Spatial representations of the body
2.5 ERP data analysis
2.5.4 Artefact rejection
2.5.4 Artefact rejection
After the data had been offline filtered, segmented and baseline corrected, each trial from each participant was visually inspected for noise and artefacts. This process is required because the EEG net not only records electrical activity from the brain, but also picks up electrical activity from
other sources (such as the skin and muscles). There are four main artefacts of concern: blinks, eye movements (saccades), alpha activity and movement potentials. Although, it must be stated that it takes a considerably larger number of these artifacts to contaminate infant ERP data than it does adult ERP data (Nelson, 1994).
2.5.4.1 Blinks
Participants’ eye blinks can greatly contaminate ERP recordings (particularly at the frontal regions). This contamination can have an experimentally confounding effect especially if participants blink systematically in response to a specific stimulus and/or experimental condition. Eye blink artifact contamination of EEG occurs due to the fact that the eyeball itself acts as a dipole, with the positive pole at the front of the eye and the negative pole at the back of the eye. When individuals blink, as the eyelid passes over the eyeball, this results in a sharp negative-going deflection in voltage, which can be observed across frontal electrodes.
The Hydrocel Geodesic Net allows blinking to be monitored as it has electrodes that are placed around the ocular socket. Whilst the adult EEG nets contain electrodes that are fixed both above and below the eye regions, due to issues of comfort and irritation the infant EEG nets only contain electrodes that sit around the top half of the eye region. Using recordings from these electrodes, it is still possible to inspect the data for
eye blink contamination and to then discard contaminated channels and/or trials. A reprieve to the problem of eye blink contaminations is granted to infancy researchers, given that not only do infants tend to blink much less than adults, but infant ERPs tend to be much greater in voltage amplitude (in comparison to adult ERP recordings, contributing to a higher signal to noise ratio). Additionally, researchers have shown that a considerably greater amount of eye activity is needed to contaminate the neural activity of interest in infant ERPs (Nelson, 1994).
2.5.4.2 Eye movements
One of the crucial constraints of infant EEG is that researchers cannot instruct participants to behave in a certain way. Specifically, we cannot instruct infants to fixate on a stimulus in their visual field, so as to minimise eye movement artifacts in the data. Additionally, infants (compared to adults) may be more inclined to visually explore the novel environment of a testing room. As such, saccadic activity is more prominent in infant ERP data than in adult data.
We undertook a number of measures to minimize the occurrence of eye movements. Firstly, the testing room was dimly lit so as to reduce the extent to which features in the room attracted the infants’ attention. Also, throughout testing an experimenter was facing the infant in all of the trials. This researcher adopted a direct eye gaze so to capture the infant’s gaze and employed engaging facial expressions and speech to sustain their
attention. The implementation of this procedural consideration differed slightly across experiments, for reasons that are discussed in more detail in the respective experimental chapters (see Chapters 4 and 6).
Nonetheless, even with these measures in place, it was not possible to eliminate eye movements entirely. Therefore, upon visual inspection of trials, if saccadic activity was present in neural recordings, any electrode channel that contained this activity was marked as ‘bad’. Again, due to the physiological differences between adults and infants (thinner skulls and less dense cell packing in the brain), it takes much more saccadic activity to contaminate infant ERPs as infant ERPs give rise to a higher signal to noise ratio (see Nelson, 1994).
2.5.4.3 Alpha waves
Alpha waves are characterized in the EEG by their sine wave morphology with a frequency of approximately 10 Hz (for infants in the first year of life it is around 7 Hz; Smith, 1941) and usually occur when participants are tired. Whilst investigating the continuous alpha rhythm (in the form of oscillatory activity) has been of interest, especially in infants (for a review see Marshall, Bar-Haim & Fox, 2002), the alpha waveform can be problematic if it is present in ERP data. This is due to the fact that the alpha rhythm can become entrained to the stimulus onset and thus averaging across trials in a specific condition does not reduce alpha contamination.
A simple way to minimise alpha contamination is to ensure that participants are well rested. This is not straightforward with infant participants as tiredness can occur quite suddenly and without warning.
Nonetheless, I scheduled testing sessions around infant’s natural schedule of nap times in order to reduce the incidence of alpha.
However, in some participants, alpha rhythm may not be related to tiredness and these participants tend to have substantial alpha activity even when alert. Due to the low frequency of alpha waves (between 7 and 10 Hz), it is difficult to eliminate alpha activity using filters alone.
However, a further simple consideration that can greatly reduce alpha activity relates to the design of the study. In order for the alpha rhythm to become entrained to stimulus onset, the stimulus needs to occur at a constant rate. If, however, a jitter (inter-stimulus interval) is introduced so that there is a random time difference between stimulus presentations, this reduces the likelihood of alpha waves becoming stimulus locked (Luck, 2005). As such, I used a temporal jitter in all ERP experiments (for specific details of this, see Chapter 4, Sections 4.2.1.3, 4.3.1.5 and 4.4.1.4 and Chapter 6, Section 6.2.2).
2.5.4.4 Movement potentials
These artefacts arise from any type of muscle activity and are particularly common and problematic in infant ERP research. In order to reduce the occurrence of movement artefacts, parents were instructed to securely
hold their child around the waist and not to bounce their legs.
Additionally, the researcher took hold of the infant’s arms and gently held them in place, discouraging any arm movements.
Of course, even with these procedures in place, infants would still engage in head turning or body movement behaviours. Upon visual inspection of the data, trial-by-trial, it was particularly obvious when infants had shifted their body posture as this would result in recorded activity with disproportionally large amplitudes (usually below -100 or above 100 microvolts) and thus did not reflect true neural activity.
Typically in these situations entire trials were marked as ‘bad’ and eliminated from further analyses. Whilst entire trials could be contaminated by, for example, movement artifacts and then rejected, there were certain instances in which only a few electrodes in a trial were noisy and had to be removed. I discuss below how I dealt with this situation.