||For mapping football (soccer) player information by using multidimensional
scaling, and for clustering football players, we construct a distance
measure based on players’ performance data. The variables are of mixed type,
but the main focus of this paper is how count variables are treated when deﬁning
a proper distance measure between players (e.g., top and lower level variables).
The distance construction involves four steps: 1) representation , 2) transformation,
3) standardisation, 4) variable weighting. Several distance measures are
discussed in terms of how well they match the interpretation of distance and
similarity in the application of interest, with a focus on comparing Aitchison and
Manhattan distance for variables giving percentage compositions. Preliminary
outcomes of multidimensional scaling and clustering are shown.