rank() vs SAS proc rank

<White.Denis <at> epamail.epa.gov> writes:

SAS proc rank has ties options of high and low that would allow
rank (c(1,1,2,2,2,2,3)) == 1 1 3 3 3 3 7
Could R support these ties.methods?

Don't know how SAS works but for your vector:

z <- c(1,1,2,2,2,2,3)
match(z,z)

[1] 1 1 3 3 3 3 7

SAS proc rank has ties options of high and low that would allow
producing ranks of the type found in the sports pages, e.g.,

rank (c(1,1,2,2,2,2,3)) == 1 1 3 3 3 3 7

Could R support these ties.methods?

Yes, it is possible to program them in R.  Here is one way:

1 + rowSums(outer(x, x, ">"))

which at least generalizes your single unexplained example.

Brian D. Ripley,                  ripley at stats.ox.ac.uk
Professor of Applied Statistics,  http://www.stats.ox.ac.uk/~ripley/
University of Oxford,             Tel:  +44 1865 272861 (self)
1 South Parks Road,                     +44 1865 272866 (PA)
Oxford OX1 3TG, UK                Fax:  +44 1865 272595

This afternoon (EST) there were solutions to two different problems,
which on reflection have a similarity:

Prof Brian Ripley's rank variation:

  1 + rowSums(outer(x, x, ">"))

and my unique row counts:

  "%all.==%" <- function(a,b)apply(b,2,function(x)apply(t(a) == x,2,all))
  colSums( A %all.==% t(unique(A)) )

It occurred to me that if we define the APL-style
generalized matrix multiply like this:

inner <- function(a,b,f,g){ 
		f <- match.fun(f)
		g <- match.fun(g)
                apply(b,2,function(x)apply(g(t(a),x),2,f))
}

then both problems can be put into a similar form:

   1+inner( t(w), t(w), sum, "<" )

and

   colSums( inner( A, t(unique(A)), all, "==" ) )