Considering the important and timely work by Fahimi et al. and Boyle et al. on the issue of residential numbers in zero listed telephone banks, there are, I believe, three key take-a-ways from this exchange.
First, science proceeds through a series of tests and retests and a single study does not a definitive conclusion make. While not a novel observation, the iterative approach to hypothesis testing is one of the oldest precepts of science, it does from time-to-time help for us to be reminded of this. Advancement of our understanding in critical and changing areas necessitates multiple points of inquiry. Where we stand now with this particular issue is at a vexing crossroads, the direction of which has major implications for a large segment of the survey research community. This is one time when “more research is needed” is not a simple tagline at the end of an article, but a call for real and necessary action.
Second, it reminds us that some of the foundations upon which we build our industry are not ours, but borrowed from others. In this instance I’m referring to the use of the U.S. telephone system database as the critical frame upon which many of our studies are built. We sometimes forget that this seemingly simple system of number assignment is not structured with surveys in mind and numbers are not allocated in ways to necessarily make it easier for us as survey researchers. Rather structural changes in the telecommunication system are occurring at an increasing rate to meet the demand for numbers and as a result of, as Fahimi et al. note, the breakup of the larger telecommunications system into a series of smaller, yet interconnected networks. This lesson is one we should heed not just for telephone sampling, but for many of the newer approaches being championed as well, including online strategies and address-based approaches. It is crucial that we maintain a critical eye on the origins and characteristics of our sampling frames to ensure we understand the limits of these sources.
Third, the studies highlight the continued trouble in which our industry finds itself with respect to landline-based random digit dialed (RDD) surveys (as an aside, it is important to separate out the concept of “telephone surveys” from “RDD landline surveys” as the former will continue to be with us for quite some time as a critical component of multimode and alternative survey designs, while the latter’s days may be numbered). This is a discussion which cannot be had without including the implications of cell phone only (as well as cell phone mostly) households – to do so misses much of the real problem. When we consider the cell phone issue in the context of the zero banks debate we are left with one of two unpleasant alternatives: if Boyle et al. are correct, then the situation for landline RDD is bad; if Fahimi et al. are correct the situation is horrible, indeed untenable. That’s not to say there are no RDD solutions to be had. Fahimi et al. propose some potential approaches for the landline portion of the sample, which combined with an RDD-based cell phone sampling could provide a solution (a costly and inefficient one, but a solution nonetheless). The notion, however, that landline-based, list-assisted RDD is a viable solution in terms of coverage error without placing that in the context of all residential households, which includes cell phone only homes, significantly misses the mark and may lead astray those who don’t consider, as Paul Harvey might put it, “the rest of the story.”