loglin on zero-diagonal contingency tables

["F. Lee De Cola" <ldecola@usgs.gov>]

I am analyzing a 'map' with 1024 'countries' each of which has one of 7
'colors.'

The following table gives the frequencies of the 1626 'neighbor' pairs.

         water forest agric woodwet emergwet urban bare
   water     0    110    32      16       26    17    2
  forest     0      0   516     130       30   291  135
   agric     0      0     0      41       12    78   39
 woodwet     0      0     0       0       11    42   16
emergwet     0      0     0       0        0     5    1
   urban     0      0     0       0        0     0   76
    bare     0      0     0       0        0     0    0


I know that loglin() will analyze these kinds of pair frequencies, but I
need the model to be constrained so that:
1. there is symmetry among the color pairs - hence the upper-triangular
grouping, and
2. no 2 countries can have the same color - hence the zeros in the
diagonal.

I seek a model that will provide residuals based on the marginal totals.

Reference(s) will be sufficient.

Lee De Cola, PhD, MCP
U.S. Geological Survey
521 National Center
Reston VA 20192 USA
Room 2D324
703-648-4178, fax 4603
W77°22'10" N38°56'48"
703-709-6972 home
LDECOLA@USGS.GOV
ldecola.home.comcast.net