Date of this Version
Published (as Chapter 2) in Experimental Methods in Survey Research: Techniques that Combine Random Sampling with Random Assignment, First Edition. Edited by Paul J. Lavrakas, Michael W. Traugott, Courtney Kennedy, Allyson L. Holbrook ,Edith D. de Leeuw, and Brady T. West. John Wiley & Sons, Inc. 2019.
Probability samples are necessary for making statistical inferences to the general population (Baker et al. 2013). Some countries (e.g. Sweden) have population registers from which to randomly select samples of adults. The U.S. and many other countries, however, do not have population registers. Instead, researchers (i) select a probability sample of households from lists of areas, addresses, or telephone numbers and (ii) select an adult within these sampled households. The process by which individuals are selected from sampled households to obtain a probability-based sample of individuals is called within-household (or within-unit) selection (Gaziano 2005).Within-household selection aims to provide each member of a sampled household with a known, nonzero chance of being selected for the survey (Gaziano 2005; Lavrakas 2008). Thus, it helps to ensure that the sample represents the target population rather than only those most willing and available to participate and, as such, reduces total survey error (TSE).
In interviewer-administered surveys, trained interviewers can implement a prespecified within-household selection procedure, making the selection process relatively straightforward. In self-administered surveys, within-household selection is more challenging because households must carry out the selection task themselves. This can lead to errors in the selection process or nonresponse, resulting in too many or too few of certain types of people in the data (e.g. typically too many female, highly educated, older, and white respondents), and may also lead to biased estimates for other items. We expect the smallest biases in estimates for items that do not differ across household members (e.g. political views, household income) and the largest biases for items that do differ across household members (e.g. household division of labor).
In this chapter, we review recent literature on within-household selection across survey modes, identify the methodological requirements of studying within-household selection methods experimentally, provide an example of an experiment designed to improve the quality of selecting an adult within a household in mail surveys, and summarize current implications for survey practice regarding within-household selection. We focus on selection of one adult out of all possible adults in a household; screening households for members who have particular characteristics has additional complications (e.g. Tourangeau et al. 2012; Brick et al. 2016; Brick et al. 2011), although designing experimental studies for screening follows the same principles.