A scatter plot of thousands of fused points still shows the shape of the distribution, not any one point

What it is

Crossing the density limit is not all-or-nothing: a scatter plot with thousands of fused points still expresses the distribution's shape faithfully — where it is dense, where sparse, which way it runs; what is lost is the ability to identify a single point. As a "distribution chart" it remains qualified; as a "detail chart" it has already failed. Judging whether a high-density chart may ship starts with one question: does it answer a shape question or an individual question?

Why it happens

Shape information and individual information travel on different perceptual routes: overall shape (density, direction, clusters) is carried by the aggregate texture of dense marks and read through pattern recognition, indifferent to any single point being hidden; individual information (which order is this point) depends on every mark remaining independently legible, which any occlusion destroys directly. This explains why a density chart can even beat a sparse chart of the same data on "see the distribution" tasks — density is itself information, and the aggregate texture makes it explicit — and why the same chart fails completely at "look up the detail": the individual identity has been swallowed by texture.

Studying it

The task dissociation can be demonstrated with a dual-task experiment: the same set of dense scatter plots, one reader group judging distributions (which region is denser, are the two distributions offset), the other identifying individuals (find the point with this ID), each recording accuracy. Distribution judgements stay stable or even benefit at high density, while identification accuracy falls off a cliff — the divergence of the two curves is the evidence. Control for familiarity: analysts who know their data will overestimate the chart's readability on identification tasks.

Where it stops holding

The "shape survives" conclusion rests on two premises. First, the aggregate texture must be homogeneous — a texture of same-class, same-sized points genuinely encodes density, but once heterogeneous marks mix into one layer the texture's meaning muddles and shape readings lose reliability. Second, transparency or darkness must not be misread — "darker means more" without a density scale is only a hint, and readers who interpret it as emphasis derive skewed conclusions. The lab finding that distribution judgements stay stable holds for specific density ranges and homogeneous stimuli; extrapolating to any density fails — once the texture saturates (an unbroken dark mass), the distribution's internal structure is lost with it.

Applying it

• Keep a high-density chart when the task genuinely is shape and distribution, and label it clearly: "shows the distribution; individual points not identifiable."
• For mixed tasks (shape plus detail), layer the implementation: an aggregate view carries the shape, hover or drill-down carries the individuals.
• Verification: readers must complete the distribution task with high accuracy for the chart to keep its shape duty; anyone who tries to find "a specific point" in it and fails has exposed a task-chart mismatch — add the interaction layer.

Related

• Same group: U5.06.1 One chart's mark count is limited by overlap and occlusion · U5.06.2 Past the density ceiling, aggregate, sample, or facet · U5.06.3 The density ceiling shifts with final size and resolution · U5.06.5 Solve density in data processing, not by shrinking marks
• Nearby: U5.06.2 Past the density ceiling, aggregate, sample, or facet · U2.05.2 Dense point clouds need transparency or binning
• Search terms: density estimation · statistical graphics · texture perception