2.2 Histograms, Frequency Polygons, and Time Series Graphs

The display the rest of the book uses. What a histogram is and why its boxes are contiguous rather than separated; how to choose a starting point one decimal place finer than the data so that no value lands on a boundary, and how to compute the class width once the number of bars is chosen; the discrete case, where a width of one centres each value in its own bar; the frequency polygon, which plots each class midpoint at its frequency and joins them, anchored at empty classes so the shape closes, and which can be overlaid to compare two distributions; and the time series graph, which plots a measurement against its date and so preserves the chronological order every other display discards.

Subject: Statistics · 70 slides · symbolic lesson

Open the interactive version of this deck

What this lesson covers

The lesson, slide by slide

1. Histograms, Frequency Polygons and Time Series Graphs

Title

Statistics · Section 2.2

Drawing a measurement so its shape, its comparisons and its history can be seen

2. What this lesson gives you

Objectives

Section 2.1 drew data you could list by category or read off a stemplot. This section draws the display the rest of the book leans on, the histogram, and two relatives of it: the frequency polygon and the time series graph.

The arithmetic in this lesson is gentle: subtract, divide, round up, count. The decisions are where the marks go, because a histogram is something you design. Where does the first bar start? How wide is each bar? Which bar owns a value that sits on an edge?

Keep one question running through every slide: if someone else drew this picture from the same data, would they get the same bars? Most of the rules in this section exist so that the answer is yes.

OpenStax Introductory Statistics 2e, §2.2 Histograms, Frequency Polygons, and Time Series Graphs §2.2, p. 75 — the section opening

3. From a bar graph of districts to a hundred heights

Warm-up

Start from the last thing you drew. In Section 2.1, Try It 2.6 gave the percent of Park City's registered voters living in each of six districts, and you drew a bar graph with one bar per district.

district123456
registered voters15.5%12.2%9.8%17.4%22.8%22.3%

Discussion prompt

Now you are handed the heights of 100 soccer players, recorded to the nearest half inch. Could you draw them the same way, one bar per recorded value? Say what would go wrong, and what decision a bar graph never asked you to make.

Notice the check in the answer: the district percentages total exactly 100, so every registered voter is in exactly one bar. A histogram makes the same promise about every data value, and keeping that promise is harder than it looks.

OpenStax Introductory Statistics 2e, §2.2 Histograms, Frequency Polygons, and Time Series Graphs §2.1, pp. 74-75 — Try It 2.6 and Table 2.12

4. What a histogram is

Section

Part 1

5. A histogram: bars that touch, over a number line

Concept

The book defines a histogram by its look before its use. It consists of contiguous, that is adjoining, boxes. The horizontal axis is labelled with what the data represent, such as height in inches; the vertical axis is labelled frequency or relative frequency.

Figure (svg): Two panels. Left: a bar graph of the percent of registered voters in six districts of Park City, with a gap between every bar and district numbers 1 to 6 underneath. Right: a histogram of the 100 players' heights, eight bars touching one another along a number line from 59.95 to 75.95 inches.

The gaps on the left say that district 2 and district 3 have nothing in between. On the right there is no gap to leave: a height of 65.95 inches is perfectly possible, so the bars must meet.

Put the two pictures side by side and the reason for touching bars is visible. A bar graph's horizontal axis holds categories, and nothing lies between district 2 and district 3, so gaps are honest. A histogram's horizontal axis is a number line, and every point on it is a possible value.

\[ \text{histogram} = \text{intervals of a number line} + \text{a count for each} \]

A gap in a histogram would therefore be a claim: that no data value falls in that stretch. If a class really is empty, you draw it as a bar of height zero sitting on the axis, not as a hole in the picture.

The book adds two practical notes. A histogram can display large data sets readily, and a rule of thumb is to use one when there are 100 values or more. And like the stemplot, it shows you the shape, the centre and the spread of the data at a glance.

OpenStax Introductory Statistics 2e, §2.2 Histograms, Frequency Polygons, and Time Series Graphs §2.2, p. 75 — the definition of a histogram

6. Frequency and relative frequency

Concept

The vertical axis of a histogram can count in two ways. Frequency is the number of times an answer occurs. Relative frequency is that count as a share of all the data.

relative frequency — The frequency of a value or class divided by the total number of data values in the sample.

\[ RF = \dfrac{f}{n} \]

The book's example is an English class of 40 students in which three scored from 90 to 100 percent. The frequency of that class is 3 and the total is 40, so you divide.

\[ RF = \dfrac{3}{40} = 0.075 \]

Then turn the share into words a reader understands: 0.075 is 7.5 hundredths, so 7.5 percent of the class scored in the top interval. Notice that the interval 90 to 100 percent is itself a range of a quantitative measure, which is exactly what a histogram bar represents.

\[ 0.075 \times 100\% = 7.5\% \]

Why bother? A frequency of 3 means little without the class size: 3 of 40 is a small group, 3 of 5 would be most of the room. Relative frequency carries its own denominator, so it can be compared across groups of different sizes.

OpenStax Introductory Statistics 2e, §2.2 Histograms, Frequency Polygons, and Time Series Graphs §2.2, p. 75 — relative frequency

7. Decoding the relative frequency formula

Notation

The formula is short, and each letter is worth a sentence. Read it once slowly so that you never divide by the wrong total.

Annotate

On: \( RF = \dfrac{f}{n} = \dfrac{3}{40} = 0.075 \)

  • Relative frequency: the fraction of all the data that falls in one class. It is always between 0 and 1, and the relative frequencies of all the classes add to 1.
  • The frequency of this one class: how many data values landed in it. Here, the three students who scored from 90 to 100 percent.
  • The total number of data values, which is also the sum of all the frequencies. Here, the 40 students in the whole class, never just the students in nearby classes.
  • A proportion, read as 7.5 percent. Multiplying every relative frequency by 100 gives the percent frequency that the book also allows on the vertical axis.

Because n is the sum of the frequencies, there is a built-in check whenever you make a relative-frequency table: the column must add to 1, apart from rounding. If it adds to 0.9, a class is missing; if it adds to 1.1, something was counted twice.

\[ \sum \dfrac{f}{n} = \dfrac{\sum f}{n} = \dfrac{n}{n} = 1 \]

OpenStax Introductory Statistics 2e, §2.2 Histograms, Frequency Polygons, and Time Series Graphs §2.2, p. 75 — the definitions of f, n and RF

8. One picture, two rulers

Picture it

The book says something that sounds surprising: the histogram has the same shape whether the vertical axis says frequency or relative frequency. The picture below shows why, using the 100 players' heights you will build in Part 2.

Figure (svg): The 100 heights drawn as a histogram with two vertical scales. The left scale, in blue-grey, counts players from 0 to 40; the right scale, in orange, reads relative frequency from 0 to 0.4. The counts 5, 3, 15, 40, 17, 12, 7 and 1 are written above the bars.

One picture, two rulers. Dividing every count by 100 relabels the vertical axis and changes nothing else, because every bar is divided by the same number.

To switch from counts to relative frequencies you divide every bar by the same number, 100. Dividing everything by the same amount shrinks the scale, not the shape: the tallest bar is still the tallest, and it is still 8 times the height of the bar holding 5.

\[ \dfrac{40}{100} = 0.40 \qquad \dfrac{5}{100} = 0.05 \qquad \dfrac{0.40}{0.05} = \dfrac{40}{5} = 8 \]

So the choice of label is about the reader, not the shape. Frequency answers how many; relative frequency answers what share, and it is the version to use when you compare two groups of different sizes, or when the next chapters start reading areas as probabilities.

OpenStax Introductory Statistics 2e, §2.2 Histograms, Frequency Polygons, and Time Series Graphs §2.2, p. 75 — the same shape with either label

9. Why every bar must be the same width

Concept

Your eye does not read a bar's height. It reads how much ink the bar covers, its area. A histogram is honest only if the area of each bar is in proportion to the number of values it holds.

\[ \text{area} = \text{width} \times \text{height} \]

When every bar has the same width, the width is a common factor, and the areas are in exactly the proportion of the counts. For the heights, with width 2, the tallest bar covers 80 square units and the bar of 15 players covers 30.

\[ 2 \times 40 = 80, \qquad 2 \times 15 = 30, \qquad \dfrac{80}{30} = \dfrac{40}{15} \approx 2.67 \]

The ratio of the areas is the ratio of the players, so the picture tells the truth. Change one bar's width and that breaks: a bar three times as wide covers three times the ink for the same count.

That is why the recipe in Part 2 fixes one width for all the bars before any counting starts, and why the book's section on misleading graphs singles out varying the width of the categories. Part 5 shows what that trick looks like.

OpenStax Introductory Statistics 2e, §2.2 Histograms, Frequency Polygons, and Time Series Graphs §2.2, pp. 85-86 — histograms and varying widths

10. Histogram or bar graph?

Discrimination

Before drawing anything, decide which display the data need. The test is the horizontal axis: is it a list of categories, or a stretch of a number line?

Sort into buckets

Sort each data set by the display it calls for.

Histogram (number line)
heights of 100 soccer players, to the nearest half inch; times for 200 runners to finish a mile; the number of siblings of each of 120 students
Bar graph (categories)
the favourite colour of each of 30 students; percent of registered voters living in each of six districts; the blood type of each of 80 donors
h
Heights, mile times and sibling counts are all quantitative. Each has a number line underneath it, so the classes are intervals of that line and the bars touch.
b
Colours, districts and blood types are names of categories. There is nothing between one category and the next, so the bars stand apart.

The sibling counts are the case to slow down on. They are discrete, whole numbers only, but they are still numbers on a number line, so they get a histogram. Part 3 shows how to centre each whole number in its own bar.

The districts are numbered 1 to 6, and that does not make them quantitative. The numbers are labels, like jersey numbers; the data are the districts, and a bar graph is right.

OpenStax Introductory Statistics 2e, §2.2 Histograms, Frequency Polygons, and Time Series Graphs §2.2, p. 75 — what a histogram displays

11. Read a share off the histogram

Estimation

A histogram is a reading instrument as well as a picture. Use the two-ruler figure of the heights to estimate before you count.

Predict first

Roughly what share of the 100 players is shorter than 65.95 inches?

  • about 0.05
  • about 0.23
  • about 0.40
  • about 0.63

Correct: about 0.23

Why: Shorter than 65.95 means the first three bars, with relative frequencies 0.05, 0.03 and 0.15. Their sum is 0.23, so about 23 of the 100 players. The tall bar of 0.40 starts at 65.95, so it is not included.

\[ 0.05 + 0.03 + 0.15 = 0.23 \]

Shares of a histogram add, because the classes do not overlap and every value lands in exactly one of them. The same move answers questions such as the share between 63.95 and 67.95: two bars, 0.15 and 0.40, so 0.55.

Notice that the question asked about 65.95, a class edge. That is the kind of question a histogram answers exactly. A question about 65 inches would cut through the middle of a bar, and the histogram alone could not tell you how that bar's players are split, which is one of the prices of grouping.

The tempting slip is 0.40, which reads the tallest bar as if it were the answer to every question. Always find which bars the question covers before you read any heights.

OpenStax Introductory Statistics 2e, §2.2 Histograms, Frequency Polygons, and Time Series Graphs §2.2, p. 77 — Figure 2.5

12. Choosing the start and the width

Section

Part 2

13. Why the edges carry one more decimal place

Concept

Every histogram begins with a decision the data cannot make for you: where the first bar starts. The book's rule is to start below the smallest value, at a number carried to one more decimal place than the most precise value in the data.

Figure (svg): A number line from 63.5 to 66 inches with tick marks at every half inch. Stacks of dots show how many players were recorded at each value: 3 at 63.5, 7 at 64, 8 at 64.5 and 10 at 66. A solid red line at 64 runs straight through the stack of seven. Dashed green lines at 63.95 and 65.95 pass between the ticks, touching no dots.

The red edge has to decide which bar owns seven players. The green edges never face that question, because nobody can be recorded at 63.95.

The picture shows the reason. The heights were recorded to the nearest half inch, so every value ends in .0 or .5. An edge drawn at 64 sits on top of seven players, and you would have to decide which bar owns them.

An edge at 63.95 has two decimal places, one more than any height. No recorded height can equal it, so the question of ownership never arises. The book puts it in one sentence: when the starting point and other boundaries are carried to one additional decimal place, no data value will fall on a boundary.

\[ \text{start} = \text{smallest value} - \tfrac{1}{2}\text{ of the next decimal place} \]

Half of the next place is the convenient amount to subtract: 0.5 for whole numbers, 0.05 for one decimal place, 0.005 for two. The book calls such a starting point more precise than the data, and that extra precision is exactly what keeps it off them.

OpenStax Introductory Statistics 2e, §2.2 Histograms, Frequency Polygons, and Time Series Graphs §2.2, p. 75 — choosing a starting point

14. The histogram recipe, in five moves

Concept

Before the details, see the whole construction at once. Every histogram of continuous data in this section is built by the same five moves, in the same order, and each later slide fills in one of them.

  1. Read the data: the smallest and largest values, the precision of the most precise value, and how many values there are
  2. Set the starting point and ending value: half of the next decimal place below the smallest value and above the largest
  3. Choose the number of bars, usually five to fifteen, and divide the stretch by it to get the width
  4. Round the width up, then add it repeatedly to the start to list every boundary
  5. Count the values in each class, draw touching bars of that height, and check the counts add to n

\[ \text{width} = \dfrac{\text{ending value} - \text{starting point}}{\text{number of bars}} \]

Notice which moves are calculations and which are choices. The start and end follow from the data by rule. The number of bars is yours to choose, and the width follows from that choice. Two people who choose the same number of bars and follow the rules will draw identical histograms.

Notice too that the last move ends with a check. A histogram is a count, and a count can be verified: the bars must hold every value exactly once, no more and no fewer.

OpenStax Introductory Statistics 2e, §2.2 Histograms, Frequency Polygons, and Time Series Graphs §2.2, pp. 75-76 — constructing a histogram

15. The book's four starting points

Worked example

The book gives four quick cases before its main example. In each, find the most precise value, step one decimal place beyond it, and subtract half of that place from the smallest value.

Smallest and most precise value 6.1

Why: one decimal place in the data, so the edge gets two; half of 0.1 is 0.05.

\[ 6.1 - 0.05 = 6.05 \]

Most precise value 2.23, smallest 1.5

Why: the precision comes from 2.23, two decimal places, so subtract half of 0.01; the smallest value supplies only the starting height.

\[ 1.5 - 0.005 = 1.495 \]

Most precise value 3.234, smallest 1.0

Why: three decimal places, so subtract half of 0.001.

\[ 1.0 - 0.0005 = 0.9995 \]

All integers, smallest 2

Why: whole numbers have no decimal places, so subtract half of 1.

\[ 2 - 0.5 = 1.5 \]

The second case is the one people get wrong. The smallest value, 1.5, has only one decimal place, and it is tempting to subtract 0.05. But 1.45 is a value the data could contain, since other values carry two places. Precision is set by the most precise value in the whole set, not by the smallest one.

Check each start against the data

Why: each start is below its smallest value and has one more decimal place than any value in its data set, so no value can sit on it.

\[ 6.05 < 6.1, \quad 1.495 < 1.5, \quad 0.9995 < 1.0, \quad 1.5 < 2 \]

Those two properties are the whole rule: below everything, and finer than everything. Any number with both would do; these are simply the most convenient ones.

OpenStax Introductory Statistics 2e, §2.2 Histograms, Frequency Polygons, and Time Series Graphs §2.2, p. 75 — the four starting points

16. Match each data set to its starting point

Matching

Four new data sets. For each, read the precision from the most precise value and the starting height from the smallest, then match.

Match the pairs

  • a. smallest 3.2, every value to one decimal place
  • b. all whole numbers, smallest 15
  • c. smallest 0.84, values to two decimal places
  • d. smallest 12, most precise value 12.25
  • w. 3.15
  • x. 14.5
  • y. 0.835
  • z. 11.995

Why: One decimal place means subtracting 0.05, so 3.2 becomes 3.15. Whole numbers mean subtracting 0.5, so 15 becomes 14.5. Two decimal places mean subtracting 0.005, so 0.84 becomes 0.835. The last set has a whole-number smallest value but a most precise value with two places, so it also needs 0.005, and 12 becomes 11.995.

The last pair is the one to check. If you matched the smallest value 12 with 11.5, you read the precision from the smallest value rather than from the most precise one, and 11.5 is a value a two-decimal data set could easily contain.

\[ 12 - 0.005 = 11.995 \]

Each answer ends in a 5 one place beyond the data. That trailing 5 is the signature of a well-chosen edge, and it is a quick thing to look for on any histogram someone else has drawn.

OpenStax Introductory Statistics 2e, §2.2 Histograms, Frequency Polygons, and Time Series Graphs §2.2, p. 75 — choosing a starting point

17. Example 2.7: start, end and width for 100 heights

Worked example

The book's main example: the heights, in inches to the nearest half inch, of 100 semiprofessional soccer players. Height is measured, so the data are continuous. The smallest value is 60 and the largest is 74.

Find the starting point

Why: the most precise values, such as 61.5, have one decimal place, so subtract half of 0.1 from the smallest value.

\[ 60 - 0.05 = 59.95 \]

Find the ending value

Why: the same half step, added to the largest value, gives an edge above everyone.

\[ 74 + 0.05 = 74.05 \]

Choose the number of bars

Why: the book chooses eight. This is a choice, not a calculation; five to fifteen bars is the usual range.

\[ \text{bars} = 8 \]

Divide the stretch by the number of bars

Why: subtract the start from the end, then share that stretch equally among the eight bars.

\[ \text{width} = \dfrac{74.05 - 59.95}{8} = \dfrac{14.1}{8} \]

Do the division

Why: the exact width is an awkward number.

\[ \dfrac{14.1}{8} = 1.7625 \]

Round the width up

Why: the book rounds up to 2, even though ordinary rounding would give 1.8; rounding up can only make the bars cover more, never less.

\[ 1.7625 \;\to\; 2 \]

Check that eight bars of width 2 cover everyone

Why: start at 59.95 and add eight widths; the last edge must be above the tallest player.

\[ 59.95 + 8 \times 2 = 75.95 > 74 \]

So eight bars of width 2 run from 59.95 to 75.95 and contain every height, with room to spare at the top. The book notes that rounding up is often necessary even if it goes against the standard rules of rounding, and the next slides show what happens if you do not.

OpenStax Introductory Statistics 2e, §2.2 Histograms, Frequency Polygons, and Time Series Graphs §2.2, pp. 75-76 — Example 2.7

18. Slide the number of bars and watch the width

Tweak it

The width is the stretch divided by the number of bars, so the two move in opposite directions. The curve below plots the width for every number of bars from 3 to 16; the sliders set the smallest and largest data values, recorded to one decimal place.

Parameter explorer

With the heights' values 60 and 74, read the width at 8 bars. Then find the number of bars at which the width first drops below 1. Finally set the values to 9 and 14, the shoe sizes, and read the width at 6 bars.

\[ \text{width} = \dfrac{({hi} + 0.05) - ({lo} - 0.05)}{\text{bars}} \]

  • lo — from 0 to 70: smallest value
  • hi — from 5 to 80: largest value

What to notice: at 8 bars the curve passes through 1.7625, the width you just computed. More bars means thinner bars; at 14 bars the width is still just over 1, and at 15 it falls to 0.94.

\[ \dfrac{14.1}{14} \approx 1.007, \qquad \dfrac{14.1}{15} = 0.94 \]

The shoe sizes give a stretch of 5.1 and a width of 0.85 at 6 bars. Keep that number in mind: it looks perfectly innocent, and later in this part it puts an edge straight onto a shoe size.

OpenStax Introductory Statistics 2e, §2.2 Histograms, Frequency Polygons, and Time Series Graphs §2.2, p. 76 — calculating the width

19. How many bars? The square-root guideline

Estimation

The book mentions a guideline some people follow for the number of bars: take the square root of the number of data values and round to the nearest whole number.

\[ \text{bars} \approx \sqrt{n} \]

Predict first

A data set has 150 values. About how many bars does the square-root guideline suggest?

  • about 7
  • about 12
  • about 15
  • about 75

Correct: about 12

Why: The square root of 150 lies between 12 and 13, since 12 times 12 is 144 and 13 times 13 is 169. It is about 12.25, which rounds to 12 bars, the book's own example.

\[ \sqrt{150} \approx 12.25 \;\to\; 12 \text{ bars} \]

Try it on the other data sets in this section. For the 100 heights the guideline gives 10 bars, where the book chose 8; for the 25 video-game values it gives 5, the number the book's figure uses.

\[ \sqrt{100} = 10, \qquad \sqrt{25} = 5 \]

It is a guideline, not a law. Too few bars hide the shape in a couple of blocks; too many leave most bars holding one or two values. Five to fifteen bars, and a look at whether the picture shows a shape, is the real test.

OpenStax Introductory Statistics 2e, §2.2 Histograms, Frequency Polygons, and Time Series Graphs §2.2, p. 76 — the square-root guideline

20. Example 2.7: nine edges, and the players counted into them

Worked example

With a start of 59.95 and a width of 2, the boundaries are found by adding 2 again and again. Eight bars need nine edges, because each bar shares its upper edge with the next bar's lower edge and the last bar needs one of its own.

Add the width to get each edge

Why: start at 59.95 and keep adding 2.

\[ 59.95, \; 61.95, \; 63.95, \; 65.95, \; 67.95 \]

Keep going to the ninth edge

Why: four more additions of 2.

\[ 69.95, \; 71.95, \; 73.95, \; 75.95 \]

Figure (svg): A dot plot of all 100 heights from 60 to 74 inches, each dot one player, stacked up to 12 high at 67. Nine dashed vertical lines mark the class edges 59.95, 61.95 and so on up to 75.95. The number of dots between each pair of lines is written near the top: 5, 3, 15, 40, 17, 12, 7 and 1.

Read across the yellow numbers and you have the frequency table. Every dot lies strictly between two orange lines, because no dashed line passes through a value anyone could have been recorded at.

Count the players in the first four bars

Why: 60 to 61.5 go in the first bar, the three 63.5s in the second, 64 and 64.5 in the third, and 66 to 67.5 in the fourth.

\[ 1 + 1 + 2 + 1 = 5, \quad 3, \quad 7 + 8 = 15, \quad 10 + 11 + 12 + 7 = 40 \]

Count the players in the last four bars

Why: 68 to 69.5 in the fifth, 70 to 71 in the sixth, 72 to 73.5 in the seventh, and the single 74 in the last.

\[ 2 + 10 + 5 = 17, \quad 6 + 3 + 3 = 12, \quad 3 + 2 + 1 + 1 = 7, \quad 1 \]

Look at the dot plot while you count: every orange line falls in an empty gap between half inches, so you never have to decide where a player belongs. That is the starting-point rule paying off.

Check that every player was counted once

Why: the eight frequencies must add to the 100 players in the data set.

\[ 5 + 3 + 15 + 40 + 17 + 12 + 7 + 1 = 100 \]

The total is exactly 100, so nobody was missed and nobody was counted in two bars. Run this check on every frequency table you make; it catches most counting slips.

OpenStax Introductory Statistics 2e, §2.2 Histograms, Frequency Polygons, and Time Series Graphs §2.2, p. 76 — Example 2.7, the boundaries

21. Example 2.7: from counts to the relative-frequency histogram

Worked example

The book draws the heights with relative frequency on the vertical axis, Figure 2.5. Each bar's height is its count divided by the 100 players.

Divide the first four counts by 100

Why: dividing by 100 moves the decimal point two places.

\[ \dfrac{5}{100} = 0.05, \quad \dfrac{3}{100} = 0.03, \quad \dfrac{15}{100} = 0.15, \quad \dfrac{40}{100} = 0.40 \]

Divide the last four

Why: the same move.

\[ \dfrac{17}{100} = 0.17, \quad \dfrac{12}{100} = 0.12, \quad \dfrac{7}{100} = 0.07, \quad \dfrac{1}{100} = 0.01 \]

Figure (svg): A relative-frequency histogram of the heights with eight touching bars over the edges 59.95 to 75.95. The bar heights, written above them, are 0.05, 0.03, 0.15, 0.40, 0.17, 0.12, 0.07 and 0.01.

The tallest bar says that 0.40 of the players, forty in every hundred, stand between 65.95 and 67.95 inches. The eight heights add to exactly 1, the whole squad.

Now read the picture in context. Forty percent of the players are between 65.95 and 67.95 inches tall, and the bars fall away on both sides, more slowly towards the tall end: 20 players are taller than 69.95 inches, but only 8 are shorter than 63.95.

\[ 12 + 7 + 1 = 20 \qquad 5 + 3 = 8 \]

Check that the relative frequencies add to 1

Why: every player is in exactly one bar, so the shares must make up the whole squad.

\[ 0.05 + 0.03 + 0.15 + 0.40 + 0.17 + 0.12 + 0.07 + 0.01 = 1.00 \]

They add to exactly 1, and each one agrees with the label printed on the book's Figure 2.5. So the typical player in this squad is about 66 to 68 inches tall, a little over five and a half feet.

OpenStax Introductory Statistics 2e, §2.2 Histograms, Frequency Polygons, and Time Series Graphs §2.2, pp. 76-77 — Example 2.7 and Figure 2.5

22. Trap: rounding the width to the nearest value

Trap

The trap

A student computes the width exactly as the book does, then rounds it the ordinary way, to one decimal place:

\[ 1.7625 \to 1.8 \text{ or } 1.7 \]

Rounding down to 1.7 looks harmless. Build the edges:

\[ 59.95 + 8 \times 1.7 \]

\[ = 73.55 \]

The eighth bar ends at 73.55, and the tallest player, at 74 inches, is in no bar at all.

The fix

Round the width up, never down.

\[ 1.7625 \to 2 \]

\[ 59.95 + 8 \times 2 = 75.95 \]

Rounding up makes the bars cover more of the line than the data need, which is harmless; rounding down covers less, which loses data.

Keeping the exact width is no rescue either: its fifth edge is 59.95 plus four widths, which is exactly 67, where twelve players stand.

23. The book's note: a width of 1.76 also works

Worked example

The book adds a remark worth testing rather than trusting: for this example, using 1.76 as the width would also work. Two things must be true for that, and you can check both.

Build the edges with width 1.76

Why: start at 59.95 and add 1.76 each time.

\[ 59.95, \; 61.71, \; 63.47, \; 65.23, \; 66.99 \]

Continue to the ninth edge

Why: four more additions.

\[ 68.75, \; 70.51, \; 72.27, \; 74.03 \]

Ask whether any edge equals a height

Why: every height ends in .0 or .5, and none of these edges does: the endings are .95, .71, .47, .23, .99, .75, .51, .27 and .03.

\[ 68.75 \ne 68.5, \; 68.75 \ne 69 \]

Ask whether the bars reach the tallest player

Why: the last edge must be above 74.

\[ 74.03 > 74 \]

Both conditions hold, so 1.76 is a legal width. It is still a worse choice for a reader, because edges such as 63.47 and 70.51 are hard to read off an axis, and that is why the book prefers the round number 2.

Check what would break the remark

Why: 1.76 is slightly less than the exact width 1.7625, so the last edge falls just short of the ending value 74.05; it works only because 74.03 still clears 74.

\[ 59.95 + 8 \times 1.76 = 74.03 < 74.05 \]

Shave the width a little further, to 1.75, and the last edge becomes 73.95, below the tallest player. Rounding down is a gamble that happened to pay off here; rounding up never needs the gamble.

OpenStax Introductory Statistics 2e, §2.2 Histograms, Frequency Polygons, and Time Series Graphs §2.2, p. 76 — the Note to Example 2.7

24. Error analysis: four class schemes for the heights

Error analysis

Four students each propose eight classes for the 100 heights. One scheme is sound and three have a flaw. Find each flaw before opening the notes.

Annotate

On: \( 60 + 2k, \quad 59.5 + 2k, \quad 59.95 + 1.7k, \quad 59.95 + 2k \)

  • Every edge is a whole number: 60, 62, 64 and so on. Each of those is a possible height, and in fact 7 players stand at 64 and 10 at 66, so the edges sit on data and the class of those players is ambiguous.
  • This start is below the data, but it has only one decimal place. The edges 59.5, 61.5, 63.5 and so on are half inches, exactly the values heights were recorded in, and one player stands at 61.5 and three at 63.5.
  • The start is right but the width was rounded down. Eight bars end at 73.55, and the player who is 74 inches tall falls outside every bar.
  • The book's scheme. Every edge ends in .95, finer than any height, and the last edge, 75.95, is above the tallest player. Nothing sits on an edge and nothing is left out.

The three flaws fall into two families. Two schemes put edges on possible values, which makes the counts depend on a tie-breaking rule nobody stated. One scheme loses a value off the end. Both families are caught by the same two questions: does any edge equal a recorded value, and does the last edge clear the largest value?

Notice also what the sound scheme does not need: it does not need the start to be a round number, or the width to equal the exact quotient. It needs edges that no recorded value can equal and bars that together reach past the largest value. Those two conditions are what you are really checking when you mark someone else's histogram.

\[ \text{last edge} = 59.95 + 8 \times 2 = 75.95 > 74 \]

Note that the start of 59.5 fails for a subtle reason: it is one decimal place, not two. Subtracting 0.5 is the rule for whole-number data, and these data are recorded to the half inch.

OpenStax Introductory Statistics 2e, §2.2 Histograms, Frequency Polygons, and Time Series Graphs §2.2, pp. 75-76 — Example 2.7

25. Try It 2.7: before any arithmetic

Step zero

The book's Try It gives the shoe sizes of a group of students, measured in whole and half sizes, and asks for a histogram with six bars. Before you subtract anything, look at the data.

\[ 9, \; 9, \; 9.5, \; 9.5, \; 10 \times 6, \; 10.5 \times 8, \; 11 \times 14, \; 11.5 \times 7, \; 12 \times 8, \; 12.5 \times 4, \; 14 \]

Discussion prompt

What is the precision of these data, what will the starting point and ending value be, and how many students are actually listed? Count them rather than trusting the heading.

This is a misprint in the book: the list and the heading disagree by two. Trust the data you can count. Every relative frequency for this data set uses 52 as its denominator, and a table that divides by 50 will add to more than 1.

Step zero, before a histogram, is always the same: find the precision, find the smallest and largest values, and count the data. Each of the three feeds a different line of the recipe.

OpenStax Introductory Statistics 2e, §2.2 Histograms, Frequency Polygons, and Time Series Graphs §2.2, p. 77 — Try It 2.7

26. Try It 2.7: six bars for the shoe sizes

Worked example

Now build the classes. The data are 52 shoe sizes from 9 to 14 in half sizes, and the book asks for six bars.

Start and end

Why: half of 0.1 below the smallest size and above the largest.

\[ 9 - 0.05 = 8.95, \qquad 14 + 0.05 = 14.05 \]

Divide the stretch by six

Why: the same move as for the heights.

\[ \text{width} = \dfrac{14.05 - 8.95}{6} = \dfrac{5.1}{6} = 0.85 \]

List the edges

Why: add 0.85 six times; the width came out exact, so no rounding is needed.

\[ 8.95, \; 9.80, \; 10.65, \; 11.50, \; 12.35, \; 13.20, \; 14.05 \]

Figure (svg): A dot plot of 52 shoe sizes from 9 to 14, stacked up to 14 high at size 11. Seven vertical lines mark the edges from a width of 0.85: 8.95, 9.8, 10.65, 11.5, 12.35, 13.2 and 14.05. The line at 11.5 is drawn in red and runs straight through the stack of seven students who wear size 11.5.

Six of the seven edges fall in the empty space between half sizes. The fourth lands exactly on 11.5, because three widths of 0.85 make 2.55, and 8.95 plus 2.55 is a size people actually wear.

Look for an edge on a value

Why: the fourth edge is 8.95 plus three widths.

\[ 8.95 + 3 \times 0.85 = 8.95 + 2.55 = 11.5 \]

And 11.5 is a shoe size seven students wear. The starting point had two decimal places, but adding 0.85 three times landed the edge back on a half size. The extra-decimal rule protects the start; it does not protect every edge after it, unless the width is chosen with care too.

Check the damage by counting the dots on that edge

Why: the size-11.5 stack sits on the line, so the fourth and fifth bars depend on a tie-breaking rule.

\[ 11.5: \; 7 \text{ students on the edge} \]

With the book's later rule, a value on a boundary counts in the bar to its right, the classes hold 4, 14, 14, 15, 4 and 1 students. That adds to 52, but seven of those placements were a decision, not a reading.

OpenStax Introductory Statistics 2e, §2.2 Histograms, Frequency Polygons, and Time Series Graphs §2.2, p. 77 — Try It 2.7

27. Does one extra decimal place guarantee clear edges?

Counterexample

Here is a claim you might take away from Example 2.7: if the starting point has one more decimal place than the data, no value can ever fall on a boundary. The shoe sizes are a counterexample.

Discussion prompt

Explain why the claim fails for the shoe sizes, and find a width that repairs it while keeping the starting point 8.95 and roughly six bars.

Figure (svg): A histogram of the 52 shoe sizes with six touching bars of width 1 from 8.95 to 14.95. The counts are 4, 14, 21, 12, 0 and 1; the fifth bar, 12.95 to 13.95, is empty and drawn as a gap of height zero on the axis.

Every edge now ends in .95, so neither a whole size nor a half size can sit on one. The empty fifth class stays in the picture as a bar of height zero: nobody in this class wears size 13 or 13.5.

\[ 4 + 14 + 21 + 12 + 0 + 1 = 52 \]

Six bars of width 1 reach 14.95, past the largest size 14, and the counts add to all 52 students. The fifth class is empty: nobody in this group wears size 13 or 13.5. It stays on the axis as a bar of height zero, because leaving it out would shift the last bar and put size 14 in the wrong place.

So the full rule has two parts. Start one decimal place finer than the data, and choose a width, rounded up, whose multiples keep every edge at that finer ending. A whole-number width does both. Even the exact heights width fails this test: 1.7625 times 4 is 7.05, which carries 59.95 to exactly 67, a height twelve players share.

OpenStax Introductory Statistics 2e, §2.2 Histograms, Frequency Polygons, and Time Series Graphs §2.2, p. 77 — Try It 2.7

28. Discrete data and the boundary rule

Section

Part 3

29. Discrete data: centre each value in its own bar

Concept

When the data are counts, the starting-point rule gets simpler and the width often chooses itself. Counts are whole numbers, so half of one, 0.5, is the step that takes an edge off the data.

Figure (svg): A number line with the whole numbers 1 to 6 marked by dots. Above each whole number is a box of width 1 running from half a unit below it to half a unit above it, so the edges are 0.5, 1.5, 2.5, 3.5, 4.5, 5.5 and 6.5. The box around 3 is highlighted, with arrows showing half a book on each side.

Each count sits in the exact middle of its own bar. The bar's label and the data value are the same number, which is what makes a discrete histogram easy to read.

Start half a unit below the smallest count and use a width of 1. Then every edge is a half-number, and every whole number sits in the exact middle of its own bar: 1 between 0.5 and 1.5, 2 between 1.5 and 2.5, and so on.

\[ \text{bar for } k: \quad k - 0.5 \text{ to } k + 0.5 \]

The book recommends this when the data are discrete and there are not too many different values. Its advantage is that nothing is grouped: each bar is one possible answer, the tick under the middle of the bar names it, and the bar's height is simply how many people gave that answer.

The number of bars is then the stretch divided by the width of 1, which is just the number of possible values from the smallest to the largest. If some value in that range never occurs, its bar is still there, at height zero.

\[ \text{bars} = \dfrac{\text{ending value} - \text{starting point}}{1} \]

OpenStax Introductory Statistics 2e, §2.2 Histograms, Frequency Polygons, and Time Series Graphs §2.2, p. 77 — discrete data and a width of one

30. Example 2.8: books bought by 50 students

Worked example

The number of books bought by 50 part-time students at ABC College. Books are counted, so the data are discrete, with values from 1 to 6. Eleven students bought one book, ten bought two, sixteen three, six four, five five and two six.

Find the starting point and ending value

Why: the data are integers, so step half a book below the smallest value and above the largest.

\[ 1 - 0.5 = 0.5, \qquad 6 + 0.5 = 6.5 \]

Choose a width that centres each value

Why: with a width of 1, each whole number lands in the middle of a bar.

\[ \text{width} = 1 \]

Count the bars

Why: divide the stretch by the width.

\[ \text{bars} = \dfrac{6.5 - 0.5}{1} = 6 \]

Fill in the book's blanks

Why: the 4 sits between 3.5 and 4.5, the 5 between 4.5 and 5.5, and the 6 between 5.5 and 6.5.

\[ 4: 3.5 \text{ to } 4.5, \quad 5: 4.5 \text{ to } 5.5, \quad 6: 5.5 \text{ to } 6.5 \]

Set each bar's height to its count

Why: the frequencies come straight from the list.

\[ 11, \; 10, \; 16, \; 6, \; 5, \; 2 \]

Figure (svg): A frequency histogram of books bought by 50 part-time students. Six touching bars of width 1 centred on 1 to 6, with heights 11, 10, 16, 6, 5 and 2.

The ticks sit under the middles of the bars, not under their edges, because each bar stands for exactly one count of books. Three books is the most common answer.

Read the result in context: the most common purchase was three books, by 16 of the 50 students, and the counts fall away quickly above that. Only 2 students, 4 percent, bought six books.

Check the frequencies against the number of students

Why: every student bought some number of books from 1 to 6, so the six frequencies must add to 50.

\[ 11 + 10 + 16 + 6 + 5 + 2 = 50 \]

They do. The book's calculator steps put the values 1 to 6 in one list and the six frequencies in another, then draw the histogram with Xmin 0.5, Xmax 6.5 and a scale of 1: exactly the start, end and width found here.

OpenStax Introductory Statistics 2e, §2.2 Histograms, Frequency Polygons, and Time Series Graphs §2.2, pp. 77-78 — Example 2.8 and Figure 2.6

31. Try It 2.8: fill in the intervals

Fill the middle

Try It 2.8 records the number of sports played by 50 student athletes: 20 play one sport, 22 play two and 8 play three. The book asks you to finish its sentence about the intervals.

Fill in the blanks

A width of one places the 1 in the middle of the interval 0.5 to 1.5, the 2 in the middle of the interval from 1.5 to 2.5, and the 3 in the middle of the interval from 2.5 to 3.5.

Why: Each interval runs from half a unit below the count to half a unit above it, so 1 is centred in 0.5 to 1.5, 2 in 1.5 to 2.5, and 3 in 2.5 to 3.5. Neighbouring intervals share their edges, which is why 1.5 and 2.5 each appear twice.

Notice that 1.5 is the upper edge of the first bar and the lower edge of the second. Shared edges are what make the bars touch, and since no athlete can play one and a half sports, nobody ever has to be assigned to an edge.

\[ \text{bars} = \dfrac{3.5 - 0.5}{1} = 3 \]

Three possible answers, three bars. The heights come next, and the check is the same as for the books: the three frequencies must add to the 50 athletes.

OpenStax Introductory Statistics 2e, §2.2 Histograms, Frequency Polygons, and Time Series Graphs §2.2, pp. 78-79 — Try It 2.8

32. Try It 2.8: the sports histogram, with shares

Worked example

Finish Try It 2.8 by drawing the bars and turning each count into a relative frequency, so the picture answers what share of athletes play each number of sports.

Write down the frequencies

Why: 20 athletes play one sport, 22 play two and 8 play three.

\[ f = 20, \; 22, \; 8 \]

Divide each by the 50 athletes

Why: the relative frequency of each bar.

\[ \dfrac{20}{50} = 0.40, \quad \dfrac{22}{50} = 0.44, \quad \dfrac{8}{50} = 0.16 \]

Figure (svg): A frequency histogram of the number of sports played by 50 student athletes: three touching bars of width 1 centred on 1, 2 and 3 with heights 20, 22 and 8. Relative frequencies 0.40, 0.44 and 0.16 are written inside the bars.

With only three possible answers there are only three bars. The middle bar is the tallest, but only just: 22 athletes play two sports against 20 who play one.

The picture is lopsided: most athletes play one or two sports, and fewer than a fifth play three. With only three bars the histogram looks like a bar graph, but the horizontal axis is still a number line, and the bars still touch.

Check the shares

Why: the three relative frequencies must make up the whole group.

\[ 0.40 + 0.44 + 0.16 = 1.00 \]

They add to 1, and the counts add to 50. So 44 percent of these athletes play exactly two sports, the most common answer by a narrow margin over the 40 percent who play one.

OpenStax Introductory Statistics 2e, §2.2 Histograms, Frequency Polygons, and Time Series Graphs §2.2, pp. 78-79 — Try It 2.8

33. Trap: starting a discrete histogram at the smallest value

Trap

The trap

For the books data, a student starts the first bar at the smallest value, 1, with width 1:

\[ 1, \; 2, \; 3, \; 4, \; 5, \; 6, \; 7 \]

Every edge is now a whole number, and every data value sits on an edge.

\[ 3 \text{ books: bar 2 to 3, or 3 to 4?} \]

The sixteen students with three books could go in either bar.

The fix

Step half a unit below the smallest value.

\[ 1 - 0.5 = 0.5 \]

\[ 0.5, \; 1.5, \; 2.5, \dots, \; 6.5 \]

Every edge is a half-number, which no count of books can be, and each value sits in the middle of its bar.

The tick for 3 then names the bar it stands under.

Check it on the largest value too: 6 plus 0.5 is 6.5, the last edge.

34. Try It 2.9: restaurants, grouped by tens

Worked example

Try It 2.9 gives the number of employees at 22 New York City restaurants and asks for a histogram with 10 to 19 as the first interval. The data are counts, but there are too many different values, from 15 to 42, for one bar per value.

\[ 22, 35, 15, 26, 40, 28, 18, 20, 25, 34, 39, 42, 24, 22, 19, 27, 22, 34, 40, 20, 38, 28 \]

Sort the values

Why: ordering makes the counting reliable.

\[ 15, 18, 19, \; 20, 20, 22, 22, 22, 24, 25, 26, 27, 28, 28 \]

Continue the sorted list

Why: the thirties and forties.

\[ 34, 34, 35, 38, 39, \; 40, 40, 42 \]

Write the classes as half-number edges

Why: 10 to 19 contains the whole numbers 10 through 19, so its edges are 9.5 and 19.5, and each later class is 10 wider.

\[ 9.5, \; 19.5, \; 29.5, \; 39.5, \; 49.5 \]

Count each class

Why: three values in the teens, eleven in the twenties, five in the thirties and three in the forties.

\[ 3, \quad 11, \quad 5, \quad 3 \]

Figure (svg): A histogram of the number of employees at 22 New York City restaurants, with four touching bars over 9.5 to 19.5, 19.5 to 29.5, 29.5 to 39.5 and 39.5 to 49.5. Their heights are 3, 11, 5 and 3. Under the axis the 22 raw values are shown as small dots.

The classes are labelled 10 to 19, 20 to 29 and so on, but their edges are the half-numbers between them. Half of all the restaurants, 11 of the 22, employ between 20 and 29 people.

Grouping is the right move here. With one bar per value you would have 28 thin bars from 15 to 42, most of them empty or holding a single restaurant, and no shape at all. Four bars show at once that the twenties are where half of these restaurants sit.

Check the total

Why: every restaurant must be counted once.

\[ 3 + 11 + 5 + 3 = 22 \]

All 22 are there. The classes are labelled 10 to 19, 20 to 29 and so on for the reader, while the half-number edges are what guarantee that a restaurant with exactly 20 employees has only one possible bar.

OpenStax Introductory Statistics 2e, §2.2 Histograms, Frequency Polygons, and Time Series Graphs §2.2, pp. 79-80 — Try It 2.9

35. Centre each value, or group into classes?

Sorting

Discrete data come in two kinds for histogram purposes: a few possible values, which each get their own centred bar, and many possible values, which are grouped into wider classes. Continuous data are always grouped.

Sort into buckets

Sort each data set by how its histogram's bars should be built.

Width 1, each value centred
books bought by 50 students, values 1 to 6; sports played by 50 athletes, values 1 to 3; goals scored per game by a team over a season, values 0 to 5
Group into wider classes
employees at 22 restaurants, values 15 to 42; heights of 100 players, to the nearest half inch; points scored by 200 basketball players in a season, values 0 to 1,800
c1
Books, sports and goals each take only a handful of whole-number values, so a bar for each value is readable and loses nothing.
g
Restaurant staff and season points are discrete but take dozens or hundreds of values, and heights are measured, so these need classes several values wide with half-step or finer edges.

The basketball points are the clearest case: one bar per possible total would mean 1,801 bars for 200 players, nearly all of them empty. Grouping in classes of, say, 200 points gives about nine bars and a visible shape.

Whichever kind you build, the edge rule is the same in spirit: edges one step finer than the data, so no value can sit on one. For whole numbers that means edges at the halves, whether the classes are 1 wide or 10.

OpenStax Introductory Statistics 2e, §2.2 Histograms, Frequency Polygons, and Time Series Graphs §2.2, pp. 77-80 — Examples 2.8 and Try It 2.9

36. When the data refuse the extra decimal

Concept

Sometimes the starting-point rule cannot be applied. The book's Example 2.9 lists hours spent playing video games on a weekend, and the values come in every precision at once: 0, 10, 7.5, 12.75, 21.9, 9.95.

\[ 0, \; 2.25, \; 5.5, \; 7.5, \; 9.95, \; 10, \; 10, \; 11, \; 12.75, \; 15, \; 15, \dots \]

The book's solution simply uses round edges, 0, 5, 10, 15, 20 and 25, and accepts that some values will fall on them. What makes the histogram well defined is a rule, stated once and applied every time.

\[ \text{a value on an edge goes in the class to its right} \]

In the book's words: a value is counted in a class interval if it falls on the left boundary, but not if it falls on the right boundary. So each class includes its lower edge and excludes its upper one, and a value of exactly 10 belongs to the class from 10 to 15.

\[ 10 \le x < 15 \]

The book adds a sentence worth remembering: different researchers may set up histograms for the same data in different ways, and there is more than one correct way. Correct means every value in exactly one bar, equal widths, and the rule said out loud.

OpenStax Introductory Statistics 2e, §2.2 Histograms, Frequency Polygons, and Time Series Graphs §2.2, p. 79 — Example 2.9, the boundary rule

37. Example 2.9: video-game hours and the left-boundary rule

Worked example

Table 2.13 lists the hours 25 classmates spent playing video games on a weekend. The smallest value is 0 and the largest is 24, and the book's Figure 2.7 uses five classes of width 5 starting at 0.

Set the edges

Why: start at the smallest value, 0, and add 5 until the edge passes 24.

\[ 0, \; 5, \; 10, \; 15, \; 20, \; 25 \]

Count the first class

Why: 0 to 5 holds 0 and 2.25; the 0 is on the left edge, so it counts here.

\[ [0, 5): \; 0, \; 2.25 \;\to\; 2 \]

Count the second class

Why: 5.5, 7.5 and 9.95; the two 10s are on this class's right edge, so they move up.

\[ [5, 10): \; 5.5, \; 7.5, \; 9.95 \;\to\; 3 \]

Count the third class

Why: the two 10s on its left edge, plus 11 and 12.75; the two 15s move up.

\[ [10, 15): \; 10, 10, 11, 12.75 \;\to\; 4 \]

Count the fourth class

Why: the two 15s, 16.75, 17.5, 18, 18.8 and 19.5; the 20 moves up.

\[ [15, 20): \; 15, 15, 16.75, 17.5, 18, 18.8, 19.5 \;\to\; 7 \]

Count the last class

Why: 20, 20.5, 20.75, 21.9, 22.5, 22.9, 23, 23.75 and 24.

\[ [20, 25): \; 20, 20.5, 20.75, 21.9, 22.5, 22.9, 23, 23.75, 24 \;\to\; 9 \]

Figure (svg): Two stacked panels sharing an axis from 0 to 25 hours. Top: a dot plot of the 25 classmates' hours, with dashed lines at 0, 5, 10, 15, 20 and 25; the six values sitting exactly on a line, 0, 10, 10, 15, 15 and 20, are drawn in yellow. Bottom: the histogram of width 5 with bar heights 2, 3, 4, 7 and 9.

This data set has values of every precision, so no edge can dodge them all. The left-boundary rule settles each tie: 10 counts in the bar from 10 to 15, and 15 counts in the bar from 15 to 20.

In context: most of these classmates play a lot, with 16 of the 25 spending 15 hours or more on a weekend, and only 5 playing less than 10 hours. The histogram climbs steadily to the right, the opposite of the heights' hump.

Check the counts against Figure 2.7 and the total

Why: the book's bars have heights 2, 3, 4, 7 and 9, and they must hold all 25 classmates.

\[ 2 + 3 + 4 + 7 + 9 = 25 \]

The counts match the book's figure exactly, which confirms the left-boundary rule is the one Figure 2.7 used: had the 10s, 15s and 20 gone left instead, the bars would read 2, 5, 4, 6 and 8.

OpenStax Introductory Statistics 2e, §2.2 Histograms, Frequency Polygons, and Time Series Graphs §2.2, p. 79 — Example 2.9, Table 2.13 and Figure 2.7

38. Where does exactly 15 hours go?

Prediction

Two classmates in Example 2.9 reported exactly 15 hours, and 15 is an edge shared by two classes. Predict before the rule is applied.

Predict first

Using the book's boundary rule, which class holds the two values of exactly 15 hours?

  • the class from 10 to 15
  • the class from 15 to 20
  • one in each class
  • neither: values on an edge are left out

Correct: the class from 15 to 20

Why: A value on a boundary is counted in the class whose left boundary it is. Fifteen is the left boundary of the class from 15 to 20 and the right boundary of the class from 10 to 15, so both values go up into the class from 15 to 20.

\[ 10 \le x < 15 \quad \text{and} \quad 15 \le x < 20 \;\Rightarrow\; 15 \in [15, 20) \]

Splitting the pair, one each way, is the tempting compromise, and it is wrong: the rule is about the value, not about fairness between bars, so equal values always go to the same class. Leaving them out is worse, since a histogram must account for every value.

The same rule settles the other ties in this data set. The single 0 sits on the left edge of the first class and stays there; the two 10s go up into the class from 10 to 15; the 20 goes up into the class from 20 to 25. Six values in all are placed by the rule rather than by reading, and the counts on the next slide depend on every one of them.

The bracket notation is a compact way to write the rule. A square bracket means the edge is included and a round one means it is excluded, so every class here is closed on the left and open on the right.

OpenStax Introductory Statistics 2e, §2.2 Histograms, Frequency Polygons, and Time Series Graphs §2.2, p. 79 — the boundary rule

39. Two correct histograms of one data set

Picture it

The book insists there is more than one correct way to set up a histogram. Here is the video-game data drawn twice, with width 4 and with the book's width 5, both using the left-boundary rule.

Figure (svg): Two histograms of the same 25 video-game values. Left, width 4 from 0 to 28: seven bars of heights 2, 2, 4, 3, 5, 8 and 1. Right, width 5 from 0 to 25: five bars of heights 2, 3, 4, 7 and 9.

Different details, the same story: few classmates play little, and the bulk play 16 to 24 hours. A choice of width is a choice of how much detail to show, and both of these are correct.

The width-4 picture has seven bars and shows more detail: a dip from 12 to 16 hours and a final bar holding the one classmate at exactly 24. The width-5 picture smooths those wiggles into a steady climb. Both account for all 25 classmates.

\[ 2 + 2 + 4 + 3 + 5 + 8 + 1 = 25 \qquad 2 + 3 + 4 + 7 + 9 = 25 \]

So what is not a matter of taste? Equal widths within one histogram, every value in exactly one bar, and a stated rule for edges. What is a matter of judgement is the width itself: narrower bars show more detail and more noise, wider bars show the overall shape and hide the detail.

When two people's histograms of the same data look different, check their widths and starting points before deciding that the data disagree. Usually they are two honest views of one shape.

OpenStax Introductory Statistics 2e, §2.2 Histograms, Frequency Polygons, and Time Series Graphs §2.2, p. 79 — more than one correct way

40. Frequency polygons

Section

Part 4

41. A frequency polygon: one point per class

Concept

A frequency polygon carries the same information as a histogram, drawn as a line instead of bars. The book compares it to a line graph: just as line graphs make continuous data easy to interpret, so do frequency polygons.

The construction has three moves. Plot one point for each class, at the class midpoint across and the class frequency up. Add an empty class at each end. Join the points with straight segments.

Figure (svg): One class of the calculus scores, 79.5 to 89.5, drawn as a box of height 40. A dashed yellow line rises from the midpoint 84.5, five units from each edge, to a yellow dot at the top labelled (84.5, 40).

Five points up from the lower edge and five down from the upper edge meet at 84.5. A single point has to represent forty scores spread across ten points of score, and the middle is the fair place for it.

Why the midpoint? A single point has to stand for every value in the class, and you do not know where inside the class those values sit. The middle is the fair compromise, the place where the class balances, so it is never further than half a width from any value it represents.

\[ \text{point for a class} = \left(\text{midpoint}, \; \text{frequency}\right) \]

Why a line? Because a line makes the shape easy to see, and, as Part 4 shows, two lines can share one pair of axes without hiding each other, which two sets of bars cannot do.

OpenStax Introductory Statistics 2e, §2.2 Histograms, Frequency Polygons, and Time Series Graphs §2.2, p. 80 — frequency polygons

42. Decoding the class midpoint

Notation

Every point of a frequency polygon begins with this calculation, so take it apart once.

Annotate

On: \( \text{midpoint} = \dfrac{\text{lower bound} + \text{upper bound}}{2} = \dfrac{79.5 + 89.5}{2} = 84.5 \)

  • The class's left edge, 79.5 here. In the book's tables it is the first column, and it carries the extra half-step that keeps it off the data.
  • The right edge, 89.5, which is also the next class's lower bound. The difference between the bounds is the class width, 10 here.
  • Averaging the two edges finds the point equally far from both: 5 above 79.5 and 5 below 89.5.
  • The horizontal coordinate of this class's point. Its vertical coordinate is the class frequency, 40 here, so the point is (84.5, 40).

A shortcut follows from equal widths: once you have one midpoint, the next is one width further along. For the calculus scores the midpoints are 54.5, 64.5, 74.5, 84.5 and 94.5, each 10 more than the last.

\[ 54.5 + 10 = 64.5, \quad 64.5 + 10 = 74.5, \quad \dots \]

The same shortcut finds the anchor points: one width below the first midpoint and one width above the last.

OpenStax Introductory Statistics 2e, §2.2 Histograms, Frequency Polygons, and Time Series Graphs §2.2, pp. 80-81 — Example 2.10

43. Example 2.10: the polygon of calculus final test scores

Worked example

Table 2.14 gives the calculus final test scores of 100 students in five classes of width 10, from 49.5 to 99.5, with frequencies 5, 10, 30, 40 and 15.

Find the first midpoint

Why: average the first class's bounds.

\[ \dfrac{49.5 + 59.5}{2} = 54.5 \]

Step along by the width

Why: each later midpoint is 10 more.

\[ 54.5, \; 64.5, \; 74.5, \; 84.5, \; 94.5 \]

Pair each midpoint with its frequency

Why: these are the five real points.

\[ (54.5, 5), \; (64.5, 10), \; (74.5, 30), \; (84.5, 40), \; (94.5, 15) \]

Add an anchor class at each end

Why: an empty class one width below, 39.5 to 49.5, and one above, 99.5 to 109.5, each with frequency zero.

\[ (44.5, 0) \quad \text{and} \quad (104.5, 0) \]

Figure (svg): The calculus final test scores drawn as a pale histogram with edges 49.5, 59.5, 69.5, 79.5, 89.5 and 99.5 and heights 5, 10, 30, 40 and 15. A frequency polygon joins the points at the class midpoints 54.5, 64.5, 74.5, 84.5 and 94.5 at those heights, starting from zero at 44.5 and returning to zero at 104.5.

Each point sits on the middle of a bar's top. The two yellow anchors come from empty classes added at either end, so the line comes down to the axis instead of stopping in mid-air.

The book writes that the first label, 44.5, is there because the lowest test score is 54.5. That is a slip: 54.5 is the first class midpoint, and the table only tells you the lowest score lies somewhere from 49.5 to 59.5. The anchor at 44.5 is there because the class below the first is empty.

Check the points against the table

Why: the five real heights must add to the 100 students, and the anchors add nothing.

\[ 0 + 5 + 10 + 30 + 40 + 15 + 0 = 100 \]

So the polygon accounts for all 100 students. In context, the scores pile up in the 70s and 80s, with 70 of the 100 between 69.5 and 89.5, and the line rises slowly on the left and falls fast on the right, which is why the book calls the distribution skewed.

OpenStax Introductory Statistics 2e, §2.2 Histograms, Frequency Polygons, and Time Series Graphs §2.2, pp. 80-81 — Example 2.10, Table 2.14 and Figure 2.8

44. Why add the empty classes?

Socratic

The anchor classes are the part of a frequency polygon students most often leave off. Reason out what they are for before you decide they are optional.

Discussion prompt

Explain what the frequency polygon of the calculus scores would look like without the anchor points at 44.5 and 104.5, why the anchors are allowed even though no student scored in those classes, and whether they change the total frequency.

There is a further payoff you will meet in later chapters. With the anchors in place, the polygon and the axis enclose a region, and with equal widths the area of that region equals the total area of the histogram's bars. That is the first glimpse of reading areas as amounts of data.

If you said the anchors are optional decoration, notice the overlay slides coming up: two polygons that both touch down at the same anchors can be compared shape against shape.

OpenStax Introductory Statistics 2e, §2.2 Histograms, Frequency Polygons, and Time Series Graphs §2.2, p. 81 — the anchor intervals

45. The polygon and the histogram enclose the same area

Intuition

Why is a frequency polygon a fair stand-in for its histogram? Because, with equal widths and the anchors in place, the two pictures cover exactly the same amount of area. You can check it for the calculus scores.

The histogram's area is easy: five bars of width 10, whose heights add to the 100 students.

\[ 10 \times (5 + 10 + 30 + 40 + 15) = 10 \times 100 = 1000 \]

The polygon's area is a row of trapezoids, one between each pair of neighbouring points, each 10 wide. A trapezoid's area is its width times the average of its two heights.

\[ 10 \times \left(\tfrac{0 + 5}{2} + \tfrac{5 + 10}{2} + \tfrac{10 + 30}{2} + \tfrac{30 + 40}{2} + \tfrac{40 + 15}{2} + \tfrac{15 + 0}{2}\right) \]

\[ = 10 \times (2.5 + 7.5 + 20 + 35 + 27.5 + 7.5) = 10 \times 100 = 1000 \]

The same 1000. Geometrically, each segment of the polygon cuts a triangle off the corner of one bar and adds an equal triangle to its neighbour, so nothing is lost or gained. The anchors matter here: without the two end trapezoids the polygon would be missing 25 + 75 square units.

This is why polygons can be compared by their shapes. In later chapters, areas under curves will stand for shares of the data, and this is the first place that idea quietly holds.

OpenStax Introductory Statistics 2e, §2.2 Histograms, Frequency Polygons, and Time Series Graphs §2.2, pp. 80-81 — Example 2.10

46. Trap: plotting the points at the class edges

Trap

The trap

A student plots each class's frequency at its lower bound:

\[ (49.5, 5), \; (59.5, 10), \; (69.5, 30) \]

\[ (79.5, 40), \; (89.5, 15) \]

The shape looks similar, but every point has slid half a class to the left.

The peak now sits at 79.5, suggesting the most common scores were in the high 70s.

The fix

A point stands for a whole class, so it goes at the class's middle.

\[ \dfrac{79.5 + 89.5}{2} = 84.5 \]

\[ (84.5, 40) \]

The busiest class runs from 79.5 to 89.5, and its point belongs in the middle of that stretch.

Edges are for histogram bars; midpoints are for polygon points.

Plotted at midpoints, the peak sits at 84.5, in the middle of the 80s, where the forty scores actually are.

47. Try It 2.10: presidents' ages at inauguration

Worked example

Table 2.15 groups the ages at inauguration of U.S. presidents into six classes of width 5, from 41.5 to 71.5, with frequencies 4, 11, 14, 9, 4 and 2.

Find the first midpoint

Why: average the bounds of the first class.

\[ \dfrac{41.5 + 46.5}{2} = 44 \]

Step along by the width of 5

Why: the midpoints are whole numbers this time, because the bounds end in .5 and the width is odd.

\[ 44, \; 49, \; 54, \; 59, \; 64, \; 69 \]

Add the anchors

Why: one width below the first midpoint and one width above the last.

\[ 44 - 5 = 39, \qquad 69 + 5 = 74 \]

List the points

Why: midpoint across, frequency up, with zero at the anchors.

\[ (39, 0), (44, 4), (49, 11), (54, 14), (59, 9), (64, 4), (69, 2), (74, 0) \]

Figure (svg): A frequency polygon of 44 U.S. presidents' ages at inauguration. Points at the class midpoints 44, 49, 54, 59, 64 and 69 have heights 4, 11, 14, 9, 4 and 2, with anchor points at zero at 39 and 74.

The peak is in the class 51.5 to 56.5, plotted at 54. The right side runs out further than the left: a few presidents were well over 61 at inauguration, and the line takes longer to come down on that side.

In context: the most common age class is 51.5 to 56.5, with 14 presidents, and 29 of the 44 were inaugurated at 56 or younger. The right side of the polygon stretches further than the left, because a handful of presidents, 6 in all, were older than 61.5.

\[ 4 + 11 + 14 = 29, \qquad \dfrac{29}{44} \approx 0.66 \]

Check the total

Why: the table's frequencies are the number of presidents it covers; the anchors add nothing.

\[ 4 + 11 + 14 + 9 + 4 + 2 = 44 \]

So the polygon describes 44 presidents. Every point is a class midpoint, and the line begins and ends on the axis, so the picture is complete.

OpenStax Introductory Statistics 2e, §2.2 Histograms, Frequency Polygons, and Time Series Graphs §2.2, p. 81 — Try It 2.10 and Table 2.15

48. Put the frequency polygon steps in order

Ranking

The construction of a frequency polygon is a short recipe, and the order matters because each step uses the one before. Arrange the steps.

Put in order

Order the steps for drawing a frequency polygon from a frequency table.

  1. Find each class midpoint by averaging its lower and upper bounds
  2. Add an empty class one width beyond each end of the table
  3. Plot each point at its midpoint across and its frequency up, with the anchors at zero
  4. Join consecutive points with straight segments
  5. Check that the frequencies add to the total number of data values

Why: The midpoints come first, since every point needs one. The anchor classes are added next, so their midpoints are ready before plotting. Then the points are plotted, including the two at height zero, and joined in order. The total check comes last, and it is the anchors' zero frequencies that let the polygon's points add to the same total as the table.

Two orderings are worth comparing. Some people plot first and add the anchors at the end, which works on paper but often leaves the polygon with a loose end in practice. Adding the anchor classes to the table before plotting means they cannot be forgotten.

The final check is cheap and worth the habit. The heights of the polygon's points, anchors included, must add to the total frequency, exactly as the heights of the histogram's bars do.

OpenStax Introductory Statistics 2e, §2.2 Histograms, Frequency Polygons, and Time Series Graphs §2.2, p. 80 — constructing a frequency polygon

49. Overlaying polygons to compare two distributions

Concept

Frequency polygons earn their keep when you have two data sets to compare. Two histograms on one pair of axes would hide each other, bar behind bar. Two polygons are lines, and lines can cross without hiding anything.

The book's method is to overlay the polygons: draw both on the same axes, with the same classes, so that every point on one has a partner directly above or below it on the other.

\[ \text{same classes} \;\Rightarrow\; \text{same midpoints} \;\Rightarrow\; \text{points line up} \]

Two conditions keep the comparison fair. The classes must be identical, or the points will not line up. And the two groups should be the same size, or else plotted as relative frequencies, because a group of 200 would tower over a group of 100 even if their shapes were identical.

Both of the book's comparisons have exactly 100 students in each table, so raw frequencies are fair. When group sizes differ, divide each frequency by its own group's total before plotting.

\[ \dfrac{f}{n_1} \text{ for the first group}, \qquad \dfrac{f}{n_2} \text{ for the second} \]

OpenStax Introductory Statistics 2e, §2.2 Histograms, Frequency Polygons, and Time Series Graphs §2.2, p. 81 — overlaying frequency polygons

50. Example 2.11: final test against final grade

Worked example

The book overlays the calculus final test scores from Example 2.10 with the same students' final numeric grades, Table 2.17. Both tables use the classes 49.5 to 99.5 in steps of 10, and both cover 100 students.

Reuse the midpoints and anchors

Why: the classes are identical, so the horizontal coordinates are the same for both polygons.

\[ 44.5, \; 54.5, \; 64.5, \; 74.5, \; 84.5, \; 94.5, \; 104.5 \]

Read the test frequencies

Why: from Table 2.16, with zero at both anchors.

\[ \text{test: } 0, 5, 10, 30, 40, 15, 0 \]

Read the grade frequencies

Why: from Table 2.17, again with zero at both anchors.

\[ \text{grade: } 0, 10, 10, 30, 45, 5, 0 \]

Compare point by point

Why: subtract the test frequency from the grade frequency at each midpoint.

\[ +5, \quad 0, \quad 0, \quad +5, \quad -10 \]

Figure (svg): Two frequency polygons on one set of axes. Blue, the calculus final test scores: 0, 5, 10, 30, 40, 15, 0 at 44.5 to 104.5. Orange, the final course grades: 0, 10, 10, 30, 45, 5, 0 at the same points. The orange line is higher at 54.5 and 84.5 and much lower at 94.5.

Both lines pass through 30 at 74.5, so the middle class is identical. The difference is at the top: the grades pile into the 80s and almost vanish in the 90s, where the test scores still had 15 students.

Read the differences in context. The grades have 5 more students in the lowest class and 5 more in the 80s, and 10 fewer in the 90s. The final grades are less spread out at the top: fewer students reached 90 or above for the course than on the final test, and more landed in the 80s.

Check both tables

Why: each set of frequencies must add to 100, and the differences must add to zero, since the same 100 students are in both.

\[ 5 + 0 + 0 + 5 - 10 = 0 \]

Both columns total 100 and the differences cancel. So the overlay shows students moving between classes, not appearing or disappearing, which is exactly what you would expect when one group is measured twice.

OpenStax Introductory Statistics 2e, §2.2 Histograms, Frequency Polygons, and Time Series Graphs §2.2, pp. 81-82 — Example 2.11, Tables 2.16 and 2.17, Figure 2.9

51. Try It 2.11: line up the two tables

Comparison

Try It 2.11 overlays the calculus final test scores with the students' final test scores in algebra, Table 2.18. Fill in the table so that each row is one point on each polygon.

Comparison matrix

classmidpointcalculusalgebraalgebra minus calculus
39.5 to 49.544.5000
49.5 to 59.554.55105
59.5 to 69.564.5105-5
69.5 to 79.574.5304010
79.5 to 89.584.54035-5
89.5 to 99.594.51510-5
99.5 to 109.5104.5000

The midpoint column is shared, which is the whole point of using the same classes. The last column is the comparison in numbers: algebra has more students in two of the lower classes and fewer in the top two.

\[ 5 - 5 + 10 - 5 - 5 = 0 \]

The differences cancel, as they must when two tables each describe 100 students. If your column does not add to zero, one of the frequencies was misread.

OpenStax Introductory Statistics 2e, §2.2 Histograms, Frequency Polygons, and Time Series Graphs §2.2, pp. 82-83 — Try It 2.11 and Table 2.18

52. Try It 2.11: algebra against calculus

Worked example

Now draw the two polygons from the table and say what the picture shows. Both groups have 100 students, so plotting raw frequencies is fair.

Plot the calculus points

Why: the same points as Example 2.10.

\[ \text{calculus: } 0, 5, 10, 30, 40, 15, 0 \]

Plot the algebra points on the same axes

Why: the frequencies from Table 2.18 at the same midpoints.

\[ \text{algebra: } 0, 10, 5, 40, 35, 10, 0 \]

Locate each peak

Why: the midpoint with the largest frequency in each list.

\[ \text{algebra peak: } 74.5 \; (40), \qquad \text{calculus peak: } 84.5 \; (40) \]

Count the students scoring above 79.5

Why: add the top two classes for each subject.

\[ \text{algebra: } 35 + 10 = 45, \qquad \text{calculus: } 40 + 15 = 55 \]

Figure (svg): Two frequency polygons on one set of axes. Blue, the calculus final test: 0, 5, 10, 30, 40, 15, 0. Purple, the algebra final test: 0, 10, 5, 40, 35, 10, 0, at the midpoints 44.5 to 104.5. The algebra line peaks at 74.5, one class to the left of the calculus peak at 84.5.

The peaks are one class apart: algebra's most common class is 69.5 to 79.5, calculus's is 79.5 to 89.5. On one pair of axes, that shift is visible in a glance.

The overlay makes the comparison immediate: the algebra polygon is shifted one class to the left. Its most common class is 69.5 to 79.5, while calculus's is 79.5 to 89.5, and 55 calculus students scored above 79.5 against 45 in algebra.

Check the totals before trusting the comparison

Why: both tables must cover the same number of students, or the heights are not comparable.

\[ 10 + 5 + 40 + 35 + 10 = 100 = 5 + 10 + 30 + 40 + 15 \]

Both groups are 100 students, so the raw-frequency overlay is fair. On these final tests, this class's calculus scores ran about one class higher than its algebra scores.

OpenStax Introductory Statistics 2e, §2.2 Histograms, Frequency Polygons, and Time Series Graphs §2.2, pp. 82-83 — Try It 2.11

53. Which way does the calculus polygon lean?

Prediction

The book looks at the calculus polygon and says the distribution is skewed, because one side of the graph does not mirror the other. Predict which side is the long one before reading on.

Predict first

In the calculus scores polygon, on which side does the line take longer to come down to the axis?

  • the low-score side, on the left
  • the high-score side, on the right
  • neither side; it is symmetric
  • it cannot be told without the raw scores

Correct: the low-score side, on the left

Why: From the peak at 84.5 the line falls to the right in two classes, through 15 to zero. To the left it falls through 30, 10 and 5 before reaching zero, three classes of gradually thinning scores. The longer, thinner stretch is on the low-score side.

\[ \text{left of the peak: } 30, \; 10, \; 5, \; 0 \qquad \text{right of the peak: } 15, \; 0 \]

A long thin stretch like that is called a tail, and Section 2.6 names the shapes properly. For now, notice that the tail points towards the unusual values: a few students scored much lower than most, while nobody could score far above the top class.

The frequency table alone already contains this, but the polygon makes it visible in a second. That is the point of drawing data rather than only listing it.

OpenStax Introductory Statistics 2e, §2.2 Histograms, Frequency Polygons, and Time Series Graphs §2.2, p. 81 — Example 2.10, a skewed distribution

54. Time series graphs

Section

Part 5

55. Order is part of the data

Concept

Suppose you record the temperature at noon every day for a month. You could find the mean or the median, or draw a histogram of how many days fell in each range. The book points out that every one of those methods ignores part of what you collected: the date.

Figure (svg): Two panels for the ten annual CPI values. Left: a histogram with classes 180 to 190, 190 to 200, 200 to 210, 210 to 220 and 220 to 230 and heights 2, 1, 2, 3 and 2. Right: the same ten values plotted against year 1 to 10 and joined in order, rising steadily with one small dip at year 7.

The histogram keeps how many years fell in each range and throws away which year was which. The time series keeps the order, and with it the trend: prices rose every year but one.

A histogram treats the values as a pile with no order. When each value is paired with a time, the order can carry the most important information of all, whether the measurement is rising, falling or repeating.

time series graph — A graph of paired data in which the horizontal axis shows the date or time and the vertical axis shows the value of the variable measured, with the points usually joined by straight lines in time order.

Building one uses both halves of each pair. Time goes across, the measurement goes up, so each point is one date and one value, and the points are joined in the order in which they occurred.

\[ \text{point} = (\text{time}, \; \text{value}), \quad \text{joined in time order} \]

Shuffle the years in the left-hand picture and the histogram does not change at all. Shuffle them on the right and you get a different, false story. That is the test of whether order matters for your question.

OpenStax Introductory Statistics 2e, §2.2 Histograms, Frequency Polygons, and Time Series Graphs §2.2, p. 83 — constructing a time series graph

56. Example 2.12: the annual Consumer Price Index

Worked example

Tables 2.19 and 2.20 give the Consumer Price Index for every month of ten years, plus an annual figure for each year. The book asks for a time series graph of the annual values only.

Find out what the annual column is

Why: add year 1's twelve monthly values and divide by 12.

\[ \dfrac{181.7 + 183.1 + \dots + 184.3}{12} = \dfrac{2207.5}{12} \]

Finish the division

Why: and compare with the printed annual value for year 1.

\[ \dfrac{2207.5}{12} \approx 183.96 \;\to\; 184.0 \]

Choose the axes

Why: year across, from 1 to 10; the index up.

\[ (1, 184.0), \; (2, 188.9), \; (3, 195.3), \; (4, 201.6), \; (5, 207.342) \]

Plot the second half and join in order

Why: the remaining five years.

\[ (6, 215.303), \; (7, 214.537), \; (8, 218.056), \; (9, 224.939), \; (10, 229.594) \]

Figure (svg): A time series graph of the Consumer Price Index over ten years. A thin grey line traces all 120 monthly values; a thicker blue line joins the ten annual values plotted at years 1 to 10, rising from 184.0 to 229.594 with one dip, from 215.303 in year 6 to 214.537 in year 7. The monthly line peaks at 219.964 in July of year 6 and falls to 210.228 by December.

Each blue dot is the average of the twelve grey months around it. The grey line shows what the averages smooth away: a sharp autumn fall in year 6 that drags the year-7 average below year 6.

Read the graph in context. Prices rose in nine of the ten years, by between about 3.5 and 8 index points a year, and the index climbed about 25 percent overall. The one exception is year 7, the only point lower than the one before it.

\[ 229.594 - 184.0 = 45.594, \qquad \dfrac{45.594}{184.0} \approx 0.248 \]

Check the dip against the monthly data

Why: the year-7 average should be lower because year 6 ended low; the monthly values fell from 219.964 in July of year 6 to 210.228 in December.

\[ 214.537 - 215.303 = -0.766 \]

So the annual graph's single dip is real, and the monthly line shows where it came from: a fall late in year 6 that left year 7 starting below year 6's average. Every annual value in years 5 to 10 matches the mean of its twelve months to three decimal places.

OpenStax Introductory Statistics 2e, §2.2 Histograms, Frequency Polygons, and Time Series Graphs §2.2, pp. 83-84 — Example 2.12, Tables 2.19 and 2.20, Figure 2.10

57. What a time series shows: trend, change and dips

Concept

The book says time series graphs make trends easy to spot: when you record the same variable over a long period, a pattern that is invisible in a column of numbers often jumps out once the points are drawn.

Three readings are worth practising on every time series. The first is the trend, the overall direction across the whole period: for the CPI, steadily upward.

The second is the size of the changes from one time to the next. A table of year-to-year differences is the time series' own arithmetic, and it is where the graph's steepness comes from.

\[ 188.9 - 184.0 = 4.9, \quad 195.3 - 188.9 = 6.4, \quad 201.6 - 195.3 = 6.3 \]

\[ 215.303 - 207.342 = 7.961, \qquad 214.537 - 215.303 = -0.766 \]

The third is the exceptions: points that break the pattern. Year 7 is the only negative change in the CPI series, and an exception like that is usually the most interesting thing on the graph, here an actual fall in prices.

None of these three readings is possible from a histogram of the same ten numbers. That is what the book means by a portion of the data that other methods ignore.

OpenStax Introductory Statistics 2e, §2.2 Histograms, Frequency Polygons, and Time Series Graphs §2.2, p. 85 — uses of a time series graph

58. Try It 2.12: United States CO2 emissions

Worked example

Table 2.21, from the World Bank, gives CO2 emissions for three countries over seven years. The Try It asks for a time series of the United States column only.

\[ 5{,}681{,}664, \; 5{,}790{,}761, \; 5{,}826{,}394, \; 5{,}737{,}615, \; 5{,}828{,}697, \; 5{,}656{,}839, \; 5{,}299{,}563 \]

Choose a readable scale

Why: the values are in the millions, so plot them in millions of the table's units, to two decimal places.

\[ 5.68, \; 5.79, \; 5.83, \; 5.74, \; 5.83, \; 5.66, \; 5.30 \]

Compute the year-to-year changes

Why: subtract each year from the next, in the table's units.

\[ +109{,}097, \; +35{,}633, \; -88{,}779, \; +91{,}082 \]

Finish the changes

Why: the last two years.

\[ -171{,}858, \qquad -357{,}276 \]

Figure (svg): A time series graph of United States CO2 emissions over seven years, in millions of the table's units. The points are 5.68, 5.79, 5.83, 5.74, 5.83, 5.66 and 5.30, joined in order; the line wobbles near 5.8 for five years and then drops steeply in years 6 and 7.

Five years of small ups and downs, then two years of steep decline. The last drop, from year 6 to year 7, is more than twice the size of any earlier change.

The graph shows two phases. For five years emissions hover around 5.7 to 5.8 million units, going up and down. Then they fall two years running, and the last fall is the biggest change in the series by far, more than twice the year-5-to-6 drop.

\[ \dfrac{5{,}299{,}563 - 5{,}828{,}697}{5{,}828{,}697} \approx -0.091 \]

Check the overall change against the graph

Why: from year 1 to year 7 the emissions fell, so the last point must sit below the first.

\[ 5{,}299{,}563 - 5{,}681{,}664 = -382{,}101 \]

The last point is lower than the first by 382,101 units, about 6.7 percent. In context, United States emissions in this table were roughly flat for five years and then fell by about 9 percent in two years.

OpenStax Introductory Statistics 2e, §2.2 Histograms, Frequency Polygons, and Time Series Graphs §2.2, p. 85 — Try It 2.12 and Table 2.21

59. How NOT to lie: where the vertical axis starts

Picture it

The book closes the section with a warning drawn from Darrell Huff's How to Lie with Statistics: a graph can say much that is distracting, confusing and simply wrong. The most common trick is on the vertical axis.

Figure (svg): Two copies of the annual CPI time series. Left, with the vertical axis running from 0 to 240: the line is a gentle, almost flat slope near the top of the panel. Right, with the axis running from 180 to 240: the same ten points climb steeply from the bottom of the panel to the top.

Same ten numbers, same ten years. On the left the rise is about a fifth of the height of the prices; on the right it fills the whole panel. Neither is false, but only one lets you judge the size of the change against zero.

The book's own Figure 2.10 starts its vertical axis at 180, not 0. That is a reasonable choice for seeing year-to-year detail, but look at what it does to the impression: the right-hand panel shows a line climbing from floor to ceiling, as if prices had multiplied.

\[ \dfrac{229.594}{184.0} \approx 1.25 \]

The left-hand panel tells the truth about size: the last year's index is about 1.25 times the first, a rise of a quarter, not a tenfold explosion. The book's advice is blunt: if the axis does not begin at zero, the reader must be told, because otherwise the shape exaggerates every change.

Two related tricks from the same passage: changing the units of the axis to make a drop look large or small, and stretching or squeezing the time axis to make growth look sudden or gentle. Neither changes a number; both change what the picture says.

OpenStax Introductory Statistics 2e, §2.2 Histograms, Frequency Polygons, and Time Series Graphs §2.2, pp. 85-86 — How NOT to Lie with Statistics

60. How NOT to lie: unequal widths and pies across years

Concept

The book names a matching trick for histograms: vary the width of the categories. A wider bar covers more area, and since the eye reads area, the wide category looks more important than its count deserves.

Figure (svg): Two histograms of the heights from 65.95 to 75.95. Left, honest: bars of width 2 with heights 40, 17, 12, 7 and 1. Right, misleading: the last three classes merged into one bar from 69.95 to 75.95 drawn at height 20, a block three times as wide as a normal bar that looks bigger than the 40 bar.

The red block covers 120 square units of ink against the tall bar's 80, so the eye reads 20 tall players as more than 40 middle ones. Equal widths are what let area stand for a count.

Here the last three height classes, 12, 7 and 1 players, are merged into one bar from 69.95 to 75.95 and drawn at their total, 20. Compare the ink.

\[ 6 \times 20 = 120 \quad \text{versus} \quad 2 \times 40 = 80 \]

The 20 tall players now cover half again as much ink as the 40 players in the busiest class. If a wide class is ever needed, its height has to be scaled down by its width, so that area, not height, matches the count.

\[ \dfrac{20}{3} \approx 6.7 \text{ per normal width} \]

The book's other example is the pie chart. Pies cannot show change over time, yet two years of a city budget are often drawn as two pies of equal size. Because the total changes from year to year while the pie stays the same size, the wedges cannot be compared, and the picture hides the one thing a reader wanted to know.

OpenStax Introductory Statistics 2e, §2.2 Histograms, Frequency Polygons, and Time Series Graphs §2.2, pp. 85-86 — pie charts and varying widths

61. Error analysis: four graphs from one annual report

Error analysis

A company's annual report contains four graphs. Each one breaks a rule from the book's How NOT to Lie passage. Name the problem in each before reading the notes.

Annotate

On: \( \text{pies}, \quad \text{wide bar}, \quad \text{time axis}, \quad \text{price axis} \)

  • Revenue by source for two years drawn as same-sized pies. Total revenue grew, but equal pies hide that, and a wedge that shrank may stand for more money than before. Pies cannot show change over time.
  • A histogram of customer ages whose top class spans 50 to 90 while the others span 10 years. The wide bar's area inflates the older customers; with unequal widths the area no longer tracks the count.
  • A sales time series whose horizontal axis switches from years to months part way across. The book says such an axis should never change part way, because the slope in one half cannot be compared with the slope in the other.
  • Share price plotted on an axis from 990 to 1,010, so a move from 995 to 1,005 fills the panel. The rise is 10 on a base of about 1,000, about 1 percent, and starting at zero would show it.

\[ \dfrac{1005 - 995}{995} \approx 0.010 \]

What the four have in common is that every number in them may be correct. The deception, deliberate or careless, lives in the drawing: in sizes, widths and axes. The book is clear that purposeful manipulation is unethical, and that even accidental versions leave the reader confused.

The defence is a short routine for any graph you read: where does each axis start, are the bars or wedges the same width or size, and are the units the same all the way across?

OpenStax Introductory Statistics 2e, §2.2 Histograms, Frequency Polygons, and Time Series Graphs §2.2, pp. 85-86 — How NOT to Lie with Statistics

62. A time series from this week's news

Real world

Time series graphs are everywhere outside the classroom: prices, temperatures, unemployment, stock indexes, case counts, sports rankings. Practise the habits of this part on a real one.

Discussion prompt

Find a time series graph in a news article or report. Write down what is on each axis and in what units, where the vertical axis starts, whether the time steps are equal all the way across, and one trend and one exception the graph shows. Then say one question a histogram of the same values could answer and one it could not.

The last part of the question is the heart of this section. Each display throws something away: the histogram throws away the order, and the time series makes it hard to see how often each value occurred. Choosing a display means choosing what you are willing to lose.

OpenStax Introductory Statistics 2e, §2.2 Histograms, Frequency Polygons, and Time Series Graphs §2.2, p. 85 — uses of a time series graph

63. Putting it together

Section

Part 6

64. Three displays, three questions

Pattern

Every display in this section answers a different question about quantitative data. Start from the question, choose the display, and then follow its recipe.

Figure (svg): A flow chart. The top box asks what question you are asking of quantitative data. One arrow, labelled how are the values spread, leads to a histogram box listing start, width, boundaries, count. A second arrow, labelled compare two distributions, leads to overlaid frequency polygons with midpoints and anchors. A third, labelled how does it change over time, leads to a time series graph with time across and the measurement up, in order.

Three displays, three questions. Choose the display from the question you are asking, and then follow its recipe line by line.
  1. How are the values spread? Histogram: start below the smallest value, one decimal place finer than the most precise value; width from the stretch over the number of bars, rounded up; touching bars of equal width; check the counts add to n
  2. Discrete with few values? Width 1, edges at the halves, each value centred in its bar
  3. Values on edges anyway? State the rule: a value on a boundary goes in the class to its right
  4. Comparing two groups? Overlaid frequency polygons on identical classes: points at the midpoints, an empty anchor class at each end, relative frequencies if the groups differ in size
  5. How does it change over time? Time series: time across, value up, points joined in time order, and check where the vertical axis starts

The checks are the part people skip, and they are the part that catches mistakes: frequencies that add to n, relative frequencies that add to 1, a last edge above the largest value, no edge on a recorded value.

65. Check: the starting point

Check

One data set, four candidate starting points. Find the most precise value first.

Check your understanding

A data set of reaction times has smallest value 0.4 seconds, and its most precise values are recorded to two decimal places, such as 0.47. Which starting point follows the book's rule?

  • A. 0.395 (correct)
  • B. 0.35
  • C. 0.4
  • D. 0.39

Answer: A

Why: The most precise values have two decimal places, so the starting point needs three, and the convenient amount to subtract is half of 0.01, which is 0.005. Subtracting it from the smallest value 0.4 gives 0.395, below every value and finer than all of them.

Why B tempts people
This subtracts 0.05, the step for one-decimal data. It reads the precision from the smallest value, 0.4, instead of from the most precise value, and 0.35 is a time this data set could contain.
Why C tempts people
Starting exactly on the smallest value puts that value on an edge. A start must be below every value and unreachable by any of them.
Why D tempts people
This is below the smallest value but has only two decimal places, so it is a value the data could contain; a reaction time of 0.39 would sit on the edge.

The trap here is the smallest value's appearance. It is written as 0.4, one decimal place, but the data set as a whole is recorded to two, so 0.4 means 0.40. Precision belongs to the data set, not to whichever value happens to be smallest.

You can always check a proposed start by asking two quick questions. Is it below the smallest value? Could any value in this data set, recorded to its precision, be equal to it? Only a start that passes both is safe, and 0.395 is the one choice here that does.

Every correct start in this lesson ends in a 5 one place beyond the data: 59.95, 8.95, 0.5, 0.395. If a proposed start does not, ask whether some value could equal it.

66. Check: building a frequency polygon

Check

A frequency table and four proposed sets of points. Decide which follows the recipe.

Check your understanding

A table has classes 19.5 to 29.5, 29.5 to 39.5 and 39.5 to 49.5 with frequencies 6, 12 and 4. Which list gives the points of its frequency polygon, including the anchors?

  • A. (14.5, 0), (24.5, 6), (34.5, 12), (44.5, 4), (54.5, 0) (correct)
  • B. (19.5, 6), (29.5, 12), (39.5, 4)
  • C. (24.5, 6), (34.5, 12), (44.5, 4)
  • D. (19.5, 0), (24.5, 6), (34.5, 12), (44.5, 4), (49.5, 0)

Answer: A

Why: Each class is plotted at its midpoint: 24.5, 34.5 and 44.5. Then an empty class of the same width is added at each end, with midpoints one width beyond the first and last, 14.5 and 54.5, and frequency zero, so the polygon reaches the axis at both ends.

Why B tempts people
These points sit at the lower class boundaries, which shifts the whole shape half a class to the left, and there are no anchors, so the line floats.
Why C tempts people
The midpoints are right, but without the anchor classes the polygon starts and stops in mid-air instead of coming down to the axis.
Why D tempts people
The zero points are placed at the outer class edges, half a width away, instead of at the midpoints of the empty anchor classes, so the end segments are too steep.

Two of the wrong lists get the midpoints right and fail only at the ends, which shows how easy the anchors are to forget or misplace. An anchor is a whole empty class, so its point is at that class's midpoint, a full width beyond the last real midpoint.

A quick way to test any proposed polygon: the horizontal gaps between consecutive points must all equal the class width, here 10, including the two gaps out to the anchors. A gap of 5 at either end means the anchor was put on a class edge rather than at an empty class's midpoint, and the shape at the ends is distorted.

\[ 44.5 + 10 = 54.5, \qquad 24.5 - 10 = 14.5 \]

67. Check: which display?

Check

A question about data, four displays. Ask what information the question needs before choosing.

Check your understanding

A city has recorded its rainfall every month for twenty years and wants to know whether its summers have been getting drier. Which display answers that question?

  • A. a time series graph of summer rainfall against year (correct)
  • B. a histogram of all 240 monthly rainfall values
  • C. a frequency polygon of the monthly rainfall values
  • D. a pie chart of rainfall by month for the twenty years combined

Answer: A

Why: Getting drier is a question about change over time, so the display must keep the years in order. A time series of summer rainfall against year shows directly whether the line trends downward, and any unusual years appear as exceptions to that trend.

Why B tempts people
A histogram shows how often each amount of rain occurred, but it throws away the dates, so a dry decade and a wet decade could produce the same bars in any order.
Why C tempts people
A frequency polygon is a histogram drawn as a line, and it discards the order of the values just as the histogram does.
Why D tempts people
Combining twenty years into one pie erases every year-to-year difference, and pies cannot show change over time, as the book warns.

Notice that two of the wrong answers are perfectly good displays for a different question, such as how often a month gets more than 100 millimetres of rain. The question decides the display, never the data alone.

If you hesitated over the frequency polygon because it is a line, remember what its horizontal axis is: classes of rainfall, not years. A line across time and a line across values look alike and mean different things.

A good habit before choosing any display is to say the question out loud with its key word stressed. How often, how spread out: a histogram. Compared with what: overlaid polygons. Over what period, rising or falling: a time series. The stressed word names the display.

68. Explain why the edges carry an extra decimal place

Explain it to yourself

One idea underlies most of the rules in Parts 2 and 3. Putting it into your own words is what lets you apply it to data sets the book never showed you.

Discussion prompt

In three or four sentences, explain to someone who missed this lesson why a histogram's boundaries are written with one more decimal place than the data, and why the class width should be rounded up. Use one example of your own.

If your explanation used the shoe sizes, it probably also noticed the limit of the rule: the start's extra decimal protects only the start, and a width such as 0.85 can carry a later edge back onto a recorded value.

If you found it hard to explain why rounding up is safe, picture the bars as a ruler laid along the data. A ruler longer than the data is fine; a ruler shorter than the data leaves someone off the end.

69. Exit ticket

Exit ticket

One data set, three decisions from the lesson. Answer in full sentences, naming the evidence for each answer.

Twenty students' typing speeds, in words per minute, are recorded as whole numbers from 23 to 61. You want six bars.

Discussion prompt

Give the starting point and ending value, compute the class width and say how you rounded it, and list the boundaries. Then say whether any boundary can equal a typing speed, and what you would check after counting.

The most common slip is keeping the width 6.5. It covers the range exactly, but 22.5 plus 6.5 is 29, a whole number and a possible typing speed, so the second edge would sit on the data. Rounding up to 7 avoids that and still reaches past 61.

Notice that the exit ticket ran the same five moves as Example 2.7: read the data, set the start and end, divide by the number of bars, round up, and list the edges. The only thing that changed was the precision, whole numbers rather than half inches, and so the amount subtracted was 0.5 rather than 0.05. If you can say why that one number changed, you have the idea of the whole lesson.

\[ 22.5 + 6.5 = 29 \qquad 22.5 + 6 \times 7 = 64.5 > 61 \]

70. What you can do now

Recap

This section built three displays for quantitative data, and a rule for each decision inside them. Each row of the table is one of those decisions with the mistake it prevents.

the decisionwhat to dothe trap
where does the first bar start?below the smallest value, one decimal place finer than the most precise valuestarting on the data, or reading precision from the smallest value
how wide is each bar?stretch divided by the number of bars, rounded uprounding down and losing the largest value; a width whose multiples land on data
discrete data with few values?width 1, edges at the halves, each value centrededges on whole numbers
a value on an edge?it goes in the class to its rightsplitting equal values between two bars
comparing two groups?overlaid polygons: midpoints, anchors, same classespoints at class edges; no anchors; unequal group sizes
change over time?time series: time across, value up, in orderan axis that does not start at zero, unannounced

Underneath the table is one promise: every data value in exactly one bar, and every bar the same width, so that area tells the truth. The checks you ran on every slide, counts adding to n and shares adding to 1, are how you know the promise was kept.

Section 2.3 uses the ordered data you have been counting here to find quartiles and percentiles, measures of location you can mark on a histogram's axis.

OpenStax Introductory Business Statistics 2e, §2.1 Display Data §2.1 — histograms, frequency polygons and time series graphs, in business settings

Sources

  1. OpenStax Introductory Statistics 2e, §2.2 Histograms, Frequency Polygons, and Time Series Graphs — Illowsky & Dean, OpenStax / Rice University, CC BY 4.0, pp. 74-86
  2. OpenStax Introductory Business Statistics 2e, §2.1 Display Data — Illowsky & Dean, OpenStax / Rice University, CC BY 4.0

Want this taught 1-on-1? Alexander tutors Statistics — $55/session, free consultation.

Book on Wyzant · Text (657) 465-8108