Coordinator

• Dr Pawlowsky-Glahn, V.  (UdG)

Researchers

Dr Barceló-Vidal, C.  (UdG)
Dr Daunis-i-Estadella, J.  (UdG) 
Dr Martín-Fernández, J.A.  (UdG)
Dr Mateu-Figueras, G.  (UdG)
Dr Thió-Henestrosa, S.  (UdG)

The core of the compositional data analysis group of Girona University lies in researching statistical methods and models for data that can only take values in the simplex or other constrained sample spaces such as the positive real line, the interval (0.1), the positive quadrant of the plane, or the unit square, to name but a few. The application of these methods in the sphere of medical science is very common, due to the frequency with which researchers are faced in their everyday practice (as regards observation, diagnosis, prognosis and treatment of clinical cases) with data that are either strictly positive or that represent parts of a whole, like parts per one, percentages, parts per million, or similar. All the constrained sample spaces mentioned have one problem in common: classical statistical methods may present difficulties or inconsistencies. Thus, it is not unusual to obtain predictions outside the sample space, such as negative values when studying strictly positive phenomena. Neither is it strange to see incoherent results when two scientists are working with proportions but with one of them using a proportions vector with more components than the other. Recent results show a solution to these problems based on a new approach. This approach is equivalent to using in the constrained sample space operations, metric and measure different from the usual Euclidean metric and Lebesgue measure. The application of usual statistical methods by means of this new approach solves the incoherences and difficulties mentioned in the classical methodology, in general making the results easier to interpret. 

 
In addition to the central subject matter that gives the research group cohesion, their activity also includes other fields of investigation in classic multivariate statistics. Succinctly, the researchers of the group work in methods of automatic classification (cluster analysis), both parametric and non-parametric; in dimension reduction methods through factorial analysis techniques; and in techniques useful for the design and treatment of questionaires.