Eye-Gaze Pointing (Jacob, 1990)
Measured by Hansen, Rajanna, MacKenzie & Bækgaard · COGAIN @ ETRA 2018 (2018)
Inputs
The measured or assumed values behind the calculations, each with its source.
- TP = 2.127 bits/s
- Fitts' law throughput for eye-gaze pointing on a head-mounted display, computed per ISO 9241-9 (effective width). Same study, same task: head pointing 2.472 bits/s, mouse 3.239 bits/s. The system date comes from Jacob's 1990 eye-movement interaction paper; the throughput is measured in the 2018 HMD study.
- MT = 812 ms
- Mean time to activate a target by gaze (head pointing was 850 ms, mouse 642 ms in the same study).
Strictest ITR
Each scoring method is an upper bound on the channel, so the headline is the strictest (smallest) one for this entry. Use the score selector on the home page to view any single method across entries.
-
Information per movement (index of difficulty)
ID = log2(A/W + 1); amplitudes (160, 260 px) × widths (50, 100 px) → ID = 1.4–2.6 bits (mean ≈ 2.0 bits/movement)
Each eye-gaze acquisition selects among the endpoints set by the distance-to-width ratio; that ratio, in bits, is the information the movement carries. Because it depends on target geometry, this is a genuine pointing task.
-
Accuracy folded in via effective width
W → We per ISO 9241-9, from the observed 1.53% error rate / endpoint scatter
Effective width discounts fixations that landed off-target, so throughput is net of the user's real gaze precision.
-
Throughput = information ÷ movement time
mean MT = 812 ms → TP = 2.127 bits/s for eye gaze (head 2.472, mouse 3.239 in the same study)
What counts as a bit depends on the action space. The number of distinguishable actions and how likely each one is are design choices of the task, not the sensing hardware. The same modality can present a fixed set of targets, a set pruned per step by a grammar or language model, or a continuous control space. Each of these changes how many actions are live and how the probability mass is spread, and therefore the information per selection. Read the action space below before comparing headline numbers across entries.
Action space
What the user can produce at each step, and how those options are distributed.
- Structure
- Continuous control space
- Size
- Continuous
- Prior
- Uniform: all actions assumed equally likely
- Notes
- Eye gaze used as a POINTING channel: the user fixates a spatially-placed target and confirms by dwell or click. Fitts' throughput folds in speed and spatial accuracy (effective width), so it is directly comparable to the mouse, trackball, stylus, tongue and head-pointing entries; eye gaze lands at ~66% of this study's mouse throughput. Crucially, the rate scales with target size and distance (Fitts), which is what makes it a pointing task, in contrast to an SSVEP speller, where the same eyes feed a discrete frequency classifier: the decoder emits a genuine 1-of-N decision, so log2(N) is the decoded information. (SSVEP is not perfectly geometry-free: foveation, crowding, eccentricity and a gaze shift all matter, but that spatial sensitivity is incidental, not Fitts-scaled difficulty, so it stays a classifier.) The realized text-entry application of gaze pointing is the separate Eye-Tracking Keyboard (dwell) entry; this entry is the bare eye-gaze pointing channel. Head pointing from the same Hansen et al. study is split into its own entry.
Comparability The strictest bound here is Fitts throughput: the index of difficulty, log₂(A/W + 1), per movement. Directly comparable to the other continuous-pointing entries (mouse, trackball, stylus, gaze and the cursor BCIs). Set against the text entries (keyboards, spellers, speech) it crosses methods: both report bits/s, but one measures movement difficulty and the other text information, so compare within the family first.
Source
- Authors
- Hansen, Rajanna, MacKenzie & Bækgaard
- Publication
- COGAIN @ ETRA 2018, 2018
- Paper
- 10.1145/3206343.3206344