How to choose the right chart type

Every chart answers a question. When the chart type does not match the question, the chart can still look good — and still be wrong. The fix is to decide what the reader should be able to say after five seconds, and only then pick the chart.

Step 1: Finish this sentence

“After looking at this chart, I want the reader to see that…” Most answers fall into one of six shapes. The chart type follows from the sentence.

Step 2: Check the shape of your data

The same numbers support different charts depending on how they are arranged. In ChartPaste, the first row is treated as column headers, the first column as category labels, and every other column as a series. That layout suits most spreadsheet data — but it means a few arrangements cause trouble:

• One row per observation, one column per variable — this is scatter and bubble territory, not a bar chart. A bar chart of 200 rows is unreadable.
• A “Total” column at the end — useful in a spreadsheet, dangerous in a stacked chart. If the total is stacked together with its parts, the bar is counted twice. If your data has a total column, drop it before choosing a stacked chart.
• Mixed units in one table — revenue in dollars next to margin in percent. Plotting both on one axis makes one of them invisible. Either split the chart or use a combo chart with two clearly labelled axes.

Step 3: Match the chart to the comparison

Comparing categories

Use a bar chart. If the category names are long, or there are more than about ten categories, use a horizontal bar chart — the labels stay readable and the ranking is obvious.

Use a stacked bar only when the total matters as much as the parts. Remember that the middle segments of a stacked bar do not share a common baseline, so their heights are hard to compare across bars. If comparing the segments is the point, use grouped bars instead.

Showing change over time

Use a line chart for trend, pace and turning points. Use an area chart when the volume under the line matters too. Use a stacked area chart when you need both the total over time and its composition — but keep the series count small, because the same floating-baseline problem applies.

A bar chart can show change over time for a handful of periods (e.g. five years). For many periods, bars become thin lines and the trend is harder to read than with a line chart.

Parts of a whole

A pie chart works for two to five slices when the message is “one slice dominates”. With more slices, the angles become impossible to compare. A donut chart has the same limitation — the hole does not make the angles easier.

For many categories, use a treemap (area encodes size) or a 100% stacked bar (shares are directly comparable across bars).

Relationships

Use a scatter plot to see whether two measures move together, and to find outliers. Use a bubble chart when a third value should control the dot size. A scatter plot shows association, not cause — say “moves with”, not “caused by”, unless you have more than a scatter plot.

Distribution

Use a histogram to see the shape of one variable — how many values fall into each range. Use a heatmap when two dimensions both matter (e.g. weekday × hour) and colour can encode the value.

Changes in a total

Use a waterfall chart when the story is “we started here, these things happened, we ended there” — profit bridges, headcount movements, budget changes. It makes the net effect visible without requiring the reader to do the arithmetic.

Four mistakes that make a chart misleading

1. A bar chart that does not start at zero. Bar length encodes the value, so cutting the axis exaggerates differences. Line charts may start elsewhere; bars should not.
2. Too many pie slices. If the reader cannot rank the slices without reading every label, use a bar chart.
3. Two y-axes used to imply a relationship. A combo chart can put two measures on one canvas, but the scaling is arbitrary — two unrelated lines can be made to look correlated. Label both axes and avoid implying causation.
4. Decoration over information. 3D effects, shadows and textured fills make values harder to compare. Flat colour and clear labels are not a style choice; they are how the chart works.

A quick decision checklist

• What should the reader be able to say after five seconds?
• Is the first column labels, or is every row an observation?
• Are all the numbers on the same scale and unit?
• Does the table contain a total column? Drop it before stacking.
• Can the chart still be read in greyscale, and by someone with colour blindness?
• Would a simpler chart answer the same question?

When in doubt, the bar chart and the line chart cover most business questions. Everything else is a special case — and a special case is only worth it when it makes the answer clearer, not when it makes the slide prettier.