Opening the book…
Although it’s important to look at the single-participant ERP waveforms, it will be easier to see subtle effects by averaging the waveforms across participants to create grand average waveforms for each bin. To do this, first make sure that the ERPsets from all 10 participants from Chapter_3 are loaded (which you can verify by looking at the ERPsets panel). If you have an other ERPsets loaded (like 30_N400, the "bad" data from the previous section), select those ERPsets and clear them (by clicking Clear in the ERPsets panel).
Also, make sure that Chapter_3 is the current folder (by clicking Current Folder in the ERPsets panel) and select the 10 ERPsets in the ERPsets panel.
Now go to the Average across ERPsets (Grand Average) panel and make sure it is set up as shown in Screenshot 3.4. Select all 10 ERPsets in the ERPsets panel and then click Run in the Average across ERPsets panel to create a grand average of the 10 selected ERPsets. You’ll then see the usual window for saving the new ERPset that you’ve created. Name it Grand_N400 (and save it as a file named Grand_N400.erp so that you have it for the subsequent exercises). You should now see Grand_N400 in the ERPsets panel, and it should be the only selected ERPset (so that you'll see the waveforms from this ERPset).

Select Bins 3 and 4 from the Bin & Channel Selection panel, and make sure that you're using the custom layout specified by GridLocations_Custom.tsv. And let’s add something new: Tick the box labeled Show standard error in the Plot Settings panel (and make sure that the transparency level is set to 0.2). You should see something like Screenshot 3.5, with a more negative voltage for the unrelated targets than for the related targets at CPz and surrounding electrode sites. The light shading around the waveforms shows the standard error of the mean (SEM) at each time point (see the box below for more information).

Now plot Bins 1 and 2 (the prime words). The waveforms for these bins should be lying right on top of each other, with any differences being small relative to the SEM. Remember, these are the bins for primes that are followed by related versus unrelated targets, and unless the participants have ESP, they can’t differ as a function of something that happens later. As a result, any differences between them are simply a result of noise.
Finally, take a look at the aSME data quality values for the grand average by going to the View Data Quality Metrics panel and clicking the Show in a table button. When you made the grand average, the default settings caused the aSME values from the individual participants to be combined together using something called the root mean square (RMS). This is like taking the average of the single-participant aSME values, except that the RMS of the single-participant values is more directly related to the impact of each participant’s data quality on the effect size (see Luck et al., 2021).
These aSME values are like standard error values, but they’re the standard error of the mean voltage over a 100-ms time window rather than the SEM at a single time point (see the box below). If you look at the table of aSME values, you’ll see that the values in the N400 time range (300-400 and 400-500 ms) at the CPz electrode site are between 1.4 and 1.9 µV. That’s reasonably small relative to the large difference in mean voltage between the unrelated and related targets. In other words, the data quality is quite good for our goal of detecting differences between these two types of targets.
Plotting the Standard Error
Plotting the standard error of the mean (SEM) at each time point in an ERP waveform, as in Screenshot 3.5, can be helpful in assessing whether the differences between conditions are reasonably large relative to the variability among participants. These standard error values are just like the error bars that you might see in a bar graph. At each time point, the grand average ERP waveform is simply the mean of the voltage values across participants at that time point. The SEM is just the standard deviation (SD) of the single-participant values divided by the square root of the number of participants (which is exactly how the SEM is usually calculated in other contexts).
You can also see the SEM when you plot a single participant’s averaged ERP waveforms. In this case, the waveform shows the mean across trials rather than the mean across participants, and the SEM reflects the SD across trials divided by the square root of the number of trials (rather than the SD across participants divided by the square root of the number of participants).
Although the SEM can be useful, it has some downsides. First, imagine that the overall voltage at 400 ms was much more positive in some participants than others, leading to quite a bit of variability in the voltage for each condition. Because of this variability, the SEM for each waveform would be quite large at 400 ms. This would make it look like the difference in means between conditions was small relative to the SEM. However, imagine that the voltage at 400 ms was exactly 3 µV more negative for unrelated targets than for related targets in every participant (i.e., the experimental effect is extremely consistent across participants). Thus, the SEM is large even though the difference for each participant was extremely consistent. In behavioral research, this problem is often addressed by using the within-subjects SEM (Cousineau, 2005; Morey, 2008). ERPLAB doesn’t have this version of the SEM built in, but you can achieve the same result by making a difference wave between the conditions (e.g., unrelated targets minus related targets) and getting the SEM of the difference wave. This is exactly what was done in the grand averages from the full study (see Figure 2.1C in Chapter 2).
Another downside of the SEM is that it can be very large if there is a lot of high-frequency noise in the data, even though this noise has minimal impact when we quantify the N400 as the mean voltage between 300 and 500 ms (as we will do later in this chapter). The aSME value provided in our Data Quality measures does not have this downside, because it provides the standard error of the mean voltage over a time period rather than the standard error of the values at individual time points. See Luck et al. (2021) for a more detailed discussion.
One last thing: Open up the ERP & Bin Information panel. In the table at the bottom of the panel, you can see that there are hundreds of trials per bin. When a grand average is made, we sum together the trials across the ERPsets, so this table shows you the total number of trials that contributed to the grand average waveforms.
This page titled 3.5: Exercise- Making a Grand Average was last modified on Thu, 01 Oct 2026 18:52:35 GMT and is shared under a CC BY-NC 4.0 license and was authored, remixed, and/or curated by Steven J Luck directly on the LibreTexts platform.
The book could not be loaded just now.