10.10: Exercise- Collapsing Across Channels and Correlating Latencies with Response Times

So far, we’ve been measuring and analyzing the data only from the C3 and C4 channels, where the LRP is largest. However, it’s often valuable to include the data from multiple channels. One way to do that is to measure from multiple different channels and include channel as a factor in the statistical analysis. However, that adds another factor to the analysis, which increases the number of p values and therefore increases the probability of getting bogus-but-significant effects. Also, interactions between an experimental manipulation and electrode site are difficult to interpret (Urbach & Kutas, 2002, 2006). In most cases, I therefore recommend averaging the waveforms across the channels where the component is large, creating a cluster, and then obtaining the amplitude or latency scores from the cluster waveform. Averaging across channels in this manner tends to produce a cleaner waveform, which decreases the measurement error (as long as we don’t include channels where the ERP effect is substantially smaller). Also, it avoids the temptation to “cherry-pick” the channel with the largest effect. A very careful analysis of the ERP CORE by Zhang and Kappenman (2024) provides quantitative support for this advice. For the ERP CORE flankers experiment, they found that a cluster including FC3/4, C3/4, and C5/6 yielded the best signal-to-noise ratio. They also found that this cluster yielded the best statistical power when they simulated a comparison of two groups (e.g., a patient group and a control group) that differed in LRP amplitude.

I’ve already created this cluster in the difference waves in the Chapter_10 folder. They’re in Channel 12, which is labeled cluster. Let’s try measuring the peak latency from this channel in the filtered data, which you can load from the Chapter_10 > Data > ERPsets_CI_Diff_filt folder (if they're not already loaded). You can just repeat the measurement and analysis procedures from the earlier exercise where we measured the peak latency from the filtered data, but changing the Channel from 5 to 12 as shown in Screenshot 10.18. As usual, take a look at the measures using the Preview tool, and then save them to a file named Peak_Latency_Cluster.txt.

 

An example of why previewing the measures is important

Many times in this chapter, I've asked you to use the Preview tool to look at the measures along with the actual ERP waveforms. Did you do it every time? Probably not. But you really should. After I wrote the text in the preceding paragraph, I got everything set up in the Measurement Tool panel, and then I clicked the Preview button to make sure everything looked OK, and I noticed something strange.

We're now measuring from the new cluster channel, and in the Preview window I saw that some of the ERP waveforms had a ton of high-frequency noise. I double-checked to make sure that I was in the folder with the filtered ERPsets (ERPsets_CI_Diff_filt), and I was. Then I looked at the other channels, and they didn't have the high-frequency noise. It was only in the new cluster channel. I then went back and looked at the script that created the ERPsets for this exercise, and I saw that the filtering was applied only to channels 1-11 (the original channels), not channel 12 (the cluster channel). I then fixed the script (which required changing only one number), ran it again, and everything was fine.

This shows the value of previewing the measures along with the ERP waveforms. If I hadn't clicked the Preview button, I would have been scoring noisy data and not getting optimal results. So, you should believe me when I tell you to preview the measures with the waveforms every time.

Now that you have the peak latency scores, you can perform the usual paired t test. You should see that the effect size is about the same as before (-0.613 for the cluster analysis relative to -0.630 when we measured from the C3/C4 channel). Collapsing across channels doesn’t always increase the effect size, but it doesn’t usually hurt, and it avoids the need to include channel as a factor in the analysis or the need to select one channel to use. My lab now measures from a cluster in virtually all of our studies.

Our last step will be to ask whether the peak latency values are correlated with response times (RTs). That is, do participants with later LRP peaks also have slower RTs? Unfortunately, it takes some additional work to extract RTs using ERPLAB. The ERPLAB documentation has a whole section devoted to behavioral analysis, and it's not super difficult, but it's too much for me to explain here. For the present exercise, I wrote an ERPLAB script named LRP_RTs.m to obtain the mean RTs for Compatible and Incompatible trials for each participant. You can look at the script (in the Chapter_10 folder) to see how I extracted the RTs from the EEGsets.

This script saves mean RTs for each participant in each condition to an Excel spreadsheet named RT.xlsx. These mean RTs were calculated after excluding trials that were rejected because of artifacts, so the ERP data and the behavioral data are based on exactly the same set of trials.

Let’s look at the correlations. You’ll need to combine the RT values from the spreadsheet with the peak latency values that you just created from the waveforms that were collapsed across the three pairs of electrode sites. And then you’ll need to read them into your statistical package or into a spreadsheet program. I turned them into an Excel spreadsheet, with scatterplots and correlations (see Peak_Latency_Cluster_with_RTs.xlsx in the Chapter_10 folder). The spreadsheet computes the correlation between peak latency and RT, separately for the Compatible condition and the Incompatible condition.

Screenshot 10.19 is a screenshot from the Excel file. The peak latency scores were correlated fairly well with the RTs, especially for the incompatible trials. Some outliers in the LRP peak latency scores for the Compatible trials clearly reduced the correlation in that condition. Nonetheless, these correlations show that the timing of the brain activity is related to the timing of the behavioral response.

 

---