Knowledge Base

Interpretation of Phasic EDA Responses

Phasic EDA responses are invoked by the sympathetic branch of the ANS. This part of the ANS regulates adaptive involuntary responses to physiological and psychological stressors in preparation of a fight-or-flight response. Sweating is one such adaptive physiological response. The frequency of phasic EDA responses indicates the magnitude of sympathetic ANS activation. This explains why momentary phasic response frequency is associated with self-reports of momentarily experienced anxiety (Strohmaier et al, 2020). As the rate of involuntary phasic EDA responses is proportional to levels of experienced emotion or anxiety, it has gained popularity in psychological studies of unconscious or unexpressed emotional responses to anxiety triggers. However, it should be noted that EDA does not relay information on the quality of the emotion that is experienced.

The phasic EDA peak constitutes an isolated single peak. The number of phasic EDA responses per minute is used as a measure for the magnitude of the ANS activation during the measurement period. In practice, multiple EDA responses are stacked if these are triggered at high frequencies. Stacking of phasic responses may be observed for phasic responses triggered at high frequencies. Subsequent stacked peaks typically have lower amplitudes because of the physiological constraint posed by the maximum achievable diameter of sweat ducts. However, these smaller stacked peaks reflect similar elevations of ANS arousal as any other phasic peak.

Signals Present in NoldusHub

In NoldusHub, the available EDA signals are:

• EDA peak rate - the number of phasic skin conductance responses per minute (unit-free).
• EDA tonic - the skin conductance baseline (in µS).

NoldusHub EDA Algorithms

NoldusHub automatically performs a number of data processing steps to obtain tonic EDA and peak rates:

1. Short-lived data artefacts in the raw data are removed by median filtering.
2. The tonic EDA signal is assessed.
3. To obtain baselined phasic responses, the tonic EDA signal is subtracted from the raw signal.
4. The theoretical stimulus onset times that invoked phasic EDA responses are identified on the basis of phasic response shapes. To achieve this, the prototypical shape of a human phasic EDA response to known visual stimuli (Bach et al., 2010) is used as the point-spread function in one-dimensional Wiener deconvolution.
5. For each theoretical stimulus onset time, the subsequent EDA peak-time is obtained at maximum amplitude.
6. The average phasic peak rate per minute is calculated from the identified peak times.

Why Deconvolution as a Peak-Detection Algorithm?

Deconvolution allows for the detection of stacked EDA peaks and reveals the point stimulus event times that invoked the phasic EDA responses. These stimulus events may have been external (stimuli) or endogenous (psychological events). Therefore, not every identified phasic response has a stimulus onset time that coincides with an actual external stimulus event.

An additional advantage of deconvolution is that it is not sensitive to differences in peak amplitude. This also allows for better detection of trains of stacked EDA peaks.

Artefact peaks that do not resemble phasic EDA responses are not identified as peaks by deconvolution, in contrast with algebraic peak-detection algorithms.

Keep in Mind

To align with common practice in the scientific community, the EDA peak rate calculated in NoldusHub uses peak times, not stimulus onset times.

To identify phasic peaks using deconvolution, a peak should be fully revealed in the conductance measure because the peak is recognized on the basis of its entire shape. NoldusHub offers a real-time analysis of phasic EDA responses. However, the 30-second duration of typical phasic EDA responses explains the delay observed for EDA signals in the Live data viewer. The same EDA visualizations in Replay mode are in-sync with all other signals.

EDA References

• Bach, D.R.; Flandin, G.; Friston, K.J. and Dolan, R.J. (2010). Modelling Event-Related Skin Conductance Responses. International Journal of Psychophysiology, 75: 349–356.
• Strohmaier, A.R., Schiepe-Tiska, A. and Reiss, K.M. (2020). A Comparison of Self-Reports and Electrodermal Activity as Indicators of Mathematics State Anxiety. An Application of the Control-Value Theory. Frontline Learning Research, 8: 16–32.

Heart Rate Data

The availability of different heart rate data signals depends on your version of NoldusHub (v1.4 or v1.6).

Two Methods to Measure Heart Rate Data

NoldusHub measures heart rate data based on two methods that give similar results but have important differences. The two methods are:

• Photoplethysmography (PPG)
• Electrocardiography (ECG)

ECG

ECG determines the heart rate by measuring the electrical signals coming from the heart. It is a reference signal for health assessment and for research. For measurement, several electrodes are placed on the chest. Each electrode has its own signal, which can be found in NoldusHub under the names Lead I, Lead II, and Lead III.

PPG

PPG determines heart rate by measuring changes in light reflection or transmission through the skin due to blood volume changes. It is generally measured with a single sensor on a finger or an ear lobe.

Raw and Derived Signals

Due to the fact that blood must travel through the body, the peaks in the PPG signal are later than the peaks in the ECG signal.

• Heart rate is the number of peaks per minute.
• Interbeat interval (IBI) is the number of milliseconds between two heartbeats.

Differences Between ECG and PPG Data

• ECG: Direct measurement via electrical current from the heart; reference signal for health; invasive with several sensors on the chest; accurate; sharp peak enabling accurate pulse detection; data obtained accurately using a small time window and real-time signal.
• PPG: Indirect measurement via reflection by blood volume; uses ECG as ground truth for research; non-invasive with one sensor on ear lobe or finger; sensitive to noise and movement; broad peak making pulse detection less accurate; requires a time window of several minutes for accurate data.

Signals Present in NoldusHub - Heart Rate

PPG Signals

• Heart rate (BPM): The number of peaks (heartbeats) per minute.
• Interbeat interval (ms): The number of milliseconds between two heartbeats.
• Raw data: The raw PPG signal.

ECG Signals

• Heart rate (BPM): The number of peaks (heartbeats) per minute.
• Interbeat interval (ms): The number of milliseconds between two heartbeats.
• Lead 1 (mV): The ECG vector signal measured from RA to LA.
• Lead 2 (mV): The ECG vector signal measured from RA to LL.
• Lead 3 (mV): The ECG vector signal measured from LA to LL. This is derived by subtracting Lead 1 from Lead 2.
• Vx (mV): The ECG vector signal measured from the average of the RA, LA, and LL to the V electrode.

For more details on the Lead and Vx signals, consult the Shimmer ECG user guide.

Heart Rate Data Processing

Heart rate data are preprocessed in the following way:

1. High pass filter: Data below 0.5 Hz is not included.
2. Notch filter: To remove artefacts caused by the power frequency, the 50 Hz (Europe) or 60 Hz (USA) frequency band is not included.

Eye Tracking Data

The availability of different eye tracking data signals depends on your version of NoldusHub (v1.4, v1.6, or v1.7).

Gaze Overlay

Gaze data is visualized in NoldusHub through the gaze overlay on the screen video in the Screen Video pane inside the Record and Replay page. Gaze data consists of two parts: fixations and saccades. Fixations are visualized with a pink dot; the line following the dot represents the saccades.

What Are Fixations and Saccades?

The moments that our eyes focus on something are called fixations. The duration of fixations varies between roughly 50 ms and 1 second. During a fixation, gaze points sampled by an eye tracker are close together. Fixations are alternated with saccades, which are rapid eye movements from one fixation to another. Saccades last much shorter than fixations, roughly 30 to 80 ms.

How Does NoldusHub Calculate Fixations?

NoldusHub uses the I-DT (Identification by Dispersion-Threshold) fixation algorithm published by LC Technologies, which is a variant of the fixation algorithm described by Salvucci and Goldberg (2000). In NoldusHub, a fixation is recorded if gaze points are within 20 pixels of the fixation center for a minimum duration of 5 samples. For an eye tracker that samples at 60 Hz, this corresponds to 83 ms.

The I-DT Fixation Algorithm in NoldusHub

The I-DT fixation algorithm makes use of two thresholds:

• Dispersion threshold: Determines the maximum distance between a gaze point and the average of all gaze points in a fixation. The default dispersion threshold in NoldusHub is 20 pixels.
• Duration threshold: Determines the minimum duration that gaze points should be within the dispersion threshold to become a fixation. The default duration threshold in NoldusHub is 5 samples. For an eye tracker sampling at 60 samples per second, this means 83 ms.

The algorithm proceeds as follows:

1. At t1, five gaze points are recorded. They are all within the dispersion threshold. The duration threshold is exceeded and a fixation is recorded.
2. At t2, another gaze point is recorded. The average gaze point moves slightly. All gaze points are still within the dispersion threshold and belong to the same fixation.
3. At t3, another gaze point is recorded. The average gaze point moves. The distance between the new gaze point and the average gaze point exceeds the dispersion threshold. The fixation ends and a saccade is recorded.

Gaze Point Validity

NoldusHub calculates the gaze point using the average of both eyes and takes into consideration the validity of samples received from the eye tracker. Validity is defined as either Valid or Invalid. NoldusHub discards samples where the X or Y value is NaN:

• If samples for both eyes are valid and none of the values are NaN, the average is calculated.
• If the validity for one eye is Invalid or there are NaN values, the other eye (valid and without NaN values) is used.
• If both eyes have NaN values or are invalid, the gaze point is set to a fixed [NaN, NaN] point.

About Gaze Data Coordinates

Fixation and gaze points are positioned in a 2D field on the user display area. The user display area can be a computer screen or, for example, an area of a shopping shelf as seen through mobile glasses. NoldusHub uses the Tobii Pro eye tracker reference system. The 2D field is mapped to normalized coordinates for X and Y (floating point values between and including 0.0 and 1.0). Point (0, 0) denotes the upper left corner and point (1, 1) indicates the lower right corner.

• Fixation X: The X coordinate of the fixation center, a floating point value from 0.0 to 1.0.
• Fixation Y: The Y coordinate of the fixation center, a floating point value from 0.0 to 1.0.
• Fixation status: Can be undefined, fixation, or saccade. Indicates whether the eye is moving or not. If the eye is moving, the status is saccade. If the eye is not moving, the status is fixation. Undefined is shown if it is unknown whether the eye is moving or not.
• Gaze X: The X coordinate of the gaze point, a floating point value from 0.0 to 1.0.
• Gaze Y: The Y coordinate of the gaze point, a floating point value from 0.0 to 1.0.
• Validity: Can be valid, invalid, or indeterminate. Indicates the validity of the gaze origin data.

Pupil Size Data

Besides gaze coordinates, Tobii eye trackers also produce pupil diameter data. Tobii eye trackers calculate pupil diameters (in millimeters) from images of the eyes. NoldusHub uses these pupil diameters, in combination with light intensity measures, to calculate the relative contributions of light and cognitive processes to the pupil diameter. All these measures are available in the Cognitive Load signal group.

Eye Tracking References

• Salvucci, Dario D., and Joseph H. Goldberg. "Identifying fixations and saccades in eye-tracking protocols." Proceedings of the 2000 Symposium on Eye Tracking Research & Applications. ACM, 2000.
• Voßkühler, Adrian, et al. "OGAMA (Open Gaze and Mouse Analyzer): open-source software designed to analyze eye and mouse movements in slideshow study designs." Behavior Research Methods 40.4 (2008): 1150–1162.
• Blignaut, Pieter, and Tanya Beelders. "The effect of fixational eye movements on fixation identification with a dispersion-based fixation detection algorithm." Journal of Eye Movement Research 2.5 (2009).
• Camilli, M., Nacchia, R., Terenzi, M. and Di Nocera, F. ASTEF: A simple tool for examining fixations. Behavior Research Methods, 40.4 (2008): 373–382.
• Shic, F., Chawarska, K. and Scassellati, B. The incomplete fixation measure. Proceedings of the 2008 Symposium on Eye Tracking Research & Applications, ACM, 111–114, 2008.

Cognitive Load

Cognitive Load in NoldusHub

NoldusHub uses pupil sizes and ambient light fluctuations to estimate cognitive load. Ambient light fluctuations are estimated from the light intensities of the participant screen when using a remote eye tracker bar. When using head-mounted eye trackers, the forward-facing camera is used to measure ambient light intensity. Eye trackers are also used to measure pupil sizes. Therefore, cognitive load is only calculated when using an eye tracker and capturing the participant screen or using forward-facing cameras. Cognitive load calculations require heavy usage of system resources.

Source: NoldusHub 1.8 Reference Manual