Reciprocal sister group comparisons. A schematic illustration of the application of reciprocal sister group comparisons to evaluate questions of causation for associated traits in situations where only correlational analyses are possible, but where the correlation is not absolute. (a) Trait 1 and 2 are positively correlated across the phylogeny, so that a difference in one trait is associated with a correlated difference in the other. (b) Provided that the correlation is not perfect, there will be situations where a difference in trait 1 is not associated with a correlated difference in trait 2, and (c) conversely, where a difference in trait 2 is not associated with a correlated difference in trait 1. This can be interpreted to mean that the traits, although statistically influenced by each other, sometimes evolve for external reasons. If (b) is more common than (c), trait 1 can sometimes change without an associated change in trait 2, while trait 2 rarely evolves without an associated change in trait 1. Hence, trait 2 does not necessarily follow the evolution of trait 1, but trait 1 appears to follow the evolution of trait 2.