Abstract

When making use of MNO data, there exist many obstacles to achieving the quality that is necessary of official statistics. In particular, the National Statistical Office typically does not have access to the micro-level MNO data that can be linked to population registers or other relevant sources, whereas the MNO aggregates may be subject to measurement errors, population domain misclassifications, device duplication noises, user ambiguity and target population coverage errors. Few official statistics based on MNO data exist today, despite the great attention they have received over the last decade or more. The methodology development of ESSnet MNO-MINDS project is aimed to guide and facilitate the integration of MNO data with other non-MNO data to produce regular official statistics, including a methodology reference frame which provides a common basis for examining all the methods relevant to utilising MNO data. When it comes to the associated uncertainty, we generally distinguish whether the basis of inference is a known sampling design or an assumed statistical model and, for model-based methods, whether the assumed model is about the target-agnostic observation mechanism or specific outcome variables, to be referred to as randomisation, quasi-randomisation and super-population modelling, respectively. Several methodological strands have been studied in the context of some of the most relevant methodological applications, which had been identified through a landscaping exercise and external consultations, such as transfer learning, statistical calibration and network flow modelling. Both the methodological applications and the applicable methods to them will be outlined, highlighted and illustrated.