3.1 Introduction

Researchers who study dolphins can’t ask them questions. So, dolphin researchers are forced to study behavior. Researchers who study humans, on the other hand, should take advantage of the fact that our participants can talk. Asking people questions has been an important part of social research for a long time, and the digital age both enables and requires certain changes in survey research. Despite the pessimism that some survey researchers currently feel, I expect that the digital age is going to be a golden age of survey research.

The history of survey research can be roughly divided into three overlapping eras, separated by two contested transitions (Groves 2011; Converse 1987). Right now we are in a period of transition between the second and third eras, but the first and the second eras—as well as the transition between them—provide insight into the future of survey research.

During the first era of survey research, roughly 1930 - 1960, developments in scientific sampling and questionnaire design gradually resulted in a modern understanding of survey research. The first era of survey research was characterized by area probability sampling and face-to-face interviews.

Then, a technological development—the widespread diffusion of landline phones in wealthy countries—eventually led to the second era of survey research. This second era, roughly from 1960 - 2000, was characterized by random digit dialing (RDD) probability sampling and telephone interviews. The change from the first era to the second era resulted in major increases in efficiency and decreases in cost. Many researchers regard this second era as the golden age of survey research.

Now, another technological development—the digital age—will eventually bring us to a third era of survey research. This transition is being driven by both push and pull factors. In part, researchers are being forced to change because approaches from the second era are breaking down in the digital age (Meyer, Mok, and Sullivan 2015). For example, more and more homes do not have landline telephones and non-response rates—respondents who are sampled but do not participate in surveys—have been increasing (Council 2013). Simultaneous with this breakdown of second-era approaches to sampling and interviewing, there is increasing availability of big data sources (see Chapter 2) that appears to threaten to replace surveys. In addition to these push factors, there are also pull factors: the third era approaches offer incredible opportunities, as I’ll show in this chapter. Although things are not totally settled yet, I expect that the third era of survey research will be characterized by non-probability sampling and computer-administered interviews. Further, although the earlier eras were characterized by their approaches to sampling and interviewing, I expect that the third era of survey research will also be characterized by the linkage of surveys with big data sources (Table 3.1).

Table 3.1: Three eras of survey research. This chapter will focus on the third era of survey research: non-probability sampling, computer-administered interviews, and surveys linked to other data.
Time Sampling Interviewing Data environment
First era 1930 - 1960 Area probability sampling Face-to-face Stand-alone surveys
Second era 1960 - 2000 Random digit dialing (RDD) probability sampling Telephone Stand-alone surveys
Third era 2000 - present Non-probability sampling Computer-administered Surveys linked to other data

The transition between the second and third eras of survey research has not been completely smooth, and there have been fierce debates about how researchers should proceed. Looking back on the transition between the first and second eras, I think there is one key insight for us now: the beginning is not the end. That is, initially many second-era methods were ad-hoc and did not work very well. But, through hard work, researchers solved these problems, and second-era approaches eventually were better than first-era approaches. For example, researchers had been doing telephone random digit dialing for many years before Mitofsky and Waksberg developed a random digit dialing sampling method that had good practical and theoretical properties (Waksberg 1978; Brick and Tucker 2007). Thus, we should not confuse the current state of third-era approaches with their ultimate outcomes. The history of survey research makes clear that the field evolves, driven by changes in technology and society. There is no way to stop that evolution. Rather, we should embrace it, while continuing to draw wisdom from earlier eras. In fact, I believe that the digital age will be the most exciting age yet for asking people questions.

The remainder of the chapter begins by arguing that big data sources will not replace surveys and that the abundance of data increases—not decreases—the value of surveys (Section 3.2). Given that motivation, I’ll summarize the total survey error framework (Section 3.3) that was developed during the first two eras of survey research. This framework enables us to understand new approaches to representation—in particular, non-probability samples (Section 3.4)—and new approaches to measurement—in particular, new ways of asking questions to respondents (Section 3.5). Finally, I’ll describe two research templates for linking survey data to big data sources (Section 3.6).