Take a look at the above output from PROC CORRESP. This table is partial contributions to inertia for the column dimensions. This output is part of a work where the CORRESP is asked to fit a 9 dimensions to a data set which was 70 rows by 15 columns (actual maximum data space dimension is 15 but we are looking for reduced 9 dimensions)However the approach is applicable for row observations (note the difference that i am assigning the word dimension here)
The rules to assign are the following:
(1) high light all the high inertia (loadings) for each of the column until the cumulative sum just passes 0.8 (the default total interia explained by that dimension).
(2) Start with Dimension 1. The first highlighted cell is from row “Child Presence”. Look for maximum among high lighted values in that row. Here the maximum is occurring in dimension4. So all the high lighted values in that row is given 4. The rest are assigned dimension 0. In the next high lighted value of first column is in row – Region, and all the high lighted values in that row is assigned to dimension(coordinate) 1, where the maxium of the row occurred. Next charity index, which gets 8, and finally political index gets 4 and all the high lighted columns in that row get 4; the rest of the columns get 0. Follow this for every column, once a row in a column has been assigned by the previous steps, skip that row in that column.
The completed assignment is given below.
From Data Monster & Insight Monster