Comprehensive Guide to Event-Related Potentials in Cognitive Psychology

Hello and welcome to the course basics of external design for cognitive psychology. I am Dr. Arkwarma from the

department of cognitive science at ID Kur. This is week eight of the course and we are talking about event related

potentials. In this lecture we will go through the over uh you know and go through an

overview of the basic steps that you need to put together an ERP experiment. We have seen through examples how are

ERPs useful to uh study mental processes. How are they useful to uh underline and basically carve out

specific aspects of temporal processing in the brain with respect to particular events or similars. In this experiment

in this lecture we will basically look at how an ERP experiment is basically structured.

Now the first step is recording in setting up the recording of the EEG. So the EEG is typically recorded from

electrodes on the scalp with a conducive gel or liquid between each electrode and the scalp so that there is maximum

conductivity and it basically makes for a stable electrical connection. The electrical potential or the voltage can

then be recorded from each electrode resulting in a separate waveform from each electrode site. This waveform will

be a mixture of actual brain activity, biological and electrical potentials produced from the outside of the brain,

by the skin, eyes, muscles, etc. and induced electrical activity from the external electrical devices that are

also picked up by the head, the electrodes and the electrode wire. So this is actually a lot of noise that you

capture from each electrode site. If precautions are taken to minimize these non-neural potentials, the voltages

produced by the brain that is the EEG will be relatively large compared to the non-neural or noise voltages. So the one

of the first and the more important things is to basically minimize all kinds of noise that can be captured here

and to maximize the neural activity or the you know the EEG that is arising or because of the neural activity of the

brain. Now typically this EG signal is relatively small under 10 molts uh 10

microvolts sorry. Uh so the signal from each electrode is usually uh amplified by a factor of 1,000 to 10,000 times.

The ampli amplification factor is called the gain of the amplifier. previous experiments if you remember the oddball

experiment that we talked about where we were working with the P3 wave it used a gain around two 20,000 and the next

experiment that we were talking about the one with schizophrenic patients used the gain around 5,000 so that is

basically by the factor of which you multiply the basic signal the continuous voltage signal is then turned into a

series of discrete signal uh discrete digital values for storing on a computer in most experiments the voltage is

sampled led from each channel at the rate of between 200 and and 1,000 evenly spaced samples per second, which is

basically 200 to 1,000 hertz. The EEG is typically recorded from multiple electrode sites across the scalp. And

there are different studies which have different preferences. For example, some studies use only around 5 to six

electrodes whereas a lot whereas some studies may use up to 256 electrodes depending upon the quality of the data

that can be obtained the research question the areas of the brain and the kind of processing that you are actually

looking for. Now after the setup of uh you know the recording the EG is uh uh created the

next uh very important step is the artifact rejection and correction. The raw EG recordings as we were just

talking about pick up several common artifacts which require special treatment which requires us to handle

them. The most common of these artifacts arise from eyes. When the eyes blink, a large voltage deflection is observed

over much of the head, the entire scalp. And this artifact is usually much larger than the ambient ERP signals that you

are actually interested in. Sometimes eyelinks are systematically triggered by tasks and may vary across groups and

conditions yielding to a systematic distortion of data which again can be interesting for the kind of question

that you're asking. Large potentials are also produced by eye movements when the eyes move and these potentials can

confound experiments that are interested in presenting lateralized stimula or they focus on lateralized ERP responses.

So in all of these lateralization studies you want to have minimal eye movement and minimal artifacts based on

eye movements. So trials containing blinks or eye movements or other artifacts are typically excluded from

the averaged ERP waveform. This approach however has two shortcomings. First, a fairly large

number of trials will need to be rejected because they contain these artifacts and it reduces the number of

trials contributing to the average ERP waveform. Secondly, the mental effort involved when you instruct your

participants to you know not blink their eyes. The mental effort involved in suppressing the eyeblinks may also

sometimes uh play in and uh you know it may impair the task performance of a participant when you are asking them to

do a given task. These problems are especially acute in individuals with neurological or psychiatric problems who

may blink on almost every trial or may perform the task poorly because of the effort that they devoted to blink

suppression. So it's it's a basic uh you know trade-off that ERP researchers have to be extremely careful about.

Fortunately, methods have been developed to estimate the artifactual activity and subtract it out, clean the data, leaving

artifact free EEG data that can be included in the average ERP waveform. The third and a very important step is

filtering. Filters are usually used to remove very slow voltage changes which are under 01 to 0.1 hertz and very fast

voltage changes which are between 15 to 100 Hz because scalp recorded voltages in these frequencies are ranging likely

from noise and non-neural sources. So these kinds of signals have to be filtered out. Frequencies between 0.11

Hz and above 18.5 Hz are filtered from the waveforms. they are removed. Uh but also it has a trade-off that filters can

distort sometimes the time codes of an ERP waveform and can induce artifactual oscillations when the low cutff is

greater than the approximate 0.5 Hz or when the high cutff is uh less than the approximate 10 hertz. So obviously uh

ERP researchers when they are analyzing this data they have to be extremely careful when these filters are being

utilized. Filters can obviously be applied to EG data also and then they can be applied to uh ERP uh you know

average ERPs or sometimes at both steps. Now how do you compute the average ERP waveform? ERPs are typically small in

comparison to the rest of the EEG activity and are usually isolated from the ongoing EEG by a simple averaging

procedure. To make this possible, it is necessary to include event codes in the EEG recordings that basically mark when

is your stimulus coming. So mark the events that happened at specific times such as the onset of each stimulus. That

is why you will be able to get the best data. These event codes are then used uh for time locking uh you know to extract

segments of the EG around each event just before the stimulus presentation and just after the stimulus onset. We

have seen in the oddball experiment that the EEG was recorded over a 9-second period in an oddball task with frequent

X stimula that were coming around 80% and infrequent O stimula that were coming around 20%. Each rectangle

basically the area that you take off each rectangle highlights a 900 millisecond segment of EEG that begins

100 milliseconds before the event code and extends until 800 milliseconds after the event code to capture processing

post stimulus. The 100 milliseconds uh before the event code is also used to provide a pre-stimulus baseline period

as to what was happening before the stimulus even appeared. Now these segments of EG around 9

seconds of worth of EG data are lined up in time. They are chronologically arranged. Stimulus onset time is set at

zero. There is quite a bit of variability in the ERP waveform from trial to trial. And this variability

basically reflects the fact that the EEG is a sum of many different uh sources of electrical activity in the brain. Many

of which are not which we are interested in and they're not involved in processing of the stimulus.

Now to extract the activity that is related to stimulus processing from the unrelated EEG, the EEG segments

following each of the X's are averaged together into a different waveform and the EEG segments following each of the

O's are averaged together into a different waveform. Any different brain activity that is not time logged to the

stimulus will be positive at a given latency on some trials and negative at a different latency on another trials. And

if these are averaged together, they sort of go away. They cancel out and they approach zero. So what you're left

with is averaged waveforms that are contingent to X's and contingent to O's after your analysis.

Now any brain activity that is consistently elicited by the stimulus across trials with approximately the

same voltage at a given latency across trial to trial will be captured will remain in the average. So by averaging

together many trials of the same type, the brain activity that is consistently time blocked to processing this

particular stimulus across trials can be extracted from other sources of voltage. So you can clean rest of the junk away

and basically capture this clean signal. Other types of and this is this will also include EG activity that is

unrelated to the stimulus and is arising from the non-neural sources of electrical noise. So you have to be very

sure of that. Other types of events that can be used as time locking point in the averaging process. For example, button

press responses, vocalizations, psychotic eye movements. You can actually work with what kind of uh

events are causing what kind of waveform. The number of trials that must be average for each ERP waveform depends

upon several factors. For example, the size of the ERP effect that you're looking for. Uh the amplitude of the

unrelated EG activity, the amplitude of non-neural activity. All of that basically depending on how much noise

there is, it'll basically uh decide how many trials you will need to average. So for instance, for a large ERP component

such as the P3 wave, very clear results can usually be obtained by averaging around 10 to 50 trials. For smaller ERP

components such as the P1 wave, it is usually necessary to average together at least 100 to 500 trials for each type of

condition for each trial type to be able to see reliable differences between groups and between conditions.

Now uh another uh very important step in this whole thing is the quantification of amplitudes. So what is the amplitude?

What is the latency at which that amplitude is peing? So the most common way to quantify the uh magnitude in the

timing of a given ERP component is to measure the amplitude and the latency of the peak voltage within the uh same time

window. For example, to measure the peak of the P3 wave, one might define a measurement window, let's say between

400 to 700 millconds or a little bit earlier and find the most positive point in that window. uh peak amplitude would

be defined as the voltage at this most positive point and peak latency would be defined as the time of this point. So as

the wave is rising the peak amplitude will be required as the uh you know main amplitude and the time at which this

peak amplitude is reached is known as the peak latency. Finding peaks was the simplest approach to measuring ERPs

before the advent of uh inexpens inexpensive computers. Basically earlier what people used to do was there was a

ruler only available to quantify the waveform. So an approach sometimes it is uh also used when you just want to

manually look at the data but it has several drawbacks and better methods for quantifying ERP amplitudes and latencies

have now been developed and are available for researchers. For example, the the magnitude of a

component can be quantified by measuring the mean voltage over a given time window and the mean amplitude is usually

superior to the peak uh amplitude as a measure of a co component's uh magnitude. So for example, once you

average the uh mean amplitude, you basically get a better and more stable measure than just a single peak over a

particular trial. Finally, the most important step is the statistical analysis. As in all kinds of

experimental methods, in most ERP experiments, an average ERP waveform is constructed at each electrode uh site

for each subject in each condition. So, you basically get subject wise data. The amplitude or the latency of a component

of interest is then measured in each of these different waveforms. And these measured values are then entered into

stat statistical analysis just like any other variable just like you would do reaction times or accuracy. Hence the

statistical analysis of ERP data is often quite similar to the analysis of other traditional behavioral measures.

However, ERP experiments provide extremely rich data sets usually consisting of several gigabytes of data

which can lead to both implicit and explicit use of many statistical comparisons. People basically want to,

you know, really see the data, really beat the data up uh in a single study which sometimes can dramatically

increase the probability of a type one error because you might be looking for effects where there are none just by

arranging the data in several ways. The explicit use of multiple comparison arises for example when separate

statistical analysis are conducted for each of these separate several different components. The implicit use of multiple

comparisons occurs when researchers first look at the waveforms and then they decide upon the time windows and

the electro sites to basically uh magnify or the quantify this component amplitude and latency. So if you have

your theory uh beforehand if you have everything in place then the uh probability of these errors is is much

less and it is reduced. Now if a time window is chosen uh such that because the difference between conditions is

greatest in that window then this would bias the result in favor of a statistical significance because you are

already seeing that even if the difference may just have been caused by noise. A similar problem would arise if

the researcher finds the electrode sites with the largest difference between conditions and then uses those sites for

statistical analysis. Unless your analysis is theorydriven, uh it is based on previous research.

There are these two different ways both of which can lead you to making errors. With enough electrode sites, it is

almost always possible to find statistically significant difference between two groups or two conditions uh

at a few electrode sites just simply due to random noise. So for example, if you have a 128 channel setup or a 256

channel setup, some or the other pairs of electrodes will always be differing with each other and if you perform

enough statistical analysis, some kind of significance will be seen. So that is why it is encouraged to have a prior

theory before you get into the analysis so that you have a sense of what is it exactly that you are looking for which

amplitudes uh which uh latencies and which sites you are actually interested in analyzing. So one should be very

careful with the choice of electrode sites chosen for analysis and the conditions for the comparison. So this

is the overall setup how you make uh you know and what are the different components of an ERP experiment. Now we

can discuss while parting briefly a little bit about the benefits and the advantages of the ERPs. For example, the

most common advantage of the ERP technique is its high temporal resolution. ERPs provide a continuous

measure of processing which begins prior to the stimulus and extends after the response. So it gives you a very good uh

you know window in time how a given uh stimulus is getting processed when the it is presented to the brain what kind

of process goes on and how does the brain uh react after the stimulus has offset in a behavioral experiment for

example ERPs give us a measure of the momentby-moment activity during the period between the stimulus and

response. ERPs actually show us the action in the brain that is how a given stimulus is processed and also provide

information about brain activity uh that occurs after a response has occurred after a feedback stimulus has been

presented which basically reflects the executive processes and determine how the brain will operate on next trial.

Say for example there are phenomena like post error slowing you've made an error the brain reacts in a particular way it

prepares to respond to other trials in a different way. These are all what can be accessed by the ERPs. Determining which

process is influenced by an experimental manipulation. Remember we were looking at the schizophrenia study in the

previous lecture. So the ERP methodology given that it has this uh continuous temporal information. It allows us to

understand the exact cognitive processes that are affected by our experimental manipulations as opposed to a bunch of

other processes as well. Say for example in a true paradigm subjects slowed responses reflect a slowing of

perceptual processing or a slowing of response selection process. Remember in the schizophrenia example we were able

to basically delineate that it was not due to slowed perceptual processing but basically due to response selection

processes. All right. Also being able to identify multiple processes that go on when a stimulus is presented. Multiple

neurocognitive processes. So ERP recordings uh actually provide us with a very rich data set that makes it clear

that a given experimental manipulation may actually influence not one and not the only one with which we are

interested in but several different uh cognitive processes which may in turn influence several different ERP

components and a given pattern of behavior might be caused by different mechanisms in different experiments. So

task based effects and so on. For example, behavioral studies often treat selective attention as one mechanism,

but different manipulations of attention have actually been shown to influence different ERP components. So for

example, you may you might refer to the attentional network phenomena by a fan and you can see that uh there are

different ways in which attention uh you know gets affected. So alerting, orienting and executive networks are all

there and they might uh differentially affect the ERP uh components in the brain.

Also uh ERPs are able to be uh you know reveal to us covert measures of processing. For example, ERPs can be

used to provide an online measure when the particular uh person is processing the stimulus when a behavioral response

is sometimes impossible or problematic. For example, ERPs can be recorded from infants who are too young to be

instructed to make a response. They are also recorded for covert measurement or cot monitoring of people who have some

kinds of neurological disorders and they are unable to make proper behavioral responses. So those covert measurements

are also possible through ERPs. Finally, ERPs can be actually used as biomarkers in several medical

applications, clinical applications as they uh can be used to measure aspects of brain function that are impaired in

neurological and psychiatric diseases providing more specific information than an individual's individual patients

brain function that could be obtained by just traditional clinical measures. So, a lot of people actually use ERP

measures as biomarkers for various diseases. Now the disadvantages disadvantage the

major disadvantage is that there is in the ERP signal there is a lot of information. This is basically the

single ERP component or the EG uh waveform uh represents a sum of many underlying components. It is it is often

times difficult to decompose this mixture of components into individual underlying uh components. This is called

the superimposition problem or superposition problem sorry. And at the same time it is difficult to determine

where is the generator of these where is the natural generating locations of these components. These two problems are

the most common impediments for successfully conducting ERP research. Another uh key limitation of the ERP

technique is that a given mental or neural process may sometimes have no ERP signature. So there are obviously these

dozens of distinct ERP components but there are also uh surely hundreds or thousands of distinct cognitive

processes that do not have a signature ERP component. So sometimes people may be misled into uh you know indicating an

ERP component as measure of a particular process. Another limitation that arises from the from the fact uh is that ERPs

are small relative to the noise level and many trials are actually required to uh accurately measure the ERP effect.

that also it takes a lot of effort to get the clean signal out of the ERP uh waveform and uh it basically makes it

difficult to conduct experiments with very long intervals uh between simul and experiments that require sometimes even

surprising the participants. Another important thing is that you know because we are not sure of where the ERP

waveforms where the EG waves are actually arising from they have a very poor spatial resolution. you can never

be sure unless there is a lot of uh channels and now you're using uh you know specialized software it is

difficult to be sure of the source of the ERP signal. So because it is a summation of so many activities over the

scalp it is difficult to know exactly where in the brain which particular uh points in the brain the ERP signals are

arising from. That is why while ERPs have an extremely good temporal resolution, they have a suffering they

have a poor spatial resolution which make it a poor method to study the uh space in the brain rather than uh of

studying uh you know temporal processing in the brain. Finally, costly. Yes, uh the setting up an ERP experiment is

costly. Uh it takes around $50,000 to more to set up an ERP experiment. So that is also something that is a

disadvantage. uh but once you have a proper setup and you can sort of get on collecting good data it is an extremely

important extremely useful behavioral method to conduct experiments. So that is all from me in this uh uh lecture and

I will see you in the last when we are talking about other topics. Thank you.