MathemagiciansLibraryExplorationsDeep Dives
← Mathematics · Upper Secondary
MathematicsUpper Secondary

Reading histograms — estimating frequencies

U4 — Statistics Advancing › U4.2 — HistogramsLecture 3 of 4~2910 words

A question the bars do not answer

A courier company weighed 56 parcels and drew this histogram of their masses.

00.511.522.533.544.501020304050Mass (kg)Frequency density
The masses of 56 parcels. The classes are 0 to 2, 2 to 5, 5 to 10, 10 to 20 and 20 to 50 kg. Height is frequency density, in parcels per kilogram. One small square is 1 kg across and 0.1 up.

Here are two questions about it. How many parcels weighed less than 5 kg? And how many weighed between 3 kg and 15 kg?

The last lecture went from a table to a picture. Now you go backwards, from the picture to the numbers. The vertical axis says "Frequency density", so no bar height is a count. As the first histogram lecture showed, the count is the area of the bar.

The second question is harder, and it is where this lecture lives. The value 3 kg is in the middle of the 2–5 bar, and 15 kg is in the middle of the 10–20 bar. The histogram tells you how many parcels are in each class. It does not tell you where they are inside the class. So no reading of this picture can give the exact answer. It can give a good estimate, if you know what assumption you are making.

First, read the scale

Before you multiply anything, find out what one small square means on each axis. On the vertical axis, the labelled lines are 0.5 apart, and there are 5 small squares between them. So one small square is 0.5 ÷ 5 = 0.1 of a unit of frequency density. On the horizontal axis, the labelled lines are 10 kg apart with 10 squares between them, so one square is 1 kg.

Now read each bar. The top of the 5–10 bar is 6 small squares above the line for 2, so its height is 2 + 6 × 0.1 = 2.6. Put every reading in a table, with the widths from the class boundaries.

Mass (kg)Class widthHeight read (parcels per kg)Frequency = height × width
23.53.5 × 2 = 7
34.04.0 × 3 = 12
52.62.6 × 5 = 13
101.21.2 × 10 = 12
300.40.4 × 30 = 12
Total56

The frequencies add up to 56, the number of parcels ✓. This check matters even more when reading than when drawing. A misreading of one small square on the 20–50 bar changes its count by 0.1 × 30 = 3 parcels.

Now the first question needs two bars: 7 + 12 = 19 parcels weighed less than 5 kg. This answer is exact. Both bars lie completely below 5 kg, and the area of a bar is its count, whatever happens inside it.

"More than" questions work in the same way, from the other end. The number of parcels of 10 kg or more is 12 + 12 = 24. You could also find it as 56 − (7 + 12 + 13) = 24. Two routes to the same number give you a free check.

Stop and think

Which class holds the most parcels?

Show the answer

The 5–10 kg class, with 13 parcels. The tallest bar, 2–5 kg, holds only 12. The 5–10 bar is lower, which means it is less crowded. But it is wider, so it covers more kilograms, and its area is larger. Three other classes hold exactly 12 parcels each, and their bars look completely different. Height shows crowding. Only area shows the count.

Part of a bar: the evenly spread assumption

Now the second question. Between 3 kg and 15 kg, three classes are involved. The whole 5–10 class lies inside the range, so all of its 13 parcels count. The 2–5 class lies only partly inside, from 3 to 5. The 10–20 class also lies partly inside, from 10 to 15.

How many of the 12 parcels in the 2–5 class weigh more than 3 kg? The honest answer is that the histogram does not know. All 12 could weigh 2.1 kg. All 12 could weigh 4.9 kg. To go further, you must assume something about how the parcels are spread inside the class.

This is the hard part of the lecture. Spend time here. The simplest assumption is that the values are spread evenly through the class. Now look at the bar again. Its top is flat. A flat top says "the same crowding all the way across". This means that the histogram has already drawn this assumption. When you use it, you are reading what the picture shows.

If the crowding is the same everywhere in the class, the count in any slice of the class is the area of that slice:

estimated count = frequency density × width of the slice

The slice from 3 to 5 kg is 2 kg wide, under a height of 4.0. That gives 4.0 × 2 = 8 parcels. The slice is two thirds of the class, so it gets two thirds of the 12 parcels, and 8 is two thirds of 12 ✓. The slice from 10 to 15 kg is 5 kg wide, under a height of 1.2. That gives 1.2 × 5 = 6 parcels: half of the class, and half of its 12 parcels.

8 + 13 + 6 = 27 parcels (an estimate)

That is it. That is the whole idea. In a histogram, area is count. The evenly spread assumption lets you cut a bar anywhere and keep that rule.

Reading a histogram

Count in a whole class = frequency density × class width. This is exact.

Count in part of a class ≈ frequency density × width of that part. This is an estimate: it assumes the values are spread evenly through the class.

For a range, add the pieces. For the whole histogram, check that the areas add up to the total.

How good is the estimate of 27? Look at the worst cases. Suppose none of the 2–5 parcels and none of the 10–20 parcels were in the range. Then only the 13 parcels from 5 to 10 kg would count, so the true answer cannot be lower than 13. Now suppose all of them were in the range. Then the count would be 12 + 13 + 12 = 37. So the true count is somewhere from 13 to 37, and 27 is the best single guess the picture allows. Try other ranges below, and watch the two worst cases.

Explore — estimate the parcels in any range

Move the two ends of the range, or choose a preset. The shaded area is the estimate. Watch which bars are cut, and how far apart the smallest and largest possible true counts are.

Presets:

From: 3 kg

To: 15 kg

    The figure shows the range from 3 kg to 15 kg shaded on the parcel histogram. Two bars are cut and one is whole.

    00.511.522.533.544.501020304050Mass (kg)Frequency density
    The shaded area is the estimate: 8 + 13 + 6 = 27 parcels. The true count could be anything from 13 (none of the cut classes inside the range) to 37 (all of them inside).
    ClassPart inside 3 to 15 kgEstimated parcels
    2 to 5 kg (cut)3 to 54.0 × 2 = 8
    5 to 10 kg (whole)5 to 102.6 × 5 = 13
    10 to 20 kg (cut)10 to 151.2 × 5 = 6
    Estimate27

    Other ranges: less than 5 kg gives 7 + 12 = 19, exact, because 5 is a class boundary. From 12 to 25 kg gives 1.2 × 8 + 0.4 × 5 = 11.6, about 12, and the true count could be anything from 0 to 24.

    Three things should stand out. First, when both ends of the range sit on class boundaries, nothing is cut, and the answer is exact. Second, every cut bar adds doubt equal to its whole count: a cut through the 20–50 bar leaves up to 12 parcels in doubt, however small the slice. Third, an estimate need not be a whole number. From 12 to 25 kg it is 1.2 × 8 + 0.4 × 5 = 11.6, so you write "about 12 parcels".

    You have met this assumption before

    On a cumulative frequency curve, some courses join the plotted points with straight lines. The cumulative frequency lecture gave the reason: a straight line between two points means the values are spread evenly in that class. The histogram's flat top and the curve's straight join are the same assumption, drawn in two ways.

    Check it with the parcels. The running totals at 10 kg and 20 kg are 7 + 12 + 13 = 32 and 32 + 12 = 44. A straight line from 32 at 10 kg to 44 at 20 kg is halfway up at 15 kg, so it gives 32 + 6 = 38 parcels below 15 kg. From the histogram, the parcels below 15 kg are 32 + 1.2 × 5 = 38. That is the same number, by two routes. This is because the gradient of that straight line, 12 ÷ 10 = 1.2 parcels per kilogram, is exactly the height of the histogram bar. The first histogram lecture showed that the histogram was hiding inside the cumulative frequency curve. Here that link does real work.

    Completing a half-drawn histogram

    A very common question gives you half of each picture. Some bars are drawn, some frequencies are in a table, and you must complete both. The method uses the same rectangle in two directions:

    From a drawn bar: frequency = height × width.
    From the table: height = frequency ÷ width.

    Here is an example. 100 students recorded how long they spent on homework one evening (invented data). Two bars have been drawn. The table gives two frequencies.

    Time (minutes)FrequencyBar height (students per minute)
    10not drawn
    ?2.8
    32not drawn
    ?0.8
    ?not drawn
    Total100

    One class, 60 to 90 minutes, has neither a frequency nor a bar. Only the total can fill it, so it must come last. Step through the working below.

    Step through — completing the homework histogram

    Press Next to add one step. Read the reason before you go on.

    Each step gives one line of working and the reason it is allowed.

    1. 20 to 30: frequency = 2.8 × 10 = 28. The bar is drawn, so its area is its count. The class is 30 − 20 = 10 minutes wide.
    2. 40 to 60: frequency = 0.8 × 20 = 16. The same rule. This bar is 20 minutes wide, so a low bar still holds 16 students.
    3. Known so far: 10 + 28 + 32 + 16 = 86. Four of the five frequencies are now known. Add them before going on.
    4. 60 to 90: frequency = 100 − 86 = 14. The total is 100, and this is the only class left. The total is the one piece of information that reaches this class.
    5. 0 to 20: height = 10 ÷ 20 = 0.5. Now go the other way. A frequency from the table becomes a height by dividing by the width.
    6. 30 to 40: height = 32 ÷ 10 = 3.2. This is the tallest bar: 32 students crowded into 10 minutes.
    7. 60 to 90: height = 14 ÷ 30 ≈ 0.47. 14 ÷ 30 = 0.4666…, so draw it at about 0.47, two thirds of the way from 0.4 up to 0.5.
    8. Check: 10 + 28 + 32 + 16 + 14 = 100 ✓ Add the area of every bar once. The last area is exactly 14, because its height came from 14 ÷ 30.
    00.511.522.533.510283216140102030405060708090Time (minutes)Frequency density
    The completed homework histogram. Solid bars were given; dashed bars were calculated from the table and the total. The number above each bar is its frequency.
    Stop and think

    In another version of this question, the 40–60 bar is drawn at 1.5 and the 60–90 bar is also drawn, at 0.8. The total is still 100. What happens when you work out the frequencies?

    Show the answer

    The frequencies become 10, 28, 32, 1.5 × 20 = 30 and 0.8 × 30 = 24. They add up to 124, not 100. So the information does not fit together: a bar has been misread, or the question has a mistake. Go back and check each reading before you change anything.

    Worked examples

    Worked Example 1 — Exact counts from whole bars

    A histogram shows the journey times of 90 students (invented data). The bar heights read from it are 1.5 for minutes, 4.2 for , 5.4 for , 1.8 for and 0.3 for . (a) How many students took 20 minutes or longer? (b) What percentage took less than 15 minutes?

    First find every frequency, so the total can be checked. The widths are 10, 5, 5, 10 and 30 minutes. The frequencies are 1.5 × 10 = 15, 4.2 × 5 = 21, 5.4 × 5 = 27, 1.8 × 10 = 18 and 0.3 × 30 = 9. They add up to 90 ✓.

    (a) 20 minutes is a class boundary, so two whole bars answer it: 18 + 9 = 27 students. This is exact.

    (b) Less than 15 minutes means the first two bars: 15 + 21 = 36 students. As a percentage, 36 ÷ 90 = 0.4, which is 40%.

    Check by the other route: 90 − 27 (20 minutes or more) − 27 (15 to 20 minutes) = 36 ✓.

    Worked Example 2 — Estimates from cut bars

    A histogram shows the masses of 150 mangoes (invented data). The heights are 0.6 for g, 2.5 for , 1.5 for and 0.5 for . (a) Estimate how many mangoes weigh between 190 g and 230 g. (b) Estimate how many weigh more than 280 g. (c) What are the smallest and largest possible true answers to (a)?

    The frequencies are 0.6 × 50 = 30, 2.5 × 20 = 50, 1.5 × 30 = 45 and 0.5 × 50 = 25, which add up to 150 ✓.

    (a) The range cuts two bars and contains one whole bar. From 190 to 200 g: 0.6 × 10 = 6. From 200 to 220 g: the whole bar, 50. From 220 to 230 g: 1.5 × 10 = 15. The estimate is 6 + 50 + 15 = 71 mangoes.

    (b) From 280 to 300 g is 20 g of the last class: 0.5 × 20 = 10 mangoes. As a check, 20 g is two fifths of the 50 g class, and two fifths of 25 is 10 ✓.

    (c) If none of the cut classes were in the range, only the 50 would count. If all of them were, the count would be 30 + 50 + 45 = 125. So the true answer is somewhere from 50 to 125. The doubt is wide because both cuts are in big classes.

    Worked Example 3 — No numbers on the vertical axis

    The histogram below shows the heights of some seedlings (invented data). The vertical axis has no scale. You are told that the 10–15 cm class has 30 seedlings. (a) Find the frequency of every class, and the total. (b) Estimate how many seedlings are between 18 cm and 25 cm tall.

    0510152025303540Height (cm)Frequency density
    Seedling heights. The bars are 3, 12, 8 and 2 small divisions tall, over the classes 0 to 10, 10 to 15, 15 to 20 and 20 to 40 cm.

    When the scale is missing, use the bar you know to find what one unit of area is worth. Measure area as width in centimetres times height in divisions. The 10–15 bar is 5 cm wide and 12 divisions tall, so its area is 60 units. Those 60 units hold 30 seedlings. So one unit of area is worth 30 ÷ 60 = 0.5 seedling.

    (a) The 0–10 bar: 10 × 3 = 30 units, so 15 seedlings. The 15–20 bar: 5 × 8 = 40 units, so 20 seedlings. The 20–40 bar: 20 × 2 = 40 units, so 20 seedlings. The total is 15 + 30 + 20 + 20 = 85 seedlings.

    A second route labels the axis. The 10–15 class has frequency density 30 ÷ 5 = 6 seedlings per cm, shown by 12 divisions, so one division is 0.5. Check: the 0–10 bar is then 1.5 tall, and 1.5 × 10 = 15 ✓.

    (b) From 18 to 20 cm: 2 × 8 = 16 units, so 8 seedlings. From 20 to 25 cm: 5 × 2 = 10 units, so 5 seedlings. The estimate is 8 + 5 = 13 seedlings.

    Trap 1 — Misreading the scale

    One small square is not always 1, and not always 0.1. On another histogram it may be 0.2 or 0.25. Count the squares between two labelled lines, and divide, before you read any bar. A slip of one square on a wide bar becomes a large error: on the 20–50 kg parcel bar, it is 3 parcels.

    Trap 2 — Cutting the frequency instead of the area

    For "between 10 and 15 kg", some students multiply the class frequency, 12, by the slice width, 5, and get 60. That is more parcels than the whole class holds. Multiply the height by the slice width: 1.2 × 5 = 6. Or take the same fraction of the frequency as of the width: five tenths of 12 is 6. A slice can never hold more than its whole class, so that is your check.

    Trap 3 — Writing an estimate as a fact

    A count that cuts through a bar is an estimate, because it rests on the evenly spread assumption. Write "an estimated 27" or "about 12", and keep the unrounded value, such as 11.6, until the end. The other way round, when both ends of a range are class boundaries, the answer is exact, and you should say so.

    Try it yourself

    Work each one on paper first. Open a hint only when you are stuck.

    1. A histogram shows the lengths of 64 leaves (invented data). The heights are 2 for cm, 7 for , 9 for , 4 for and 1 for . (a) How many leaves are 6 cm or longer? (b) Which class has the most leaves? (c) What percentage of the leaves are shorter than 4 cm?

      Hint 1

      Find every frequency first. Each one is height × width.

      Hint 2

      The widths are 4, 2, 2, 4 and 8 cm. The first frequency is 2 × 4 = 8.

      Full answer

      Frequencies: 2 × 4 = 8, 7 × 2 = 14, 9 × 2 = 18, 4 × 4 = 16, 1 × 8 = 8. Total 64 ✓. (a) 6 cm is a boundary, so this is exact: 18 + 16 + 8 = 42 leaves. (b) The 6–8 cm class, with 18. It is also the tallest bar here, but you only know that it holds the most by comparing areas: the 8–12 class is close, with 16. (c) 8 ÷ 64 = 0.125, which is 12.5%. Check: 8 + 14 + 42 = 64 ✓.

    2. Using the leaves in question 1, estimate how many are between 5 cm and 10 cm long. Then give the smallest and largest possible true counts.

      Hint 1

      The range cuts two bars and contains one whole bar. Which ones?

      Hint 2

      From 5 to 6 cm is 1 cm of the 4–6 class, under a height of 7.

      Full answer

      5 to 6 cm: 7 × 1 = 7. 6 to 8 cm: the whole bar, 18. 8 to 10 cm: 4 × 2 = 8. The estimate is 7 + 18 + 8 = 33 leaves. Smallest possible: only the whole bar, 18. Largest possible: all three classes, 14 + 18 + 16 = 48. Check: 7 is half of 14 and 8 is half of 16, and each slice is half its class ✓.

    3. The waiting times of 120 customers are shown in a half-drawn histogram (invented data). The table gives 18 customers for minutes and 35 for . Bars are drawn at height 8 for and 2.5 for . The last class, , has neither. Complete the frequencies and the bar heights.

      Hint 1

      Multiply for the drawn bars, then use the total for the last class.

      Hint 2

      The drawn bars hold 8 × 3 = 24 and 2.5 × 10 = 25 customers.

      Full answer

      Known: 18 + 24 + 35 + 25 = 102. So the last class has 120 − 102 = 18 customers, and its height is 18 ÷ 10 = 1.8. The missing heights are 18 ÷ 2 = 9 for 0 to 2 minutes and 35 ÷ 5 = 7 for 5 to 10 minutes. Check: 9 × 2 + 8 × 3 + 7 × 5 + 2.5 × 10 + 1.8 × 10 = 18 + 24 + 35 + 25 + 18 = 120 ✓.

    4. On a histogram of reaction times, the bar for seconds is 4 cm wide and 6 cm tall, and it represents 48 people. The bar for seconds is 6 cm wide and 2.5 cm tall. (a) How many people are in the 30–45 class? (b) Estimate how many people took between 25 and 40 seconds. (c) Suppose you learn that most times in the 30–45 class are close to 30 seconds. Is your estimate in (b) too high or too low? Explain.

      Hint 1

      Work in the paper's own units. What is 1 cm² of bar worth?

      Hint 2

      The first bar covers 4 × 6 = 24 cm² for 48 people, so 1 cm² is worth 2 people. On the horizontal axis, 4 cm stands for 10 seconds.

      Full answer

      (a) 6 × 2.5 = 15 cm², so 15 × 2 = 30 people. (b) From 25 to 30 seconds is half of the first class: half of 48 = 24. From 30 to 40 seconds is 10 of the 15 seconds of the second class: two thirds of 30 = 20. The estimate is 24 + 20 = 44 people. Check in paper units: 25 to 30 is 2 cm wide, 2 × 6 = 12 cm², which is 24 people ✓. (c) Too low. The estimate assumed the 30 people were spread evenly from 30 to 45 seconds, which puts 20 of them below 40. If most of them are close to 30 seconds, more than 20 are below 40, so the true count in the range is larger than 44.

    The bigger idea

    A histogram is a picture in which area means "how many". Everything in this lecture follows from that one rule. The evenly spread assumption is the simplest model of what happens inside a class, and question 4 showed how it can fail.

    Now imagine a very large set of data, grouped into narrower and narrower classes. The flat tops become small steps, and the steps start to look like a smooth curve. The rule does not change: the area under the curve between two values is the number of values between them. Divide by the total, and the area becomes a proportion, or a probability. This is exactly how the normal distribution is used later: as a smooth curve in which area means probability. Finding areas under curves exactly is what integration does.

    Key concepts in this lesson