1.3 Peak Amplitudes and Latencies Are Measurable Independent of the SNR
A common argument against measuring peak amplitude and latency at the ST level is that the SNR is too low to identify the components of interest (Blankertz, Lemm, Treder, Haufe & Müller, 2011; Chamanzar, Malekmohammadi, Bahrani, & Shabany, 2015; Jaśkowski & Verleger, 1999). In other words, there is the assumption that the signal in ST ERPs representing the brain activity elicited by a stimulus event is generally smaller than the random background brain noise and, therefore, cannot be directly seen or measured in the single trial. Although viewed as a descriptive indication of the quality of the data, there is little information in the literature specifying the minimum SNR needed for obtaining reliable and valid measures of brain activity in the averaged ERP or in single trials. This may be due to the fact that efforts to understand SNR have often been conducted by combining synthesized signals with recordings of resting brain signals, then studying computational methods retrieving the synthesized signal (Hu et al., 2011; Jaśkowski & Verleger, 1999).
Furthermore, SNR has been calculated through a variety of methods (Luck, 2005; Spencer, 2005). For example, one method is to calculate the ratio of the largest evoked potential peak-to-peak amplitude and the mean amplitude of the baseline period prior to the presentation of the stimulus (Pfurtscheller et al., 1987). Another method is to divide the standard deviation of the averaged ERP by the standard deviation of the noise as estimated from amplitudes of oscillatory signals recorded during the resting states (Ouyang et al., 2015). Alternatively, SNR has been defined as the signal being the amplitude of a peak in ST ERP, while the noise, or measurement error, defined as uninformative intra-subject variability created by neural activity not activated by the stimulus as well as non-neural artifacts (Blankertz et al., 2011; Ribeiro et al., 2016). While this latter approach can be viewed as more compatible with test and measurement theory, this definition of noise may contain elements of meaningful variability across single trials.
Independent of how SNR is calculated, a low SNR may not be problematic if the effect of the baseline noise can be shown to be relevant to brain processing post-stimulus (for review see Rey et al., 2015). Pre-stimulus brain activity has been shown to consistently influence later ERP components after the presentation of a task-relevant stimulus in both young adults and older adults, even when significant age-related changes in post-stimulus neural processing are found. For example, a greater pre-stimulus gamma power correlated with larger P300 amplitudes and longer response times (Reinhart, Mathalon, Roach, & Ford, 2011), and pre-stimulus alpha power influenced errors on a task (Shou, Dasari, & Ding, 2015). In another study, McNair, Kayser, and Kayser (2019) found that the amplitudes of P1 and N1 were significantly larger and the latencies of N1 and P2 were greater in the older adults compared to young adults, while the pre-stimulus rhythmic brain activity influence on post-stimulus brain activity remained constant across age groups.