Choosing a chart is not an art project. It is a lookup. You have a question, the data has a shape, and one or two chart types answer that question better than the rest. Everything else is noise you talked yourself into because it looked nice in the gallery. Start from the question, not the picture. Get the question right and the chart almost picks itself. Get it wrong and you can decorate a bad choice forever without it ever reading clearly.
There are four questions a business chart usually answers: how do things compare, which way is a thing moving, how is a thing spread out, and how does a whole break into parts. Match your question to one of those four and you have narrowed seventeen chart types down to two. Here is how each bucket works, and where the popular wrong answer lives.
Comparison charts: the sorted bar
A bar chart compares categories. That is its entire job, and it does it better than anything else invented. Revenue by region, tickets by agent, conversion by channel: any time you are ranking a set of named things against one measure, the bar is the answer. Length is the easiest visual quantity for a person to judge, far easier than angle or area, which is exactly why the bar beats the pie for the same data almost every time.
Two rules turn a passable bar into a fast one. Make it horizontal when the category labels are words, because “Enterprise Northeast” fits on a horizontal bar and gets crushed under a vertical one. And sort it. A sorted bar chart speeds up comparison because the ranking is the answer, and leaving bars in alphabetical or random order forces the reader to rebuild the ranking in their head. A ranking belongs on a horizontal bar chart, sorted longest to shortest, full stop. The one exception is when the category has a natural order you must preserve, like months or age brackets, where alphabetical-by-value would scramble the meaning.
Start bars at zero. Always. A bar encodes value by length, so if the axis starts at 40 instead of 0, a bar that is twice as long no longer means twice as much, and you have quietly told a lie. This is not a style preference. It is the difference between a chart and a misleading picture.
Trend charts: the line
Is your x-axis time? Then you almost certainly want a line. A line chart shows a trend over time, and the slope carries the meaning: rising, falling, flattening, spiking. A time series belongs on a line chart because the connected points tell the eye that these values are one continuous story, not a set of separate bars to compare. People read slope without thinking. That is the whole advantage.
Keep the series count low. One line is clear. Two is fine if they are genuinely related. Five overlapping lines is a bowl of spaghetti, and the reader stops trying. If you must show many series, small multiples beat one crowded chart: a grid of little identical line charts, one per series, same axes, so the shapes compare side by side. It uses more space and reads in a tenth of the time.
Resist two temptations here. Do not put a second measure on a second y-axis to imply the two move together. A dual axis misleads the reader because you can slide the two scales until any pair of lines appears to dance in step, and the reader has no way to catch you. If two measures relate, prove it with a scatter plot, where a scatter plot reveals correlation between two variables honestly and lets the reader see how tight the relationship actually is. And do not truncate the y-axis to inflate a flat trend into a mountain. A truncated y-axis exaggerates small differences, which is fine if you label it loudly and dishonest if you do not.
Distribution charts: histogram and box plot
Averages lie by omission. “Average order value is $60” hides whether that is a tidy pile of $60 orders or a heap of $12 orders and a handful of $900 ones. Two very different businesses, one identical mean. Distribution charts show the spread the average hides, and most dashboards skip them entirely, which is a mistake worth fixing.
The histogram is the plain-language option. A histogram shows a distribution by slicing the range into bins and counting how many values fall in each. You see the shape immediately: one hump or two, tight or wide, skewed left or right. Bin width is the one decision that matters. Too few bins and you flatten the shape into a blur, too many and every bin holds one value and the pattern shatters. Try a handful of widths and keep the one where the shape is honest and legible.
The box plot is the compact option, and it earns its keep when you compare distributions across groups. A box plot summarizes spread and outliers in one small mark: the median line, the box holding the middle half of the data, the whiskers, and the dots hanging out past them. Line up one box plot per region or per product and you can compare five distributions in the space a single histogram would need. It asks a little more of the reader, so label it or add a one-line note on how to read it. For an audience that has never seen one, that note is the difference between insight and a shrug.
Part-to-whole: the stacked bar, and when a pie fails
Now the fight everyone wants to have. Part-to-whole means showing how a total splits into pieces: revenue by product line, traffic by source, headcount by department. The reflex is to reach for a pie. The reflex is usually wrong.
A pie chart shows parts of a whole, and it is fine in exactly one situation: two or three slices where one clearly dominates. “68 percent paid, 32 percent organic” reads instantly as a pie. Beyond that it collapses. A pie chart breaks down above five slices, because the human eye ranks angles and areas badly, and once several wedges sit close in size nobody can tell which is bigger without reading the percent labels. And if the reader has to read the labels to rank the slices, the pie did no work. A sorted bar would have shown the ranking directly.
So the honest default for part-to-whole is a bar. A part-to-whole split suits a stacked bar better than a pie, especially when you want to show composition changing over time. A stacked bar chart shows composition within categories, so you can line up one bar per quarter and watch the mix shift, something a row of pies can never do cleanly. For a single snapshot where you also want the ranking, a plain sorted bar with each part as its own bar beats both. Percentages next to sorted lengths, no angle-guessing required.
One more offender on the pile: the 3D chart. A 3D effect distorts the values by adding perspective the data does not have, tilting the near slice larger than the far one. There is no data question a 3D chart answers better than its flat version. Skip it. If you want the deeper argument on why some of these fail, the dashboard widgets guide covers which tiles to delete on sight, and it lands on the same short list.
Color, legends, and the honest defaults
A quick detour that saves more charts than any chart-type choice. Color should encode one variable at a time. The moment you use color for both “which product” and “did it hit target,” the reader cannot tell which meaning applies to a given hue, and the chart stops communicating. Pick one job for color per chart.
Drop the red-green pairing where you can. A red-green palette excludes color-blind readers, roughly one in twelve men, who cannot reliably tell your “good” from your “bad.” Blue and orange carry the same contrast and everyone sees it. And prefer labels on the marks over a legend off to the side. A legend slows the reader by forcing the eye to bounce between the key and the chart, decoding as it goes. Direct labels on the lines or bars beat a separate legend nearly every time. The full accessibility and formatting checklist lives in the data visualization best practices page.
A lookup table: question to chart
Print this and tape it to the wall. When you are unsure, read the question, not the data, and the row tells you the chart.
| The question you are answering | Use this | Not this |
|---|---|---|
| How do these categories rank? | Horizontal sorted bar | Pie, radar |
| Which way is this moving over time? | Line chart | Bar per date, dual axis |
| How are these two variables related? | Scatter plot | Two lines on two axes |
| How spread out are the values? | Histogram | A single average number |
| How do distributions compare across groups? | Box plot, side by side | Overlapping histograms |
| How does the whole split into parts? | Stacked or sorted bar | Pie above five slices |
| How does the split change over time? | Stacked bar per period | A row of pies |
| What are the exact values? | Table | Any chart |
| How does this vary by location? | Map | A table of place names |
The pattern across every row is the same. Pick the mark that turns your specific question into a length, a slope, or a position the eye reads without effort, and refuse the chart that looks clever but makes the reader do arithmetic. A table shows precise values when precision is the point. A map encodes data by geography when place is the point. Everything else is a bar or a line wearing a costume. Once the chart is chosen, where it sits on the screen decides whether anyone reads it, and that is the job of dashboard layout. The whole system, from KPI selection to color, ties together in the dashboard design guide.