Variance estimation for population attributable risk in a complex cross-sectional animal health survey

Abstract

Population attributable risk estimates offer a method of combining information on population exposure and disease risk factors into a single measure. Univariate and multivariable methods exist for calculating point estimates and variances under the assumption of equal sampling probabilities. National Animal Health Monitoring System national studies typically use a complex survey design (where selection probabilities vary by design strata), which makes use of these methods of calculating variance inappropriate. We suggest the use of a method called "delete-a-group" jackknife to estimate the variance of population attributable risk when a complex survey design has been implemented. We demonstrate the method using an example of Johne's disease. Advantages of the "delete-a-group" jackknife method include simplicity of implementation and flexibility to estimate variance for any point estimate of interest.