Low and declining response rates in household surveys provide the potential for nonresponse bias, but assessing the nonresponse bias is a challenging problem when characteristics of nonrespondents are unknown. For example, for random-digit-dial (RDD) surveys, information about nonrespondents may be limited to just the telephone number. To address this issue, the census geocoding (CG) method has sometimes been used. This method appends aggregate census information for small geographic units (for example, census blocks, block groups, or tracts) to nonresponding cases located in those units. These census data can then be used to evaluate nonresponse bias or adjust for it. Little is known about the effectiveness of the CG approach, but it is often used in nonresponse evaluations, particularly for RDD surveys. While the CG method is an important tool for recovering information from nonrespondents, it relies on the degree to which its assumptions are met. This research project will investigate the efficacy of the CG method using RDD and face-to-face survey data and will develop guidelines for implementing it so as to maximize its effectiveness for use in household surveys. One of the goals of the proposed project is to investigate the minimum amount of information required to apply the CG approach effectively. As an example, if only a telephone number is known, census aggregate data closer to the telephone exchange-level must be used rather than at the block-level. Finding a method that can be used effectively to evaluate and correct for nonresponse bias in household surveys will be an important contribution to survey research.
Much of social science relies on probability-based surveys, which in turn rely on the ability to obtain survey measures from all sample members. Nonresponse threatens probability-based inference. This project will advance the knowledge about methods for evaluating and adjusting for nonresponse bias in surveys. The project will not only evaluate the CG method for use in telephone surveys but for virtually any survey where the CG method can be used. This includes any survey where information on nonrespondents is limited to an address, ZIP code, or telephone number and very little else. Many household surveys use address frames where all that is known for a frame unit is the household family name, the address and, in some cases, a telephone number. The CG approach can be applied in these situations to evaluate the bias due to nonresponse. The research is supported by the Methodology, Measurement, and Statistics Program and a consortium of federal statistical agencies as part of a joint activity to support research on survey and statistical methodology.
