72 – Developing Statistical Methods in Setting Environmental Exposure Limits
Indirect Sampling in Case of Asymmetrical Link Structures
Torsten Harms
ISS Hamburg Int. Business School
Indirect sampling extends the application of sampling theory for settings when the sampling frame and target population are not identical, but rather connected through specified links. This approach has found widespread application, particularly through the generalized weight share method (GWSM) and can be applied generally to sample from networks.
In order to ensure unbiased estimates, the inbound links for any sampled unit in the target population must however be known. While this is not a problem in many types of link structures (such as family relationships or household association), it prevents important applications in networks with asymmetrical link structures, such the world-wide web or lose social relationships.
We will extend the application of the generalized weight share method (GWSM) to yield design-unbiased estimators even for such situations, where the inbound links for each element are generally unknown. This is achieved by adequately reflecting the link-structure which is partially revealed during the sampling process. An empirical study as well as a comparison with other estimation methods, particularly Monte-Carlo approaches, evaluates the proposed method.