[R] Measures of association for ordinal data
Christof Bigler
christof.bigler at colorado.edu
Tue Sep 14 01:47:06 CEST 2004
In a classification problem with ordinal data (classes 1 - 4), I used
multidimensional optimization to maximize gamma (Goodman's measure of
association) between observations and predictions. This resulted in the
following frequency table (rows = observations, columns = predictions):
1 2 3 4 sum
1 16993 1 4854 27 21875
2 1308 0 1691 120 3119
3 1427 1 4587 434 6449
4 289 0 1965 593 2847
sum 20017 2 13097 1174
Predictions for class 1 are fairly good, however, classes 2 and 4 are
underrepresented and class 3 is overrepresented, as shown by the
marginal sums.
Is there any measure of association implemented in R that takes into
account the prevalence of certain classes?
Thanks for your help!
Christof
More information about the R-help
mailing list