Large-Sample Rank Distance Critical Values in the Single Variable Two-Group Case.
Rept. for May-Nov 72,
SCHOOL OF AEROSPACE MEDICINE BROOKS AFB TEX
Pagination or Media Count:
The basic technique of using rank distance for classifying was suggested by Zimmerman in 1970. A method for accepting or rejecting a proposed discriminator variable, based on the number of misclassifications of a training set by this technique, was described by these authors in 1971. Tables of null distribution were given for the two-group case up to 12 observations per group in the training set. A supplementary algorithm was later described which gives useful critical values up to about 25 observations and some smaller significance values up to 40 observations. The paper uses the Monte Carlo technique to extend the previous tables up to 250 observations after comparing the Monte Carlo results against exact calculations. An approximate formula is derived for various significance levels which provides interpolation and will permit some extrapolation above 250. Author
- Statistics and Probability