Poster A75, Saturday, March 24, 1:30–3:30 pm, Exhibit Hall C
Having your cake and eating it too: Flexibility and power with mass univariate statistics for ERP data
Eric C. Fields1,2, Gina R. Kuperberg3,4; 1Boston College, 2Brandeis University, 3Tufts University, 4Massachusetts General Hospital
Event-related potential studies generate large amounts of data across time and space. Statistical analyses in the ERP literature often do not sufficiently address the multiple comparisons problem that this creates. ERP researchers therefore face a catch-22: pre-specifying time windows and spatial ROIs for analyses requires knowing in advance where effects will appear, but choosing analysis parameters based on the observed data is biased and can significantly inflate the Type I error rate. This problem is often exacerbated by low power, leading to studies and statistical tests that provide little evidence for true effects despite reporting significance. Mass univariate statistics have been proposed as one solution, but it is often assumed that this approach sacrifices power to maintain flexibility and Type I error rate. However, simulation studies comparing mass univariate approaches to traditional mean time window approaches have not tested this assumption. We present such simulations using the newly released Factorial Mass Univariate Toolbox (https://github.com/ericcfields/FMUT/wiki). Our results show that when spatial and temporal assumptions are matched, mass univariate approaches actually yield greater power than the mean amplitude approach. This was true for both broadly-distributed and focal ERP components. In addition, whereas the mean amplitude approach requires knowing where effects will appear in space and time, our simulations indicate that mass univariate approaches show only modest decreases in power when broad spatial and temporal ROIs are used. These results suggest that mass univariate statistics offer the ideal balance of power and flexibility for many or most ERP studies.
Topic Area: METHODS: Electrophysiology