Alternative Methods of Unit Nonresponse Weighting Adjustments: An Application from the 2003 Survey of Small Business Finances

摘要

In business surveys, as in all surveys, firms selected to participate often fail to respond to the survey. When left unaccounted for, this unit nonresponse can lead to biased estimates (both relative and absolute), inappropriate variances in the weights, as well as invalid confidence statements. However, there is no single accepted method to account for unit nonresponse. Whenever nonresponse adjustments are made, there is a risk of introducing further or different problems to the data. The 2003 Survey of Small Business Finances (SSBF) screening interview had significant unit nonresponse and therefore some type of nonresponse adjustment was deemed necessary. The approach used in the 2003 survey differed from that used in previous surveys. The current paper uses the data from the 2003 SSBF to investigate the impact of various nonresponse adjustment methods on the weights, point estimates and variance of the estimates by comparing the approach ultimately implemented for the 2003 survey to alternative approaches, including one similar to what was done in the 1998 SSBF. The remainder of this paper is organized as follows. We begin in Section 2 with a brief description of the SSBF, followed in Section 3 by a discussion of different nonresponse adjustment methods, highlighting the general advantages and the disadvantages of each method. We provide a detailed description of the four methods examined in this study in Section 4, explaining how each was applied to the 2003 SSBF data. We also describe the evaluation methodology. Section 5 provides the results of the four methods using the 2003 SSBF screener data and Section 6 concludes.