Most “best practices” lists are decoration tips dressed up as principles. These are not that. Each rule below has a reason underneath it, and the reason is always the same: the reader has to get the point in seconds, and anything that slows them down is working against you. Break a rule on purpose when you have a better reason. Break it by accident and you have just made your data harder to trust.
Five things separate a chart that lands from one that gets a polite nod: the data-ink ratio, color that everyone can read, labels that beat legends, numbers rounded to the decision, and the handful of mistakes that quietly distort the truth. For how these fit into a full screen, see the pillar on dashboard design; for which chart to reach for in the first place, the chart type guide.
The data-ink ratio (Tufte)
Edward Tufte coined the data-ink ratio in 1983, and it is still the sharpest single test for a chart. The idea: of all the ink on the page, what share actually encodes data? The data-ink ratio measures useful pixels against total pixels, and the practice is to push that ratio up by erasing everything that is not carrying information.
Go looking for the ink that is just there for looks. The heavy chart border. The gray background fill. The bold gridlines every ten units. Drop shadows. A legend repeating what a label already says. All of it is chart junk, and chart junk distracts the reader from the one line you wanted them to follow. Gridlines especially: they should fade into the background, a faint gray at most, because their job is a gentle reference, not a cage around your data.
A word of caution, because purism has a failure mode too. Tufte taken to the extreme produces charts so stripped they are hard to read, gridlines gone entirely, axes barely there. The goal is not minimum ink, it is maximum clarity. Keep the light gridline if it genuinely helps someone read a value off the axis. Erase it if it does not. The ratio is a lens, not a religion.
Color and accessibility (red-green, 4.5:1 contrast)
Roughly one in twelve men cannot cleanly tell red from green. Sit with that, because red and green are the single most common way dashboards encode “bad” and “good.” A red-green palette excludes color-blind readers, which means a meaningful slice of your audience sees your status column as an undifferentiated smear. They are not confused. Your chart is.
The fix is not to ban color, it is to never let color be the only signal:
- Add a second cue. Pair the color with a shape, an icon, position, or a direct label. Up arrow and down arrow read the same to everyone.
- Prefer blue and orange over red and green when you need two contrasting states. It survives every common form of color blindness and still photocopies in black and white.
- Encode one variable at a time. Color should encode one variable, so do not color bars by region and then also by status. The reader cannot tell which dimension the color is answering.
Then there is plain legibility. Text and its background must reach a contrast ratio of 4.5 to 1 to meet the accessibility standard, and pale gray labels on a white card routinely fail it. Run your colors through any contrast checker before you ship. And use color with restraint: a single accent color draws the eye to the thing that matters, so gray out the supporting series and save the bright color for the one line the reader is supposed to watch. A chart where everything is colorful is a chart where nothing is emphasized.
Labels over legends
Put the label on the thing. A legend forces a round trip the reader has to make for every single series: read the line, drop their eyes to the legend, match the color, carry it back, and hope they remembered which line they were on. A legend slows the reader, and it slows them once per data point, which adds up fast on a busy chart.
Direct labels beat a separate legend nearly every time. On a line chart, write each series name at the end of its line, where the line actually is. On a bar chart, label the bars directly and drop the axis clutter. The reader’s eye never has to leave the data. That round trip you just deleted is the difference between a chart someone reads and a chart someone skims and misremembers.
Legends are not banned. Sometimes you have eight series and direct labels would collide into mush. Fine. When you must use a legend, order it to match the visual order of the data (top line, top legend entry) so the matching is at least fast. But reach for the label first. It is almost always the better tool, and the dashboard widgets that read best are the ones that label themselves.
Formatting numbers (round to the decision)
Precision you cannot act on is just noise wearing a lab coat. Number formatting rounds to significant digits, and the right number of digits is set by the decision, not by whatever the calculation spat out. A conversion rate of 47.32891% and a conversion rate of 47% lead to the exact same action, so the extra four digits buy you nothing and cost the reader a pause.
Round to the decision, every time:
- Percentages to whole numbers or one decimal. 47%, or 47.3% if the tenth genuinely moves someone.
- Currency to thousands or millions on a dashboard. $4.8M, not $4,823,911.47. The cents are an insult to the reader’s time.
- Big counts abbreviated. 1.2M sessions reads instantly; 1,204,882 makes the eye count commas.
- Tables with decimals aligned on the decimal point, so the numbers form a clean column your eye can run down and compare.
One caution the other direction: do not show more precision than your data actually supports. If your source is accurate to the nearest thousand, reporting a figure to the dollar is not precision, it is a lie with good posture. Match the displayed digits to the real confidence in the number.
The common mistakes (truncated y-axis, dual axes, 3D)
Which chart lies to you most often? Usually the one that looks the most impressive. Three mistakes do the bulk of the damage, and all three are easy to make by accident and easy to fix on purpose.
The truncated y-axis
Start a bar chart’s axis at 90 instead of 0 and a 2% difference looks like a landslide. A truncated y-axis exaggerates small differences, and on a bar chart it is simply wrong, because a bar encodes value by length and half a bar should mean half the value. Bars start at zero. Non-negotiable. Line charts have a little more room to zoom, since they encode by position rather than length, but if you crop the axis, say so plainly on the chart so nobody mistakes a molehill for a mountain.
Dual axes
Two measures, two different scales, one chart, and suddenly revenue and complaints appear to move together in a way you secretly chose by sliding the axes around. A dual axis misleads the reader because the crossing point and the apparent correlation are artifacts of scaling, not the data. Split it into two small charts stacked together, or index both series to 100 at the start and plot them on one honest axis. If you truly need to show two measures at once, a combo chart with a clear, labeled second axis is the careful version, but the default answer is: do not.
3D effects
There is no such thing as a helpful 3D chart. A 3D effect distorts the values, full stop: the perspective tilts the bars, the back ones shrink, the front ones loom, and the reader can no longer trust the heights against the axis. It adds ink that carries no data, which is the data-ink ratio failing in the most literal way. Flat, every time.
Two smaller sins round out the list. A pie chart breaks down above five slices, dissolving into a ring of similar wedges nobody can rank; past five categories, use a sorted bar chart instead. And leave bars unsorted at your peril: a sorted bar chart speeds up comparison because the ranking is the message, and alphabetical order throws that message away. Sort by value unless there is a natural order the reader expects, like months.
None of this is about making charts pretty. It is about not making the reader work, and not accidentally lying to them. Get those two right and your charts will be trusted, which is the only reputation a chart can actually earn. Where these rules meet the full screen, the dashboard layout guide shows how to arrange the honest charts so the eye reads them in the right order.