Allstat Users,
To quantify the amount of information in a RxC table iteratively built
up by reducing either rows or columns I used Cramer's V coefficient as
measure. This leads to local maxima instead of a global maximum. The RxC
table has some empty cells. It seems that all chisquare-based measures will
behave like V coefficient. On the other hand, empty cells make use of
log-likelihood=2SUM(OijlogOij/Eij) (Eij where is computed from the
marginal frequencies) returns undefinite values.
What statistic(s) should I be considering for optimization? Does it
exist other measure(s) someone might suggest?
Thanks,
Ray Haraf.
|