If a subset of training points is used, the prior probability is correct
in the Set Analysis Parameters window, but in the report and the response
theme the prior probability (and therefore the posterior probabilities) is
incorrect. The prior for the whole training set is used instead of the
prior for the subset, so the prior is too large.
One side effect of this error is that the conditional independence ratio
will be in error, because the predicted number of deposits (= sum of area *
posterior prob over all polygons) will be larger than it should be.
The map patterns of the response theme will be unaffected in a relative
sense. The map of the sum of the weights is unaffected, and the map of
posterior logits is too large by a contant amount .
|
Create a new training point file as a subset before running
Arc-SDM. |