Abstract:
|
Integration of surveys with alternative data sources (sometimes described as "big data" or "organic data") requires evaluation of the bias and variance properties of each prospective source. In many such cases, directly computed variance estimators may be unstable, and use of estimated variance functions may be preferred. This paper extends previous literature on generalized variance functions (GVFs) for complex sample surveys to: (1) develop variance-functions estimators intended to reflect dominant error components in a given set of surveys and alternative data sources; (2) evaluate the convergence and stability properties of the estimators from (1); and (3) present some diagnostics based on (2). The primary ideas in this paper are illustrated with examples based on (a) an establishment survey that depends heavily on an administrative record source; and (b) a sensitivity analysis for prospective linkage of a household survey with commercial or administrative sources.
|