Technical Characteristics and Features
Sampling Frame of Working Residential Blocks
The RDD Express sampling frame is comprised of working residential hundred blocks. This has been the de facto standard in telephone sampling for about 25 years. A hundred block is a set of 100 potential telephone numbers defined by Area Code, Exchange, and the next two digits. A working residential block is a hundred block that contains some minimum number of residential directory listings.
The motivation for using working residential blocks is yield. It turns out that about half of hundred blocks in working exchanges do not contain any residential listings. A few of these “empty” blocks do contain some unlisted residential households, but most are entirely devoted to business, government, or educational institutions, or have never been assigned at all. Wholesale elimination of these blocks has obvious yield benefits, with virtually no loss in real coverage.
RDD Express offers a choice of two listings thresholds -- 3+ and 15+. The 3+ threshold corresponds to Survey Sampling’s Super A (most of the apparent 1 and 2 listing blocks result from errors in directory compilation rather than the presence of real listed households). The 15+ threshold is our version of a higher yield sample. It doesn’t quite match Survey Sampling Super B yields, but has better statistical properties (equal probability selection). As a matter of information, Arbitron uses a threshold of 10+.
Geographic and Demographic Assignments
Each Exchange is assigned to a Zip Code and county based on the plurality location of listed residential numbers in the Exchange. Each Exchange is also assigned median household income and Black, Hispanic, and Asian household incidence, based on updated Census information.
Data Sources and Updating
All information outlined above comes from Experian (formerly Metromail), and is updated annually. Between annual updates area code splits are handled as they arise, using information from Bellcore.
RDD Express users always have the latest version of the system.
Sample selection is straightforward.
The user specifies:
Geography. This may be defined by Area Code/Exchange, Zip Code, County, DMA, Met, State, or National, or combinations thereof.
Demographic filters, if any (e.g., Median HH Income $50K+). Over, under, and range filters are available for HH Income, Black, Hispanic, and Asian, and may be used in combination.
Total sample size and number of replicates.
RDD Express then:
Selects the Exchanges that are in the specified geography and meet the demographic filter.
Sorts/stratifies the working blocks by State, County, Zip Code, Area Code/Exchange, Block.
Makes a simple n’th selection of working blocks.
Appends a two-digit random number to the end of each selected block, to form complete telephone numbers. In the event a block is selected twice (possible if the total sample size is more than the number of blocks), the first selection is assigned a random number between 00 and 49 and the second is assigned a random number between 50 and 99. The same idea is extended if a block is selected 3, 4, .. times.
Samples are output to ASCII files. In addition to telephone number, each record contains replicate, time-zone, and all geographic identifiers.
In addition to the basics outlined above, RDD Express has some features that make it more powerful and easier to use. These include:
Storage of geographic/demographic specifications, so they can be used again later.
Storage of prior sample output and sample de-duping.
Incidence/coverage reports, to aid in deciding demographic filters.
Sample tabulations by State, County, Zip Code, or Area Code/Exchange.
Supplemental data files providing Zip/Exchange crossover information and geographic/demographic data.
Comments on Zip Code Assignments
County geographic assignments are very accurate, because Exchanges rarely cross county boundaries. This is not the case for Zip Code assignments, because the typical Exchange contains listings in 4 or 5 Zip Codes. Consequences depend on the type of sample:
For sample geographies defined by counties or larger entities, incorrect Zip Code assignments have no effect on sample coverage or selection. All the numbers that genuinely belong in the sample area are included and have appropriate chance of selection. However, Zip Code labels attached to sample numbers may be incorrect. If exact Zip Code is important to end study purposes, ask respondents for actual Zip Codes.
For sample geographies defined by a large number of geographically adjacent Zip Codes (e.g., a city), errors in Zip Code assignment usually have no more effect than for larger, county-based geographies, because Zip Code crossovers generally occur within the Zip Codes in the sample area. Again, however, if exact Zip Code is important, put it in the questionnaire.
For sample geographies defined by a few or non-adjacent Zip Codes, errors in Zip Code assignments can be an important issue. When sample Zip Codes happen to be the plurality location of many Exchanges, coverage is fine but incidence may be low. When sample Zip Codes happen not to be the plurality location of the Exchanges that serve them, coverage can be a problem. In fact, about 1/3 of all Zip Codes are not the plurality location for any Exchange, meaning no Exchanges are assigned to them. RDD Express offers two ways to handle this issue. One is the supplemental file of listing counts for all Zip/Exchange combinations, which users may examine and use however they see fit. The second is a built-in RDD Express feature. Whenever a Zip Code is specified to which no Exchange is assigned, the system reports a) whether or not the Zip Code contains any residential listings, and if so, b) the Exchanges that contain residential listings in that Zip Code, along with the Zip Codes to which those Exchanges are assigned (along with corresponding listing counts).
The issues outlined above are not unique to RDD Express. They are present in all RDD sampling systems and arise because postal and telephone geographies do not coincide.