On high-pass filter artifacts (they’re real) and baseline correction (it’s a good idea) in ERP/ERMF analysis

In Tanner, Morgan-Short and Luck (2015; henceforth TMSL) we demonstrated how commonly-used high-pass filter settings can distort ERP (and analogously ERMF) data, and that these distortions can lead to spurious conclusions about the nature of the cognitive processes engaged during the experimental task. We appreciate Maess, Schröger, and Widmann’s interest in our work, and we thank them for their thoughtful commentary. Indeed, we feel that open discussion of these issues – and importantly empirical demonstration of the benefits and pitfalls of high-pass filtering, baseline filtering correction, and other issues – will benefit the field by helping establish a set of best practices for signal processing in ERP research. Establishing a consistent best-practices approach to filtering and ERP analysis more generally will help ensure cross-study comparability within sub-fields of ERP research and lead more reliable, consistent, and replicable results. Maess et al. raise two major points in response to our article.

First, they argue that our original test data were not optimally suited to show the benefits of high-pass filtering because they simply did not contain enough low-frequency noise. Second, they argue that high-pass filtering should replace the common practice of baseline correction in ERP research, contra our recommendations. We will respond to both of these arguments here, as well as a point they raise about criteria for detecting filter-induced artifacts.