.641043e-8,0.0006780443,0.01083273,0.000005868279,0.7199103,0.000499824,0.0004079679],"study_design_scores_gemma":[0.003083578,0.0005953484,0.462592,0.00036547,0.00000801054,0.000001928868,0.0007665263,0.2086834,0.0000581636,0.3208207,0.002698845,0.0003259576],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.929493,0.0005324246,0.04982258,0.01382287,0.001190118,0.002139197,0.0007682553,0.00002108343,0.002210502],"genre_scores_gemma":[0.9365428,0.00005292171,0.06209567,0.001027322,0.0001199383,0.00008670872,0.00002697503,0.00002098167,0.00002669021],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.3990896,"threshold_uncertainty_score":0.9977671,"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."}}