METHODOLOGICAL NOTE

A basic review of the Mitofsky-Waksberg sampling methodology.

Since the late-1970's, the Mitofsky-Waksberg sampling procedure has been the preferred telephone sampling methodology for governmental and social science survey research. This was an outgrowth of the limitations inherent in proprietary RDD methods being used prior to that time. More importantly, this preference continued due to basic problems with most major commercial sampling suppliers' methodologies.

First generation proprietary RDD methodologies utilized two-stage designs in order to minimize costs associated with sample frame development. A typical design utilized a 1st Stage sample of telephone exchanges selected with equal probabilities. The portions of those sample exchanges, typically "hundred series" banks, that did not serve residences were then identified; this was a time consuming, expensive, and not altogether reliable process. The 2nd Stage sampling procedure involved the generation of an equal number of four-digit suffixes within each exchange; those numbers which corresponded to the previously identified non-residential series or banks were eliminated. These methods were clearly epsem, and the expected distribution of residential numbers was proportional to the actual residential assignments within the 1st Stage sample of telephone exchanges.

However, the potential non-coverage bias of these RDD samples was large because the 1st Stage exchanges and their associated "non-working" hundred series' banks became quickly outdated. As one can appreciate, the process of developing, maintaining and updating such an RDD system was prohibitively expensive. Consequently, only the largest research companies could afford to invest the money and staff time required to maintain their own proprietary RDD system.

An alternative RDD methodology was eventually proposed by Joseph Waksberg of WESTAT, and it is now known as the Mitofsky-Waksberg (M-W) RDD procedure. The two-stage M-W procedure utilizes 1st Stage sampling units (PSUs) selected PPS, while 2nd Stage sampling requires an equal sample taken from each PSU. The essential brilliance of this method was that by simply reversing the probabilities of selection, one eliminated the substantial investments required to secure the critical working bank information. The M-W method answered criticisms relating to non-coverage and the reliability of working bank information. Most importantly, it provided a universally accepted, theoretically superior RDD methodology that any research firm could implement.

However, the "no free lunch" maxim is evident when one actually utilizes a M-W design. The practical problems of implementation, and their theoretical implications are well documented in the literature. These difficulties have engineered a number of approximations, lessening the original sampling structures at both stages.

Although GENESYS is recognized as the industry's only source for a single stage epsem RDD sampling methodology, our wide range of M-W capabilities are less well known. The flexibility of the GENESYS system provides the user with the ability to fulfill both standard M-W design requirements as well as more complex strategies using variable primary unit selection probabilities, or dual frame single stage epsem/M-W mixed-method designs. Although GENESYS has begun to reverse this fifteen year trend of "efficiency-first" sampling, client needs continue to demand a wide range of M-W capabilities. The inherent flexibility of the GENESYS system, in combination with our internal sampling and database resources has enabled us to routinely adapt procedures to the myriad of M-W variations encountered.

Some of the M-W options available through GENESYS are described below:

Sample Frame Definition. Geographic options for known "residential" telephone exchanges are quite extensive, from FIPS Codes to use of GENESYS ZIP code-to-exchange and Census Tract-to-exchange correspondence files. For "non-residential" exchanges, geographic correspondence is currently limited to standard BELLCORE locational references, but additional geographic coding is being under taken to provide more precise service area descriptions.

Stratification. Exchange-level, or 1st Stage Unit stratification schemes can utilize any of the standard GENESYS geo-demographic codes/estimates, nominal exchange-types, and/or measures of size(MOS). For example, exchanges could be stratified by various status codes (known, or unknown residential service, regency of establishment, etc.), or independent vs. Bell. Exchanges/1st Stage Units can be stratified by density, number of listed households, household income levels, racial/ethnic characteristics, etc.

1st Stage Unit Definition. Complete flexibility in defining the size of 1st Stage units: hundred-series, two-hundred, thousand-series, etc.

1st Stage Sample Selection. Selection options include unrestricted SRS (with or without replacement), nth, SRS within fixed intervals, etc. Within known "residential" exchanges, PPS selection schemes can utilize a variety of variables including listed households, estimates of total households, and exchange-level estimates of demographic variables(i.e., race/ethnicity, household income, etc.)

Mixed-RDD Methodologies. Combinations of M-W and single stage epsem RDD methodologies can be specified to provide more efficient data collection protocols. For example, "residential" strata could utilize a single stage epsem RDD procedure, while low density residential exchange strata can be sampled more efficiently with an M-W procedure.

Either 1st or 2nd Stage sample telephone numbers can utilize GENESYS-ID, a minimally intrusive process developed to identify business numbers and non-working/disconnected telephone numbers.

GENESYS Sampling Systems can also provide longitudinal frame definition/update services necessary for maintaining tracking study consistency.

Although the above options are currently available only on a custom basis, the next generation of GENESYS will incorporate the above capabilities through a new Mitofsky-Waksberg module. This module will enable the user to replicate virtually any current methodology while providing significant design flexibility, including mixed-method applications.