Zero-Number Banks

Telephone numbers in the United States consist of 10 digits: The first three are the area code; the next three are the prefix or exchange; the final four digits are the suffix or local number. The 10,000 possible numbers for a suffix can be subdivided into banks of consecutive numbers: 1,000-banks (Nnnn); 100-banks (NNnn); or 10-banks (NNNn). Zero-number banks, or zero-listed banks, are banks that do not contain directory-listed residential numbers. Although including zero-number banks in a random-digit dialing frame allows for 100% coverage of landline residential telephone numbers, their inclusion can substantially reduce sample efficiency.

Based on regular analyses conducted by Survey Sampling International, only 29% of 100-banks in POTS (Plain Old Telephone Service) prefixes, and other prefixes shared by different types of service, have at least one directory-listed number. In prefixes that have at least one directory-listed telephone number, only 47% of the possible 100-banks contain at least one directory-listed number. Because of these inefficiencies, most random-digit dialing surveys today use list-assisted, truncated frames—that is, frames truncated to include only listed-banks or those 100-banks that contain at least one directory-listed residential telephone number.

This truncation introduces some coverage error by excluding 100-banks that contain residential numbers but are missing from the truncated frame. The most common reason for this omission is that newly opened banks have not yet been published in a telephone directory. Because directories are published only once a year, the time lag between number assignment and directory publication can result in new blocks not being represented. Another source of error is common in rural areas, particularly those serviced by small, local telephone companies. A formal directory may not be readily available to compilers, but numbers are listed in a paperback book that looks like the local real estate listings available at the supermarket.

Alternative sample designs are available for researchers that opt to include zero-listed banks. One approach, the Mitofsky-Waksberg method, takes advantage of the tendency of telephone numbers to cluster in 100-banks. It starts with a sample of primary numbers in prefixes available for landline residential use. If a primary number is determined to be a working residential number, a cluster of additional numbers in generated in the same 100-bank. Another approach is to use disproportionate stratified samples of both zero-listed banks and listed banks. For many years, this design was the sampling protocol for all surveys conducted as part of the U.S. Department of Health Behavioral Risk Factor Surveillance System.

Research by J. Michael Brick and others suggests that this coverage error is 3–1% of telephone households. However, work by Brick and by Clyde Tucker and Jim Lepkowski confirms that the efficiency gains of list-assisted designs make them preferable in most cases. In fact, in 2003 the Behavioral Risk Factor Surveillance System protocol was revised to include only the listed-bank stratum.