Heatmaps hide temporal order
Aliases: fixation heatmap · scanpath · time to first fixation
What it is
A fixation heatmap stacks dwell over a time window, and often over people, into spatial intensity. A hot patch says “this place accumulated looking,” not what was looked at first, whether gaze looped, or which fixation solved the task. Two opposite scanpaths can burn the same image: one person reading a form top to bottom, another ping-ponging between Submit and an error. A heatmap is a spatial integral; order is the dimension it integrates away.
Why it happens
Each fixation has a location and a timestamp. The heatmap adds on location, drops the timestamp, and averages people into a blob. Time to first arrival, regressions, turning angles, and loops are exactly the material that separates fluent reading from “cannot find the next cue.” Aggregation is also rhetorically attractive: a bright patch looks like an answer, and teams stop at “users look here” instead of asking “in which second, and what did they do next.” On dynamic interfaces the problem is worse—the same pixel is a different state at different times, and a static heatmap stacks those states too.
Studying it
Lead with timed measures: time to first fixation, area-of-interest order, scanpaths, regressions. Use heatmaps only as an exploratory sketch. When a display is required, switch to time-sliced heatmaps, overlaid scanpaths, or a few typical and atypical paths, and name which people and which task segment were pooled. Segment dynamic pages by state or step rather than stacking a whole session. When comparing designs, differences in order metrics outrank a heatmap that “looks more focused.”
Where it stops holding
For a very short capture check—whether a sudden warning is looked at within a few hundred milliseconds—spatial aggregation can suffice because order barely unfolds. When the stimulus is fully static and the question is only “what is more conspicuous,” the loss is smaller. With a badly unbalanced sample, a heatmap is dominated by a few long dwells and even the spatial claim is unstable. Static posters arguing visual weight can live on heatmaps; interactive flows cannot.
Applying it
- Default the readout to path diagrams or step slices; if a heatmap appears, label its time window and headcount.
- After a hot patch, ask three order questions: when was first arrival, what was visited before, and did action follow immediately.
- Do not set information hierarchy from one whole-session heatmap; split the task into steps and inspect order per step.
- If two paths share hotspots but one succeeds and one fails, keep the path difference; do not treat the heatmap as proof they are equivalent.