Histogram

Best for

Seeing the shape of one numeric measure across many records, such as delivery times, order values or test scores.

Avoid when

You want to compare totals across named categories (use a bar chart), or you have fewer than about 30 values.

Data you need

One numeric column with many rows. The chart counts how many values fall in each range.

A histogram shows how the values of one measure are distributed. It splits the range into equal intervals, called bins, counts how many values fall into each, and draws a bar for each count. The result shows where most values sit, how spread out they are, and whether there are unusual bumps or long tails.

Averages hide all of that. "Average delivery time is 31 minutes" sounds fine until a histogram shows that most orders arrive in 25 minutes and a second large group takes 45. The histogram is how you find out what a summary number is summarizing.

When to use a histogram

Use one when you have many values of a single numeric measure and the question is about shape, such as how long customers wait, how much they spend per order, what test scores look like, how long visitors stay on a page.

It needs volume. With 15 values, the bars are mostly noise; a simple dot plot of every value is more honest. From a few dozen values up, a histogram starts to show real structure, and it handles millions of rows without changing at all.

If your values are counts of named groups ("orders by country"), you have categories, not a distribution. That's a bar chart.

Histogram vs. bar chart

They look alike, which causes most of the confusion. The difference is in the x-axis.

Never sort histogram bars by height. The order of the bins is the whole point, and sorting would destroy the shape you are trying to show.

Choosing the bin width

Bin width changes what the chart says more than any other setting. Too few bins merge separate features into one lump. Too many split the data so finely that every bar is a random spike.

Most tools pick a default bin count with a rule like Sturges' or Freedman-Diaconis. Treat that as a first draft. Try a few widths, and pick the one where the main features stay stable when you nudge the width up or down a little. A peak that appears at one width and vanishes at the next is probably not real.

Prefer round, meaningful widths. Bins of 5 minutes, $10 or 1 year of age are easier to read than bins of 4.37. And keep the bin edges consistent: say whether a value of exactly 10 falls in the 5-10 bin or the 10-15 bin, and stick to it.

Rule of thumb: if changing the bin width by one step changes your conclusion, the conclusion was not solid enough to publish.

Reading the shape

A few shapes come up again and again:

• Symmetric, one peak. Values cluster around the middle and thin out evenly on both sides. Heights, measurement errors, many test scores.
• Right-skewed. Most values are small, with a long tail stretching to the right. Incomes, order values, response times, likes per post.
• Left-skewed. The tail stretches left. Exam scores on an easy test, where most people do well and a few struggle.
• Two peaks (bimodal). Usually means two different groups are mixed together, like the lunch and dinner rush above, or mobile and desktop users.

Skew matters most when someone quotes an average. In a right-skewed distribution, a handful of large values pulls the mean well above what a typical value looks like.

Mark the median on the chart for skewed data, and use it in your text. "Half of all orders are under $36" describes customers better than "the average order is $44". If the long tail itself is the story (a few big spenders driving revenue), say so directly.

Comparing distributions

Comparing two or more distributions, before and after a change, or two customer segments, is one of the most common reasons to draw histograms. It's also where they get messy.

Overlaying two histograms in transparent colors works for two groups at most, and even then the overlap creates a third color that readers have to puzzle out. Stacking the histograms one above the other, with exactly the same x-axis and bin edges, is easier to read.

When the groups have very different sizes, show the share of each group per bin instead of raw counts. Otherwise the bigger group always looks taller, whatever its shape. For more than three groups, box plots or strip plots compare the middle and spread of each group more compactly than a stack of histograms.

Labels and axes

The y-axis of a histogram starts at zero, because bar height is a count. Label the x-axis with the measure and its unit ("Delivery time, minutes"), and put the total number of values in the subtitle ("400 orders, March"). Readers judge how much to trust the shape partly by how many values made it.

Keep the bars in one color. Use a second color only to point at a range that matters, like orders taking longer than your 40-minute promise, and label it.

Common mistakes

A histogram is often a working chart before it is a published one. Draw one for every important measure before you summarize it, and you will catch the second peak or the long tail before your readers do.