You are a data analyst at a logistics company. You have created a dashboard to monitor delivery performance. The dashboard includes a scatter plot showing delivery time (hours) vs. distance (miles) for each delivery, with points colored by delivery region (A, B, C, D, E). Users have reported that the scatter plot is cluttered because there are over 10,000 points, making it hard to see patterns. Additionally, the color legend for the five regions uses similar shades of blue, making it difficult to distinguish which region a point belongs to. You need to improve the scatter plot to reduce overplotting and improve region differentiation. Which approach is most effective?
Reducing point opacity to 30% lets overlapping marks accumulate into darker density, revealing clusters across 10,000 points. A qualitative palette assigns each of the five regions a perceptually distinct hue, replacing the near-identical blues that currently prevent region differentiation.
Why this answer
Reducing opacity (alpha blending) mitigates overplotting by making overlapping points more transparent, while switching to a qualitative color palette (e.g., distinct hues like red, green, blue) ensures each of the five regions is easily distinguishable. This directly addresses both user complaints without losing the overall distribution context.
Exam trap
The trap here is that candidates often choose small multiples (Option B) thinking they reduce clutter, but the question specifically asks to improve differentiation and reduce overplotting in a single view, and small multiples fragment the data, making cross-region comparison harder.
How to eliminate wrong answers
Option A is wrong because a hexagonal binning plot aggregates points into hexagons based on density, but it cannot encode a categorical variable like region with color inside a single plot—hexbin typically uses a single color gradient for density, not multiple regions. Option B is wrong because creating five separate scatter plots (small multiples) forces users to mentally compare across separate charts, losing the ability to see relative spatial patterns and interactions between regions in a single view. Option D is wrong because converting to a bubble chart by adding package weight as bubble size does not reduce overplotting (it can worsen it with larger circles) and does not improve region differentiation; it introduces a third continuous variable that is not related to the reported issues.