LAUSR

In this article, the most commonly used algorithms for causal search on fMRI data are reviewed and discussed, with particular attention paid to aspects of the algorithms useful for substantive neuroimaging researchers. Classic algorithms, such as PC and GES, as well as more contemporary algorithms, such as IMaGES–LOFS and GIMME, are compared and contrasted. One major difference between algorithms, the use of lagged variables to infer direction vs. the assumption of non-normality to infer direction is discussed, with eye on the impact those choices have on substantive findings. The algorithms share some commonalities as well as differences. For instance, some use lagged variables to assist in ascertaining directionality, whereas others rely on aspects of non-normality in the data to infer directionality. Attention is given to the impact of these choices on the reliability of results.