Editor's note: Linda Piekarski is vice president, database and research at Survey Sampling International, Fairfield, Conn.

The accompanying table shows the correct way to calculate the WPR for a sample. It is important to note that B is the number of sample records dialed not the number of dialings.

Working phone rates can be adversely affected by a number of factors, singly or in combination:

  • sample design;
  • sample geography;
  • listed rates;
  • population density;
  • mobility rates;
  • number of businesses;
  • telephone numbering plan changes;
  • number demand;
  • number assignment policies;
  • advances in telephony; and
  • length of time between the drawing of the sample and the fielding of the sample. (SSI recommends that a sample not be kept unused for more than three months.)

WPR will vary according to sampling methodology. A sample of directory-listed phone numbers will usually have a significantly higher WPR than a random-digit-dial (RDD) sample but be less representative, while an EPSEM or random A RDD sample will have a lower WPR but be more representative.

Average WPR by sample type:

Listed - 80%

Random B (unscreened) – 52% (range 40% to 65% by state)

Random A (unscreened) – 39% (range 27% to 48% by state)

A non-U.S. sample should have a WPR similar to a U.S.  random B sample, but will vary by country.

For directory-listed samples, the WPR is primarily impacted by the mobility of the population. In the U.S., 18 percent of households move in a year (the average life span of a telephone directory) meaning that an average of 18 percent of directory-listed numbers might have been disconnected by the time a sample is selected in an area where a new directory is about to be published. Higher mobility rates in urban areas can mean higher disconnect rates for listed samples from urban areas while lower mobility rates in rural areas can mean lower disconnect rates for listed sample.

For RDD samples, the WPR will be significantly lower than for directory-listed samples. RDD samples improve coverage of telephone households over directory-listed samples by including numbers that are not found in directories (ex-directory) but are in 100-blocks that contain directory-listed numbers. However, many of these RDD numbers that are not listed will be either non-working or ineligible, resulting in a lower WPR.

The WPR for an RDD sample is directly proportional to the number of residences present in the frame of possible RDD numbers. Over the past decade, in an effort to meet the demand for more telephone numbers, telephone companies in the U.S.  and around the world have modified their numbering systems. The introduction of new area codes, overlays and exchange partitioning in the U.S., and the standardization of telephone number lengths in many countries have negatively affected working phone rates by substantially increasing the pool of possible RDD numbers in RDD frames.

Since 1995, the U.S. RDD frame has grown 44 percent while the number of telephone households has only increased 17 percent. The probability of an EPSEM RDD number being a working residential number has declined an average of 1 percent per year from over 51 percent in 1995 to only 39 percent in 2006. To use a favorite analogy, the number of fish in our pond has not changed much over the years but the pond has doubled in size. It is simply getting more difficult to find our fish.

After a decade of expanding telephone numbering frames, the resulting decline in RDD WPR in the U.S.  and around the world has been exacerbated by recent declines in the demands for telephone numbers. More and more households and businesses are replacing their multiple lines with a single broadband connection while still others are giving up their landline numbers altogether, relying solely on wireless phones. Although some of these numbers may be recycled to new subscribers, many remain unused or not-in-service.

Adverse effect

Local number portability (LNP) has also had an adverse effect on sample efficiencies. LNP is the ability of a subscriber to change his telephone service provider within a particular rate center without changing his telephone number. By December 2004, almost 31 million landline numbers had been ported to another landline provider. Porting requires two 10-digit numbers for each telephone line. One is the original subscriber number and the other is the number associated with the switch belonging to the new carrier to which the call is connected. In most cases today, the new numbers - sometimes referred to as ghost numbers - do not connect if dialed, but their presence in RDD frames contributes to the overall decline in WPR.

The way in which local telephone companies assign telephone numbers within their allotted prefixes may also affect WPR. In areas where number conservation (1,000-block pooling) has not been mandated, random assignment of numbers is more common and working blocks in these prefixes will have fewer working numbers than in the past.

Small area samples may have unusually low or unusually high working phone rates. Urban areas generally tend to have a lower WPR because they have higher mobility rates, more renters, more businesses and more wireless-only households. Listed samples in rural areas tend to have a higher WPR because they have lower mobility rates, fewer privacy concerns and fewer businesses. However, RDD samples in rural areas may have a lower WPR because there are fewer new listings or unlisted numbers to be found among RDD numbers that are not listed numbers. Suburban communities, with their high growth rates and high unlisted rates, generally have a higher WPR.

Using a file of 12.5 million business telephone numbers, SSI identifies and removes known business numbers from its listed and RDD databases. However, most businesses have multiple voice lines (rollover lines, direct inward dial lines) which are not listed in directories or business files. This means that RDD samples will always contain unlisted business numbers as well as unlisted fax and modem numbers. The percent of unidentified business numbers in the sample is naturally higher in urban areas and lower in rural areas.

Screening can improve the WPR of a sample by identifying about 50 percent of the non-working and modem/fax numbers in a sample. Clients should expect to find roughly the same number of disconnects in their delivered sample as was removed by a screener. The more bad numbers identified by the screener, the more bad numbers will remain in the sample. For example:

 Sample 1,000 1,000
 Screened out        150 (15%) 300 (30%)
 Delivered 850   700
 Remaining bad 150 (17.6%) 300 (42.9)
 Sample WPR 70% 40%

SSI has monitored this situation for more than a decade. We will continue to do so and periodically report our findings in order to help researchers set realistic expectations and allocate fieldwork resources appropriately.