My work built on the earlier work of Thomas Funkhouser, who proposed a predictive level of detail optimization algorithm as part of his PhD. My contribution was to extend his algorithm to scene hierarchies — hierarchical decompositions of scene objects into parts (and parts of parts). I showed that it was possible to apply the same approach to optimize the replacement of collections of objects by simple representations of their wholes. In the process I found and corrected an important mistake in the algorithm proposed by Funkhouser. I provided a counter-example, proposed a fixed version, and eventually proved its correctness mathematically.