AGAT Method

Published As:

Brooks, J.L., Zoumpoulaki, A., & Bowman, H. (2017). Data-driven region-of-interest selection without inflating Type I error rate. Psychophysiology, 54(1), 100-113.

****Please cite the above paper if you use our method****

Abstract:

During ERP data analysis, it is often difficult to know, a priori, precisely where effects will occur on the scalp, in time, and in frequency (for oscillations). To overcome this, researchers often identify regions-of-interest (ROIs), but have been criticized for sometimes using biased, data-driven methods which inflate Type I error rates. Through simulations and analysis of human ERP data, we demonstrate a data-driven ROI-selection method using the aggregated-grand average (AGAT) wave which controls Type I error rate. Furthermore, it relaxes the precision necessary for a priori ROI specification. We demonstrate that the AGAT is orthogonal to the experimental contrast and, importantly, show that other common methods for computing orthogonal waveforms for ROI selection can inflate Type I error rate under some conditions. Finally, we show that the AGAT method has superior statistical power over common a priori ROI selection methods by up to 60%. Our results demonstrate a simple, unbiased and data-driven ROI selection method which is relevant for replicable and powerful ERP analysis in studies of sensory and cognitive ERP components.

****Before using the AGAT, please see our list of assumptions for usage in Table 4 of our 2017 paper****

Downloads & Instructions

• EEGLab Plugin for AGAT
  • This plugin can be used with EEGLab to compute AGAT waveforms for ERP analyses.
  • Instructions: Unzip the file and add the folder AGA1.00 to the "plugins" directory of your EEGLab folder. If EEGLab is already open, then close it an open it again. You should now see "AGAT" listed as an option within your "Tools" menu. Select this and it will bring up a dialog box to allow you to create an AGAT from one or more SET files. There are options to select a subset of channels and one continuous time period. Currently, the tool only allows one to select max or min features of the AGAT waveform.
• AGAT in BrainVision Analyzer
  • Instructions: You can compute the AGAT in BrainVision Analyzer software using the Grand Average function with the “Calculate Weighted Average” box ticked. In this case, the Grand Average computation will take into account all of the individual trials that contribute to each ERP in the grand average rather than just averaging the ERPs themselves.
• Brooks, et al. (2017) AGAT Simulation Code
  • The simulations for our 2017 paper were conducted in the R environment. Below are links to the R program files which you can use as a basis to replicate our simulations or as a basis for modification to simulate your own experimental designs.
  • Core Functions & Data: All of the following rely on these functions and data to be in root directory from which the below are run.
  • Simulation 1 - AGAT Type I Error Rate
  • Simulation 2 - Condition Trial Number Asymmetry
  • Simulation 3 - Condition Noise Asymmetry
  • Simulation 4 - AGAT Power

Further/Ongoing Work

Our simulation work on the AGAT is ongoing and we are currently working on various questions related to how AGAT can be used to safely, and with high power, localise ROIs in ERP and other data sets. We will post more information here as this work is completed. However, if you have a particular usage that you are considering, please feel free to get in touch with Joseph Brooks to ask about whether we have any results for that approach. If not, you are welcome to use our simulation code (see above) as a basis for simulating your design.