December 5, 2019
More accurate solution for cross-border internet purchases
How can cross-border internet purchases be accurately estimated? Researchers Quinten Meertens, Cees Diks, Jaap van den Herik and Frank Takes of the Leiden Institute of Advanced Computer Science (LIACS), Statistics Netherlands (CBS) and the University of Amsterdam (UvA) present their solution to this problem with an animation.
Leiden University, the University of Amsterdam and Statistics Netherlands have collaborated to develop a new method to estimate cross-border internet purchases, within the EU, more accurately. The method improves upon earlier methods by leveraging new data sources as well as advanced data analytics.
The paper indicates a general statistical bias of machine learning algorithms that has until now been overlooked by most of the machine learning community. The estimates produced by the new method are six times as high as earlier estimates. The method has been implemented, in adjusted form, at Statistics Netherlands, where it is used each quarter to report on cross-border internet purchases within the EU.
The article belonging to this research was published in the Journal of the Royal Statistical Society.