Skip to main content

Advertisement

Table 3 Prior and posterior estimates for cluster sampling coverage survey and SQUEAC conjugate Bayesian analysis

From: Estimating program coverage in the treatment of severe acute malnutrition: a comparative analysis of the validity and operational feasibility of two methods

  Final prior coverage estimate Likelihood coverage estimate Posterior coverage estimate Strength of evidence for conflictd
Scenario for prior estimation   Required sample size No. SAM cases (No. villages)    
Cluster sampling coverage survey 96 (46) 25.7% (17.6–33.7%)
Standard uncertainty in prior estimation (prior ±25%)
 Scenario 1a: broad program implementationa 55% 63 (30) 25.5% (15.4–34.6%) 34.7% (26.3–43.9%) Strong (p = 0.0033)
 Scenario 2a: basic program implementationb 52% 63 (31) 25.5% (16.0–34.4%) 33.8% (25.8–42.8%) Strong (p = 0.0076)
 Scenario 3a: external Implementationc 42% 62 (30) 25.4% (15.2–34.5%) 30.3% (22.5–39.6%) Weak (p = 0.1165)
High uncertainty in prior estimation (prior ±35%)
 Scenario 1b: broad program Implementationa 55% 80 (39) 25.6% (18.2–31.6%) 30.0% (22.3–39.3%) Strong (p = 0.021)
 Scenario 2b: basic program Implementationb 52% 81 (39) 25.6% (18.2–31.6%) 29.6% (21.9–38.8%) Strong (p = 0.0369)
 Scenario 3b: external Implementationc 42% 79 (38) 25.7% (17.9–32.1%) 28.2% (20.8–37.3%) Weak (p = 0.2582)
  1. aScenario 1: broad program implementation is the mean of six prior estimates (all with the exception of weighting and histogram provided by the external support team)
  2. bScenario 2: basic program implementation is the mean of four prior estimates including simple scoring, product of program performance, histogram of belief and previous SQUEAC coverage estimate
  3. cScenario 3: external implementation is the mean of two prior estimates including weighted scoring and histogram of belief by external support team
  4. dThe Z test is employed to test the null hypothesis of no conflict between the prior and likelihood coverage estimates