I can only think of one situation in which the NAs might represent unknown but existing p-values. This would be when a large experiment has been conducted leading to many p-values. Instead of inputing all the p-values to the p.adjust() function, you decide to enter only the smallest p-values and represent the others using NAs. The trouble with this approach is that it can only be used with method="bonferroni".
Actually this could be done with "holm" as well as "bonferroni". Holm's method enforces monotonicity of the adjusted p-values by working up from the low end rather than working down from the high end, so just knowing the small ones is still ok.
Gordon
All the other adjustment methods are step-up or step-down methods or involve closure like Hommel's method. For these methods, you simply have to know all the p-values before you can adjust any of them.
______________________________________________ R-devel@stat.math.ethz.ch mailing list https://stat.ethz.ch/mailman/listinfo/r-devel
