A variety of techniques have been introduced over the last decade for abstracting search graphs and then using these abstractions for search. While some basic work has been done to predict the value of an abstraction mechanism, the results have not been validated in practice. In this paper we analyze a variety of old and new abstraction mechanisms in a pathfinding testbed and show that the work done in abstraction-based refinement-style search can be predicted by the diameter and size of abstract nodes.