If you are planning to buy a eye tracker, you might have come across the varying sampling frequencies that different eye trackers are capturing gaze data at. Like Accuracy and Precision, higher sampling rate is often seen as better for data validity. However, this is not entirely the case.
Sampling frequency refers to how many times per second the eye position is measured. For example, a 120 Hz eye tracker collects 120 data points for each of the tracked eyes. In order to understand which sampling rate you need for your eye tracking study, let us first take a look at a few types of eye movements and events that are typically studied.
- Fixation lengths (or duration) typically varies from 100 – 800 milliseconds. During these fixations, the brain starts to process the visual information received from the eyes. This is the time when most information from the scene where a person is looking at is acquired. Fixation lengths usually gives an indication of information processing and cognitive activities.
- Fixation patterns can also indicate certain cognitions e.g., many short sporadic fixations can illustrate confusion, random searching or a lack of content deemed interesting or useful.
- Saccades are extremely fast movement (“jumps”) of the eyes from one fixation to
another. During a saccade, vision is largely suppressed. This is known as “saccadic
- The end point of a saccade is “decided” before it occurs and it cannot be changed in the middle of the movement.
- Saccades typically range in duration of 10-40 milliseconds, depending on the distance travelled by the saccade.
- Smooth pursuit is one of the 2 ways that visual animals can voluntarily shift their gaze, besides saccades. Smooth pursuit allows the eyes to closely follow a slowly moving target, while preserving the stability of the visual image of the target during eye movement. Any other objects beside the target are poorly processed.
- Smooth pursuit typically range from 90 – 150 milliseconds, but is largely limited to the angular velocity of the moving target. The angular velocity limit of smooth pursuit is typically ~30o/sec, and if the target is moving beyond this limit, “catch-up saccades” would occur.
- Smooth Pursuit is asymmetrical. We typically perform better when we follow objects moving horizontally, than vertically. Even for vertical pursuits, humans are typically better at downward than upward pursuit.
- Vergence occurs when both the left and right eye move in opposite directions
- Vergence can involve either a convergence (when object is getting closer) or divergence (when object is getting further) of the lines of sight of each eye.
- Vergence is the slowest speed eye movement (about 20 times slower than saccades) and are rather small in angular amplitude (typically only a few degrees for each eye)
Vestibular Ocular Reflex (VOR)
- VOR is a reflex eye movement that goes contrary to head movements, so as to keep the image of the object in fixation more or less the same place on the retina.
- VOR is one of the fastest reflexes in the human body, with eye movements lagging the head movements by less than 10 milliseconds.
With the main eye events covered, let’s go back to the sampling frequency of eye trackers.
Higher sampling frequency produces better temporal resolution (or accuracy). Lower temporal resolution would mean higher sampling errors. In fact, the average sampling error approximately equates to half the duration of time between samples.
For example, a sampling rate of 1000Hz (1 data point every millisecond) would produce an average error of approximately 0.5 milliseconds, while a sampling rate of 60Hz (1 data point every 16.67 milliseconds) would produce an average error of approximately 8.3 milliseconds. Based on the typical duration of the eye events above, an 8.3 milliseconds error might be considered too large to study saccades and VOR, but not too significant for fixations.
Be mindful that more sampling errors would lead to more “noise” in your data. There are 2 main ways to mitigate data “noise”. One is to collect more data (participants), or alternatively, you can use a higher sampling frequency for your eye tracking study.
Now let us zoom a bit into fixations. For fixations, you can be interested in where people look (fixation location), and also for how long (fixation duration). For the former, a higher sampling rate does not necessarily gives you better accuracy for identifying gaze location than a lower sampling rate. However, for the latter, a higher sampling rate can give you more precision in identifying when a fixation starts and when the fixation stops.
Given that eye tracking systems cost higher for higher sampling rates, how high do you really need to go? As you might have known, sampling errors can never be completely removed entirely, but they reach insignificance below a certain point. The level of this point depends on the type of eye events you are studying, and the research rigor in your domain of research. For example, researching saccades, VOR or other micro eye movements would require sampling rates of 250Hz and higher, but they are usually only of interest in the academic domain such as neuroscience. For vergence and smooth pursuit, a sampling rate of at least 120 Hz would be preferable.
In the UX domain, the general rule of thumb is to use sampling frequency of 60 – 120 Hz, as most of the time we are only interested in fixations, and sampling errors of around 10 milliseconds for fixation duration are still considered acceptable. If you are only interested in measuring where people are looking at, a 30Hz eye tracker would most likely suffice.
For a more in-depth discussion on eye tracking sampling frequency, watch the Tobii Pro Webninar below:
For more eye tracking essentials, visit Tobii Pro’s Learning Centre to optimize what you can do with your Tobii Pro eye tracker!
This is a list of eye trackers that Tobii Pro have to offer:
Screen-based Eye Trackers
Wearable Eye Tracker
Alternatively, you can visit Objective Eye Tracking for a comparative glance.
If you would like to know more about eye tracking and how it can help you in your research, drop us an email at firstname.lastname@example.org or call us at +65 67374511.