What do the ``late'' cases tell us? Evidence from the 1998 Survey of Consumer Finances

Article type: Research Article

Authors: Kennickell, Arthur B.

Affiliations: Board of Governors of the Federal Reserve System (Ret.), Washington, DC 20551, USA. E-mail: [email protected]

Abstract: As is the case in most other government surveys, the Survey of Consumer Finances (SCF) makes strenuous efforts to maintain response rates. These efforts are quite costly. The argument for pursuing the relatively difficult ``late'' cases is two-fold: First, sample size is important for more efficient estimation. Second, there is an implicit assumption that higher response rates lessen the possibilities of bias. The former argument is straightforward, but the latter is less so. This paper investigates the information contained in the later observations of the SCF. The data presented here suggest that there are differences in some of the economic and other characteristics of respondents and nonrespondents, and that these differences are present in a weaker form in the contrast between the cases that are early and those that are late. However, for general purposes, the differences between the early and late cases are not dependably strong. Because nonresponse is affected both by respondent-specific characteristics and management decisions about the deployment of interviewers, it unlikely that we can make substantial progress without separating these effects.

Keywords: Unit nonresponse, paradata

DOI: 10.3233/SJI-160302

Journal: Statistical Journal of the IAOS, vol. 33, no. 1, pp. 81-92, 2017