ABSTRACT
Some have described the period spanning 1960–1980 as the “golden age of survey research”. Neyman’s sampling theory was universally accepted, response rates were high, data collection modes were few and highly standardized, costs were reasonable and some error sources were not yet well known. Also surveys and censuses were allowed to take time, sometimes years. Beginning around 1980, major new developments in survey design, methods and practice appeared. New technology gave us more advanced computers, software and new data collection modes. Since that time, changes have continued, quite dramatically in the past two decades.
We might talk about a changing survey landscape. Users want products that are more advanced than just estimates of parameters. They sometimes want interactive features associated with the products, more analytics and datasets that allow them to perform their own secondary analyses. There is increased demand for rich, timely data that are inexpensive. The changing landscape also includes new data sources such as types of big data, administrative data, paradata and other process data. For example, there are attempts at using probabilistic and algorithmic models more frequently to address information needs. Survey samples and methods often play a role in such modeling efforts – both model-assisted surveys and survey-assisted models. Combinations of probability and nonprobability sampling are being investigated. Artificial intelligence (AI) and machine learning are used to complement and enrich traditional survey work using actively or passively collected nonsurvey data. Alternatives to the dominant frequentist inferential paradigm used in survey statistics production, such as calibrated Bayes (CB), have been suggested. These alternatives explicitly model not only the observational data from multiple sources but also the selectivity of inclusion and other sources of error associated with such data.
In this chapter, we provide an overview of issues related to this new survey landscape, where a traditional survey can meet the information needs users have to a lesser extent than before and there will be increasing focus on statistical data integration that combines multiple sources of survey and other forms of data.
