Online Program

Pseudo-clustering for Combining Data Sets with Multiple Hierarchies (307841)

Keywords: Pseudo-clustering, hierarchical structure, data integration, complex sampling design

National health care surveys collect data using complex sampling designs for the purpose of reducing time and cost or enhancing the efficiency of the data collecting procedure. However, it may be advantageous to allow different sampling designs to be used for different segments of the population. When working with survey data collected under multiple sampling designs such that the resulting data have multiple distinct hierarchical structures, it is not straightforward to integrate the analysis. We propose an approach considering pseudo-clustering of observational units including pseudo-clusters of singletons into multi-level models in order to handle such data. This study is motivated by the National Survey of Healthcare Organizations and Systems (NSHOS) which employed three levels of healthcare businesses: complex systems, simple systems, and independent practices. In the NSHOS, physician practices are the observational units in each, which are sampled using three different sampling designs depending on types of healthcare business. We show that the pseudo-clustering estimation approach performs well at integrating datasets and allowing inferences that combine all the data.