Visual Analytics for Survey Statistics: Interactive Benchmarking and Calibration

• Activity Number: 678
• Type: Topic Contributed
• Date/Time: Thursday, August 8, 2013 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistical Learning and Data Mining
• Author(s): Matthew R Williams*+
• Companies: USDA - National Agricultural Statistics Service
• Keywords: Official Statistics ; Nonlinear Constraints ; User Feedback

Abstract

We motivate a need for Interactive Visual Analytics in survey and official statistics through applications in Benchmarking/Calibration. In particular, we discuss a framework for constraining estimates from the survey process to meet known targets (i.e. aggregated state values must reproduce published national values). Fully automated procedures provide wholesale adjustments which minimize measures of discrepancy between the original estimates and their final adjusted versions. However, an analyst may have additional knowledge of a subset of the estimates based on the survey design and quality measures of the survey process. We demonstrate methods to combine these approaches and discuss implications for implementation. In this context, Visual Analytics takes the form of simple tables and heat maps (stylized tables as figures) which the analyst can easily interpret. Analysts provide feedback in the form of basic decisions/judgments about the current estimates, which are used by the automated procedure to update the tables and displays.