I think the following weighted approach is more reasonable:
A function is defined F(N,P) = avg(f(games, points, ...)) for past picks, i.e. there is a known value for D-men picked at position 7 F(7,D).
Then the expected value E(N,P) = 0.1(F(N-2,P)) + 0.2(F(N-1,P)) + 0.4(F(N,P)) + 0.2(F(N+1,P)) + 0.1(F(N+2,P))
For pick #1 use 0.7, 0.2 and 0.1
For pick #2 use 0.3, 0.4, 0.2 and 0.1
This is just a sample weighting, others can be used, but this is much better than the arbitrary strict grouping of picks.