Are Neighbour Methods Preferable to Analysis of Variance for Completely Systematic Designs? `Silly Designs are Silly!'