The report shows a mound. The next export shows a slope. Nobody changes a duration.
The chart changes its bins.
Before describing a distribution, inspect the intervals that collect the observations. Keep the original values within reach. A shape deserves a sentence only after its counts reconcile.
NIST's histogram guidance defines the common histogram through equal-width intervals and the number of observations inside each interval. Its guidance also distinguishes counts, proportions, and density.
The review below turns those definitions into a proposed reporting worksheet. All requests, durations, and reporting conditions are fictional. The example establishes no service benchmark or measured business result.
Define What Enters the Distribution
One observation in this example is one completed request's elapsed review duration, in minutes. Timing begins when review starts. It ends at the first completed response. Waiting before review is outside the interval. Waiting after review starts remains inside it.
The fictional population contains twenty eligible standard requests completed during a seven-day reporting window. Reopened requests are outside this definition. The extract includes every eligible completion in that window, with no missing durations.
That population excludes unfinished work. A report about completed durations cannot answer how long every open request will eventually take. Keep that limitation beside the population statement.
Name the extract version before opening the chart settings. Here it is synthetic-duration-v1. The source records use whole elapsed minutes. No rounding or unit conversion occurs between the source table and either bin scheme.
In a real review, record timestamp precision and the duration calculation too. A value calculated from date-only fields needs a different explanation from a value calculated from event timestamps.
The chart title can stay short. Its working note must preserve the observational unit, eligibility rule, population, reporting window, and measurement unit. Another reader should be able to rebuild the same input table.
Keep All Twenty Synthetic Values Available
The table is sorted by duration for inspection. Request identifiers retain the connection to each fictional record. Sorting this table supplies no evidence about calendar order.
| Request | Review duration, minutes | Request | Review duration, minutes |
|---|---|---|---|
| H-01 | 4 | H-11 | 20 |
| H-02 | 6 | H-12 | 22 |
| H-03 | 8 | H-13 | 24 |
| H-04 | 9 | H-14 | 26 |
| H-05 | 10 | H-15 | 28 |
| H-06 | 12 | H-16 | 29 |
| H-07 | 14 | H-17 | 30 |
| H-08 | 16 | H-18 | 32 |
| H-09 | 18 | H-19 | 38 |
| H-10 | 19 | H-20 | 48 |
Count the records first: ten pairs of requests give twenty observations. The minimum is four minutes. The maximum is forty-eight minutes. Neither fact depends on bin width.
Keep forty-eight in the table. Its distance from neighboring observations gives you a record to inspect. It does not supply permission to remove it. Use an outlier review when source validity needs investigation.
If a duration later proves unavailable, preserve the request and its missing-value status. Reconcile eligible records separately from plotted observations. A chart with nineteen measured durations should not silently retain a twenty-record denominator.
Count Two Equal-Width Schemes
Both schemes start at zero and end at sixty minutes. Scheme A uses ten-minute intervals. Scheme B uses twenty-minute intervals. Each scheme uses equal widths within its own chart.
Use a lower-inclusive, upper-exclusive rule for every interval except the last. The last includes both its lower edge and the final upper edge. In this example, ten belongs to ten through under twenty. Twenty belongs to twenty through under thirty.
The exception at the final upper edge prevents an observation of exactly sixty from falling outside the displayed range. There is no sixty-minute observation in this dataset. The rule still belongs in the specification.
| Scheme A interval, minutes | Included values | Count |
|---|---|---|
| 0 to under 10 | 4, 6, 8, 9 | 4 |
| 10 to under 20 | 10, 12, 14, 16, 18, 19 | 6 |
| 20 to under 30 | 20, 22, 24, 26, 28, 29 | 6 |
| 30 to under 40 | 30, 32, 38 | 3 |
| 40 to under 50 | 48 | 1 |
| 50 through 60 | None | 0 |
| Total | Every observation once | 20 |
Scheme A reconciles as four plus six plus six plus three plus one plus zero. Every source value appears once. The zero-count interval stays visible because both charts use the declared zero-to-sixty range.
| Scheme B interval, minutes | Included values | Count |
|---|---|---|
| 0 to under 20 | 4, 6, 8, 9, 10, 12, 14, 16, 18, 19 | 10 |
| 20 to under 40 | 20, 22, 24, 26, 28, 29, 30, 32, 38 | 9 |
| 40 through 60 | 48 | 1 |
| Total | Every observation once | 20 |
Scheme B reconciles as ten plus nine plus one. Its first bar combines Scheme A's first two bars. Its second combines the next two. Its third combines the final two.
This nesting makes the comparison easy to audit. It does not establish these edges as the best choice for another dataset.
State What the Vertical Axis Means
A count histogram answers how many observations fall inside each interval. In Scheme A, six requests occupy ten through under twenty minutes.
A proportion histogram divides each count by the number of plotted observations. That interval holds six divided by twenty, or thirty percent. Scheme A's percentages are twenty, thirty, thirty, fifteen, five, and zero. They sum to one hundred percent.
A density histogram divides each count by the observation total times bin width. For the same interval, six divided by (twenty times ten) gives 0.03 per minute. Multiplying that height by its ten-minute width gives an area of 0.30.
NIST explains these two normalizations: proportions sum to one, while density gives total area one. Label density with its reciprocal measurement unit.
For Scheme B's first interval, density is ten divided by (twenty times twenty), or 0.025 per minute. Its area is 0.50. A lower density height can therefore accompany a larger interval count.
Do not compare heights across charts until you know the normalization and interval widths. Put the formula in the report note when readers need to reproduce it. Keep the axis review attached to the comparison.
Describe What the Bins Reveal and Conceal
Scheme A has two neighboring intervals tied at six observations. Scheme B has a first bar of ten and a second of nine. One view emphasizes the middle intervals. The other emphasizes the combined lower interval.
Both descriptions are correct for their stated bins. Neither description changes the original twenty values.
The broad second interval hides the division between six observations below thirty and three from thirty through under forty. The narrow scheme exposes that difference. Even Scheme A hides each observation's exact position inside its interval.
A bar containing four values does not tell you whether those values sit near one edge or spread across the interval. Return to the record table when that distinction affects a decision.
Also inspect the starting edge. A different interval origin can regroup values even when width stays constant. For recurring reports, record both the width and every edge. “Automatic bins” is not a reproducible specification.
Appearance alone cannot establish normality, a separate population, or a business cause. NIST points to additional distribution checks for model verification. This review stops at describing observed counts and identifying source questions.
The forty-eight-minute record remains a candidate for review. The chart supplies no explanation for its duration. A large gap between that record and the next could reflect the particular sample. A histogram cannot identify the missing operational evidence.
Keep Histograms Separate From Category Charts
A histogram places numeric intervals along an ordered measurement axis. The intervals partition a duration scale here. Keep their numeric order intact.
A categorical bar chart would count request types such as standard, expedited, and exception. Those labels do not describe adjacent numeric intervals. Sorting those categories by count would answer a different question.
For this worksheet, use touching interval bars and keep the boundaries visible. Put request-type counts in a separate labeled category chart. Avoid changing the chart type while leaving a title that still claims to show durations.
Neither chart preserves the sequence in which requests happened. If the question concerns changes over time, retain the original order in a separate view. The distribution chart answers a different reporting question.
Copy the Histogram Review Worksheet
Use this proposed worksheet with the source extract. Replace each example entry with your own documented definition.
| Review field | Entry for this synthetic report | Acceptance question |
|---|---|---|
| Dataset version | synthetic-duration-v1 | Can the same twenty records be retrieved? |
| Observation and population | Completed standard request; seven-day completion window | Are unfinished and reopened requests explicitly outside scope? |
| Measurement | Elapsed review minutes, review start to first completed response | Does every row use the same events and unit? |
| Missing values | Twenty eligible; twenty measured; zero missing | Do eligible and plotted counts reconcile? |
| Bin edges | A: 0, 10, 20, 30, 40, 50, 60; B: 0, 20, 40, 60 | Are widths and starting edges recorded? |
| Boundary rule | Lower included, upper excluded; final upper edge included | Does each value enter exactly one interval? |
| Counts | A: 4, 6, 6, 3, 1, 0; B: 10, 9, 1 | Does each scheme total twenty? |
| Axis meaning | Count of completed requests | Are proportion and density versions labeled separately? |
| Supported description | Bin width changes visible grouping of these twenty values | Does the sentence stay within the inspected data? |
| Next question | Verify H-20's timing events before explaining its duration | Is the requested evidence specific? |
Keep any rejected chart export with its settings when the interpretation changes during review. That record lets another reader understand which visual choice produced the earlier wording.
A summary measure can sit beside the chart, but it needs its own definition. Use the mean and median review to decide what a typical-value statement should mean.
The histogram's job here is narrower: make the interval counts inspectable. State the edges. Reconcile the records. Describe only what survives that check.
Sources
- NIST/SEMATECH e-Handbook: Histogram. Equal-width bins, counts, normalization, and distribution review. Checked 2026.10.28. The dataset, chart, worksheet, and operating rules are editorial examples.