In the era of “big data,” it can feel as though there is a persistent search for some previously untapped wealth of naturally occurring information that will replace the sometimes-expensive surveys.
Survey researchers believe that surveys can provide information that we cannot obtain in other ways—such as drug use in the U.S., new moms’ access to healthcare services for their newborns, and boating activity. Tom Smith’s point at BigSurv18 that we need to focus on high-quality surveys that tell us things we don’t already know was not new information for survey folks in the crowd—if anything, it was a relief that other data will not completely replace surveys.
Carefully designed data, a type of survey, in many cases is the best and only way to get the information we want. To paraphrase Bob Groves on “designed data,” the data in the survey world tend to be designed because we are asking people about themselves and their household for a specific purpose, driven by what the user wants to do with the data. Designed data may not be going anywhere; however, that does not mean we should disregard big data altogether. There are ways to take more naturally occurring big data and utilize their strengths to better access designed data.
BigSurv18 provided a learning opportunity beyond my normal fare and showed potential for mixing different methods. Often when attending professional conferences, we quickly find our way to the topics of interest and do not deviate much into uncharted territory. As a survey statistician, I am either at survey-specific talks at the annual Joint Statistical Meetings (JSM) conference or statistical methods talks at the American Association for Public Opinion Research (AAPOR) conference.
At BigSurv18, I attended presentations that I likely would not have seen at JSM and AAPOR. It’s refreshing to see the view from other perspectives—not just what can be done with machine learning and big data but the problem-solving strategies of researchers in different fields.
For example, I attended different presentations to learn about bandit algorithms for evaluating strategies in adaptive survey design and machine learning techniques to identify windmills in satellite imagery. I did not need to be sold on the idea that machine learning techniques can be used to cut down on menial and sometimes lengthy human undertakings—but after hearing about these kinds of approaches from computer science and data science researchers, I walked away from the conference thinking about how I can apply these methods to the survey work I do.
I’m inspired by the success of BigSurv18—not only because of the insights I gained from presentations but also because of what the future holds for this conference and others.