When one tries to sequence a large number of candidates  
from the voting data by using DEA model, he/she often 
observes the fact that  it is extremely difficult for 
low-vote gainers to take full advantage of DEA  model. 
In addition, one encounters a problem of sequencing 
a huge number of  candidates on a scale of `1'. One can 
overcome such problems, however, by  sequencing candidates 
in each cluster with DEA model after breaking down 
the candidates into clusters. In this paper, the authors 
make two proposals as  measures as such. One is to stratify 
the candidates in advance so as to make clear the sequence 
among the clusters. The other is to seek the alignment 
between clusters rank order relations and within cluster 
rank order relations of the candidates.