Pick a chart by the question it has to answer: comparison, change over time, distribution, relationship, part-to-whole or ranking. Includes a decision table.
By DataVisuals TeamUpdated 6 min read
Pick the chart by the question the reader is asking, not by the data you happen to have. Comparing amounts between categories points to a bar chart. Change over time points to a line chart. The shape of one variable points to a histogram, and the link between two variables points to a scatter plot. Parts of a whole go to a stacked bar, a treemap or, with very few parts, a pie chart. Two dimensions at once, like hour by weekday, go to a heatmap.
The same spreadsheet can feed all of these. A table of monthly sales by region can become a line (how did sales change?), sorted bars (which region sold most?) or a stacked bar (what share came from each region?). Which one you draw depends on the question.
Write the question down first
Before opening any tool, write one sentence that starts with a question word. "Which store had the most returns last quarter?" "How did weekly web traffic change after the redesign?" "Do customers who open more emails spend more?" If you can't write that sentence, you aren't ready to pick a chart, because you don't yet know what the reader should take away.
The sentence usually contains the answer to the chart choice. Words like "which" and "most" signal a comparison or ranking. "How did ... change" signals time. "Do ... more ... more" signals a relationship between two measures. "What share" signals part-to-whole.
Fig. 1 Six questions, six default charts. Each sketch is the shape that answers its question at a glance.
The six questions
Most everyday charts answer one of six questions.
Comparison asks how big each category is relative to the others: sales by region, survey answers by option, page views by browser. Bars do this best because readers compare lengths from a shared baseline very accurately.
Change over time asks how a value moved: daily active users, monthly rainfall, a team's points per season. A line connects the points, so the reader sees the shape (rising, flat, seasonal, a sudden drop) before reading any single value.
Distribution asks how values are spread: order sizes, response times, player heights. A histogram groups the values into ranges and shows where most of them fall and how long the tails are. An average alone hides all of that.
Relationship asks whether two measures move together: ad spend and sales, temperature and ice cream orders, minutes played and points scored. A scatter plot puts one measure on each axis and lets the cloud of dots show the pattern, including the outliers.
Part-to-whole asks what share each piece contributes: traffic by channel, budget by department. Stacked bars work well, especially when you compare the split across several groups. A pie can work for two or three parts of a single total.
Ranking asks who is first, second and last: top products, best-performing posts, slowest pages. Sorted horizontal bars are the default, with the item you care about highlighted.
Rule of thumb: if two charts seem to fit, choose the one that uses position or length on a common scale. Readers judge those far more accurately than angles, areas or color shades.
How much data you have changes the answer
The question picks the chart family. The size and shape of the data then narrows it down.
A comparison of five categories is a simple bar chart. A comparison of forty categories needs horizontal bars sorted by value, or a cut to the top ten plus an "other" bar. Time series with four quarterly points can be bars; time series with 365 daily points must be a line, and probably a smoothed one with the daily noise shown faintly behind it.
Scatter plots with tens of thousands of points turn into a solid blob. At that size, make the dots smaller and semi-transparent, or switch to a binned heatmap of the two variables. Histograms need enough values to have a shape at all. With twenty observations, a strip of dots (one per value) is more honest than bins.
Common wrong picks
Some mismatches show up again and again, and each has a straightforward replacement.
A pie chart with many similar slices is the classic. Angles are hard to compare, so a pie with six shares between 15% and 19% tells the reader almost nothing. Sorted bars make the ranking obvious.
Fig. 2 The same six shares. In the pie the slices look alike; in sorted bars the order and the one-point lead of A are visible immediately.
Bars for a long time series are another. Fifty-two weekly bars make the reader compare heights one at a time. A line shows the trend in one look.
A line chart for unordered categories implies a progression that isn't there. Joining "Chrome", "Safari" and "Firefox" with a line suggests something flows from one browser to the next. Use bars.
Stacked bars with many segments make every segment except the bottom one hard to compare, because only the bottom one shares a baseline. If readers need to compare the middle segments, split them into small multiples or use grouped bars.
A dual-axis chart, two measures with two different y-axes, lets the chart maker decide where the lines cross by choosing the scales. Readers then see a relationship that the axis settings created. Show two charts stacked vertically with a shared time axis, or plot one measure against the other in a scatter plot.
When the answer is a table or a number
Sometimes the right chart is no chart. If the reader needs exact values to look up (prices by plan, a schedule, a list of thresholds), a well-formatted table beats any picture. If the whole story is one number, such as "conversion was 3.4% this month, up from 3.1%", write the number large and put the comparison next to it. A chart with two bars would add nothing.
Tables and charts also work well together. A chart shows the pattern, and a small table below gives the exact figures for readers who want them. That pairing also helps with accessibility.
Check your pick before you polish
Once the chart is drafted, show it to someone for five seconds and ask what they saw. If they answer your original question, the chart type is right and the remaining work is labeling and color, covered in labels and annotations and using color. If they describe something else, go back to the question. Usually the fix is a different chart type, and more styling won't help.
Questions
What is the most versatile chart type?
The bar chart. It handles comparisons, rankings and, with stacking, simple part-to-whole questions, and almost everyone reads it correctly. When in doubt, start with sorted bars and switch only if the question needs something else.
When should I use a table instead of a chart?
When readers need to look up exact values rather than see a pattern, or when there are only two or three numbers. A chart earns its space when the shape of the data matters more than any single figure.
Can one chart answer two questions?
Usually not well. A chart that shows both the trend and the share of each category tends to show neither clearly. Two simple charts side by side are easier to read than one that tries to do both jobs.