Estimating the Timing of Cognitive Operations With MEG/EEG Latency Measures: A Primer, a Brief Tutorial, and an Implementation of Various Methods

Abstract

The major advantage of MEG/EEG over other neuroimaging methods is its high temporal resolution. Examining the latency of well-studied components can provide a window into the dynamics of cognitive operations beyond traditional response-time (RT) measurements. While RTs reflect the cumulative duration of all time-consuming cognitive operations involved in a task, component latencies can partition this time into cognitively meaningful sub-steps. Surprisingly, most MEG/EEG studies neglect this advantage and restrict analyses to component amplitudes without considering latencies. The major reasons for this neglect might be that, first, the most easily accessible latency measure (peak latency) is often unreliable and that, second, more complex measures are difficult to conceive, implement, and parametrize. The present article illustrates the key advantages and disadvantages of the three main types of latency-measures (peak latency, onset latency, and percent-area latency), introduces a MATLAB function that extracts all these measures and is compatible with common analysis tools, discusses the most important parameter choices for different research questions and components of interest, and demonstrates its use by various group analyses on one planar gradiometer pair of the publicly available Wakeman and Henson (2015) data. The introduced function can extract from group data not only single-subject latencies, but also grand-average and jackknife latencies. Furthermore, it gives the choice between different approaches to automatically set baselines and anchor points for latency estimation, approaches that were partly developed by me and that capitalize on the informational richness of MEG/EEG data. Although the function comes with a wide range of customization parameters, the default parameters are set so that even beginners get reasonable results. Graphical depictions of latency estimates, baselines, and anchor points overlaid on individual averages further support learning, understanding and trouble-shooting. Once extracted, latency estimates can be submitted to any analysis also available for (averaged) RTs, including tests for mean differences, correlational approaches and cognitive modeling.

Repetition Attenuates the Influence of Recency on Recognition Memory: Behavioral and Electrophysiological Evidence

“…First, multiple stimulus repetitions increased the amplitude of the left parietal old/new effect (also see Graetz et al, 2018;Van Strien et al, 2005). Second, we provide evidence that the latency of the old/new ERP effect was shortened by repetition (for analogous results, see de Chastelaine, Friedman, Cycowicz, & Horton, 2009;Liesefeld et al, 2016;Park & Donaldson, 2016) as well as recency, further indicating the viability of using robust measurement approaches to understand the precise timing of cognitive processes (Liesefeld, 2018;Ouyang, Herzmann, Zhou, & Sommer, 2011;Smulders, 2010). Critically, the relationship between recency and the latencies of the behavioral and neural correlates was shown to be attenuated with additional presentations of recognized stimuli.…”

“…The timing of the effects was defined as the percent-area latency, which is the time at which the effect of interest (here, the old/new difference) reaches a specified percentage of its area under the curve. Percent-area latency has been shown to be more robust to various sources of noise than alternative measures such as peak latency or onset latency (Liesefeld, Liesefeld, & Zimmer, 2016;Liesefeld, 2018). We defined the latency measure by first establishing the onset and offset as 30% of the peak amplitude difference within our time window of interest (500-800 ms) and then determining the time at which the area under the curve reached 50% (also see Liesefeld, 2018).…”

“…Percent-area latency has been shown to be more robust to various sources of noise than alternative measures such as peak latency or onset latency (Liesefeld, Liesefeld, & Zimmer, 2016;Liesefeld, 2018). We defined the latency measure by first establishing the onset and offset as 30% of the peak amplitude difference within our time window of interest (500-800 ms) and then determining the time at which the area under the curve reached 50% (also see Liesefeld, 2018). As shown in Figure 4B for the group of six left posterior electrodes (see above), when comparing the latencies of the old/new effects according to repetition, the effect for the Old1 condition appeared later than that for the Old2 and Old3 conditions.…”

“…The behavioral and EEG analyses were conducted in R (v. 3.5.0), using the brms (Bürkner, 2017), BayesFactor (Morey & Rouder, 2018), bridgesampling (Gronau & Singmann, 2017), and lme4 (Bates, Martin, Bolker, & Walker, 2015) packages. For the EEG data, in addition to using EEGLAB for processing, the latency toolbox (Liesefeld, 2018;in MATLAB) was used for analysis, and the MNE package (Gramfort et al, 2014; in Python v. 3.6.4) was used for plotting purposes.…”

“…Moreover, our experimental design was used in combination with ERPs, from which we assessed the latency of the left parietal positivity to characterize how the neural signature of retrieval success scaled with the presumed termination of the backwards serial search. Whereas prior ERP studies found no effect of lag on the peak latency of this ERP correlate (Friedman 1990a(Friedman , 1990bRugg & Nagy, 1989), we make use of more robust methods of percent-area latency (for discussion, see Luck, 2014;Liesefeld, 2018) to provide evidence to the contrary.…”