Total Survey Error in Practice

4The Role of Statistical Disclosure Limitation in Total Survey Error

Alan F. Karr

Center of Excellence for Complex Data Analysis, RTI International, Research Triangle Park, NC, USA

4.1 Introduction

The thesis of this chapter is that statistical disclosure limitation (SDL) ought to be viewed as an integral component of total survey error (TSE). SDL is the one step in the survey process where error is introduced deliberately, for the sake of protecting respondent privacy and dataset confidentiality. As such, SDL interacts—often subtly and not always positively—with other error reduction steps, especially editing and imputation. To date, however, there has been a disconnect. SDL error is, together with sampling error, the only controllable form of survey error, but this property has never been exploited. Typically, SDL as implemented does not account for other sources of survey error. We argue and demonstrate here that to the extent that SDL‐introduced error resembles other forms of error—especially measurement error—both risk and utility objectives are furthered when SDL is cognizant of errors and steps to reduce them. Conversely, the many and costly efforts to reduce survey error do not account for the fact that SDL will be performed.

Figure 4.1 encapsulates the argument. There, SDL is a discrete step lying between data processing (in particular, editing and imputation) and data release. Oversimplifying, what is the point of expending resources to correct other errors, when ...