Assessing Variables Importance When Clustering Enterprises

The motivation of the study lies in the fact that many Struc- tural Business Statistics (SBSs) surveys must move from considering the Legal Unit (LU) as unit of interest towards considering the Enterprise (ENT) as such. Therefore, it may be necessary to study a different strat- ification of the sample for adapting and improving the usual sample design based on LU for facing this change. By applying K-prototype clustering algorithm, we were able to identify for several input dataset which variables influence the most the clustering partition. By consider- ing the most-influential variables in the new definition of the strata, a different stratification can be implemented with the aim of reducing the dimensional complexity and preserving the efficiency of the estimates.