Figure 5.2 shows the data for a few key channels with the four references we’ve used so far (Average of P9 and P10, Cz, Oz, and the average of all the EEG sites). As you can see, the choice of reference electrode has a massive effect on the waveforms. But which of these is the correct way to reference the data? That is, which reference site will give us the true waveforms?
Figure 5.2. Grand average ERP waveforms for selected channels from the N400 ERP CORE experiment with four different references.
If you think about it, these are not meaningful questions. Voltage is the potential for electrical current to flow between two places. It makes no sense to talk about the voltage at a single electrode site. Even if we could measure (and not merely estimate) the absolute voltage, it would have no special truth status. For example, the cortical generator site of a given ERP component could be quite far away from the electrode with the largest voltage.
Even if there is no “correct” way to reference the data, is there a “best” way? I discuss this issue in detail in Chapter 5 of Luck (2014). My bottom line is that the best approach is usually to use whatever is most common in a particular area of research. If you don’t use the same reference as most other studies, then people can’t compare your data with the data from other similar studies. And you might think you’ve discovered a new effect. For example, most language ERP studies use the average of the mastoids as the reference, and if you use the average reference, you might think you’ve discovered a new “P400” component at the F7 electrode site.
If you have a good reason to use an atypical reference in a given study, you should also show what the data look like with the typical reference (e.g., by providing the waveforms with the typical reference in online supplementary materials). That way, no one will be confused about the relationship between your study and other studies.
My final piece of practical advice about the reference electrode—especially when you’re first starting out in ERP research or looking at a new component—is to look at your data with multiple different references (as in Figure 5.2). That way, you won’t be lulled into thinking that the waveform in a given channel primarily reflects brain activity at the so-called active electrode for that channel. That is, you’ll see that the reference has a big effect on your data, and you’ll realize that no matter what reference you use, you’re looking at the potential between two sites.
I’ve made this recommendation to many people, but I don’t know how many of them have followed my advice. However, I do know that the people who have followed it and then discussed their experiences with me said that it really helped them understand their data better. So give it a try!
