New Techniques and Technologies for Statistics Brussels, March 2017 Satellite session on Multi-indicator systems and partially ordered sets Views from Official Statistics Emanuele Baldacci *, European Commission, Eurostat emanuele.baldacci@ec.europa.eu @emibaldacci Dario Buono*, European Commission, Eurostat dario.buono@ec.europa.eu @darbuo *The views expressed are the author’s alone and do not necessarily correspond to those of the corresponding organisations of affiliation
Triggers for Socio-economic evaluation Data-users call for condensed information Allow multidimensional phenomena to be synthesized. Need to ensure time and regional comparisons
Dashboards proliferation Pro: Separation of Statistics from Economic Reading Cons: not always straightforward
Dimensionality Reduction Indicator systems might leads to the need of computing composite indicators and introduce some degree of arbitrariness concerning: The analytical approach The choice of the weights The aggregation technique
POSET@work This is a non-parametric approach for ranking multi-indicator data sets Poset theory can help to overcome the conceptual and computational drawbacks of the standard aggregative procedures by exploiting the relational structure of the data, so as to compute evaluation scores in purely ordinal terms.
Features for Official Statistics Based on some Eurostat pilot projects Poset permits to deal with complex multi-dimensional systems of indicators Poset allows to derive rankings and synthetic indicators, without variable aggregations Poset have already been applied in a variety of areas (Refugees' relocation in the EU, European opinions on services, Formal Concept Analysis, Fiscal policies)
Drawbacks for Official Statistics Based on some Eurostat pilot projects Poset does not have a temporal dimension (all variables refer to a fixed point in time) Poset is computationally complex and tools are currently under development
POSET@Eurostat (1) Paper on "Complementing scoreboards with composite indicators: the new business cycle clock", by Mazzi (Eurona 2015). 4 possible applications of POSET to the PEEIs (with mixed results) detect the presence of cross-sectional effects in financial markets () explain the economic phases () explain the country ranking with another variable (V) measure the diffusion of a crisis (?) The results obtained until now are still preliminary and not very conclusive. What has emerged is that the third application is the easiest to implement, even if the results could be quite obvious, so that its added value will be relatively low. By contrast, the first, second and fourth applications appear to be more challenging due to the fact that some quite complex hypotheses have to be formulated but their informational content from analysts' point of view is expected to be relatively high.
POSET@Eurostat (2) Applications of POSETs in macro-economics, by Ruggeri,Mazzi,Fattore(NTTS 2017) Case of Regional Competitiveness Index in IT set of triples; regions are ranked; use of "benchmarking profiles" The interesting point: no synthetic indicators, incomparability remains but the “distance” between two elements could be quantified Use of the R package, PARSEC (Partial Orders in Socio-Economics), for implementing the procedure for multidimensional evaluation.
Lessons learned The effectiveness of the partial order approach is particularly evident in the way the weighting and the aggregation problem is addressed and solved with an objective approach.
Challenges ahead Computational issues New skills with deep learning curve Communication and dissemination of methods and approach
Next Steps Call for scaled-up partnership between official statisticians, researchers and analysts Further research, in order to improve the efficiency of algorithms to avoid too heavy numerical computations Education of users about the possibilities and limitations of using indicators including communication of detailed information on the underlying assumption and selection method Lisbon Memorandum, (DGINS Sept 2015)
Thanks!