The EEGLAB News #14

Question: Does removing eye movement artifact from EEG recordings using ICA decomposition reduce error in EEG brain measures, or increase it!?

In 2014 Robert Thatcher posted a YouTube video on his claim that removing eye movement artifact from EEG data by subtracting out ICA components accounting for these artifacts (negatively) affects ‘phase’ measures of the data delivered by his proprietary (NeuroGuide) software: []. In 2017, Arnaud Delorme posted an EEGLAB wiki response here: []. The thread below is excerpted from a long discussion that took place at the time on the EEGLABLIST about whether and how removing artifacts using ICA decomposition methods distorts “phase” measures of the EEG (as computed by Thatcher’s software product). Spoiler alert: the artifacts themselves, when added to the actual brain source signals summed at the scalp electrodes, change the phase dynamics of the recorded scalp signals. Removing the artifact processes identified by ICA decomposition can (in some large part) restore the phase dynamics of the summed brain sources, as well as revealing the phase dynamics of the individual effective source signals that dominate the cortical contribution to the scalp EEG. In the following, Scott and Arnaud have expanded the posts of Arnaud and Jason Palmer somewhat for added clarity. Other contributions have not been edited except to correct typos or clarify punctuation.

Arnaud Delorme (with additions by Scott Makeig): We get asked often about a 2014 YouTube video by Robert Thatcher [with 5k views as of 12/22] that makes a claim that ICA adulterates EEG channel phase & coherence, and claims that we at the Swartz Center, UCSD, acknowledge this. The facts are these: ICA decomposition learns one-sample, purely spatial filters that separate out (non-brain) artifact activity from various sources (eye movements, EMG, line noise, etc.) which can be then removed from the data by subtracting the relevant independent components (ICs) from the data. It also separates the brain-generated (artifact-free) data into IC processes that typically have simple scalp projection patterns (scalp maps) compatible with generation in a single (or sometimes, a bilaterally symmetric pair of) cortical area (or areas).

Arnaud & Scott. There are of course necessary cautions in applying ICA decomposition for artifact removal. One should be mindful of seeing whether the decomposition was successful in isolating the artifacts of interest (here, Independent Component (IC) processes accounting for eye movement artifact), and carefully inspect the ICs being removed. For EEG data with a relatively small number of channels, e.g., the 19-channel scalp data used here by Thatcher, ICs accounting for eye movements may necessarily include some (relatively small) amount of brain source activity – as the scalp data may possibly have more than 19 appreciable contributions from localized brain effective sources as well as non-brain sources whose volume-conducted potential fluctuations are summed at each scalp electrode. Recall that each EEG scalp channel signal is then the difference between all the potentials summed at its anode and cathode electrodes respectively.

Removing any time varying signal from the scalp channel data (artifact or not) will naturally change the ‘phase-measurable’ properties of the remaining data [unfortunately, in his video Thatcher does not explain what his proprietary software ‘phase’ measures are computing]. Therefore, removing from the data ANY non-brain source processes separated by ICA decomposition will, naturally, change the data record!

But does it make sense to claim that whatever ‘phase’ measures of the raw recorded scalp data are measures of what THE brain is doing? – Clearly not. Cortical field potentials have quite complex spatiotemporal dynamics, and each of the effective source dynamics has spectral power (and phase) at every EEG frequency. How these sum at the scalp electrodes must differ somewhat from, for example, how they would sum at electrodes (hypothetically) placed within the skull itself, or at the CSF/skull boundary. …

Jason Palmer (+Scott): My take: ordinary ICA can’t change the phase of any oscillations because it is just an “instantaneous” linear combination of the channels without any time shifts. Just as the channel EEG is a linear combination of the sources in the brain, each IC is a certain linear sum of all the channels -- such that ICA is formulated to separate (effective) sources into different ICs. The only way to truly distort data phase is to do some kind of temporal filtering (convolution) of the data. For instance, using a normal FIR filter will change the phase of data (ideally, just adding a constant delay to each oscillation – a linear phase effect). Typically, to do high-pass filtering we use filtfilt, which runs the filter first in the forward direction, adding a constant phase, and then in the reverse direction, subtracting the phase, to leave all oscillations with zero phase shift.

But each IC is just an instantaneous combination of the channel potential values, with no convolutive filtering. So all oscillations (in the Fourier decomposition of the data) retain the same phase, it’s just that oscillations (and their various effects on phase of the summed scalp channel signals) may be separated into different ICs.

There may be some confusion if you use the term ‘phase’ to refer to a temporal shift in a general waveform, for example the peak latency of an ERP, which is itself the combination of a number of source contributions, each with a range of frequencies and phases. ICA might separate an apparent peak into the sum of two temporally distinct peaks (in two source processes), but our argument (which is testable) is that if the sources are statistically independent (distinct) of each other -- and therefore must also have some degree of functional independence as well -- then the decomposition of the channel ERP peak into these two functionally distinct (and typically individually localizable) IC processes gives more information about the nature of the ERP – including its dependence on situational and physiological condition variables. And again, the separation is accomplished using an instantaneous combination of the (non-delayed) channel signals.

This is completely different from doing a time-domain PCA or other type of linear or nonlinear decomposition. In ICA we are basically trying to design (instantaneous) spatial filters such that the output signals are statistically – and thereby likely functionally independent.

Robert Thatcher (author of the videos): ICA reconstruction distorts or adulterates the phase differences between channels for each and every time point in an EEG recording, including the artifact free sections.  This is a scientific and mathematical and verifiable fact.  There are publications on this topic and it is very easy to demonstrate for oneself.  Dr. Georges Otte and colleagues presented mathematical and empirical proof at the recent ANT meeting in Beau France and they are planning to submit another detailed publication on this topic soon.

  • While it is true that linear superposition can preserve the phase differences in an original 19-channel time series when one uses the 19 ICA components and an inverse solution, like the inverse Fourier transform. However, superposition does not hold when one uses a smaller number of ICA components to create a large number of channels, for example, use 16 or 17 ICA components to create 19 channels. One cannot create something out of nothing.

Arnaud & Scott: No matter what are the multiple sources of phase differences between the original time series, the fact is that when the contributions to the scalp data of some ICs are zeroed out, the ICA reconstructed time series, are (naturally) not the same as the original time series.

  • The reconstructed data, after removing artifactual (non-brain) ICA components differs from the original time series, and because of that there must be differences in any measure of the data. However, the original phase measures are likely misleading because they are contaminated by effects of non-brain source (artifact) processes including eye movements.
  • An example: If ICA teases apart source processes A and B from the source-mixture channel signals, and then we remove B (thereby, for example, removing some non-brain artifact contribution to data data), then yes, the phases of the channel level signals are changed (hey are now exactly the phase of A, since we removed the contributions of (effective) source process B). However, this does not mean that ICA has somehow corrupted the phase.
  • Most accurate and meaningful measures of EEG brain dynamics need to be measured at the effective source level – information that is not directly available in the individual channel signals.

Robert Thatcher: The analyses by Georges and colleagues used real human EEG and standard ICA software and more realistic simulations with much larger changes in phase differences due to ICA reconstruction.  Large changes in phase differences by ICA is best demonstrated by regular type EEG data samples and with more realistic mathematical simulations.  The EEG samples from the Australian workshop are better examples of ICA phase distortion and these examples are commonplace. [My] WinEEG is commercial software used worldwide.  Georges and colleagues as well as myself and colleagues can easily reproduce large phase distortion and will do so in a future publication.

  • ICA feature detection is excellent, the problem is limited to the mathematically "ill-posed" reconstruction when used for the purposes of artifact rejection.
  • ICA is excellent in feature detection and the brain operates by highly efficient sub-clusters of neurons extracting features, e.g., face recognition by combining features like eyebrows, head shape, ears, chin, etc. The problem with ICA is in its use in artifact rejection and then reconstruction of a new time series that results in a new time series that is disconnected from brain network connectivity dynamics of phase shift and phase lock and coherence and cross-frequency coupling and phase amplitude coupling, etc.

Arnaud & Scott: It is important to know which ICA component you removed to understand why this is the case. Are you sure these were artifactual components? Removing brain components may alter the phase of the signal recorded on the scalp (it would be as if you were removing from the scalp signal the contribution of a brain area). Without that information, it is not possible to figure out the origin of the phase difference.

Robert Thatcher: It does not make any difference which components that the scientist/clinicians removed because your own analyses confirmed phase difference distortion by ICA when you removed your [choice of] own components. Up to this point in time you have resounding proven that ICA reconstruction does distort phase differences no matter what reconstruction is used.

It is important to recognize and to publicly accept that phase or time differences between channels in the EEG is due to physiological processes like differences in synaptic rise times, differences in synaptic summation times and differences in conduction velocities, etc. Even a small amount of adulteration or distortion of EEG phase differences is not good and must be avoided at all costs.

You already showed that two ICA component removals results in more phase distortion than the removal of one ICA component. If you were to remove three and then reconstruct and then four and then five, etc and reconstruct and then attach the .edf files and share them with [this] forum then we can plot the magnitude of phase distortion of the artifact free sections of the original record due to the ICA reconstructions.

Arnaud & Scott: There is no need to remove more components - the procedure we use is to identify a (typically) handful of artifact components, remove them -- and then your data is (largely) cleaned of these artifacts. ICA is a linear decomposition that isolates sources that are maximally independent. Exact timing of blink artifact is mostly independent of other ongoing brain effective source activity -- so ICA is typically able to isolate them.

As far as “phase” distortion after removing ICA components in my decomposition, I am not sure what you are referring to. Is it the minute shift when the red and black curves in your data do not exactly superpose? I would argue that the data, after removing ICA artifacts, better reflect the brain activity than before -- and the minute shifts you observe may well be due to removal of potentials generated by small ongoing eye movement activity. (I agree that this should be demonstrated).

The data we are looking at on the scalp is a summation of the activity of billions of neurons, so any scalp signal ‘phase’ we measure is a measure of the sum of all these signals. The phase of the signal at one scalp channel at a given time is not representative of the numberless underlying brain source signals that contribute directly to it. However, phase in each brain effective source signal separated from the EEG by ICA decomposition must represent near-synchronous local field activity throughout some small cortical area (just how small may be difficult to determine), making its signal at the scalp electrodes much stronger than the numberless contributions of spatially desynchronized field potential fluctuations across the wide pattern of cortical territories each scalp electrode position is receptive to. As there are so many of these desynchrous (spatially independent) potentials arriving at each electrode, their individual characteristics are very largely cancelled out in the summed scalp channel signals through ‘phase cancellation’ (their numberless positive and negative values continually summing to near-zero).

… Properly removing artifacts (which may be 10-fold the amplitude of the actual EEG brain signals) using ICA is more important than preserving any exact ‘phase’ measure of the artifact-contaminated signals at any given time. In your analogy of looking at stars, if you have a picture of a star, would you rather remove a visual artifact that is 10-fold the size of your original signal or continue to look at your original signal (not being able to see much because of the large artifact masking most of it).

Exact ‘phase’ measures at any electrode sites cannot be informative about brain activity if these measures are dominated by non-brain source contributions to the recorded scalp data. Differences in phase between any two electrode sites may not be informative either -- there may be dozens of ways for effective source activities within the brain to generate such phase differences. One must move to the effective source level to measure the most relevant information, and this is what ICA decomposition allows us to do.

Robert Thatcher: You write,As far as phase distortion after removing ICA components (in my decomposition), I am not sure what you are referring to.” I am referring to the differences in phase between pairs of EEG channels. The alteration of phase differences [is] present no matter what measure one uses. The least reliable is a visual analysis although there are plenty of visual examples if one carefully reviews the traces.

You write, “I would argue that the data after removing ICA artifacts reflect more brain activity than before, and that the minute shift is due to removal of small eye movement activity.” Myself and many others do not disagree that elimination of artifact is important; what we disagree with is the ICA reconstruction method that adulterates the artifact free segments of the record. Why not simply delete the eye movement manually or like Neuroguide does with a signal detection algorithm that measures the voltage gradients produced by a blink or eye movement, etc? In this way all of the original digital data samples are unaltered.

You write, “The EEG signal is extremely noisy.” The vast number of EEG experts would disagree with you that the “EEG is extremely noisy”. Simply visually examine the EEG traces showing well behaved and well organized alpha rhythms or theta rhythms or beta rhythms which reflect large synchronous LFPs.

You write, “The phase of the signal at one electrode site and one given time is not representative of the underlying brain signal.” This also cannot be true because the phase difference between electrodes and/or sources are produced by the physiological foundations of the brain and networks and are due to differences in synaptic rise times, synaptic integration times, differences in conduction velocity, etc. This is the underlying brain signal and it is highly reproducible and clinically useful. If your belief were valid then there would be no clinical correlations to the EEG such as schizophrenia or ADHD or depression or epilepsy or drug effects, etc.

You write, “… If you have a picture of a star, would you rather remove a visual artifact that is 10-fold the size of your original signal or continue to look at your original signal (not being able to see much because of the large artifact masking most of it).” I agree that the 10-fold size artifact needs to be avoided or eliminated but not by using ICA reconstruction that effects the artifact-free parts of the spectrum and thereby distorts the measurement not only of the one star that you are looking at but also all other stars and planets in the universe.

Stefan Debener: We should not forget that many physiological artifacts are more or less continuous in nature. ECG artifacts for instance are usually not visible in the raw EEG, but they clearly exist - for as long as the subject you are recording from is as alive. So, the important implication is that non-brain artifacts near continuously contaminate the recording, even if you don't see them! Just because they are not visible does not mean that they don't exist. They are not simply on or off, but more or less active, thus they contribute funny activation patterns to the EEG. It follows that an artifact-rejection approach alone may be misleading, regardless of whether it is implemented by visual inspection or using objective criteria). I personally don't believe that there is any such thing as artifact-free EEG recordings. The artifact contributions are just more or less dominant

The ICA model comes with a couple assumptions (like any statistical approach), and if the data do not adhere to the model assumptions then the resulting decomposition may be of very poor quality. In my view, it does not make much sense to praise or condemn a procedure without keeping in mind this fact.

In my opinion, source-level analysis can only confirm predictions (i.e., the result matching your priori expectations); I personally don't trust EEG source activations at unexpected locations, they appear, more often than not, spurious to me. In contrast, the use of ICA for eye blink correction seems much, much better validated to me!

I don't really ‘get’ the concern of ICA messing up the phase of the continuous EEG signals. Of course, a spatial (or temporal) filter will modify the signal, that's its purpose, and if the filter attenuates some portions of the signal, the residual signal may have a different phase, amplitude and/or topography. Because for real EEG recordings nobody knows the ground truth, the question should be whether a filter makes the data better or worse [for some purpose].

The raw data do not qualify very well as the gold standard, because they may be messy (i.e., mixed!). More informative would be a comparison (of, say, two different filter approaches) with regard to a particular effect of interest (say, theta and working memory relationship, or any feature one has sufficient evidence to justify a clear prediction).

Georges Otte: ECG artefact not being visible in clinical EEG is a bold statement that awaits proof as it is quite against my (+ 40 years ) experience in neurological and psychiatric EEG. Furthermore You seem to suggest that EEG is like continuously (even when not visible) contaminated by all kind of artefact -- even EMG-artefacts. I respect this statement as your personal honest opinion, maybe supported by others, but i respectfully dare to doubt that there is a general consensus about this. It is not because EEG is sensitive to artefact due to the amplification factor that all [its] traces are necessarily or unavoidably noisy or contaminated. If care is taken, as we are thought to do in daily EEG practice and how we train our studies, then many artefacts can be prevented. It takes some clinical skills and a technical awareness.

The fact that components must be truly independent (a more stringent condition than uncorrelated), that they must be stationary, and have a non-gaussian pdf amongst others are not known to clinicians. Yet the methods are used in clinical theatre where diagnosis will depend on. Let us agree that it is not because ICA belongs to Blind Source separation that clinicians should be allowed to use it blindly.

Like yourself, I am deeply concerned with phase and phase relationships between channels and I am sure that many scientists share this deep concern. We have many publications in peer reviewed journals where techniques of phase differences, phase slope index, phase shift and phase lag have been linked to several significant neuropsychological endophenotypes such as in ADHD, autism spectrum disorders, intelligence etc.

Stefan Debener: I looked up some [of my] own data and find absolutely no evidence in favor of your ICA phase adulteration claim (see the attached pdf report). I guess you simply used a poor ICA implementation, and/or a poor component selection. The attached example is in full accordance with Arno’s reply, with the difference that I zoom into a clearly visible alpha oscillation, to have a reference brain signal. The example shows no evidence that occipital alpha phase is biased by ICA eye blink correction.

In your own published paper, you say, “If the original EEG/event-related potential (ERP) time series is transformed into a second time series by using the average reference then the original phase differences from three electrode locations may be scrambled and lost. For example, with an average reference the entire surface of the brain is not measured, thus the averaging does not create a true zero potential at each instant of time."

Your phrase, "distortion of the original time series," does make sense only if you believe that the original time series represents somehow a gold standard, something special (something "magic", forgive my poor use of English) that is magically close to the contributions from brain generators to the surface-recorded signal. In my view this is not justified, for instance because it disregards the fact that recording settings will influence how the data will be recorded -- that is, they determine phase and amplitude! Also, attributing something special to the "original" time series is highly misleading, the "original" time series is not closer to the brain signal, in contrast it may be pretty far away from it, not only because [measuring the] phase of mixed brain generators does not make much sense, but also because of all the artifactual influences not accounted for.

I argue that ANY post-recording signal processing that changes the morphology of a time series will change the phase values as well, NOT just the average reference, and NOT just ICA.

The second flaw in your reasoning is that you believe there are "artifact-free" intervals. Now, only because you don't see heart-electrical activity in your "original" 19-ch 10/20 recordings, can you seriously claim that the heart of your participant was not beating during recording? Of course not! All it says is that the influence may be stronger or weaker represented in your recordings (depending on individual differences, and, again, recording parameters).

Ramesh Srinivasan: It is my opinion that all EEG recordings are a mixture of artifact and brain activity. The observation of consistent and reliable EEG results in many experiments reflects the robustness of EEG phenomena, in spite of the presence of artifacts, which vary from laboratory to laboratory and person to person. I think you are very mistaken to think the objective is to measure something [about] relative phase very precisely when in fact the objective is to measure something robust in the face of a noisy recording situation.

Robert Thatcher: If EEG is always full of artifact then such high reliability and robustness would not be published and low statistical effect sizes would be common which they are not and EEG would not be admitted in court and certainly there would not be over 100,000 peer reviewed QEEG studies.  Also, I did not see any ICA reconstruction studies in this search, instead visual deletion of eye movement and other artifact and some used artifact detection routines are used to delete artifact without any type of reconstruction of an entire record and thereby altering the original EEG recording.   This is clear evidence of Robustness of the EEG where robustness is "the ability to withstand or overcome adverse conditions or rigorous testing".  

This list group seems unique in believing that all EEG is so full of artifact that it justifies using methods like ICA reconstruction that create more artifact because what’s wrong with adding more artifact to something that is already essentially 100% artifact? I have not come across this belief system in my career until engaging with the EEGlab list. Obviously, I do not agree with this opinion. In the short term it is ok to have such a belief because science relies upon hypothesis testing and cross-validation and verification and independent testing and therefore this belief system will not have a long-term impact on the science of EEG and hopefully will not waste taxpayer money or set back the young minds of students trying to learn about EEG.

Unique and dramatic ICA reconstruction is different than only ICA decomposition, it goes a step further and falsely assumes that the physiological sources of the EEG are “independent” and proceeds to create an alternate and artificial reality by attempting to create something out of nothing. Georges and others mean “something out of nothing” by removing one or more components and with this lower dimensional space then create a new time series that replaces the original full dimensional time series. There is a mathematical loss of information by this process that results in alterations of all the phase differences in the physiological sources of the original ground truth and thereby de-coupling the physiology of the brain (i.e., Conduction velocities, synaptic delays, synaptic integration times, etc) from the ground truth of the brain.

The defense by the users of ICA reconstruction is a false equivalence between the ground truth of the EEG and their ICA “alternate truth” that replaces the ground truth. Further justification for adulteration of the ground truth is the false claim that the EEG is random and full of artifact and therefore there is no harm done.

Arnaud & Scott: None of the articles you have mentioned [in a list not reproduced here] show that studying EEG phase on the uncorrected raw data is a better presentation of the brain activity than other approaches (involving ICA or not). In fact, none of the articles you mention compare between methods.

Marius Klug: Robert Thatcher and others on his side of argument several times spoke of "artifact-free data segments" which would be distorted by taking out an ICA component, even if it was artifact-free. The argument here is -- and I can't believe that I must write this again, since it's been written so often already -- that no such data exists, end of story.

EEG is just an electrical recording. Electrical signals are abundant and generated constantly during the recording by a wide number of sources, not only [those arising in] the brain. One example here being heart beats. Now obviously those signals do have an amplitude and a phase, and since they are super positioning the brain signals and thus part of the EEG recording, they will - no way around this! - have an impact on the phase (and amplitude) of the data set. The data set is thus _continuously_ distorted by the artifact. So by saying that a data set has artifact-free time periods, Robert implies that the heart has stopped beating

This applies not only to heart beats but to each and every single artifact source that can be recorded by the EEG! So, even if you have no super strong eye blink artifacts, the eyeballs and eye muscles will still continuously contaminate the data - a bit at least. Since the IC for eyes in a 19-channel data set does likely contain both blinks and movements, taking it out will also take out the continuous eye movement artifacts produced by all kinds of drifts, saccades, and micro saccades, so it will take out parts of the data in the complete set! Now, since the eye signal again has a phase and an amplitude, the resulting signal will be distorted - but to the better, not to the worse! The new signal does not contain the artifacts of the eyes, AT ALL TIME POINTS, at least to the degree that ICA was able to separate them. So yes, the phase will be distorted, AND THAT IS A GOOD THING!

In fact, the discussion about his data began when Arno could clearly show that taking out eye movement artifacts did NOT distort phase to a relevant degree in times other than eye artifacts occurring in one of his earlier emails. How Robert can interpret those figures to suit his own argument is a mystery to me. Arno was, so to say, not able to replicate your bug and then the list proceeded to search for other reasons, but we were stuck by the fact, that there was no movement on Roberts side, no careful examination of the facts and arguments that have been laid out.

I can't believe that I have to write this again: YES, TAKING OUT ICs WILL ALTER PHASE! This is not even an argument! The thing is a) that this is something good for the time points where the artifacts occur, because in fact the original data has been distorted by a the artifacts, eye in this case, and taking out the artifacts will restore the more correct brain signal phases and b) the extent and spatial distribution of the distortion presented by Robert led to the conclusion that something must have been done incorrectly or at least not with a lot of scrutiny.

You cannot seriously argue that physiological signals all-of-a-sudden cease to exist every now and then and then come back to life again later. Let me be clear:

As long as your subjects have eyes that move, muscles that work, sweat glands on their skin, and a beating heart, there are biological artifacts in your EEG that continuously contaminate your data by superpositioning with the brain signals. And as long as you have electricity in the house that you record in, especially if it is close to the electrodes, and no faraday cage, you have other noise and artifacts that continuously contaminate your signals.

And the best thing is: In times where this is not the case, the continuous activation of the IC components would be zero (as long as the decomposition is perfect - otherwise at least very low) and there would be no data alteration at those times if you subtract that IC!

Clayton Hickey: The idea that EEG can be free of eye movement artifacts is something that can be empirically studied… what is needed is an independent measure of eye movements. Many of us have this in concurrent eye-tracking data. For the fun of it (procrastination, get thee behind me) I just had a quick look at a few datasets, extracting intervals where participants maintained fixation. I ran infomax ICA on this data and identified ‘artifactual’ components as those that correlated with the eye-tracking data.

I consistently got at least a couple of components that strongly correlated with the tracker signal. This in spite of the fact that the data was putatively ‘artifact-free’ and collected during consistent maintenance of fixation. So... the eyes are doing something to the EEG signal during the maintenance of fixation, and ICA can pick this up. This seems to be primarily artifact from muscle / eyeball rotation.

Joseph Dien: I’ve been working on an artifact correction manuscript with an EEG plus eye-tracker dataset.  The corneo-retinal dipole (CRD) artifact reflects eye position not eye movement so it’ll be constantly present even if the eyes maintain fixation.  My take is that each artifact is best corrected with a method tailored to its unique characteristics.

Robert Thatcher: No regression method can remove the superimposition and it is best to simply use improved recording hygiene methods and delete and sections of the EEG record that have artifact no matter what the artifact is. Most scientists insist that one must preserve the physics of the brain at all costs and not conduct analyses based on an altered time series. There is no equivalence of the physics ground truth by an artificial replacement base on a hypothesis or belief that all EEG is noise or artifact.

Joseph Dien: The CRD artifact is always present.  There’s no such thing as artifact-free recordings.  It’s a matter of relative costs and benefits.  Do you lose more or gain more via artifact rejection or artifact correction?  That depends on your dataset and what you wish to do with it.