The Effects of Acute Moderate Exercise on Cognitive Functioning — Part 4
Analysis Methodology 1: Average ERP Waveforms
Neuroimaging signals generally contain noise as well as signal. EEG recordings are no different. EEG signals generally contain noise such as eye movements, eye blinks, muscle activity, and line noise (Li & Principe, 2006). The magnitude of these artifacts can sometimes be significantly larger than the ERP components of interest, distorting the amplitudes and latencies of the ERP waveform (Li & Principe, 2006). In an effort to understand the ERPs more clearly, researchers have developed techniques to remove noise artifacts in order to reduce distortions to the ERPs (Li & Principe, 2006).
After the removal of noise artifacts the data must be prepared for statistical analysis. The most common technique used to quantify differences across conditions using ERP data is to average all trials within a participant at a specific time point (Luck, 2005). This is otherwise known as creating an averaged ERP waveform.
Below, an example has been outlined to illustrate how the average ERP waveform is generated from the data that has been collected.
- Load in the artifact removed ERP data for each individual
2. Organize the data and average across participant, condition, time, and flanker task
Individuals data has been loaded into R. The first order of business is to subset out the relevant electrode data. Next, we want to create another variable that will be a count of the number of rows for that represents the length of one individual’s ERP data. This is important because we’ll want to average across time in the future. Now we can rename flanker conditions 3 and 4. Flanker conditions 3 and 4 simply represent button presses with the left hand. Considering that we have no interest in handedness, it makes sense to average across the hands. At this point, we are ready to average across all factors (participant, condition, time, and flanker). This will result in an average ERP waveform for each individual at each respective time point.
3. Plot one individual’s average ERP waveform across conditions, times, and flankers
MCT here refers to the moderate exercise condition
1 & 2 refer to the congruent and incongruent flanker conditions respectively
4. Plotting the same individual’s data with a smaller time window
Considering that we are primarily interested in the P3 component and the time window it typically falls within, we can reduce our time window to get a better sense of the neural activity within our time frame of interest.
5. Plotting the grand average ERP waveform
The grand average ERP waveform might also be of interest. This is calculated by taking the mean across individuals. Although this data will not be used in the statistical analysis, it may provide a simplistic visualization of how neural activity differs across conditions, time, and congruency.
6. Limitations of the average ERP waveform
However, the representativeness of the average ERP requires several assumptions: 1) EEG data collected on a single trial contains the ERP waveform, random noise, and stimulus related noise; 2) the ERP waveform is assumed to be more or less the same on each trial; 3) the neural activity underlying the ERP is the similar on every trial; 4) noise is assumed to vary according to the stimulus and from trial to trial; 5) the signal to noise ratio improves with more trials, where the average of the noise approaches zero with infinite trials (Luck, 2005). According to these assumptions, the ERP waveform will become clearer by averaging across many trials because the noise will be reduced due to its random nature (Luck, 2005). In addition, the average ERP would meaningfully represent a condition’s neural activity, due to the assumed consistent nature of the ERP signal (Ouyang et al., 2017; Hultsch & MacDonald, 2004).
When visualizing data across trials, however, it is clear that these assumptions are not met (Luck, 2005). If trial ERPs are plotted in comparison to the condition’s average ERP waveform, it is clear that ERP latencies and amplitudes vary from trial-to-trial (Ouyang et al., 2017; Hultsch & MacDonald, 2004). By averaging peak amplitude across trials, as is typically done with ERP processing, amplitudes are distorted because of the variation in ERP latencies. As a result, the average ERP may be insensitive to subtle changes in cortical activity (Ouyang et al., 2011). Further, the averaging protocol disregards meaningful trial-to-trial variability (i.e., intra-subject variability (ISV)) (Ouyang et al., 2011). The same average ERP (as indicated by the red lines) could be generated by larger peak amplitudes with high ISV in latency (condition A) or smaller peak amplitudes with lower ISV in latency (condition B).
