Start with the question, not the chart
Most poor charts come from choosing a style first and forcing the data into it. A better order is to write down the question in one sentence, then pick the chart that answers it. “Which region sold the most?” and “Is the number of support tickets rising?” need different pictures, even if the underlying table is the same.
Questions usually fall into a handful of types: comparing categories, showing change over time, showing parts of a whole, showing how values are spread out, and showing the relationship between two measures. Each has a few reliable chart choices.
Comparing categories
Bar charts are the dependable default for comparing amounts across groups, such as sales by product line. The eye compares bar lengths easily, provided the bars start at zero.
- Use horizontal bars when category names are long.
- Sort bars by value unless the categories have a natural order, such as age bands or months.
- Avoid more than a dozen or so bars; group the small ones into “Other” or show a table instead.
Showing change over time
Line charts suit values that are measured at regular intervals, such as daily visits or monthly costs. The line makes direction and rhythm visible: rising, falling, seasonal swings.
- Put time on the horizontal axis, running left to right.
- Limit the number of lines. Three or four are usually readable; ten become a tangle.
- For a small number of periods, or when each period is a separate total, columns can work just as well.
If the line is jagged because the data is noisy, a smoothed version or a longer time unit (weekly instead of daily) can reveal the trend, as long as you say you did so.
Showing parts of a whole
Pie and donut charts are popular but limited. They work only when there are few slices (about two to five), the slices add up to a meaningful whole, and the exact differences are not important. People judge angles and areas less accurately than lengths, so similar-sized slices are hard to rank.
A single stacked bar or an ordinary sorted bar chart of the shares is often clearer. For parts of a whole that change over time, stacked columns or a stacked area can work, though only the bottom segment has a steady baseline, so the upper segments are harder to compare.
Showing distributions
If the question is “how are the values spread?”, a histogram groups values into ranges and shows how many fall in each. It reveals the typical range, skew and outliers that a simple average hides. Box plots compress the same idea into a compact summary, handy for comparing several groups side by side, though many general audiences find them unfamiliar and need a short explanation.
Averages alone can mislead. Two groups with the same average can look very different once their spread is shown.
Showing relationships
A scatter plot places each record as a point, with one measure on each axis. It shows whether two measures tend to move together, whether the pattern is straight or curved, and whether a few points sit far from the rest. A visible trend does not prove that one measure causes the other. If you want to describe such a relationship with a line and numbers, that is the job of statistical modeling.
When there are thousands of points, make them semi-transparent or summarize them in bins so the plot does not become a solid mass.
Tables are sometimes the right answer
When readers need exact values, or the data has only a few numbers, a plain table can beat any chart. Charts show shape; tables show precision. Many interactive dashboards combine both: a chart for the pattern and a table for the detail.
Common mistakes to avoid
- Truncated bar axes. Starting bars above zero exaggerates small differences.
- Too many colors. Use color to carry meaning, such as highlighting one category, rather than decorating.
- 3D effects. They distort proportions and add nothing.
- Dual axes. Two vertical scales on one chart invite false impressions about how the lines relate.
- Missing labels. Every chart needs a clear title, axis labels, units and a note on the data period.
- Color alone. Some viewers cannot distinguish certain color pairs, so add labels, patterns or direct annotations.
A quick checklist
- What single question should this chart answer?
- How many categories or time periods are there?
- Do readers need exact numbers or the general shape?
- Would a person who has never seen the data understand it in a few seconds?
If a draft fails the last test, simplify it: fewer series, cleaner labels and a title that states the takeaway. Charts are also a form of technical communication, so explain any unfamiliar term in plain words. They depend on trustworthy data too, so confirm that data cleaning is done before you draw anything.