Poisson sampling: A sampling strategy for concurrently establishing freedom from disease and estimating population characteristics Michael S. Williams a, *, Eric D. Ebel a,1 , Scott J. Wells b a Risk Assessment Division, Office of Public Health Science, Food Safety Inspection Service-USDA, 2150 Centre Avenue, Building D, Fort Collins, CO 80526, USA b Department of Veterinary Population Medicine, University of Minnesota, 1365 Gortner Avenue, Saint Paul, MN 55108, USA 1. Introduction In both wildlife populations and animal production systems, animals are commonly clustered into groups. Examples of these groups are a herd of beef cattle, a school of fish, a flock of birds, or a consignment of culled animals arriving at slaughter. For simplicity, the general term group will be used to describe any collection of animals. Samples are drawn from a group either to substantiate freedom from disease or to estimate characteristics of the group, such as the average weight or market value of animals or the distribution of genotypes. Inferences from the sample may be at the group level (e.g., a herd-accreditation scheme) or at a regional or national level (e.g., the estimation of population parameters, such as the percent of vaccinated animals in a country). In the case of regional or national level inferences, two-stage sampling is often employed; the first stage is selection of a sample of groups and the second stage is the sampling of animals from the group (Cameron and Baldock, 1998b). This study will focus on methods for sampling animals from a group, though the methods can be used for the selection of groups as well. Preventive Veterinary Medicine 89 (2009) 34–42 ARTICLE INFO Article history: Received 15 November 2007 Received in revised form 19 December 2008 Accepted 12 January 2009 Keywords: Improved efficiency Prevalence Targeted sampling Points Estimation ABSTRACT Surveys of animal populations are often designed to either demonstrate freedom from disease or to estimate parameters that describe the population, such as disease prevalence, proportion of vaccinated animals, or average animal weight and value. Targeted surveillance is a sampling approach where animals are selected for testing based on the presence of characteristics that indicate a higher probability of disease. This approach can substantially reduce the sample size that is required to demonstrate freedom from disease, but inferences about other population parameters are generally not possible because the sample design often lacks the properties required for making inferences in a traditional survey sample. Determining which animals to sample can also be difficult when either more than one characteristic exists or the characteristic is a continuous attribute, such as age or weight. Poisson sampling is an unequal probability sampling design that can provide efficiencies similar to targeted surveillance while allowing inferences for other population parameters. The adaptation of Poisson sampling to animal surveys is described. A simulation study, based on sampling a flock of sheep, is used to demonstrate the reductions in sample size that are possible with Poisson sampling. The study showed that the sample size required for a flock-level sensitivity of 0.95 when using Poisson sampling was less than half that required when using simple random sampling. The performance of estimators for prevalence of scrapie and distribution of genotypes are also compared. Published by Elsevier B.V. * Corresponding author. Tel.: +1 970 492 7189. E-mail addresses: mike.williams@fsis.usda.gov (M.S. Williams), eric.ebel@fsis.usda.gov (E.D. Ebel), Wells023@umn.edu (S.J. Wells). 1 Tel.: +1 970 492 7187. Contents lists available at ScienceDirect Preventive Veterinary Medicine journal homepage: www.elsevier.com/locate/prevetmed 0167-5877/$ – see front matter . Published by Elsevier B.V. doi:10.1016/j.prevetmed.2009.01.005