You've seen the meeting. A slide goes up with a rainbow pie chart of six near-identical slices, and the room spends ninety seconds squinting at it instead of deciding anything. The data was fine. The chart was the problem.

The quick version

  • Choose a chart by the question it answers, comparison, trend over time, part-to-whole, or a running total, not by what looks impressive.
  • People read position and length far more accurately than angle and area. That single fact explains why a bar chart usually beats a pie chart.
  • Bars compare, lines trend, stacked bars show composition, cumulative lines show "how much so far," and a marimekko packs two dimensions into one picture.
  • Strip away anything that isn't the data. Decoration doesn't add credibility; it adds noise.

The idea in depth

A chart is a translation device. It turns numbers, which the eye reads slowly and imprecisely, into shapes the eye reads quickly. The whole craft of chart selection comes down to one question: does the shape I've chosen let a person decode the number accurately, with the least effort? That isn't a matter of taste. It has been measured.

Why bars beat pies: the perception hierarchy

In 1984, statisticians William Cleveland and Robert McGill published "Graphical Perception: Theory, Experimentation, and Application to the Development of Graphical Methods" in the Journal of the American Statistical Association (vol. 79, no. 387). The data-visualisation writer Nathan Yau has a clear plain-English walkthrough of their findings. They ran controlled experiments on how accurately people judge quantities from different visual encodings, and ranked the results. People are most accurate when they compare position along a common scale (where the tops of bars sit on a shared axis), then length, then angle and slope, then area, and least accurate of all with colour saturation.

So the move is mechanical: encode the number you most want your audience to judge accurately using the highest-ranked channel you can. A bar chart uses position-on-a-common-scale, the top of the hierarchy. A pie chart asks the eye to compare angles and slice areas, two of the weakest channels. That is the research behind a rule of thumb you may already half-believe: when slices aren't obviously different, a pie chart hides the very comparison it's supposed to show.

flowchart TB
  A(["How accurately can people read it?"]) --> B("Position on a common scale, most accurate")
  B --> C("Length")
  C --> D("Angle / slope")
  D --> E("Area")
  E --> F("Colour saturation, least accurate")
Cleveland & McGill's ranking of visual encodings, simplified. Higher up the list, the more accurately the average viewer decodes the value. Leaders Loop, after Cleveland & McGill (1984)

The data-visualisation expert Stephen Few pushed this into a memorable practitioner argument in his 2007 essay "Save the Pies for Dessert." His case: a pie chart only lets you read a slice cleanly when it lands near 0, 25, 50, 75 or 100 percent; everything in between turns into guesswork, and comparing two middling slices is harder still. Swap to a bar chart and every value sits against the same baseline, so the comparison becomes trivial.

Choosing a chart is choosing how hard you make your reader work to see your point.

The honest limitation: the hierarchy is about accuracy, not everything. Pie charts survive because a single dominant slice, "one customer is 70% of revenue", reads instantly, and because audiences are used to them. The lesson isn't "pie charts are banned." It's that you're trading some decoding accuracy for familiarity, so spend that trade deliberately, on the rare chart where part-of-a-whole is the entire message and the slices are few and unmistakable.

Match the chart to the question

Most chart mistakes are really question mistakes, the author never decided what the chart was for. Four questions cover the bulk of business reporting, and each has a default answer. This is the same instinct behind first-principles versus heuristic reasoning: a good default saves you from re-deriving the answer every time, as long as you know when to override it.

  • Comparing items? A bar chart. One bar per category, sorted by value (not alphabetically, sorting is the insight). Bars beat everything else for "which is bigger."
  • Showing change over time? A line chart. Lines exploit our reading of slope and direction to make trend, seasonality and inflection points jump out. Use a line only when the x-axis is genuinely continuous, like time.
  • Showing parts of a whole? A stacked bar, usually beats a pie, and if you stack across several periods, you see both the total and the shifting mix. Its weakness: only the bottom segment sits on the baseline, so middle segments are hard to compare across bars.
  • Showing a running total? A cumulative line (a "how much so far" curve), sales-to-date against target, or a burndown. The slope tells you the rate; the height tells you the accumulated total.
flowchart TD
  Q(["What is the chart for?"]) --> C1(["Compare items?"])
  Q --> C2(["Change over time?"])
  Q --> C3(["Part of a whole?"])
  Q --> C4(["Running total?"])
  C1 --> R1("Bar chart, sorted by value")
  C2 --> R2("Line chart")
  C3 --> R3("Stacked bar, pie only if 2–3 clear slices")
  C4 --> R4("Cumulative line vs target")
A starting decision tree, not a law. Override it when the data has a reason to. Leaders Loop

Then there's the marimekko (or "mekko") chart, a stacked bar whose column widths also carry meaning. Width might show each market's size; the segments within show each competitor's share of that market. It's a staple of strategy decks because it shows two things at once: how big each segment is, and how the mix breaks down inside it. The cost is steep, it leans on area and width, low on the perception hierarchy, so precise comparison is hard and a busy mekko quickly becomes a wall of rectangles nobody can read. Reach for it when the two-dimensional story (size and mix) is the actual point, and skip it otherwise.

Clear the clutter: data-ink

Edward Tufte, in The Visual Display of Quantitative Information (1983), gave the discipline a sharp test: the data-ink ratio. Of all the "ink" in a chart, how much actually represents data, and how much is gridlines, heavy borders, drop shadows, gradients and 3-D effects? His advice: maximise the share that's data, and erase the rest. So the move is to default to plain. Kill the 3-D pie (it literally distorts the areas you're asking people to compare), drop the chart-junk, label directly instead of forcing a trip to a legend. A clean chart isn't a stylistic preference; it's a faster read with fewer chances to mislead.

A worked example

A regional sales director opens the Monday review with a 3-D pie chart of revenue across her six territories. The slices are close in size, tilted in perspective, and colour-coded to a legend in the corner. The team debates which territory is actually second-largest for several minutes. (Figures here are illustrative.)

She rebuilds it three ways for the question she actually has. "Which territories are pulling weight?" becomes a horizontal bar chart, sorted high to low, and the under-performer that the pie had buried in the tilt is now unmistakably the short bar at the bottom. "Are we trending up or stalling?" becomes a line chart of monthly revenue per territory, and a flattening curve in one region, invisible in any snapshot, is suddenly the headline. "Are we on track for the quarter?" becomes a cumulative line of revenue-to-date against the target line; the gap between the two curves is the whole story, no spreadsheet required.

Nothing about the underlying numbers changed. The decisions got faster because each chart answered one question with the encoding the eye reads best. That's the entire job.

Frequently asked questions

Are pie charts always wrong?

No. They're a poor default but a fine special case: two or three slices, one obviously dominant, and part-of-a-whole is the only message. The moment you have many slices or need to compare similar ones, the angle-and-area encoding works against you, switch to a sorted bar chart.

Bar chart or line chart, how do I choose?

Look at the x-axis. If the categories are discrete and unordered (regions, products, teams), use bars. If the axis is continuous and ordered, almost always time, use a line, because the slope between points carries real meaning. Don't draw a line across categories that have no natural order between them.

Should I ever start a bar chart's axis above zero?

For bars, no. Bars encode value by length, so a truncated baseline exaggerates differences and misleads, it's one of the most common ways a chart lies. Line charts, which emphasise change rather than absolute magnitude, can sometimes use a non-zero axis to make a real trend visible, but label it clearly when you do.

When is a stacked bar better than several separate bars?

When the total and the composition both matter and you want them in one view, revenue split by product line across quarters, say. If your audience mainly needs to compare one component across groups, give that component its own un-stacked bar chart instead, because only the bottom segment of a stack sits on a shared baseline.

What's the single most useful habit?

Write the chart's takeaway as a sentence first, "Territory C is dragging the quarter", then pick the chart that makes that sentence obvious at a glance. If no chart makes it obvious, you may have the wrong takeaway, or the wrong data.

Related in the Toolkit

Where to go next