The master sample concept describes the process of enumerating all sample units within a finite sampling frame in random order to support flexible probabilistic environmental surveys across broad geographic extents. The concept was first proposed in the mid-twentieth century for agricultural purposes and has re-emerged with the development of spatially-balanced randomization algorithms as a framework for facilitating collaborative monitoring among multiple partners. A fundamental property of the spatially-balanced master sample is the exhaustive random ordering of all sample units such that any ordered subset of units remains spatially-balanced with known inclusion probabilities. This permits partners to scale their level of effort (i.e., subsample size) within their jurisdictional boundaries commensurate to available resources, yet still contribute to a larger statistically-valid probability sample for regional or continental analyses and reporting. The master sample concept provides solutions to a number of frequently-encountered real-world problems and has been adopted by regional and continental-scale monitoring programs for aquatic resources and bats. The concept has tremendous potential for ecological research and monitoring conducted across broad geographic scales to assess impacts of global change but adoption has been slowed by the lack of conceptual development and practical guidance. Using an example from the North American Bat Monitoring Program, we outline concept fundamentals, establish terminology, and identify challenges, pitfalls, and recommendations. We explore different implementation scenarios, comparing spatial-balance properties and environmental gradient coverages among independent-by-jurisdiction simple random samples, independent spatially-balanced samples, and master sample jurisdictional subset strategies.
Results/Conclusions
Key terminology includes those that help practitioners distinguish between sampling frame(s), sample units, and the master sample itself. In our application, practitioners are often confused by these fundamentals. Using Ripley’s k function to test for spatial balance, we found considerable flexibility among scenarios considered. However, coverage of environmental gradients of interest across the study region varied widely among scenarios depending on proportionality of jurisdictional representation, revealing how unrepresentativeness and bias can creep in to some kinds of sample design scenarios. An important emergent property of coherent master sample implementation is the identification of sample units shared among jurisdictions that reduces redundancy and facilitates collaboration and economy of scale. Efficient data management practices that enable tracking of master sample subsets and implementation details are crucial for inclusion probabilities and survey design weights to be calculated. Future needs include web-based tools that support multiple partners to communicate about master sample implementation details and firmer statistical guidance on incorporating design weights in model-based inferences.