``Re-make/Re-model'': Should big data change the modelling paradigm in official statistics?

Abstract

Big data offers many opportunities for official statistics: for example increased resolution, better timeliness, and new statistical outputs. But there are also many challenges: uncontrolled changes in sources that threaten continuity, lack of identifiers that impedes linking to population frames, and data that refers only indirectly to phenomena of statistical interest. We discuss two approaches to deal with these challenges and opportunities. First, we may accept big data for what they are: an imperfect, yet timely, indicator of phenomena in society. These data exist and that's why they are interesting. Secondly, we may extend this approach by explicit modelling. New methods like machine-learning techniques can be considered alongside more traditional methods like Bayesian techniques. National statistical institutes have always been reluctant to use models, apart from specific cases like small-area estimates. Based on the experience at Statistics Netherlands we argue that NSIs should not be afraid to use models, provided that their use is documented and made transparent to users. Moreover, the primary purpose of an NSI is to describe society; we should refrain from making forecasts. The models used should therefore rely on actually observed data and they should be validated extensively.

The Validity of Google Trends Search Volumes for Behavioral Forecasting of National Suicide Rates in Ireland

“…Nonetheless, the use of internet sources for statistical purposes should be used with caution. Selection bias is a predominant issue due to the uneven Internet penetration among and within countries, the population covered by these sources is also subject to daily changes, and often there is difficulty in linking the data to other datasets [31,32,33]. In detail, for search queries, one must also be wary of several factors such as changes to the search algorithm [34] and media events which lead to an unexpected behavior [35].…”

The Evolution of Data Quality: Understanding the Transdisciplinary Origins of Data Quality Concepts and Approaches

“…The goal is to extrapolate this decision construct to using sources of data that were collected for other reasons and repurposing them for new research purposes. For example, how would one adopt concepts such as fitness-for-use, relative measures of data quality, and timeliness (Agafitei et al 2015, Braaksma & Zeelenberg 2015, Couper 2013? One could weigh the tradeoffs of lower quality data with increased timeliness in the short run and the reverse in the longer run.…”

“…Current conditions require a wider set of data quality dimensions, including privacy, security, and complexity (UNECE 2015). In addition, the notion that data should be viewed in terms of potential new trade-offs (timeliness versus representativeness), as increasing efficiency when combining data sources, and as potentially generating new data products (Braaksma & Zeelenberg 2015) casts data quality as relational among all data sources. The use of repositories may help to ensure the quality and accessibility of these data to advance social and behavioral research (Petrakos et al 2014).…”

Does Big Data Change the Privacy Landscape? A Review of the Issues

“…These nonstatistically designed sources of data are intoxicating in that they provide easily accessible and often inexpensive information about individuals, businesses, and society. They offer possibilities for studying behavior and social drivers of population attributes and characteristics at a finer level of geographic and demographic resolution and in more frequent time intervals than do survey and census data (Braaksma & Zeelenberg 2015). They hold the promise of understanding human interactions at a societal scale, within a context of rich spatial and temporal dynamics, and for detecting complex interactions and nonlinearities among variables (Agafiţei et al 2015).…”