1.1 Multiple Approaches Are Used to Obtain Single Trial ERP
Measures
Research efforts to derive measures of brain activity from single trial
(ST) ERPs have employed a wide variety of techniques such as
peak-picking the amplitudes of raw EEG or bandpass filtered data (e.g.,
Spencer, 2005), template matching methods including use of Woody filter
(e.g., Fein & Turetsky, 1989; Gavin et al. 2019; Patterson,
Michalewski, & Starr, 1988; Pfefferbaum, Ford, Roth, & Kopell, 1980),
maximum likelihood estimation methods (e.g., Jaśkowski & Verleger,
1999; Möcks et al., 1988), regression-based estimation (e.g., Mayhew et
al., 2006), wavelet filtering or smoothing (Hu, Mouraux, Hu, &
Iannetti, 2010; Hu et al., 2011; Quian Quiroga, Atienza, Cantero, &
Jongsma, 2007; Rey, Ahmadi, & Quian Quiroga, 2015; Roth, Roesch-Ely,
Bender, Weisbrod, & Kaiser, 2007), decomposition methods such
as residue iteration decomposition (RIDE; Ouyang, Sommer, & Zhou,
2015), principal components analysis (PCA) or independent component
analysis (ICA) (e.g., De Vos et al., 2012; Saville et al., 2015), and
graph theory (Dimitriadis, Brindley, Evans, Linden, & Singh, 2018). For
partial reviews on these techniques, see De Vos et al. (2012) who
compared four of these techniques and Ouyang, Hildebrandt, Sommer, and
Zhou (2017) who compared eight of these techniques. Interestingly, De
Vos et al. (2012) stated “At present, little information exists as to
which ST estimation method is the more powerful” (p. 1197). Most of
these techniques were developed to assess a single measure, usually a
mid to late component of the ERP such as the N2 or P3, with some
focusing on measuring amplitude while other techniques focus on
determining latency variability (Spencer, 2005). Though informative,
none of these articles report the psychometric properties of both
amplitude and latency measures of multiple components obtained in single
trials commonly found in the averaged ERP. That is, these articles do
not provide quantified descriptions of ERP components from the viewpoint
of classical measurement and statistical theories for either the
averaged wave forms or averages of ST measures (e.g., measures of
central tendency, variability, and distribution) nor are reliability
coefficients of split-half or test-retest provided. Consequently, to
date none of the methods cited above have been adopted by the research
community at large.