
 5.5.9.4. Interaction effects matrix plot 
  The latter depends on the number of factors, k, the choice of the k factors, the constraints on the number of runs, n, and ultimately on the experimental design that the analyst chooses to use. 
  For those main factors along the diagonal that have flat lines (that is, are unimportant), the naive conclusion to use either setting, perhaps giving preference to the cheaper setting or the easiertoimplement setting, may be unwittingly incorrect. 
  In this case, the 2factorinteraction confounding structure may be deduced by comparing all of the 2factor interaction subplots (and effect estimates) with one another. 
 www.itl.nist.gov /div898/handbook/pri/section5/pri594.htm (2978 words) 
