How to Choose Eye Tracking Calibration Settings
Before running a webcam eye tracking study in Labvanced, a handful of calibration decisions determine both the accuracy of your gaze data and how long calibration takes for each participant. All of these settings live under Settings tab → Physiology → Eye-Tracking.
Table of Contents
Choosing a Tier: Effort, Screen-Outs, and the Trade-off Between Them
As of Physiology Toolbox Version 2.0, turning on eye tracking from any entry point, the study settings toggle, a task's own eye tracking setting, or adding an event trigger that needs gaze data, opens the same setup wizard. The wizard first asks who your participants are, Adults / Adolescents or Infants / toddlers, then asks for a tier (accuracy vs. effort). Only for the tiers that support it does a second, independent choice follow: a screening level (accuracy vs. screen-out); the rest apply the tier directly with no second step. The implications of each choice, and the numbers behind it, differ by participant type.
Turning up either knob, a higher-effort tier or a stricter screening level, improves the average accuracy of your recorded gaze data, but each spends a different resource to do it. Picking the lowest option on both, a Passive tier with no screening level at all, gives up on accuracy entirely in exchange for asking nothing of the participant. A higher-effort tier spends participant time and patience: every admitted participant sits through a longer, more demanding calibration. A participant who genuinely cannot produce a usable calibration can still fail outright once they exceed the tier's retry limit, that is a basic quality floor every tier applies, not a deliberate quota. A stricter screening level adds that quota on top: it spends your recruitment budget by deliberately turning away a target share of the people who start your study, even ones whose calibration technically succeeded. Labvanced's own setup wizard prices this directly: a screened-out participant costs only the screen-out fee (for example, 0.5 to 0.8 GBP on Prolific), and setting up a screen-out link in your study settings makes sure that participant is actually compensated rather than simply dropped.
In practice, the two choices can substitute for each other. If you need high accuracy and are recruiting on a panel where screen-outs are expensive, raising the tier (more effort, same screening level) buys accuracy without turning away more participants. If your participants are effort-sensitive (children, clinical populations, an already-long session) but your budget tolerates some data loss, keeping a low-effort tier and raising the screening level instead filters for accuracy after the fact rather than asking more of every participant up front.
Adults / Adolescents
This is the participant type most studies use: anyone old enough to follow spoken or on-screen instructions and sit through a normal calibration.
Accuracy vs. Effort (Adults / Adolescents)
Each tier bundles calibration length, chinrest strictness, and gaze control frequency into a single choice, more accuracy means asking more of the participant. The Screen-out Option column below is explained in the next section:
| Accuracy | Effort Level | Description | Best For | Screen-out Option |
|---|---|---|---|---|
Passive: Lower Accuracy | No Effort | No instructions, no calibration, no chinrest. Runs passively in the background. | Best when participants should have no effort and must never be interrupted. Eye-tracking accuracy is secondary. | ❌ |
Fast: Medium Accuracy | Minimum Effort | 30 second calibration, no gaze control, loose chinrest. | Best when setup time must be minimal, but accuracy must be better than in the passive/base model. | ❌ |
Immediate: Good Accuracy | Limited Effort | 1 minute calibration, gaze control, medium chinrest. | Best when participant effort must be rather limited, so there is no room for a 2 minute calibration, but good accuracy is still required. A good choice for studies where effort and accuracy are equally important. | ✅ |
High: High Accuracy (recommended) | Moderate Effort | 2 minute calibration, gaze control every 3 minutes, medium chinrest. | High gaze accuracy with acceptable participant effort. The right choice for studies where high gaze accuracy is crucial but participant effort should stay moderate. Recommended for most academic research. | ✅ |
Very High: Highest Accuracy | High Effort | 4 to 5 minute calibration, gaze control before every trial, medium chinrest. | Best when gaze accuracy is the most important and participant effort is secondary. Participants need to be patient and are expected to keep their head still at all times. | ✅ |
Beyond a longer calibration up front, higher tiers also run gaze control checks more often during the study and enforce a stricter chinrest. A gaze control check interrupts the session with an on-screen prompt: "Look at the black circle. Adjust your head slightly so the red dot covers it, the better the overlap, the faster the countdown," and the participant has to actively hold that position before the study continues. A stricter chinrest setting narrows how far the participant can drift from the calibrated head position before the study pauses outright, so the higher the tier, the sooner a given movement triggers that pause. A Custom tier is also available, it applies none of these bundled presets and leaves every setting on this page editable individually.
Accuracy vs. Screen-out (Adults / Adolescents)
The second question is independent of the tier: how strictly to screen out participants whose calibration accuracy falls short. Only the tiers marked ✅ above support it, Passive and Fast have no calibration accurate enough to screen on, so they skip this step entirely and finish on the tier choice alone. Immediate, High, and Very High each preselect a starting level (High, for example, starts at Light), but the level can be changed on its own. Expected Accuracy is the average gaze error reported as a percent of the screen diagonal, lower means more accurate. The figures below are shown for High, the recommended tier, the next section explains how they shift on a different tier:
| Screen-out Level | Admitted (of 100) | Expected Accuracy (% of Screen Diagonal) on High |
|---|---|---|
Level 0: No screening | 100 | ~6.5% |
Level 1: Failed calibrations only | ~85 | ~6.0% |
Level 2: Light | ~65 | ~5.6% |
Level 3: Balanced | ~50 | ~5.2% |
Level 4: Strict | ~30 | ~4.9% |
Level 5: Very strict | ~15 | ~4.4% |
Expected Accuracy: How the Two Choices Combine (Adults / Adolescents)
Choosing a tier in Step 1 (Effort) sets a baseline for your expected accuracy. If that tier supports a screening level, Step 2 (Screen-out) lets you control accuracy further, on top of that baseline, by choosing how strictly to screen participants. On High, for example, the baseline at Level 1 (Failed calibrations only) is about 6.0%; tightening to Level 4 (Strict) brings it to about 4.9%, without changing the tier at all. The tier alone also moves the baseline: at that same Level 1, Very High starts from a better baseline than High, about 5.2% instead of 6.0%.
Infants / toddlers
This participant type is for studies with infants or toddlers who cannot be instructed to sit still or fixate a dot on command, so every tier here replaces spoken instructions with animal icons, sounds, and attention-getters instead.
Accuracy vs. Effort (Infants / toddlers)
Each tier bundles calibration length, chinrest strictness, and gaze control frequency into a single choice, the same shape as the adult tiers, but the ladder itself is different: the lower infant tiers use a brief animal attention-getter before each trial instead of a timed calibration, and only the top two tiers add an enforced chinrest and gaze control. The Screen-out Option column below is explained in the next section:
| Accuracy | Effort Level | Description | Best For | Screen-out Option |
|---|---|---|---|---|
Passive: Lowest Accuracy | No Effort | No instructions, no calibration, no chinrest, nothing between trials. Runs entirely in the background. | Best for 2-alternative preferential looking paradigms and other left vs. right or two-to-four region designs. | ❌ |
Attention: Low Accuracy | Minimal Effort | No instructions, no calibration, one animal attention getter before each trial with a very loose chinrest. | Best for left vs. right or two-to-four region designs where each trial should start from a roughly known gaze position. | ❌ |
Medium: Medium Accuracy | Limited Effort | About 40 seconds of infant friendly calibration, attention getter before each trial with a loose chinrest. | Best for designs with 2 to 6 larger stimuli. The infant sees 5 animal images before the first trial to fit a correction, rather than relying on the general model. | ❌ |
Good: Good Accuracy (default for infants) | Moderate Effort | About 90 seconds infant friendly calibration, loose chinrest during trials, attention getter before each trial. | Best for designs where gaze accuracy is more important. Requires the infant to sit still and pay attention for about 90 seconds. | ✅ |
High: High Accuracy | Higher Effort | About 90 seconds infant friendly calibration, medium chinrest during trials, attention getter before each trial, gaze control. | Best for designs aiming to track smaller or many regions on the screen. Good for older kids who can follow instructions and are able to hold still through the trial. | ✅ |
Accuracy vs. Screen-out (Infants / toddlers)
Only the tiers marked ✅ above support it, the same way Passive and Fast skip it for adults, but Good and High do not support it equally. Good does not run Gaze Control, so it can only reach two of the six levels; the other four need a Gaze Control probe to score against, which only High runs. High is the only infant tier that permits the full six-level range, the same way every screenable adult tier does, see Expected Accuracy below for how that comparison plays out. Expected Accuracy is the average gaze error reported as a percent of the screen diagonal, lower means more accurate. The figures below are Good's own two options:
| Screen-out Level | Admitted (of 100) | Expected Accuracy (% of Screen Diagonal) on Good |
|---|---|---|
Level 0: No screening | 100 | ~11.7% |
Level 1: Failed calibrations only | ~85 | ~10.8% |
Expected Accuracy: How the Two Choices Combine (Infants / toddlers)
Choosing a tier in Step 1 (Effort) sets a baseline for expected accuracy, scaled to how much harder infant calibration is. If that tier supports a screening level, Step 2 (Screen-out) lets you control accuracy further on top of that baseline. Good sets a baseline of about 10.8% at Level 1 (Failed calibrations only), and since Good does not support the stricter levels, that baseline cannot be tightened any further. High starts from a better baseline at the same level, about 9.6%, and can also be tightened further: moving to Level 3 (Balanced) brings it to about 8.3%, without changing the tier.
Physiology Toolbox Version
Everything above assumes the current Physiology Toolbox Version, which is what a new study gets by default. If you are instead continuing a study built on an older version, one thing to know for calibration purposes: Fixation Detection mode in the task editor requires v1.2 or above, pre-1.2 legacy versions do not support it. See Physiology Toolbox Version for the full version history and versioning policy.
Physiology Toolbox Version 1.3 and Earlier
The settings below describe calibration on Physiology Toolbox Version 1.3 and earlier, before the tier and screening system above existed. If your study is on 2.0, picking any tier already sets these for you, Custom is the only way to reach them directly there, and not every setting below exists in the same form on 2.0, noted individually where that is the case.
Calibration Length
Calibration Length (Settings tab → Physiology → Eye-Tracking → Calibration Length) is the single biggest trade-off in eye tracking calibration: more calibration points produce more accurate gaze predictions, at the cost of a longer calibration for participants. Three options are available to researchers building a new study:
130 points, 9 poses (~5 minutes): accuracy of approximately 1.5 visual degrees55 points, 4 poses (~2 minutes): accuracy of approximately 2.0 visual degrees15 points, 1 pose (~30 seconds): accuracy of approximately 2.7 visual degrees
A fourth, longer option (175 points, 12 poses, ~7 minutes) exists in Labvanced's underlying settings but is not available to researchers building a new study through the standard interface.
How to choose: if your study depends on distinguishing between closely spaced targets (for example, reading research or small on-screen elements), the 130-point option's higher accuracy is worth the extra calibration time. If your study only needs to distinguish between a few widely separated regions of the screen, the 55-point or 15-point option keeps the study shorter without meaningfully affecting your results. Studies with children or participants who may struggle to sit through a longer calibration often favor the shorter options for this reason.
Calibration Quality Control
Two settings let you enforce a minimum calibration quality rather than accepting whatever the initial calibration produces:
Set Maximum Calibration Error, when enabled, revealsMax Calibration Error (%), the maximum calibration error allowed as a percentage of the screen diagonal. Lowering this value improves data quality but means more participants will fail calibration outright.Max Nr Calibration Attemptscontrols how many times a participant may retry calibration before the study fails completely.
Tightening both of these makes sense for studies where data quality matters more than sample size (precise timing or spatial measurements). Studies prioritizing sample size, or piloting a new task, generally leave these at their defaults.
Calibration Image and Engagement Options
Calibration Image Typelets you choose between dots and animal icons for the calibration targets.Play Calibration Sounds, when enabled, reveals a volume slider (available from v1.1 and above) to adjust the calibration sound's loudness.Show Griddisplays a grid during calibration, which can help participants anticipate where the next calibration point will appear.Allow Calibration Reuselets a participant skip calibration entirely if they already calibrated on the same device within the last few hours. This is particularly useful while developing and testing a task yourself, since it avoids re-running calibration on every test run.
Infant Friendly Mode
Infant Friendly Mode (Settings tab → Physiology → Eye-Tracking → Infant Friendly Mode) automatically adjusts the calibration settings above toward values more suitable for infant participants (typically animal icons, sounds enabled, and a shorter calibration length) when toggled on. You can still edit any of the adjusted values afterward.
Virtual Chinrest and Chinrest Constraint
During calibration, a participant's head position is recorded as a center pose, referred to as the virtual chinrest. Virtual Chinrest (Settings tab → Physiology → Eye-Tracking → Virtual Chinrest) controls when this position is enforced during the rest of the study:
Enable (check during trials): enforced continuously; the study pauses immediately if the participant moves away from the center poseEnable (check between trials): enforced only between trials, not during themEnable (check between trials) and show ignore button: same as above, but with an option for the participant to dismiss the prompt and continueDisable: not enforced at all
A related setting, Chinrest Constraint, controls how much movement is tolerated before the chinrest is considered violated, ranging from very loose to very strict. Stricter settings improve accuracy but are harder for participants, especially children, to maintain. A common pattern is to pair a loose constraint with the ignore-button mode for infant studies, and a stricter constraint with continuous enforcement for adult studies where precision matters most.
Minimum Facemesh Speed
Minimum Facemesh Speed (Settings tab → Physiology → Eye-Tracking → Minimum Facemesh Speed) sets a threshold for how many face snapshots per second a participant's device camera must be able to process to take part in the study, ranging from Very Low (0.5Hz) to Very High (15Hz). This is a data-quality floor: lower thresholds admit more participants' devices but degrade data quality, while higher thresholds restrict participation to faster devices. A medium setting (Medium Low (5Hz) to Medium (7.5Hz)) is a reasonable default for most studies.
What Participants Actually See
The settings on this page determine what the calibration process looks and feels like from the participant's side, but not the exact sequence of prompts. For that, see The Participant Calibration Experience.
Frequently Asked Questions
Further Reading
The Participant Calibration Experience
What these settings actually look like from the participant's point of view.
Setting Up Eye Tracking in a Task
Enable eye tracking for a specific task and configure recalibration during the study.
Webcam Eye Tracking Accuracy and Study Design
The evidence behind webcam eye tracking's validated accuracy, and the design decisions that affect it.