The graph construction algorithms often need to make a random choice among several equivalent locations to place the next department. Similarly, the layout improvement algorithms based on simulated annealing select random pair of departments to be tested for possible exchange. Different replications of the same algorithm can thus provide different adjacency graphs or block layouts. The higher the number of replications, the more likely it is that a good adjacency graph or block layout will be constructed. Of course, more replications require more computation time.