Using Response Rates to Adjust a Dual Sample Design

• Activity Number: 654
• Type: Contributed
• Date/Time: Thursday, August 7, 2014 : 10:30 AM to 12:20 PM
• Sponsor: Survey Research Methods Section
• Author(s): Eric Grau*+
• Companies: Mathematica Policy Research
• Keywords: dual sample design ; response rate

Abstract

When face-to-face interviews are required in a probability sample, sampling statisticians often construct a clustered sample to minimize the cost of data collection. In this type of sample design, the primary sampling units are geographic clusters, allowing interviewers to visit areas that are close together, reducing traveling costs. However, in some subpopulations, the number of cases in the primary sampling units is sometimes too small to provide enough sample cases for analysis. In several projects at Mathematica, we have resolved this issue by implementing a dual sample design, where the sample design consists of a clustered component (with face-to-face nonresponse follow-up) and an unclustered component (with no face-to-face follow-up). In this paper, we discuss a sample where budget realities during data collection made the original sample design too costly to implement. In this case, the proportion of the design within each component had to be constantly adjusted during data collection. We discuss how data collection reports were used to anticipate final response rates within various subpopulations, and how the sample design was adjusted accordingly. [presnt only Tu,We,Th]