Tom:

I found the problem.  What I want for the subcritical percolation correlation
length is

      [ sum_x |x|^q   P(x in C) ] ^ 1/q
 xi = [ ----------------------- ]
      [ sum_x         P(x in C) ]

(for q=1,2); what I had been doing was

      [ sum_x |x|^q   P(x in C) ] ^ 1/q
 xi =   -----------------------.
        sum_x         P(x in C)

(Note that sum_x P(x in C) is chi(p).)  That is, my correlation-length numbers
were too small because I was dividing by a higher power of chi(p) than I should
have.

Thanks for your help in finding it!

----------------------------------------------------------------
John Kerl
Graduate student, University of Arizona Department of Mathematics
kerl@math.arizona.edu
http://math.arizona.edu/~kerl