05-01-2013 10:24 AM
Proc Univariate provides two-sided p-value for wilcoxon signed rank test. If n>20, can I divide the two-sided p-value by 2 to get one-sided p-value? Does SAS do normal approximation of signed rank statistics when n > 20? Thanks.
05-01-2013 10:45 AM
In Common Statistical Methods for Clinical Research with SAS Examples, Third Edition (pg. 258), it states, "SAS uses the t-approximation to the Wilcoxon signed-rank test if n is greater than 20. For n <=20, SAS computes the exact probability." As for getting the 1-sided p-value from a 2-sided p-value, you should be able to simply divide by 2 as you suggested.
05-01-2013 01:00 PM
Be careful in the direction. The one-sided p-value is either P/2 or 1-(P/2), where P is the printed two-sided value. If the estimated statistic (or median, or mean, or whatever) is in the opposite direction of the alternative hypothesis, then the one-sided p-value will be a large number [1-(P/2)].