12: Multivariate Pattern Classification (Decoding)

Applied Event-Related Potential Data Analysis 2e · LibreTexts

0%

Contents

    Text

    Size
    100%
    Theme
    Typeface
    Line length
    Spacing

    Read aloud

    How

    Speed: 1×
    Light up
    Highlight

    Symbols: Mulberry Symbols, © 2018–2026 Steve Lee, CC BY-SA 4.0 (licence, the set), the site's own copies.

    From LibreTexts, under CC BY-NC 4.0, as the source states it; its equations are written in TeX, which the PDF shows typeset. This page at LibreTexts.

    12: Multivariate Pattern Classification (Decoding)
    Learning Objectives

    In this chapter, you will learn to:

    • Apply decoding to a simple binary case of distinguishing between pictures of two different faces
    • Apply decoding to the data from multiple participants at once, using multiple parallel processing cores to increase the speed of decoding
    • Apply decoding to a more complex multiclass task, distinguishing among four different face identities or among four different emotional expressions
    • Use decoding in place of conventional mean amplitude analyses to improve sensitivity, effect sizes, and statistical power (using data from the ERP CORE N170 experiment)
    • Perform temporal generalization analyses, in which you can assess the extent to which similar neural representations are active at different time points
    • Perform decoding on frequency-domain data as well as time-domain data
    • Prevent the artificial inflation of decoding accuracy that typically occurs when there are more trials in one class than another (using both simulated data and real data)
    • Quantify decoding performance in terms of both proportion correct and area under the curve (AUC) using data from the ERP CORE P3b experiment

    This chapter focuses on ERPLAB’s tools for multivariate pattern classification (decoding). Remarkably, you can decode subtle brain representations with ERPs, allowing you to determine whose face a participant just saw, the emotional content of a scene they are holding in working memory, and their planned motor responses. You can also apply decoding to traditional ERP paradigms to increase your statistical power for determining things like whether the N400 is larger for one class of words than for another.

    Decoding isn’t terribly difficult to understand, but it’s pretty different from traditional approaches to ERP analysis. I’m going to assume that you already know the basics (e.g., how a support vector machine is applied to ERP data, how cross-validation works in the context of ERPs), so I will provide only a brief review of the essence of decoding. If you don’t already understand the basics, I recommend reading Chapter 11 in the third edition of An Introduction to the Event-Related Potential Technique (Luck, in preparation). If you don’t have access to that chapter, you can instead read the first three parts of the Decoding Tutorial in the ERPLAB documentation. However, I strongly recommend that you read the book chapter before you start applying decoding to your own data, because it discusses many potential problems that you will need to avoid. (And I’m not recommending it just to earn more royalties!)


    This page titled 12: Multivariate Pattern Classification (Decoding) was last modified on Thu, 01 Oct 2026 18:52:27 GMT and is shared under a CC BY-NC 4.0 license and was authored, remixed, and/or curated by Steven J Luck.