01 — Three instruments
Gaze samples information, the mouse acts and scrolling recomposes the page.
A fixation briefly stabilizes gaze to extract information; a saccade rapidly moves the eyes to a new region. The fovea, the central retinal area providing the sharpest vision, is thereby redirected. The mouse points, clicks, selects or keeps a marker. Scrolling reveals new content and changes the relationship between screen and page coordinates.
On a search-results page, Rodden et al. observed three active uses: following gaze vertically, following it horizontally, or leaving the cursor as a marker on a promising result while the eyes continue exploring.
02 — Shared dynamics
Rhythms resemble one another more than positions do.
In our study, 151 participants explored 18 webpages during free-viewing and search tasks. Movement and dwell parameters changed over time and then stabilized. Eye and mouse trends were similar, while scrolling often evolved in the opposite direction.
This convergence suggests shared organization of exploration. A mouse pause and an eye fixation remain different events: the cursor may stay in the middle of the screen while attention moves elsewhere.
03 — Around events
Target, click and scroll reorganize exploration mode.
In the thesis protocol, exploration tended to become more focal just before the first target fixation. Processing was also focal before clicking, then moved toward a more ambient mode roughly 500 ms before the action. Near the first scroll, gaze briefly became more focal, as if preparing the content change.
These windows describe average tendencies, not universal causality. Above all, they show why signals should be aligned around meaningful events instead of averaged across an entire session.
04 — Product consequence
Measure coordination around key components.
To estimate attention from mouse behavior, a model should include user identity, content region, task and elapsed time. It may use eye-cursor distance, relative lag, regions visited by both signals and click or scroll events.
This analysis detects useful sequences: content examined then pointed at, a target seen but not acted upon, scrolling that interrupts reading, or a cursor used as external memory. Interpretation remains probabilistic and is validated in the actual use context.