enumerating patterns in a set of contingency tables

Question

I have a set of 3X3 contingency tables and want to identify the distribution of the patterns by comparing their expected and observed number of samples in each cell using a right statistical test. For example, for a given table such as the following,

             A  B  C 
  high       2  8  3
  middle     0 10  3
  low        3  3  5

I want to transform them into a tri-state discretized table.

             A  B  C
  high       0  0  0
  middle    -1  0  0
  low        1 -1  0

Here, 1 is assigned to the cells where the observed number of samples is significantly larger than expected, -1 is assigned to the cells where the observed number of samples is significantly less than expected, and 0 is assigned when the observed and expected numbers are not significantly different. I wonder what is the right statistical test for this problem.