Why Bin Width Changes What a Histogram Shows
A histogram groups quantitative observations into intervals, called bins, and uses the height of each bar to show how many observations fall in its interval. The observations do not change when the bins change, but the pattern we see can. Narrow bins can reveal fine detail, including small fluctuations that may not matter. Wide bins smooth those fluctuations but can merge separate clusters or hide gaps.
A useful way to read a histogram is to ask whether its main features persist when the bin width changes. A pattern that appears across several reasonable widths is usually more convincing than a tiny bump visible at just one width. This comparison does not establish that a feature is real in a broader population; it helps describe the data at hand without treating every bar as a meaningful discovery.
For a fair visual comparison, keep the data, horizontal scale, and bin boundaries aligned as far as possible, and change the width. Each observation should be counted once and only once. With equal-width bins, the bars can be compared by height; the counts in all the bins must add to the sample size.
Worked Example: Three Widths for the Same Measurements
Worked Example: Three Widths for the Same Measurements
Suppose a hypothetical set of 20 workshop task times, in minutes, is:
\(1.2,\ 1.4,\ 1.8,\ 2.1,\ 2.3,\ 2.6,\ 2.8,\ 3.0,\ 3.2,\ 3.4,\ 5.8,\ 6.0,\ 6.2,\ 6.4,\ 6.6,\ 6.8,\ 7.0,\ 7.2,\ 7.4,\ 7.6\).
First use width \(0.5\) minute with bins from 0 to 8. Counting observations in each interval gives:
| Bin (minutes) | Count |
|---|---|
| [0, 0.5) | 0 |
| [0.5, 1) | 0 |
| [1, 1.5) | 2 |
| [1.5, 2) | 1 |
| [2, 2.5) | 2 |
| [2.5, 3) | 2 |
| [3, 3.5) | 3 |
| [3.5, 4) | 0 |
| [4, 4.5) | 0 |
| [4.5, 5) | 0 |
| [5, 5.5) | 0 |
| [5.5, 6) | 1 |
| [6, 6.5) | 3 |
| [6.5, 7) | 2 |
| [7, 7.5) | 3 |
| [7.5, 8] | 1 |
The counts sum to \(2+1+2+2+3+1+3+2+3+1=20\); the empty bins add zero. There are two concentrations of observations, one roughly between 1 and 3.5 minutes and another roughly between 5.5 and 8 minutes, with an empty stretch between them. The narrow bins make individual changes in bar height visible.
Now combine four consecutive half-minute bins at a time to make width-2 bins. The counts are \(3, 7, 1,\) and \(9\): for example, the bin \([2,4)\) has count \(2+2+3+0=7\). The four counts still add to \(3+7+1+9=20\). The broader view retains the low-count interval between the concentrations, although the detail within each concentration is reduced.
Finally, combine two consecutive width-2 bins to make width-4 bins. The counts are \(3+7=10\) for \([0,4)\) and \(1+9=10\) for \([4,8]\). Both bars have height 10. This wide-bin display no longer makes the two concentrations apparent. The same 20 times created a detailed, somewhat irregular pattern with narrow bins, a clearer broad pattern with moderate bins, and a flat-looking two-bar display with wide bins.
The comparison does not make the narrow or wide version automatically wrong. Each summarizes the same data at a different resolution. Here the width-2 display is useful for noticing the broad separation, while the width-4 display obscures it.
Why Counts Combine When Bins Combine
When a wider bin is made by joining adjacent narrower bins, its count is the sum of their counts. This is a direct consequence of how the intervals partition the number line: each observation in the wider interval lies in exactly one of its narrower component intervals.
This identity provides a practical check. For example, in the workshop-time data, the four half-minute bins from 2 to 4 minutes have counts \(2, 2, 3,\) and \(0\), so the width-2 bin \([2,4)\) has \(2+2+3+0=7\) observations. If the component counts do not add to the wider-bin count, recheck the interval boundaries or the tally.
This addition works directly when the wider interval is made from whole narrower intervals with matching boundaries. If boundaries do not align, the bins cannot simply be combined by adding whole counts; observations would need to be assigned to the new intervals from the data.
Worked Example: Narrow Bins Can Make Small Fluctuations Prominent
Worked Example: Narrow Bins Can Make Small Fluctuations Prominent
A second hypothetical data set gives 12 task durations in minutes:
\(2.1,\ 2.2,\ 2.4,\ 2.6,\ 2.8,\ 3.0,\ 3.1,\ 3.3,\ 3.4,\ 3.6,\ 3.8,\ 3.9\).
With width \(0.5\) minute and bins from 2 to 4, the counts are \(3,2,4,3\). For instance, \([2,2.5)\) contains \(2.1, 2.2,\) and \(2.4\), while \([3,3.5)\) contains \(3.0, 3.1, 3.3,\) and \(3.4\). The total is \(3+2+4+3=12\).
With width 1 minute, the bins are \([2,3)\) and \([3,4)\). Their counts are \(3+2=5\) and \(4+3=7\), for a total of 12. The narrower display shows a local dip and rise in the bar heights; the wider display shows only that there are somewhat more observations from 3 to 4 minutes than from 2 to 3 minutes.
The narrow-bin fluctuation is a description of these 12 values, but by itself it is not evidence of a meaningful subgroup or process change. The wider display suppresses some local variation, but also loses detail. Neither view justifies a claim about why the observations differ.
What Different Widths Can Reveal or Hide
A narrow-bin histogram divides the data into many intervals. It can help locate small gaps, peaks, or clusters, but many bins may contain only a few observations or none. Those low counts can make the outline look jagged. A small change from one bar to the next may reflect only how a few individual values happened to fall into neighboring bins.
A wide-bin histogram combines observations over broader ranges. Its bars often give a smoother, less detailed view. That can make a broad overall pattern easier to see, but distinct groups can be merged into one bar, and a gap can disappear inside an interval that also contains observations. The workshop example’s two width-4 bins each had 10 observations, despite the strong separation visible at smaller widths.
A moderate-bin histogram can be a useful starting point because it often preserves broad structure without making every small count change prominent. “Moderate,” however, is relative to the number and spread of observations; there is no single width that is best for every data set or question. Compare reasonable alternatives rather than relying on one display to settle the interpretation.
When comparing widths, check whether the following features remain visible:
- Clusters: Do concentrations of observations still appear separate, or are they merged into one bar?
- Gaps: Does a low-count region persist, or does a wider interval conceal it?
- Peaks: Does a broad high region remain, or is the apparent peak just one small bar among many?
- Irregularity: Do narrow bins create alternating high and low counts that vanish when adjacent bins are combined?
A histogram is a summary, not a record of every observation. If a feature matters, check the values or use another appropriate display as well. The plotted pattern should be described cautiously: “the histogram shows two concentrations” is a statement about the displayed data, not proof that two distinct populations or causes exist.
Worked Example: Deciding Whether a Gap Is Robust
Worked Example: Deciding Whether a Gap Is Robust
Return to the 20 hypothetical workshop times. A reader notices that the width-0.5 histogram has empty bins between 3.5 and 5.5 minutes and asks whether the apparent gap is merely an effect of the narrow bins.
At width 0.5, the four intervals \([3.5,4), [4,4.5), [4.5,5),\) and \([5,5.5)\) each have count 0. The next interval, \([5.5,6)\), has count 1. Thus the count from 3.5 to 5.5 is \(0+0+0+0=0\).
At width 2, the bin \([4,6)\) has count 1: only the observation 5.8 falls there. The width-2 display therefore does not show that the entire region from 4 to 6 is empty, but it still shows much lower frequency there than in \([2,4)\), with 7, or \([6,8)\), with 9. The low-count region remains visible at both widths, although its exact location is less detailed in the wider display.
A careful conclusion is that these data have a substantial low-count region between two concentrations. The conclusion is supported across these two widths; it does not establish why the times are separated or imply that every value between the observed groups is impossible.
Common Mistake: Treating Every Bar as a Real Feature
Wrong reasoning: “The narrow histogram has a high bar beside a low bar, so there must be two meaningful subgroups.” A narrow bin can have a high or low count simply because a few observations landed on one side of a boundary rather than the other. One bar’s height alone does not establish a subgroup.
Correction: Compare the broad pattern across more than one reasonable width. Check whether a peak, cluster, or gap persists when neighboring bins are combined, and consult the individual values if the feature matters. Describe what the display shows without turning a histogram pattern into a causal explanation.
Check Your Understanding
Use the ideas in this tutorial to answer the following questions.
- What does a narrow-bin histogram show more readily than a wide-bin histogram, and what risk comes with that detail?
- Four adjacent bins have counts 2, 0, 3, and 1. What is the count in the wider bin formed by combining them?
- Why does a wide bin spanning a gap and observations on both sides fail to show the gap clearly?
- A small peak appears with one bin width but disappears when adjacent bins are combined. What is a cautious interpretation?
- In the first worked example, why does the width-4 histogram conceal the two concentrations seen with narrower bins?