ESTIMATING TELEPHONE NONCOVERAGE BIAS WITH A TELEPHONE SURVEY

Nontelephone households are implicitly treated as a static population in discussions of sampling frame noncoverage. Yet telephone service is known to be episodic for many households, who may gain or lose service as their financial situation changes or when they move. Thus the population of telephone households at any given time includes households that were recently a part of the nontelephone population. These households may be used to characterize the nature of some noncoverage errors and even to estimate their magnitude. Using a panel constructed with the 1992-93 Current Population Survey, "transient" telephone households-those who gained or lost service over the year covered by the panel-are shown to comprise over half of the panel households reporting no telephone service in either the 1992 or 1993 surveys. These households are compared with the total nonphone population and found to be similar on a variety of key demographic characteristics. Several statewide Virginia telephone surveys are used to compare households reporting "intermittent" phone service with nontelephone households surveyed through in-person interviews. Households reporting intermittent telephone service were very similar to nontelephone households in terms of health insurance coverage and other variables known to be related to telephone status. By comparison with personal interviewing, telephone surveys provide a relatively economical means of data collection. However, telephone samples are subject to coverage errors, the most significant source of which is households that do not have telephone service. Depending on SCOTT KEETER iS professor of political science and public administration and director of the Commonwealth Poll in the Survey Research Laboratory at Virginia Commonwealth University. E-mail: skeeter@vcu.edu. The author wishes to thank Mike Brick, Jim Ellis, Kevin Fisher, Neil Henry, Pete Rikard, Santa Traugott, Joe Waksberg, Cliff Zukin, the Survey Research Laboratory staff, and this journal's referees for their valuable advice and assistance. Data utilized in this article were made available, in part, by the Interuniversity Consortium for Political and Social Research, which bears no responsibility for the analyses or interpretations presented here. Public Opinion Quarterly Volume 59:196-217 ? 1995 by the American Association for Public Opinion Research All rights reserved. 0033-362X/95/5902-0002$02.50 This content downloaded from 157.55.39.231 on Wed, 05 Oct 2016 04:36:02 UTC All use subject to http://about.jstor.org/terms Estimating Noncoverage Bias from a Phone Survey 197 the population of interest and the subject of the survey, the omission of nontelephone households can lead to seriously biased estimates. The problem is especially acute when characteristics of low-income households are the focus of the research. Nontelephone households are implicitly treated as a static population in discussions of sampling frame noncoverage. However, subscription to telephone service is known to be episodic for many households, who may gain or lose service when their financial situation changes or when they move. Because of this, the population of telephone households at any given time includes households that were recently a part of the nontelephone population. By comparing data from these "transient" telephone households with those who have experienced continuous telephone service, it may be possible to characterize the nature of some noncoverage errors and even to estimate their magnitude. In order to use transient telephone households found in telephone samples to estimate characteristics of the nontelephone population, two conditions must be met. First, transient households must be reasonably representative of the nontelephone population. Second, such households need to be numerous enough to provide reliable estimates. This article uses data from two national panel surveys to estimate the size of the transient telephone population and compare it with more chronic nonphone households, and it reports on experiments using statewide (Virginia) telephone surveys in which substantive data obtained from transient telephone households are compared with data on nonphone households gathered from personal interviews. The Nontelephone Population Much is known about the size and characteristics of the nonphone population in the United States. Trend data indicate that it is slowly shrinking, from 19.7 percent of all households in 1963 to 6.2 percent in 1994 (Federal Communications Commission 1995; Thornberry and Massey 1988, p. 29). In general, the decline was much steeper in the 1960s and early 1970s than in the years since, suggesting that further changes are likely to be modest. Telephone penetration rates across households depend on both household characteristics and external factors, the most important of which is the size of the installation charge (Belinfante 1992, p. 7; Hausman, Tardiff, and Belinfante 1993). The divestiture of AT & T and the subsequent regulatory uncertainty in the telecommunications field have occasioned considerable concern about the feasibility of the national goal of universal telephone service (see, e.g., Aufderheide 1987, p. 87). This content downloaded from 157.55.39.231 on Wed, 05 Oct 2016 04:36:02 UTC All use subject to http://about.jstor.org/terms