Nontelephone Household

Telephone surveys became an acceptable mode of data collection in the United States in the 1970s, when approximately 90% of households in the United States had a telephone. According to the 2000 Census, 98% of U.S. households contained a telephone. However, the 2000 Census did not distinguish between wireline and wireless service, so having a “telephone” could mean having a landline phone, a wireless phone, or both. In each year since the 2000 Census, more and more households began to substitute wireless telephone service for their wireline or landline telephone service. This phenomenon, often referred to as “cutting the cord,” has introduced additional coverage bias in traditional wireline random-digit dial (RDD) samples, as in 2008 approximately 20% of U.S. households had only a cell phone.

By 1986, only a little more than 7% of households were without a telephone. Analysis of National Health Interview Survey (NHIS) data by Owen Thornberry and James Massey in 1988 showed that certain socio-demographic cohorts were disproportionately represented among the nontelephone population, particularly families and persons living in poverty. If information on low income, or any correlates of low income, was an important objective of a telephone survey, researchers were encouraged not to use a telephone sample for data collection. As the number of households without a telephone declined into the 1990s, concerns about bias diminished but did not disappear.

In 1995, Scott Keeter proposed a method for minimizing the bias associated with the exclusion of households without a telephone. Keeter showed that interruption of telephone service is usually episodic in nature. Based on this observation, he proposed using survey data collected from respondents reporting interruption of telephone service to make adjustments for the nontelephone population.

Based on the 2006 NHIS data and data from Media-mark's 2006 national syndicated survey, only 2% of households have no phone of any kind. Findings from the 2007 NHIS survey show 15% cell phone only and another 10% as being “cell phone mostly” (i.e. they have a landline and a cell phone but essentially do not use their landline for incoming calls). This shift in telephone ownership away from landline telephones means that a traditional RDD sample today will represent only 87% of all households. In 2006, E. Deborah Jay and Mark DeCamillo experimented with a Keeter-like approach for adjusting for cell phone only households. In their study, they asked about interruption of telephone service and whether the household had cell phone service during that interruption. As with the Keeter study, the demographics of recent cell phone only households were similar to published demographics of cell phone only households. The authors proposed using data collected from respondents that reported a recent cell phone only status to adjust estimates for noncoverage of the cell phone only population.

Including households with no telephone service of any kind is not a viable option for RDD surveys. However, cell phone only households can be included in telephone surveys. Frames exist for sampling cell phone numbers, and there has been a significant amount of ongoing research related to sampling cell phones both in the United States and around the world. Many methodological issues remain to be solved, and in the United States sampling cell phones is uniquely complicated by Telephone Consumer Protection Act (TCPA) regulations that prohibit dialing a cell phone with an autodialer.

...