ABSTRACT
CHAPTER 7
7.1 Introduction
Nonresponse is a pervasive problem with sample survey data. For most surveys the nonresponse rate can be appreciable. Even for official government surveys, full response is rarely obtained by the time estimates are required. While good survey development, testing and data collection procedures, including following up initial nonrespondents, can be used, a substantial nonresponse rate in the final data set is normal. Nonresponse is a difficult problem in the sense that we often cannot analyze data with nonresponse without making strong assumptions about the nonresponse mechanism which are difficult to check. The situation is made even more unsatisfactory by the fact that these assumptions are usually chosen to produce the simplest analysis, effectively by making the effect of nonresponse as harmless as possible, and that the assumptions are often left implicit. For example, the simplest approach is to ignore the nonresponse and analyze the data for the responding units in the same way as the complete data would be analyzed. This approach implicitly makes very strong assumptions about the nonresponse mechanism. At the very least, analysts should consider carefully the possibility of different nonresponse mechanisms and make explicit the mechanisms considered in the analysis. In this chapter, we develop some models for sample survey data with nonresponse and highlight the additional complexity nonresponse adds to the analysis.
