Sampling and Intervention Designs

点击翻译按钮获取中文摘要

The proposed research develops new methods for sampling and intervention designs for hidden and hard-to-access populations. These include ecological populations of animals and plants and human populations that are hard to find by ordinary survey methods. Populations of animals and fish, for example, can be highly mobile, camouflaged, and spatially uneven in patterns that change over time. Hard-to-access human populations include those that are hidden, mobile, and stigmatized. Members of such populations may be reachable only by following social network links from one member to another. The proposed research develops new network, spatial, and temporal sampling designs and interventions strategies for populations that have spatial, temporal, or network structure. Network sampling is necessary for surveys of hidden human populations, for understanding organizational dynamics as well as for understanding online networks and the behaviours of people who use those networks. Adaptive spatial sampling designs are needed for unevenly distributed natural populations. In many cases also, adaptive and network intervention strategies are more effective than conventional strategies for mitigating harms or bringing benefits to human populations, organizations, and populations of plants and animals. The proposed research continues a long-term effort of the investigator that has resulted in effective new methods for all of these types of situations. In recent work I introduced simple new estimators for network sampling of hidden and hard-to-access human populations. Empirical simulations show the new estimators greatly improved upon the estimators currently in widespread use. For instance, for estimating mean number of partners in the hidden population, the new methods almost totally eliminated the bias and reduced mean sqare error to between one-twenty-sixt to one-eightieth that of the currently most widely used estimator. The proposed research will extend the methods of the new estimators and make high-performance software available under open access. The key technique underlying the new network sampling estimators is to run a fast network sampling process on the sample network data. The inclusion frequencies of the fast sampling process provide estimates of the unequal inclusion probabilities for the real-world network sampling design by which the sample was selected. These estimates of the inclusion probabilities are then used in generalized unequal probability estimators to estimate population quantities. Confidence intervals are obtained by a simplified linearization method. The proposed research will build on these new results to find new and improved methods also for spatial temporal sampling and inference and for making interventions to benefit hard-to-access, at-risk, and under-served populations.