In recent years, large-scale statistical organizations have directed substantial attention toward the integration of data from sample surveys and from non-survey sources like administrative or commercial records. For some cases, this requires one to focus sample-design work on providing information for (i) sub-populations or (ii) specific variables that are not adequately covered by the non-survey sources.
This paper considers adaptive forms of sample designs intended to address issues (i) and (ii). Three topics receive principal attention:
(1) Clarification of the goals of the adaptive version of the design, including (a) improved data quality; (b) mitigation of the risks of degradation, or outright loss, of one or more non-survey sources; and (c) cost management
(2) Paradata that may inform the resulting adaptive design
(3) Approximate (conditional) bias and variance properties of the resulting estimators
The proposed concepts and methods are illustrated with two examples. The first example centers on adaptive extensions of standard multiple-frame designs. The second example involves adaptive selection of units to provide supplementary information on key variables.
|