For decades, sample surveys have provided valuable information in various fields. To survive, they must retain their advantages. A first advantage is that the surveys collect information that is generalizable from the measured sample to a well-defined population.
An important basis for the functioning of the surveys is their ability to recruit respondents. In this regard, surveys have adapted over the years to marked changes in living conditions, demographics, culture and communication technology.
Another reason that justifies the longevity of the surveys is the fact that, until recently, they did not have to face serious competition to get the attention and money of the sponsors.
Only the surveys could indeed provide a "data rich" perspective on the conditions of the populations they intended to describe.
However, the fundamentals on which the value of investigations are based is now weakened, as a decreasing number of potential respondents participate when invited. Less participation means more money to recruit large enough samples to meet the demands of the sponsors. It is clear that contact with potential respondents is more difficult than before. Technological (call screening) and physical barriers hinder attempts to persuade people to participate. If contacted, however, they are more likely to refuse. The circumstances of the participation have been the subject of much research and the secular tendency to the decrease in participation in the investigation remains a daunting problem.
Recruitment of respondents
The most radical idea is to abandon the probabilistic selection of the interviewees. Instead of building a list that (ideally) covers the population and uses a probabilistic method to select units from it, some survey organizations have recruited panels composed of interviewed volunteers, who participate in one or more surveys for some sort of remuneration. While participation in probability-based surveys has significantly diminished in recent years, a large industry, fueled by volunteer respondents, has grown at the same time with these non-probabilistic panels. Those who support the probabilistic sampling method observe that they have probability theory to support their inferences on the population, while the supporters of the volunteer groups are based on inductive models that have no comparable theoretical foundations.
Induction works until it does so, without explanation. Panel advocates stress that modeling works well enough for the needs of sponsors, and even better than probabilistic methods in some domains. They also note that, although the probability may be the basis of the project, the probabilistic samples are increasingly self-selected, as response rates continue to decline.
Mobile communication technology
Along with the recruitment of respondents, the future of the investigation will depend on their adaptation to changing communication technologies. The surveys have relied on media that are always present in the families and have selected individual respondents from those larger units.
As individuals now have increasing access to personal media, the link between the family and the individual for the purpose of investigation is being reduced. We now have significant portions of samples of telephone surveys made up of cell phone numbers. Meanwhile, the use of landline phones continues to decline.
Some studies say that interviewing people on mobile devices does not lead to inferior data quality compared to surveys completed on personal computers.
Other studies suggest that offering respondents a choice between the various survey modes offered by personal communication devices can improve the quality of survey data. The opportunities for digital communication are offered by the Internet, whose penetration has grown in recent years.
Union of data
The future of the investigation will probably involve a greater union of the data produced with information from other sources. The integration of survey and administrative information has a long methodological history. The purpose of some research was to see how well surveys produce data that matches information in the records. Survey reports are often considered bankrupt in this type of verification. This is credible in economic measurement, which has stimulated much of this work. (In other fields, for example the vote, the reports of the crimes, the presumption of infallibility of the registers can be less sustainable.)
Meyer, Mok and Sullivan (2015) have recently documented a growing division between estimates of the transfer of payments measured by self-assessment in surveys and administrative records, due to the decrease in the quality of survey data. They argue that a solution for decreasing the quality of survey estimates is to combine the information from the records into the survey results.
Investigations in a new context
As we have seen, the combination of survey data with other forms of information offers promising results and technical challenges. It is emphasized that it is necessary for companies to re-focus resources on evaluation and strategy to address problems that hinder combinations of survey data from other sources to produce statistics.
There are numerous and significant obstacles to the idea of the panel (access, representativeness, data quality and, in particular, privacy threats).
However, if the panel guidelines are followed, it is still likely that the future provenance of statistical information available to policy makers will involve survey data.
PETER V. MILLER