Data Visualization — Troubleshooting

A chart technically shows the data but nobody understands it quickly

If a chart requires real explanation before an audience gets the point, the cause is almost always one of: too many dimensions crammed into one chart (color, size, shape, and position all encoding different variables simultaneously — cognitively, a viewer can reliably track maybe two encoded dimensions at once, not five), a chart type mismatched to the actual question (a pie chart used for a trend-over-time question, which pie charts fundamentally can't show), or missing annotation leaving the viewer to independently notice and interpret a pattern the chart creator already knows the meaning of. The fix in each case is different — reducing encoded dimensions (splitting into small multiples), choosing the chart type driven by the actual question rather than what data happens to be available, or adding a direct annotation explaining the "why," not just showing the "what."

A chart is technically accurate but gives a misleading first impression

This is a more subtle problem than a factual error — the data is correct, but the visual proportions don't match the actual proportions the data represents. The most common causes: a truncated y-axis on a bar chart (covered on Fundamentals — makes a small difference look dramatic), inconsistent bin widths on a histogram (makes some ranges look artificially over- or under-represented), or a 3D chart (genuinely distorts perceived size/proportion regardless of the underlying data's accuracy, which is why Tufte's principle explicitly excludes them). The audit habit worth building: before publishing a chart, ask "if someone glanced at this for 3 seconds and only absorbed the visual proportions, would their takeaway match the actual data" — if not, the chart needs a design fix even though the underlying numbers are correct.

A scatter plot is an unreadable mass of overlapping points

Overplotting — too many points rendered at a scale where individual points become indistinguishable — has a few standard fixes depending on the actual cause: transparency (alpha=0.3 in matplotlib) lets overlapping density show through as darker regions, a cheap first fix for moderate overplotting; jittering (adding small random noise to position) helps specifically when many points share exact coordinate values (common with categorical or rounded data) and would otherwise stack perfectly on top of each other; and for genuinely large datasets (hundreds of thousands to millions of points), density-based rendering (Datashader, covered on the Advanced tab) is the right architectural fix rather than trying to tune transparency/jitter parameters on a fundamentally too-large point-per-point rendering approach.

A dashboard/chart renders correctly locally but breaks or is painfully slow in production

If a chart or dashboard that works fine on a local development sample dataset becomes slow or unusable against real production data volume, the near-certain cause is that the local test dataset was orders of magnitude smaller than production, and rendering performance that seemed fine never actually got stress-tested at real scale. Before optimizing rendering code, confirm the actual data volume difference — testing against a genuinely representative production-scale sample (not just a few hundred rows) during development is what catches this class of problem before it reaches production, rather than after a stakeholder reports a dashboard that takes 30 seconds to load.

A chart's color scheme looks fine to the creator but colleagues report it's hard to read

If feedback specifically mentions difficulty distinguishing colors (not a general "the chart is confusing" complaint), color vision deficiency is worth checking directly rather than assuming it's a subjective taste disagreement — running the chart through a colorblind-simulation tool (many are freely available online) against the specific palette used often reveals a red-green or similar distinction that's genuinely indistinguishable to a meaningful fraction of viewers, even though it looks completely clear to someone with typical color vision. The fix is switching to a colorblind-safe palette (covered on the Intermediate tab) and, more robustly, adding redundant non-color encoding (line style, direct labels, distinct markers) so the chart doesn't depend on color distinction alone to be readable.