Abstract
In this paper, we address the critical assessment of Ramsey et al. of our method for learning partially directed graphs from meta-analysis imaging data (Neumann et al., 2010). We argue that our method provides valid and interpretable results when applied to data representing a single experimental paradigm. Simulations further suggest that, despite theoretical limitations, the application of our method to mixed probability distributions yields reliable results with error rates at acceptable levels. Finally, we discuss the nature of meta-analysis data and the notion of causality in the context of functional neuroimaging.