Statistics › Section B · Descriptive statistics › Topic B.2
Histograms, frequency polygons, and time series
A bar graph's categories are labels; a histogram's axis is a measuring scale. This topic builds histograms, names the shapes they can take, adds frequency polygons and time series, and practises the four-part description every data set deserves.
Why histograms?
When a quantitative data set grows beyond a few dozen values, a stemplot becomes cluttered and a bar graph is the wrong tool, because its categories are labels rather than a scale. The histogram is the display for measured data: values are grouped into equal-width intervals, and touching bars show how many fall in each. The bars touch because the horizontal axis is a continuous scale - each bar's edge is simply where its neighbour begins. Dot plots and stemplots keep individual values; a histogram trades that exactness for a clear picture of the whole distribution's shape.
Building a histogram
Construction has four steps. 1. Choose equal-width classes covering all values - a width of 2 minutes, 5 points, whatever fits. 2. Tally each value into exactly one class. 3. Draw each bar's height equal to its class count, or to its relative frequency: the shape is the same either way. 4. Let the bars touch, since the axis is continuous. For edge cases, either choose boundaries no value can land on, as in frequency tables, or use the convention of this topic: a value exactly on a boundary counts in the interval it starts, the interval to its right. State the convention you use, and remember there is more than one correct histogram for the same data: different class widths reveal or hide different features.
Histogram versus bar graph
The two graphs are easy to confuse and easy to tell apart. A bar graph displays categorical data - blood types, transport modes - so its bars are separated by gaps: distinct groups with nothing between them. A histogram displays quantitative data on a continuous scale, so its bars touch: the right edge of one bar is the left edge of the next. The order of a bar graph's categories carries no mathematical meaning, though alphabetical or size order helps reading; a histogram's bars must run in value order and stay equal in width, because the picture is making claims about position and area.
Describing a distribution
Any distribution deserves the same four-part description: shape, centre, spread, and unusual values. Shape: is it symmetric, or does it lean, and how many peaks does it have? Centre: around what value do the data settle? Spread: over how wide a range do the values fall? Unusual: are there gaps, or values far from the rest? This checklist is how statisticians report a data set in words, and every later topic - measures of centre, of spread, of position - is a tool for one of those four questions.
Shape vocabulary
Symmetric: the two halves mirror each other around a centre. Skewed right: a long tail stretches toward the high values - incomes and waiting times often look like this. Skewed left: the tail runs toward the low values, as on an easy quiz where most students score high and a few score very low. Unimodal: one main peak. Bimodal: two separated peaks, which often means two different groups mixed into one data set - the heights of a parent-and-student group, for instance. Naming the shape is the fastest honest summary a graph can give.
Worked example: a histogram of wait times
A coffee shop records the waits of 60 customers, in two-minute bins: 0 to 2 minutes, 18 customers; 2 to 4, 15; 4 to 6, 11; 6 to 8, 8; 8 to 10, 5; 10 to 12, 3. 1. Check the total: \(18+15+11+8+5+3 = 60\). 2. Draw six touching bars of heights 18, 15, 11, 8, 5, 3. 3. The tallest bar sits at the left, and the heights fall steadily toward a long right tail. 4. Check with proportions: \(18/60 = 0.30\) of customers wait under 2 minutes while \(3/60 = 0.05\) wait 10 to 12 minutes, so the shape is real, not an illusion of scale.
Worked example: describing the histogram
Describe the wait-time distribution in four parts. 1. Shape: one peak at the lowest class - unimodal - with a long tail toward long waits: skewed right. 2. Centre: the middle (30th and 31st) waits fall in the 2-to-4-minute class, since 18 customers wait under 2 minutes and 33 under 4. 3. Spread: waits run from 0 to about 12 minutes, with most below 6. 4. Unusual values: no gaps, no isolated bars, no outlier. One caution the shape earns: the few long waits will drag the average above the typical wait - a pattern the skewness topic develops fully. Reporting only the average would mislead; the shape is part of the story.
Frequency polygons
A frequency polygon is the histogram's line-shaped cousin: mark the midpoint of each class at the height of its frequency and connect the marks with straight segments. For the wait-time bins, the midpoints sit at 1, 3, 5, 7, 9 and 11 minutes, at heights 18, 15, 11, 8, 5 and 3. The line usually drops to zero at the midpoint of an empty class just beyond the data (here at 13 minutes, on the right). Polygons have one clear advantage over histograms: several distributions overlay cleanly on the same axes, so two groups' shapes can be compared directly - morning waits against afternoon waits, on one picture.
Time series graphs
A histogram ignores when each value was collected - the time order is thrown away. When time itself matters, use a time series graph: each value is plotted at its date, and consecutive points are connected. Time series show what histograms cannot: trends (a steady rise or fall), cycles (repeated seasonal ups and downs), and sudden breaks. A shop's daily sales, a city's population each decade and a monthly temperature log all belong in time series. The misleading-graph habit carries over too: before trusting a trend, check the full range of dates, not a cherry-picked window.
Common mistakes to avoid
First, leaving gaps between histogram bars, which wrongly turns a continuous scale into categories. Second, using unequal class widths, which misleads because a wider bar looks bigger whatever its count. Third, forgetting the boundary convention, so edge values are counted twice or not at all. Fourth, calling every two-peak histogram bimodal: small ripples are noise, so look for separated, meaningful peaks. Fifth, treating the tallest bar as the centre - a full description needs all four parts. Sixth, using a histogram when time matters: a trend cannot be read from bins that discarded the order.
Key takeaways
Histograms group quantitative data into equal-width classes with touching bars, under a stated boundary convention. Describe every distribution with four parts: shape (symmetric, skewed right or left; unimodal or bimodal), centre, spread, and unusual values. Frequency polygons connect class midpoints at class heights, and overlay well for comparing two distributions. Time series graphs preserve the time order that histograms discard, revealing trends, cycles and breaks - and demanding a check for cherry-picked windows.
Flashcards
Flashcard
Touching bars showing counts in equal-width intervals of a scale
Test your understanding
Question 1 of 5
In a histogram, why do the bars touch?
Question 2 of 5
A value falls exactly on a class boundary. Under the convention in this topic, it is counted in:
Question 3 of 5
A histogram of house prices has one peak at the low end and a long tail toward high prices. Its shape is:
Question 4 of 5
A histogram of heights from a parent-and-student group shows two clear separated peaks. This most likely indicates:
Question 5 of 5
Which display best shows how a city's population changes over several decades?
