1.3 Frequency, Frequency Tables, and Levels of Measurement

The four levels of measurement as a ladder — nominal names with no order, ordinal adding order but no measurable differences, interval adding meaningful differences but no true zero, and ratio adding a true zero so that ratios become meaningful — and which operations each level permits. Then the frequency table and its two derived columns: relative frequency as each count divided by the sample size, and cumulative relative frequency as the running total that must finish at one. Closes with grouped frequency tables for continuous data, and the book's rounding convention of carrying one more decimal place than the original data.

Subject: Statistics · 66 slides · symbolic lesson

Open the interactive version of this deck

What this lesson covers

The lesson, slide by slide

1. Frequency Tables and Levels of Measurement

Title

Statistics · Chapter 1 — Sampling and Data

What you are allowed to do with data, and how to count it so the counts answer questions

2. What this lesson gives you

Objectives

Section 1.2 was about collecting data. This section is about the first two things you do once the data are in your hands: decide what kind of numbers you are holding, and organise them so that they can answer questions without anyone rereading the raw list.

The arithmetic in this section is addition and division, nothing harder. What loses marks is the reasoning around it: averaging numbers that are really labels, reading a cumulative column as if it described one value, or rounding too early and wondering why a column adds to 1.01. Every slide is aimed at one of those.

You will work with the book's own data throughout: twenty students' work hours, a hundred soccer players' heights, fifty towns' rainfall, nineteen commutes, seventy federal employees and eighteen years of traffic crashes. Every number was recomputed from those data.

OpenStax Introductory Statistics 2e, §1.3 Frequency, Frequency Tables, and Levels of Measurement §1.3, pp. 26-34 — the section these objectives come from

3. Two sorting questions you already know how to answer

Warm-up

Section 1.2 split data two ways, and both splits come straight back in this lesson. Answer this on paper before you read the reveal, because the four variables here return later as examples of the four levels.

Discussion prompt

For each variable say whether it is qualitative or quantitative, and if quantitative, whether it is discrete or continuous: a person's favourite food, a cruise passenger's rating of excellent, good, satisfactory or unsatisfactory, the number of hours a student works per day, and a soccer player's height.

Hold on to the continuous one too. Heights can take any value in a range, so when you count them, almost every player ends up with a height of their own. Part 5 of this lesson deals with that by grouping heights into intervals.

If you hesitated over the work hours, that is fair: time is continuous in principle. The book recorded whole hours, and the recorded data are what a frequency table counts, so the table treats them as discrete.

4. The four levels of measurement

Section

Part 1

5. The level of measurement decides which arithmetic is legal

Concept

The way a set of data is measured is called its level of measurement. It matters because not every statistical operation can be used with every set of data. You can count favourite foods, but you cannot average them, and working out which operations make sense is the job of this part.

level of measurement — The way a set of data is measured. From lowest to highest there are four levels: nominal, ordinal, interval and ratio.

\[ \text{nominal} \;<\; \text{ordinal} \;<\; \text{interval} \;<\; \text{ratio} \]

Read the line as a ranking of how much information the numbers carry. Each level has everything the level below it has, plus one more property. Nominal and ordinal data are the qualitative ones from Section 1.2; interval and ratio data are quantitative.

The practical rule that falls out is simple. A method that works at one level works at every level above it, so the first question to ask about any variable is how high up it sits. Ask it before you calculate, because a calculator will happily average zip codes and give you a number that means nothing.

OpenStax Introductory Statistics 2e, §1.3 Frequency, Frequency Tables, and Levels of Measurement §1.3, p. 26 — levels of measurement, the four levels in order

6. A ladder, not a list

Picture it

The four levels are easiest to hold as a staircase. You climb one step each time the data gain a property that makes a new operation meaningful.

Figure (svg): Four steps rising from left to right like a staircase, labelled nominal, ordinal, interval and ratio. Each step says what it adds: names with no order, then an order, then meaningful gaps, then a true zero. Under each step is an example: smartphone brand, cruise rating, temperature in Fahrenheit, exam score out of 100.

Read the staircase from the left. Each level can do everything the level below it can do, plus exactly one new thing.

Look at what each step adds. Ordinal data add an order to the names. Interval data add gaps you can measure, so subtraction means something. Ratio data add a true zero, a point that means none of the quantity at all, and that is what makes division meaningful.

The staircase shape is the content, not decoration. An exam score is ratio data, and it can also be put in order and subtracted, so every ordinal or interval method still applies to it. A cruise rating is ordinal, and it can be ordered but nothing more, so an interval method applied to it is a mistake however tidy the output looks.

When you classify a variable, climb from the bottom and stop at the first step it cannot reach. That is the three-question test on the next few slides.

7. Nominal data are names with no order

Concept

Data measured on a nominal scale are qualitative. Categories, colours, names, labels, favourite foods and yes or no answers are all nominal. The defining feature is what is missing: there is no order to them.

Figure (svg): Three rows, each showing the same four favourite foods, pizza, sushi, tacos and curry, in a different left-to-right order, with a note underneath that no arrangement is more correct than another.

Shuffle nominal categories however you like and nothing is lost. That is exactly what it means for data to have no order.

The book's example is favourite foods. Putting pizza first and sushi second is not meaningful, and the picture shows why: every arrangement of the categories is as good as every other, so no arrangement carries information.

Smartphone brands are another. People have preferences, strong ones, but there is no agreed order among the companies, and a personal preference is not a property of the data.

So nominal data cannot be used in calculations. What you can do is count them: how many people chose pizza, what share said yes. Counting is exactly what the frequency tables later in this lesson do, which is why a frequency table is the natural summary of nominal data and very nearly the only one.

OpenStax Introductory Statistics 2e, §1.3 Frequency, Frequency Tables, and Levels of Measurement §1.3, p. 26 — nominal scale level, favourite foods and smartphone companies

8. Ordinal data can be ordered, but gaps cannot be measured

Concept

Ordinal data are still categories, but now the categories have a definite order. The book's examples are a list of the top five national parks in the United States, and a cruise survey whose answers are excellent, good, satisfactory and unsatisfactory.

\[ \text{unsatisfactory} < \text{satisfactory} < \text{good} < \text{excellent} \]

Figure (svg): Two horizontal lines. On the top one the four cruise ratings, unsatisfactory, satisfactory, good and excellent, sit evenly spaced above the codes 1 to 4. On the bottom one the same four ratings sit at uneven positions with question marks in each gap.

The order is real on both lines. The spacing on the top line is an invention of the coding, and ordinal data never tell you which spacing is true.

What you cannot do is measure the difference between two answers. Is the step from satisfactory to good the same size as the step from good to excellent? The survey never asked, so the data cannot say. Writing the answers as 1, 2, 3 and 4 makes the steps look equal, which is the top line of the picture, but the equal spacing was supplied by whoever chose the codes.

That is why the book says that, like nominal data, ordinal data cannot be used in calculations. The same goes for the national parks: park number one is better than park number two by some amount nobody has measured, and it may be a whisker or a mile.

OpenStax Introductory Statistics 2e, §1.3 Frequency, Frequency Tables, and Levels of Measurement §1.3, p. 26 — ordinal scale level, national parks and the cruise survey

9. Interval data have meaningful differences but no true zero

Concept

Interval data are numerical. Like ordinal data they have an order, and now the gaps mean something: you can calculate differences. What they lack is a zero that means none of the quantity.

\[ 100^\circ - 60^\circ = 40^\circ \quad \text{in either scale} \]

Figure (svg): Two vertical temperature scales side by side, Celsius on the left and Fahrenheit on the right, with dashed lines joining 0 Celsius to 32 Fahrenheit, 20 Celsius to 68 Fahrenheit and 40 Celsius to 104 Fahrenheit. The zero of each scale is highlighted at a different height, and minus 15 Celsius and minus 10 Fahrenheit sit below them.

Each scale puts its zero somewhere convenient, and the two zeros are at different temperatures. Neither one means no temperature at all, which is why neither scale can support a ratio.

Temperature in Celsius or Fahrenheit is the book's example. A difference of 40 degrees is the same amount of warming wherever it happens on the scale, so subtraction is meaningful. But 0 degrees is not a minimum: minus 10 degrees Fahrenheit and minus 15 degrees Celsius exist and are colder.

The picture shows the real problem with those zeros. Each scale put its zero at a convenient place, the freezing point of water for Celsius and a cold winter day for Fahrenheit, and the two places are different. A zero that depends on which scale you chose is a label on the scale, not an absence of heat.

Calendar years are interval data for the same reason. The gap between 1990 and 2020 is thirty years on any calendar, but the year numbered zero is a convention rather than the start of time, so no year is twice another.

OpenStax Introductory Statistics 2e, §1.3 Frequency, Frequency Tables, and Levels of Measurement §1.3, p. 26 — interval scale level, Celsius and Fahrenheit

10. Is 40 degrees Celsius twice as hot as 20 degrees?

Worked example

The ratio looks obvious: 40 is twice 20. Test whether that statement is about the temperatures or only about the numbers you wrote them in. If it is about the temperatures, it has to stay true when you change the scale.

\[ \dfrac{40^\circ\text{C}}{20^\circ\text{C}} = 2 \quad ? \]

Convert both temperatures to Fahrenheit

Why: multiply by nine fifths, then add 32.

\[ 40 \cdot \tfrac95 + 32 = 104^\circ\text{F}, \qquad 20 \cdot \tfrac95 + 32 = 68^\circ\text{F} \]

Take the same ratio in the new scale

Why: the physical temperatures have not changed, only the numbers.

\[ \dfrac{104}{68} \approx 1.53 \]

Same two days, a different ratio. A statement whose truth depends on the scale you happen to use is not a fact about the temperatures, so twice as hot is not a meaningful statement here.

Now test the difference instead

Why: subtract in both scales and compare.

\[ 40 - 20 = 20 \text{ Celsius degrees}, \qquad 104 - 68 = 36 \text{ Fahrenheit degrees} \]

Subtraction asks how far apart two temperatures are, and that question has a scale-free answer: the same physical gap, just counted in smaller Fahrenheit degrees. Nine Fahrenheit degrees fit into every five Celsius degrees, so the two differences differ by a constant factor and by nothing else.

Check that the difference survived the conversion

Why: a gap of 20 Celsius degrees converts to 20 times nine fifths Fahrenheit degrees, and that is exactly the 36 you got.

\[ 20 \cdot \tfrac95 = 36 \;\checkmark \]

The difference translated faithfully and the ratio did not. That pair of results is the signature of interval data: subtraction is meaningful, division is not.

OpenStax Introductory Statistics 2e, §1.3 Frequency, Frequency Tables, and Levels of Measurement §1.3, p. 26 — interval scale: differences make sense, 0 is not a minimum

11. Move the zero and watch twice as hot change

Tweak it

The previous slide tried one other scale. This one lets you try every scale at once. Imagine a temperature scale with the same size of degree as Celsius but with its zero placed some number of degrees below the freezing point of water.

On that scale, 40 degrees Celsius reads 40 plus the offset, and 20 degrees reads 20 plus the offset. The readout shows the ratio of those two readings; the curve shows the ratio of any temperature to 20 degrees Celsius.

Parameter explorer

Drag the zero lower and read how many times hotter 40 degrees Celsius looks than 20 degrees. What does the ratio approach when the zero reaches absolute zero, about 273 degrees below freezing?

\[ \dfrac{40 + {z}}{20 + {z}} \]

  • z — from 0 to 300: degrees the scale's zero sits below freezing

With the offset at zero you are on the Celsius scale and the ratio is 2. Slide it and the ratio falls, continuously, towards 1. Any answer between 1 and 2 is available just by choosing where to put a zero, which is the clearest possible proof that the ratio was never a property of the temperatures.

At an offset of about 273 you reach the Kelvin scale, whose zero is absolute zero, the real absence of heat. There the ratio is about 1.07, and that one is meaningful, because Kelvin has a true zero. Temperature in kelvins is ratio data, while the same temperatures in Celsius are interval data.

12. Ratio data have a true zero, so ratios mean something

Concept

Ratio data give the most information. They are like interval data, with a definite order and meaningful differences, and they also have a true minimum: a zero that means none of the quantity. With that zero in place, ratios can be calculated and mean what they say.

Figure (svg): A number line from 0 to 100 with the four exam scores 20, 68, 80 and 92 marked as dots. Below it a bar from 0 to 80 is split into four equal pieces of 20, and a second bar runs from 0 to 20.

Because both bars are measured from a zero that means nothing at all, the longer one really is four copies of the shorter one.

The book's example is four machine-graded statistics exam scores out of 100: 80, 68, 20 and 92. The minimum possible score is 0 points, which means no marks earned, so a score of 80 is genuinely four times a score of 20.

Look at the bars in the picture. Both start at the same real zero, so laying four copies of the 20 bar end to end gives exactly the 80 bar. With temperature in Celsius the bars would have started at freezing, an arbitrary point, and the comparison of lengths would have been meaningless.

Heights, weights, incomes, counts of anything, times taken and distances are ratio data. A useful quick test: could the value be zero, and would zero mean none of it? A height of zero means no height. A Celsius reading of zero does not mean no temperature.

OpenStax Introductory Statistics 2e, §1.3 Frequency, Frequency Tables, and Levels of Measurement §1.3, pp. 26-27 — ratio scale level and the four exam scores

13. The four exam scores, taken up the ladder

Worked example

The book's exam scores support every operation on the ladder. Work up it one rung at a time, doing the operation each level adds, so that you see all four permissions on one data set.

\[ 80, \; 68, \; 20, \; 92 \quad \text{(out of 100)} \]

Nominal permission: count them

Why: four scores, so four observations.

\[ n = 4 \]

Ordinal permission: put them in order

Why: lowest to highest.

\[ 20 < 68 < 80 < 92 \]

Nothing on the first two rungs used the sizes of the scores. Counting and ordering would work just as well if the scores were replaced by letter grades, which is exactly why those two operations are also available to nominal and ordinal data.

Interval permission: subtract

Why: the book's comparison: the score 92 is more than the score 68 by a real number of points.

\[ 92 - 68 = 24 \text{ points} \]

Ratio permission: divide

Why: the zero is 0 points, a true absence of marks, so the ratio is meaningful.

\[ \dfrac{80}{20} = 4 \]

Check the ratio by the bar picture

Why: four lengths of 20 laid from zero reach 80, which is what the division claimed.

\[ 20 + 20 + 20 + 20 = 80 \;\checkmark \]

The book words this as a score of 80 being four times better than a score of 20. Four times the points is exactly right; whether it is four times the understanding is a question about what the exam measures, not about the data.

OpenStax Introductory Statistics 2e, §1.3 Frequency, Frequency Tables, and Levels of Measurement §1.3, p. 27 — the exam scores 80, 68, 20 and 92

14. The three questions that place any variable

Notation

Classifying a variable is a short interrogation. Ask these three questions in order, and the first answer of no tells you the level.

Annotate

On: \( \text{ordered?} \to \text{gaps measurable?} \to \text{true zero?} \)

  • If the categories have no agreed order, stop: the data are nominal. Favourite food, blood type, smartphone brand.
  • If they can be ordered but the difference between two values is not a fixed amount, stop: ordinal. Cruise ratings, park rankings, finishing places in a race.
  • If differences are meaningful but zero is just a point on the scale, stop: interval. Temperature in Celsius or Fahrenheit, calendar years.
  • Ordered, measurable gaps and a zero meaning none of the quantity: ratio. Exam scores, heights, hours worked, number of siblings.

The order of the questions matters. Each question only makes sense once the one before it has been answered yes: there is no point asking whether gaps are measurable in data that cannot even be ordered.

The third question is the one that needs care, because a scale can print a zero without having a true one. The test is whether zero means an absence. A year numbered zero is not the beginning of time, and zero degrees Fahrenheit is not the absence of heat, so both are interval.

OpenStax Introductory Statistics 2e, §1.3 Frequency, Frequency Tables, and Levels of Measurement §1.3, pp. 26-27 — the four levels

15. Place each variable on the ladder

Sorting

Ten variables. Run the three questions on each: ordered, measurable gaps, true zero. Commit to a bucket for every one before you check, and pay attention to the ones that look numerical.

Sort into buckets

Sort each variable into its level of measurement.

Nominal
blood type; zip code
Ordinal
shirt size: small, medium, large; finishing place in a race; cruise rating
Interval
temperature in °F; year of birth
Ratio
weight in kilograms; number of siblings; hours worked per day
nom
Blood type has no order at all, and a zip code is a label written in digits: a larger zip code is not more of anything, and subtracting two of them gives nothing meaningful.
ord
Sizes, race places and cruise ratings all come in a definite order, but the gap between neighbours is not a fixed amount. First and second place may be a hundredth of a second apart while second and third are a minute apart.
int
Differences in temperature and in years are real amounts, but zero degrees Fahrenheit is not the absence of heat and year zero is not the beginning of time, so ratios mean nothing.
rat
Each has a zero that means none of the quantity: no weight, no siblings, no hours. So someone with four siblings has twice as many as someone with two, and the statement means what it says.

Zip codes and finishing places are the two to learn from, because both are written in numbers. The digits in a zip code are just a label, and the numbers 1, 2, 3 for race places record an order and nothing more. Being written in digits never settles the level.

Year of birth catches people the other way. Years are numbers and can be subtracted perfectly well, so it feels like ratio data, but someone born in the year 2000 is not twice as anything as someone born in 1000.

OpenStax Introductory Business Statistics 2e, §1.3 Levels of Measurement §1.3 — Levels of Measurement: the same four-level classification in a business setting

16. Trap: averaging an ordinal scale

Trap

The trap

Four cruise passengers answer the survey. Two say excellent and two say unsatisfactory. Someone codes the answers 4, 3, 2 and 1 and averages them.

\[ \dfrac{4 + 4 + 1 + 1}{4} = 2.5 \]

The report says the average passenger found the cruise halfway between satisfactory and good.

No passenger said anything of the kind. Half loved it and half hated it, and the average has invented a lukewarm middle that nobody chose.

The fix

Report what ordinal data can support: the counts in each category, in order.

ratingfrequency
excellent2
good0
satisfactory0
unsatisfactory2

The average needed equal gaps between the codes, and the survey never measured the gaps. Change the codes to 10, 3, 2, 1, which respect the same order, and the average becomes 5.5, a different verdict from the same answers.

A frequency table cannot be fooled that way. It shows the split directly, which is the real finding.

17. The jersey number argument

Counterexample

A classmate argues that because a soccer player's jersey number is a number, jersey numbers must be ratio data: zero is a possible jersey number, and 20 is twice 10.

Discussion prompt

Find the flaw. Which of the three questions does a jersey number fail first, and what is its level of measurement? Give a specific statement about jersey numbers that sounds arithmetical but is meaningless.

The general lesson is that a zero on the list of possible values is not the same as a true zero. A true zero means none of the quantity, and a jersey number measures no quantity at all.

Keep this example to hand for any variable coded in digits: phone numbers, student ID numbers, zip codes, room numbers. All of them are nominal, and a spreadsheet will average all of them without complaint.

18. Each level and the new operation it permits

Matching

Each level adds one operation to the ones below it. Match each level to the operation it is the lowest level to allow. The matching is the whole of Part 1 in four lines.

Match the pairs

  • n. nominal
  • o. ordinal
  • i. interval
  • r. ratio
  • cnt. count how many fall in each category
  • srt. put the values in order, from lowest to highest
  • sub. subtract to find how far apart two values are
  • div. divide to say one value is so many times another

Why: Counting works at every level, so it is the one thing nominal data allow. Ordering arrives with ordinal data, subtraction with interval data because the gaps become fixed amounts, and division with ratio data because only a true zero makes one value a multiple of another.

Read the right-hand column from top to bottom and it is the ladder again: count, order, subtract, divide. Every level keeps the operations above it in the list and gains the next one.

Notice that counting sits at the bottom and is available to every level. That is why the tool for the rest of this lesson, the frequency table, can be built for any variable at all, from favourite foods to heights.

OpenStax Introductory Statistics 2e, §1.3 Frequency, Frequency Tables, and Levels of Measurement §1.3, pp. 26-27 — which operations each level supports

19. Counting with a frequency table

Section

Part 2

20. A frequency is how often a value occurs

Concept

Twenty students were asked how many hours they worked per day. Here are their answers exactly as they were collected, and in this form they are hard to read: it is not obvious which answer is the most common.

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

frequency — The number of times a value of the data occurs.

Figure (svg): A dot plot of the twenty work-hours responses: three dots above 2 hours, five above 3, three above 4, six above 5, two above 6 and one above 7, with each column's count written above it.

Each dot is one student. The height of a column is the frequency of that value, so the table you are about to build is just this picture written down as numbers.

Stack each answer as a dot above its value and the frequencies appear as heights. Three students work two hours, five work three hours, and so on up to the single student who works seven.

The frequencies must account for every student exactly once, so they add up to the sample size. That total is not a formality: it is the first check on everything that follows.

\[ \sum f = n \]

The capital Greek sigma means add up, and f stands for the frequency in each row, so the line reads: add up the frequency column and you get the number of observations. Sigma returns in Chapter 2, doing the same job for means and standard deviations.

OpenStax Introductory Statistics 2e, §1.3 Frequency, Frequency Tables, and Levels of Measurement §1.3, p. 27 — frequency, and the twenty work-hours responses

21. Building Table 1.10 from the raw responses

Worked example

Turn the twenty raw answers into the book's frequency table. Work systematically: list the possible values, then count each, then check the total.

List the distinct values in ascending order

Why: the smallest answer is 2 hours and the largest is 7, and every value between them occurs.

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

Listing every value between the smallest and the largest, rather than only the values you remember seeing, protects you from forgetting a row. If a value in the middle never occurred it gets a frequency of zero, as 27 years of service will in Example 1.18.

Tally each value through the raw list

Why: one stroke per answer, crossing each answer off as you go so none is counted twice.

\[ f: \; 3, \; 5, \; 3, \; 6, \; 2, \; 1 \]

Figure (svg): A tally chart with one row for each value from 2 to 7 hours. Row 2 has three tally strokes, row 3 has a group of five with a diagonal slash, row 4 has three, row 5 has a group of five and one more, row 6 has two and row 7 has one. The frequency column reads 3, 5, 3, 6, 2, 1 with a total of 20.

One stroke per raw value, crossed off the list as you go. The total at the foot is the check that no value was skipped or counted twice.

Crossing off the raw list is the habit that makes tallying safe. If you instead count the twos, then go back and count the threes, you read the list six times and have six chances to lose your place.

Write the table

Why: values in the first column, frequencies in the second.

data valuefrequency
23
35
43
56
62
71

Check the frequency column against the sample size

Why: twenty students were asked, so the column must add to 20.

\[ 3 + 5 + 3 + 6 + 2 + 1 = 20 = n \;\checkmark \]

The total matches, so no answer was skipped or counted twice. If it had come to 19 or 21, the table would be wrong somewhere, and you would recount before doing anything else with it.

OpenStax Introductory Statistics 2e, §1.3 Frequency, Frequency Tables, and Levels of Measurement §1.3, p. 27 — Table 1.10, Frequency Table of Student Work Hours

22. Trap: reading the frequency as the value

Trap

The trap

Asked for the most common number of hours worked, a student scans Table 1.10, finds the largest number in the frequency column, and answers 6 hours.

data valuefrequency
56
62

The 6 they found is a count of students, not a number of hours. Only two students work six hours.

The fix

The largest frequency belongs to a row. Read across that row to the value it counts.

\[ f = 6 \;\text{ belongs to the value } 5 \text{ hours} \]

So the most common answer is 5 hours, given by 6 students. Every frequency table has two columns of numbers, and answering the question means naming which column the question is about.

A quick guard: say the answer as a sentence with units. Six students worked five hours reads correctly; five students worked six hours is the swap.

23. How many rows would an ungrouped heights table need?

Estimation

The work-hours table needed six rows because answers were whole hours between 2 and 7. Now imagine the heights of 100 soccer players, measured to the nearest tenth of an inch and running from about 60 inches to about 75.9.

Predict first

If you gave every distinct recorded height its own row, roughly how many rows would the table need?

  • About 6, like the work-hours table
  • About 16, one for each whole inch
  • Several dozen, most with a frequency of 1 or 2
  • Exactly 100, one per player

Correct: Several dozen, most with a frequency of 1 or 2

Why: Between 60.0 and 75.9 there are 160 possible readings to the nearest tenth, and 100 players spread over them. Many readings are never used and most of the rest are used once or twice, so the table would run to several dozen rows of tiny counts. It would not be exactly 100, because some players do share a reading.

A table with dozens of rows of ones and twos is barely better than the raw list. It hides the shape of the data behind detail, which is the opposite of what a frequency table is for.

That is the problem continuous data pose, and Part 5 solves it by grouping the readings into eight intervals two inches wide. For now, notice where the problem comes from: it is not the sample size, it is the fact that heights can take so many values.

OpenStax Introductory Statistics 2e, §1.3 Frequency, Frequency Tables, and Levels of Measurement §1.3, pp. 28-29 — Table 1.13, the heights of 100 players

24. Two runs of twenty die rolls: will their tables agree?

Prediction

Section 1.2 closed with an experiment you can run in a minute: roll a fair six-sided die twenty times, record the frequency of each face, then do the whole thing again.

Predict first

How close will the two frequency tables be to each other?

  • Identical, because the die is fair
  • Each face about three or four times in both, with small differences
  • Noticeably different, possibly with a face missing entirely
  • Impossible to say without knowing whether the die is fair

Correct: Noticeably different, possibly with a face missing entirely

Why: Twenty rolls is a small sample, and small samples vary a lot. A fair die makes each face equally likely on every roll, which is a statement about the long run, not a promise about any particular twenty rolls. Both runs are correct records of what happened.

Figure (svg): Two bar charts of twenty simulated rolls of a fair die. In the first run the faces 1 to 6 came up 0, 5, 1, 3, 7 and 4 times; in the second run 2, 4, 7, 1, 3 and 3 times.

Same fair die, same number of rolls, two different tables. The first run never rolled a one at all.

These two runs were simulated in a computer with a fair die. In the first, face one never appeared and face five came up seven times; in the second, face three came up seven times and face four only once. Nothing was wrong with the die.

The book's question is which experiment had the correct results, and its answer is both. A frequency table records a sample faithfully; it does not tell you the truth about the population on its own, and another sample will give another table.

OpenStax Introductory Statistics 2e, §1.3 Frequency, Frequency Tables, and Levels of Measurement §1.2, pp. 25-26 — the collaborative die-rolling exercise, Tables 1.8 and 1.9

25. Relative frequency: each count as a share

Section

Part 3

26. Relative frequency divides each count by the sample size

Concept

A frequency of 6 means a lot in a sample of 20 and very little in a sample of 2,000. To compare, you need each count as a share of the whole, and that share is the relative frequency.

relative frequency — The ratio (fraction or proportion) of the number of times a value occurs to the total number of outcomes: each frequency divided by the sample size.

\[ \text{relative frequency} = \dfrac{f}{n} \]

For the work hours, n is 20, so every frequency is divided by 20. Three students work two hours, and three out of twenty is 0.15, or 15 percent of the sample.

Relative frequencies can be written as fractions, decimals or percents, and all three say the same thing. The fraction keeps the counts visible; the decimal is easiest to add up; the percent is easiest to say out loud. The book writes them as a fraction or a decimal, and so will these slides.

OpenStax Introductory Statistics 2e, §1.3 Frequency, Frequency Tables, and Levels of Measurement §1.3, p. 27 — the definition of relative frequency

27. Reading the relative frequency formula

Notation

The formula is one division, but each of its three parts is a specific number from the table. Name them before using it, so you never divide by the wrong total.

Annotate

On: \( \dfrac{f}{n} = \dfrac{3}{20} = 0.15 \)

  • The frequency of one value: here the three students who work two hours. It comes from a single row of the table.
  • The sample size: all twenty students. It is the same for every row, and it is the total of the frequency column, not the number of rows.
  • The share of the sample with that value. It is always between 0 and 1, because a part can never be more than the whole.
  • There is none. Students divided by students leaves a pure proportion, which is what lets you compare samples of different sizes.

The commonest slip is dividing by the number of rows. The work-hours table has six rows, and dividing by six would give shares that add to far more than one. The denominator is always the number of observations.

A useful check on any single relative frequency: multiply it back by n and you should get a whole number, the count you started from. 0.15 times 20 is 3 students. If the product is not a whole number, the division went wrong.

The second commonest is dividing by a different total in each row, usually after a cumulative column has been started. Every row of a relative frequency column shares one denominator.

OpenStax Introductory Statistics 2e, §1.3 Frequency, Frequency Tables, and Levels of Measurement §1.3, p. 27 — relative frequency as frequency over total

28. Adding the relative frequency column: Table 1.11

Worked example

Extend Table 1.10 with a third column. Each row divides its frequency by the same n of 20.

Divide the first two frequencies by 20

Why: two hours, then three hours.

\[ \dfrac{3}{20} = 0.15, \qquad \dfrac{5}{20} = 0.25 \]

Divide the next two

Why: four hours, then five hours.

\[ \dfrac{3}{20} = 0.15, \qquad \dfrac{6}{20} = 0.30 \]

Divide the last two

Why: six hours, then seven hours.

\[ \dfrac{2}{20} = 0.10, \qquad \dfrac{1}{20} = 0.05 \]

Every one of these divisions came out exactly, because 20 divides evenly into hundredths. With a sample of 19 or 70 they will not, and the rounding rule later in this part tells you how many places to keep.

Write the table

Why: one new column beside the frequencies.

data valuefrequencyrelative frequency
233/20 = 0.15
355/20 = 0.25
433/20 = 0.15
566/20 = 0.30
622/20 = 0.10
711/20 = 0.05

Read one row as a sentence to make sure it means something: 30 percent of the sampled students work five hours per day. That is a statement you could put in a report, which the raw count of 6 was not.

Check that the column adds to one

Why: the shares of the whole must make up the whole.

\[ 0.15 + 0.25 + 0.15 + 0.30 + 0.10 + 0.05 = 1.00 \;\checkmark \]

It must, because the frequencies add to 20 and each was divided by 20, so the relative frequencies add to 20 over 20. The book writes the same total as twenty twentieths.

OpenStax Introductory Statistics 2e, §1.3 Frequency, Frequency Tables, and Levels of Measurement §1.3, p. 27 — Table 1.11, work hours with relative frequencies

29. Relative frequency keeps the shape and changes the scale

Picture it

Draw the frequencies and the relative frequencies side by side and something important is visible at once.

Figure (svg): Two bar charts side by side for the work-hours data. The left one has frequency on the vertical axis from 0 to 7; the right one has relative frequency from 0 to 0.35. The bars have exactly the same shapes, tallest at 5 hours.

Dividing every bar by the same 20 changes the numbers on the axis and nothing else. Relative frequency keeps the shape and makes it comparable with any other sample size.

The two charts are the same shape. Every bar was divided by the same 20, so every bar shrank by the same factor, and only the numbers on the vertical axis changed. The peak is at five hours in both, and the seven-hour bar is one sixth of the peak in both.

That is exactly why relative frequency is useful. Suppose a second school surveyed 200 students. Their frequency bars would be ten times taller and impossible to compare with these by eye; their relative frequency bars would sit on the same scale from 0 to 1, and any difference in shape would be a real difference between the schools.

It also tells you what relative frequency cannot do. It cannot tell you how big the sample was. A relative frequency of 0.30 might be 6 of 20 or 600 of 2,000, and the second is much more trustworthy, so always report n beside a relative frequency.

30. Double the survey: what happens to the third column?

Prediction

Suppose a second survey asks 40 students, and it happens that every hour count comes out exactly twice as often as in the first survey: 6 students at two hours, 10 at three, and so on.

Predict first

Compared with Table 1.11, what does the relative frequency column of the 40-student table look like?

  • Every relative frequency doubles
  • Every relative frequency halves
  • The relative frequencies are exactly the same
  • The relative frequencies now add to 2

Correct: The relative frequencies are exactly the same

Why: Each frequency doubled and n doubled as well, so each fraction has its top and bottom multiplied by 2: six fortieths is the same as three twentieths. Relative frequency depends only on the shares, not on the size of the sample, and the column still adds to one.

\[ \dfrac{6}{40} = \dfrac{3}{20} = 0.15 \]

This is the property that makes relative frequencies comparable across samples. It is also a warning: the two surveys look identical in the third column even though the second rests on twice as much evidence. The frequency column and n are where the sample size lives.

If a relative frequency column ever totals 2, you have divided by half the true sample size somewhere, which is worth checking before you trust any row.

31. One row of Table 1.11, filled in

Fill the middle

The five-hour row is the tallest in the table. Fill in the sample size and the relative frequency. Type the decimal the way the book writes it, to two places.

Fill in the blanks

\dfrac200.30} = ___

Why: Six students work five hours, and the sample is all twenty students, so the share is six twentieths. Six divided by twenty is 0.30, which is thirty percent of the sample.

The denominator is the place to be careful. It is the whole sample, 20, and not the six rows of the table or the six students in the row. Six over six would say every student works five hours, which the dot plot plainly contradicts.

Once you have 0.30, say it as a sentence to check it: thirty percent of the students work five hours a day. If a sentence like that sounds absurd, the division was wrong.

OpenStax Introductory Statistics 2e, §1.3 Frequency, Frequency Tables, and Levels of Measurement §1.3, p. 27 — Table 1.11

32. Rounding: one more decimal place than the data, and only at the end

Concept

Statistics produces long decimals, so the book fixes a rounding rule once and uses it all year. Carry your final answer to one more decimal place than was present in the original data.

\[ \dfrac{4 + 6 + 9}{3} = \dfrac{19}{3} = 6.333\ldots \approx 6.3 \]

That is the book's own example: the average of the quiz scores four, six and nine is 6.3, rounded to the nearest tenth because the data are whole numbers. One extra place reports the answer honestly without claiming precision the data never had.

The second half of the rule matters as much. Round only the final answer. If you must round an intermediate result, carry it to at least twice as many decimal places as the final answer, so the rounding errors cannot build up into the digits you report.

The book also says that most fractions need not be reduced in this course, and in the probability chapter it prefers them unreduced. Three twentieths shows the count and the sample size at a glance, while its reduced form would hide them. Leave fractions like that as they are.

OpenStax Introductory Statistics 2e, §1.3 Frequency, Frequency Tables, and Levels of Measurement §1.3, p. 26 — Answers and Rounding Off

33. Rounding the mean number of hours worked

Worked example

Apply the rule to the twenty work-hours answers. The data are whole numbers, so the final answer gets one decimal place, and nothing is rounded before that.

Add the data, using the frequency table to save work

Why: each value times the number of students who gave it.

\[ 2(3) + 3(5) + 4(3) + 5(6) + 6(2) + 7(1) \]

Multiply out

Why: one product per row.

\[ = 6 + 15 + 12 + 30 + 12 + 7 \]

Add the products

Why: this is the sum of all twenty answers.

\[ = 82 \]

Divide by the number of students

Why: twenty answers in all.

\[ \dfrac{82}{20} = 4.1 \]

Here the division came out exactly, so the rounding rule has nothing to cut. Reported to one decimal place, the mean is 4.1 hours per day. It is not a whole number even though every answer was, which is normal for a mean.

Check the sum against the raw list

Why: adding the twenty raw answers directly must give the same total.

\[ 5 + 6 + 3 + 3 + 2 + 4 + 7 + 5 + 2 + 3 + \ldots = 82 \;\checkmark \]

The raw list and the table agree, which also confirms the table once more. Using a frequency table to compute a sum is a technique Chapter 2 builds on, so it is worth trusting now.

OpenStax Introductory Statistics 2e, §1.3 Frequency, Frequency Tables, and Levels of Measurement §1.3, pp. 26-27 — the rounding rule applied to the work-hours data

34. Why a relative frequency column can total 1.01

Worked example

The book warns that because of rounding, a relative frequency column may not add to exactly one, and the last cumulative entry may not be exactly one. Here is it happening, with the nineteen commuting distances from Example 1.17.

The eleven corrected frequencies are 2, 1, 1, 3, 2, 3, 2, 1, 1, 2 and 1. Dividing by 19 gives decimals that do not terminate, so every one must be rounded.

Round each share to two places

Why: the fractions are over 19, so none comes out exact.

\[ \tfrac{2}{19} \approx 0.11, \quad \tfrac{1}{19} \approx 0.05, \quad \tfrac{3}{19} \approx 0.16 \]

Count how many of each there are

Why: four rows of 0.11, five of 0.05 and two of 0.16.

\[ 4(0.11) + 5(0.05) + 2(0.16) \]

Add the rounded column

Why: multiply out and total.

\[ = 0.44 + 0.25 + 0.32 = 1.01 \]

Every individual share was rounded correctly, and the total is still off by a hundredth. Each rounding nudged its value up a little, and eleven small nudges added up to something visible.

Check with the unrounded fractions

Why: before any rounding the column is nineteen nineteenths.

\[ \dfrac{2 + 1 + 1 + 3 + 2 + 3 + 2 + 1 + 1 + 2 + 1}{19} = \dfrac{19}{19} = 1 \]

So the table was right and the rounding was right; 1.01 is what honest rounding does. A total of 1.01 is acceptable. A total of 0.95 or 1.2 is not rounding, it is a counting error.

OpenStax Introductory Statistics 2e, §1.3 Frequency, Frequency Tables, and Levels of Measurement §1.3, p. 28 — the Note on rounding in the relative frequency column

35. Cumulative relative frequency: the running total

Section

Part 4

36. Cumulative relative frequency accumulates the rows so far

Concept

Many questions are about a range of values rather than one value: what share of students work four hours or fewer? For that you want the shares added up as you go down the table, and that running total is the cumulative relative frequency.

cumulative relative frequency — The accumulation of the previous relative frequencies: add all the previous relative frequencies to the relative frequency of the current row.

\[ \text{cum}_k = \text{cum}_{k-1} + \text{relative frequency}_k \]

Figure (svg): A staircase graph for the work-hours data. The height jumps at each value: to 0.15 at 2 hours, 0.40 at 3, 0.55 at 4, 0.85 at 5, 0.95 at 6 and 1 at 7, and it never comes down.

Each step up is one row's relative frequency, and the height you stand at is everything so far. A staircase that only climbs, finishing at exactly 1, is what a correct cumulative column looks like.

The staircase is the column drawn. Each step up is one row's relative frequency, and the height after a step is the share of the sample at or below that value. It only ever climbs, because relative frequencies are never negative.

And it must finish at exactly 1, because by the last row every observation has been accumulated. One hundred percent of the data is at or below the largest value. A cumulative column that does not end at 1, give or take rounding, has an error in it.

OpenStax Introductory Statistics 2e, §1.3 Frequency, Frequency Tables, and Levels of Measurement §1.3, p. 28 — cumulative relative frequency

37. Adding the cumulative column: Table 1.12

Worked example

Build the fourth column of the work-hours table. Each row adds its own relative frequency to the running total from the row above.

The first row has nothing before it

Why: so its cumulative value is its own relative frequency.

\[ 0.15 \]

Add the second row

Why: three hours contributes 0.25.

\[ 0.15 + 0.25 = 0.40 \]

Add the third row

Why: four hours contributes 0.15.

\[ 0.40 + 0.15 = 0.55 \]

Add the fourth row

Why: five hours contributes 0.30.

\[ 0.55 + 0.30 = 0.85 \]

Notice that you never go back to the frequency column. Each new entry needs only two numbers, the cumulative value directly above it and the relative frequency in its own row, which is what makes the column quick to build and easy to check line by line.

Add the fifth and sixth rows

Why: six hours contributes 0.10, seven hours 0.05.

\[ 0.85 + 0.10 = 0.95, \qquad 0.95 + 0.05 = 1.00 \]

data valuefrequencyrelative frequencycumulative relative frequency
230.150.15
350.250.40
430.150.55
560.300.85
620.100.95
710.051.00

Read the finished column from the top and it tells a story: 15 percent of the students work two hours or fewer, 40 percent three or fewer, 55 percent four or fewer, and so on until everyone is accounted for at seven hours.

Check the running total by counting students

Why: the four-hour row claims 0.55 of the sample works four hours or fewer, which is 11 students.

\[ 3 + 5 + 3 = 11, \qquad \dfrac{11}{20} = 0.55 \;\checkmark \]

Counting the students directly gives the same share as the running total, and the last row ends at 1.00. Both checks pass, so the column is right.

OpenStax Introductory Statistics 2e, §1.3 Frequency, Frequency Tables, and Levels of Measurement §1.3, p. 28 — Table 1.12, with relative and cumulative relative frequencies

38. Reading one row: four numbers, four different questions

Notation

The five-hour row of Table 1.12 holds four numbers, and each one answers a different question. Mixing them up is the commonest error in the section, so take the row apart.

Annotate

On: \( 5 \quad | \quad 6 \quad | \quad 0.30 \quad | \quad 0.85 \)

  • The data value: an answer some students gave, five hours of work per day.
  • The frequency: how many students gave exactly that answer. A count of students, not of hours.
  • The relative frequency: the share of the sample giving exactly five hours. Six of twenty.
  • The cumulative relative frequency: the share giving five hours or fewer. It covers every row from the top down to this one.

The first three numbers describe exactly one value. The fourth describes a range, everything from the smallest value up to and including this one. That single difference is behind most wrong answers on cumulative questions.

The frequency and the relative frequency describe the same six students in two units, so 6 and 0.30 are one fact. The cumulative 0.85 is a different fact, about seventeen students, and it is the number to handle with care.

A good habit when you read the fourth column: say or fewer out loud. Eighty-five percent work five hours or fewer. Without those two words the sentence is false.

OpenStax Introductory Statistics 2e, §1.3 Frequency, Frequency Tables, and Levels of Measurement §1.3, p. 28 — Table 1.12

39. Range questions answered from the cumulative column

Worked example

Three questions about the work hours, answered without adding up any rows by hand. Each uses one or two numbers from the cumulative column.

What share work five hours or fewer?

Why: that is precisely what the five-hour row's cumulative entry means.

\[ P(\text{at most } 5) = 0.85 \]

What share work more than five hours?

Why: everyone not in the previous group, so subtract from the whole.

\[ 1 - 0.85 = 0.15 \]

The word more is the signal for a complement. The cumulative column answers at most questions directly and nothing else, so a more than question is answered by taking what the column gives you away from one.

What share work more than two and at most five hours?

Why: the five-or-fewer group minus the two-or-fewer group.

\[ 0.85 - 0.15 = 0.70 \]

Figure (svg): One long horizontal strip representing all twenty students, divided into six pieces whose widths are the relative frequencies 0.15, 0.25, 0.15, 0.30, 0.10 and 0.05 for 2 to 7 hours. Brackets mark the first 0.85 of the strip as five hours or fewer, the last 0.15 as more than five, and the stretch from 0.15 to 0.85 as more than two and at most five.

Cumulative relative frequencies are positions along this strip. Any question about a range of values is the length between two positions.

The strip shows why subtraction works. Cumulative values are positions along the whole sample, and the share between two positions is the distance between them. The complement in the second question is the stretch from 0.85 to the end.

Check each answer by adding relative frequencies directly

Why: more than five is six and seven hours; more than two and at most five is three, four and five hours.

\[ 0.10 + 0.05 = 0.15 \;\checkmark \qquad 0.25 + 0.15 + 0.30 = 0.70 \;\checkmark \]

Both routes agree. The cumulative route is faster on long tables, and the direct route is the check, which is exactly the relationship the two should have.

OpenStax Introductory Business Statistics 2e, §2.1 Display Data §2.1 — reading cumulative relative frequency

40. Trap: reading a cumulative entry as exactly that value

Trap

The trap

Asked what share of students work exactly five hours, a student reads the five-hour row of the cumulative column.

\[ P(\text{exactly } 5) = 0.85 \]

That would mean 17 of the 20 students work five hours. The dot plot shows six.

The fix

Exactly five hours is one row, so it is that row's relative frequency.

\[ P(\text{exactly } 5) = 0.30 \]

The cumulative 0.85 means five hours or fewer, and it includes the 0.55 of students who work two, three or four hours.

\[ 0.85 - 0.55 = 0.30 \]

Before reading a cumulative column, turn the question into at most, more than or between. Exactly is none of those, and the moment you see it, the answer lives in a different column.

That subtraction is the way back. Any row's relative frequency is its cumulative entry minus the one above it.

41. Put the table-building steps in order

Ranking

Six steps take you from a raw list to a finished four-column table. Some can only happen after others, and two of them are checks. Put them in the order you would actually do them.

Put in order

Order the steps from first to last.

  1. List the distinct values in ascending order
  2. Tally the raw data to get each frequency
  3. Check that the frequencies add to n
  4. Divide each frequency by n
  5. Keep a running total of the relative frequencies
  6. Check that the last cumulative entry is 1

Why: The values must be listed before they can be tallied, and the tally must be checked against n before n is used as a divisor. Relative frequencies come next, then their running total, and the last entry of the running total is the final check that nothing was lost along the way.

The position of the first check is the one that matters. Dividing by n assumes the frequencies really do add to n, so if they do not, every number in both new columns inherits the error. Check the cheap thing before building on it.

Notice also that the cumulative column cannot be started until the relative frequency column exists, because each cumulative entry is built from the one above it plus the current relative frequency.

42. Example 1.17(a): what is wrong with this table?

Error analysis

Nineteen people were asked how many miles, to the nearest mile, they commute to work each day. The book prints a table built from their answers and asks whether it is correct. Check it against the data before you trust any of it.

\[ 2;\,5;\,7;\,3;\,2;\,10;\,18;\,15;\,20;\,7;\,10;\,18;\,5;\,12;\,13;\,12;\,4;\,5;\,10 \]

Annotate

On: \( \sum f = 2+1+1+3+2+3+2+1+1+1+1 = 18 \)

  • The printed frequencies are 2, 1, 1, 3, 2, 3, 2, 1, 1, 1, 1 for the values 2, 3, 4, 5, 7, 10, 12, 13, 15, 18 and 20. They add to 18.
  • Nineteen people answered, so the frequencies should add to 19. One answer is missing from the table.
  • The value 18 appears twice in the raw list, seventh and twelfth, but the table gives it a frequency of 1.
  • The cumulative entry for 18 miles is printed as seventeen nineteenths but should be eighteen nineteenths. The entry for 20 miles is printed as nineteen nineteenths, which is right, but it does not follow from the rows above it: seventeen plus one is eighteen.

Figure (svg): A dot plot of the nineteen commuting distances from 2 to 20 miles. Most columns have one to three dots; the column above 18 miles has two dots, one of them drawn in red with a note that the printed table counted only one.

Nineteen dots, because nineteen people answered. The printed table's frequency column adds to 18, and the dot plot shows which answer went missing.

So the answer to part (a) is no. The frequency column sums to 18, not 19, and not all the cumulative relative frequencies are correct. The first check, frequencies against n, finds the error in one addition.

Part (d) of the book gives the corrected cumulative column. Only the entry for 18 miles changes, because the error is in that row and everything above it was built correctly.

\[ \tfrac{2}{19}, \tfrac{3}{19}, \tfrac{4}{19}, \tfrac{7}{19}, \tfrac{9}{19}, \tfrac{12}{19}, \tfrac{14}{19}, \tfrac{15}{19}, \tfrac{16}{19}, \tfrac{18}{19}, \tfrac{19}{19} \]

OpenStax Introductory Statistics 2e, §1.3 Frequency, Frequency Tables, and Levels of Measurement §1.3, pp. 31-32 — Example 1.17(a), Table 1.15

43. Example 1.17(b) to (d): fractions of the commuters

Worked example

With the table corrected, answer the rest of the example. The book leaves these answers as fractions over 19, unreduced, so the counts stay visible.

(b) Is it true that three percent commute three miles or less?

Why: three miles or less means the values 2 and 3, with frequencies 2 and 1.

\[ \dfrac{2 + 1}{19} = \dfrac{3}{19} \approx 0.158 \]

False. Three people commute three miles or less, but three people out of nineteen is about 15.8 percent, not three percent. The statement confused a count with a percentage.

(c) What fraction commute five or seven miles?

Why: two rows, frequencies 3 and 2.

\[ \dfrac{3 + 2}{19} = \dfrac{5}{19} \]

Five or seven is two separate rows, not a range. The value 6 sits between them, but nobody gave it, so there is no row for it; the word or simply tells you to add the two rows together.

(d) What fraction commute 12 miles or more?

Why: the values 12, 13, 15, 18 and 20, using the corrected 2 for 18.

\[ \dfrac{2 + 1 + 1 + 2 + 1}{19} = \dfrac{7}{19} \]

Less than 12 miles is the complement

Why: everyone not commuting 12 or more.

\[ 1 - \dfrac{7}{19} = \dfrac{12}{19} \]

Between 5 and 13 miles, not including either

Why: the values strictly inside are 7, 10 and 12.

\[ \dfrac{2 + 3 + 2}{19} = \dfrac{7}{19} \]

Check the complement against the corrected cumulative column

Why: less than 12 means 10 miles or fewer, whose cumulative entry is twelve nineteenths.

\[ \text{cum}(10) = \dfrac{12}{19} \;\checkmark \]

The complement and the cumulative column agree. The words strictly between in the last part matter: including 5 and 13 would add 3 and 1 and give eleven nineteenths instead.

OpenStax Introductory Statistics 2e, §1.3 Frequency, Frequency Tables, and Levels of Measurement §1.3, p. 32 — Example 1.17(b)-(d) and solution

44. Grouping continuous data into classes

Section

Part 5

45. Grouping heights into intervals: Table 1.13

Concept

Heights are continuous, so a table with one row per recorded height would be long and uninformative. The book instead groups the heights of 100 semiprofessional soccer players into eight intervals, each two inches wide.

heights (inches)frequency
59.95–61.955
61.95–63.953
63.95–65.9515
65.95–67.9540
67.95–69.9517
69.95–71.9512
71.95–73.957
73.95–75.951

Figure (svg): A histogram of Table 1.13: eight adjacent bars over the height classes from 59.95 to 75.95 inches, with heights 5, 3, 15, 40, 17, 12, 7 and 1 written above them. The bar for 65.95 to 67.95 inches is by far the tallest.

The table drawn as bars. One class holds 40 of the 100 players, and the counts thin out on either side of it.

Drawn as bars, the table shows its shape at once: most players are between about 64 and 70 inches, with the 40 in the middle class by far the largest group, and a thin tail on each side.

Grouping trades detail for clarity. You can no longer tell whether a player in the 65.95 to 67.95 class is 66.0 or 67.9 inches tall, but you can see the overall pattern, which is usually what you wanted. Chapter 2 explains how the book chose these particular intervals.

OpenStax Introductory Statistics 2e, §1.3 Frequency, Frequency Tables, and Levels of Measurement §1.3, pp. 28-29 — Table 1.13, Frequency Table of Soccer Player Height

46. Why the class boundaries end in .95

Concept

The intervals look oddly specific. Why 61.95 rather than 62? The answer is in how the heights were recorded: to one decimal place, like 61.9 or 62.0.

Figure (svg): A zoomed number line from 61.6 to 62.3 inches with ticks every tenth. Yellow dots sit on the recorded heights 61.8, 61.9, 62.0 and 62.1. A dashed red boundary at 61.95 falls halfway between 61.9 and 62.0, splitting the first class on its left from the second class on its right.

A boundary with one more decimal place than the data sits between two possible readings, never on one, so every player lands in exactly one class.

A boundary with two decimal places sits halfway between two possible readings. A player recorded at 61.9 is clearly in the first class and one at 62.0 is clearly in the second, and no recorded height can ever equal 61.95, so no player sits on a boundary.

The book states the result: all heights fall between the endpoints of an interval and not at the endpoints. If the boundaries had been whole numbers such as 62, a player measured at exactly 62.0 would belong to two classes at once, and you would need an extra rule to say which.

The general principle is to give the boundaries one more decimal place than the data. It is the same one-extra-place idea as the rounding rule, used for a different purpose: to guarantee every observation lands in exactly one class.

OpenStax Introductory Statistics 2e, §1.3 Frequency, Frequency Tables, and Levels of Measurement §1.3, p. 29 — the eight intervals, and all heights falling between endpoints

47. Completing the columns of Table 1.13

Worked example

Add the relative and cumulative columns to the heights table. The sample size is 100, so each relative frequency is the count moved two decimal places.

Check the frequencies against n first

Why: a hundred players were measured.

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

Divide each frequency by 100

Why: the eight shares, in class order.

\[ 0.05, \; 0.03, \; 0.15, \; 0.40, \; 0.17, \; 0.12, \; 0.07, \; 0.01 \]

With 100 observations every relative frequency is just the count written as hundredths, which makes this the friendliest grouped table in the section. Five players out of a hundred is 0.05; forty is 0.40.

Accumulate them, row by row

Why: each entry is the previous one plus the new share.

\[ 0.05, \; 0.08, \; 0.23, \; 0.63, \; 0.80, \; 0.92, \; 0.99, \; 1.00 \]

heights (inches)frequencyrelative frequencycumulative
59.95–61.9550.050.05
61.95–63.9530.030.08
63.95–65.95150.150.23
65.95–67.95400.400.63
67.95–69.95170.170.80
69.95–71.95120.120.92
71.95–73.9570.070.99
73.95–75.9510.011.00

Each cumulative entry now means the share of players shorter than that class's upper boundary. The 0.63 in the fourth row says 63 percent of the players are under 67.95 inches.

Check the last cumulative entry

Why: every player has been accumulated by the final class.

\[ 0.99 + 0.01 = 1.00 \;\checkmark \]

With n equal to 100 there is no rounding at all, so the column must land exactly on 1, and it does.

OpenStax Introductory Statistics 2e, §1.3 Frequency, Frequency Tables, and Levels of Measurement §1.3, pp. 28-29 — Table 1.13

48. Example 1.14: heights less than 65.95 inches

Worked example

From Table 1.13, find the percentage of heights that are less than 65.95 inches. The boundary 65.95 is the top of the third class, so the question is about the first three rows.

Figure (svg): The heights histogram with its first three bars, from 59.95 to 65.95 inches, highlighted in yellow and labelled 5 plus 3 plus 15 equals 23.

Less than 65.95 inches is everything to the left of that boundary: the three highlighted classes and nothing else.

Identify the classes entirely below 65.95

Why: the first three classes end at 61.95, 63.95 and 65.95.

\[ 59.95\text{--}61.95, \quad 61.95\text{--}63.95, \quad 63.95\text{--}65.95 \]

Add their frequencies

Why: players in those three classes.

\[ 5 + 3 + 15 = 23 \]

Divide by the number of players

Why: one hundred in all.

\[ \dfrac{23}{100} = 0.23 = 23\% \]

Say the answer in context: 23 percent of the sampled players are shorter than 65.95 inches. That describes this sample of 100; whether it describes all semiprofessional players depends on how the sample was drawn, which is part (f) of Example 1.16.

Check against the cumulative column

Why: the third row's cumulative entry is the share below that row's upper boundary.

\[ \text{cum}(63.95\text{--}65.95) = 0.23 \;\checkmark \]

The book makes this point itself: the answer is the cumulative relative frequency entry in the third row. Once the cumulative column exists, any less than question on a class boundary is a single lookup.

OpenStax Introductory Statistics 2e, §1.3 Frequency, Frequency Tables, and Levels of Measurement §1.3, p. 29 — Example 1.14

49. Example 1.15: heights between 61.95 and 65.95 inches

Worked example

From Table 1.13, find the percentage of heights between 61.95 and 65.95 inches. Now the range starts partway down the table, so it covers neither the first row nor the whole table.

Figure (svg): The heights histogram with only the second and third bars, from 61.95 to 65.95 inches, highlighted in orange and labelled 3 plus 15 equals 18.

Between two boundaries means the classes that start at the first and stop at the second. The first class is left out.

Identify the classes inside the range

Why: 61.95 to 63.95 and 63.95 to 65.95, the second and third rows.

\[ \text{rows } 2 \text{ and } 3 \]

Add their relative frequencies

Why: the book's route.

\[ 0.03 + 0.15 = 0.18 = 18\% \]

Because the endpoints 61.95 and 65.95 are class boundaries and no height sits exactly on one, the question of whether between includes its endpoints does not arise here. With whole-number data, as in Example 1.17, it does.

Check with the cumulative column

Why: below 65.95 minus below 61.95.

\[ 0.23 - 0.05 = 0.18 \;\checkmark \]

The subtraction route always works for a range of whole classes, and on a long table it is faster than adding the rows in the middle one at a time.

OpenStax Introductory Statistics 2e, §1.3 Frequency, Frequency Tables, and Levels of Measurement §1.3, p. 30 — Example 1.15

50. Before you add anything: which rows?

Step zero

Example 1.16 asks four range questions in a row, and every one of them goes wrong at the same place: choosing the rows, before any arithmetic happens. Do only that step here.

Discussion prompt

For each of these ranges in Table 1.13, list the classes it covers, by their boundaries. (a) from 67.95 to 71.95 inches; (b) from 67.95 to 73.95 inches; (c) more than 65.95 inches; (d) between 61.95 and 71.95 inches.

A quick check on your row choices: the width of the range divided by the class width of 2 inches gives the number of classes. From 61.95 to 71.95 is 10 inches, so five classes.

The words more than 65.95 run to the end of the table. Because there is a top class, the range stops there without saying so, and that is what makes the complement route on the next slide so quick.

OpenStax Introductory Statistics 2e, §1.3 Frequency, Frequency Tables, and Levels of Measurement §1.3, p. 30 — Example 1.16 (a)-(d)

51. Example 1.16: four range questions on the heights

Worked example

Now do the arithmetic for the four ranges from the previous slide. Use relative frequencies for the percentages and frequencies for the count.

(a) From 67.95 to 71.95 inches

Why: two classes.

\[ 0.17 + 0.12 = 0.29 = 29\% \]

(b) From 67.95 to 73.95 inches

Why: the same two classes and one more.

\[ 0.17 + 0.12 + 0.07 = 0.36 = 36\% \]

Part (b) is part (a) with one more class on the end, so its answer must be part (a) plus that class's 0.07. Spotting that one question extends another saves adding the same rows twice, and it checks both answers at once.

(c) More than 65.95 inches

Why: everyone except the three shortest classes, so use the complement of the cumulative entry.

\[ 1 - 0.23 = 0.77 = 77\% \]

The complement is the fast route for part (c). The long route adds five relative frequencies, 0.40, 0.17, 0.12, 0.07 and 0.01, which also gives 0.77 and takes three times as long.

(d) Number of players between 61.95 and 71.95 inches

Why: a count this time, so add frequencies.

\[ 3 + 15 + 40 + 17 + 12 = 87 \]

Figure (svg): The heights histogram with the five bars from 61.95 to 71.95 inches highlighted in green, labelled 87 players; the first bar and the last two are left plain.

Part (d) takes the five middle classes. The three plain bars, 5, 7 and 1 players, are the check: 100 minus 13 is 87.

Check part (d) by counting who is left out

Why: the players outside the range are the first class and the last two.

\[ 100 - (5 + 7 + 1) = 87 \;\checkmark \]

Both routes give 87 players. Parts (a) and (b) check the same way from the cumulative column: 0.92 minus 0.63 is 0.29, and 0.99 minus 0.63 is 0.36.

OpenStax Introductory Statistics 2e, §1.3 Frequency, Frequency Tables, and Levels of Measurement §1.3, pp. 30-31 — Example 1.16 (a)-(d) and solution

52. Example 1.16(e) and (f): what the heights are, and where they came from

Concept

The last two parts of Example 1.16 step back from the table and ask about the data themselves. They connect this section to the two before it.

Part (e) asks what kind of data the heights are. They are quantitative and continuous: numbers that measure an amount and can take any value in a range. On the ladder from Part 1 they are ratio data, because a height of zero would mean no height at all.

\[ \text{height: quantitative, continuous, ratio level} \]

Part (f) asks how the heights could be gathered so that they are characteristic of all semiprofessional soccer players. The book's answer is to get the rosters from each team and choose a simple random sample from each.

That is the stratified sampling of Section 1.2, with each team as a stratum. It guarantees every team is represented, so the sample cannot accidentally be all goalkeepers from one tall squad. A table is only as good as the sample behind it: the tidiest frequency table cannot repair data that were collected badly.

It is also worth noticing that the frequency table itself does not care how the data were collected. That judgement has to come from you, which is why a report always states how its sample was drawn.

OpenStax Introductory Statistics 2e, §1.3 Frequency, Frequency Tables, and Levels of Measurement §1.3, pp. 30-31 — Example 1.16 (e) and (f)

53. Complete the rainfall table: Table 1.14

Comparison

The book's rainfall table records the annual rainfall of a sample of 50 towns. Fill in the missing relative and cumulative entries. The sample size is 50, so each relative frequency is the count divided by 50.

Comparison matrix

rainfall (inches)frequencyrelative frequencycumulative
2.95–4.9760.120.12
4.97–6.9970.140.26
6.99–9.01150.300.56
9.01–11.0380.160.72
11.03–13.0590.180.90
13.05–15.0750.101.00

Dividing by 50 is the same as doubling and then dividing by 100, so seven towns is fourteen hundredths and eight towns is sixteen hundredths. That shortcut makes a table over 50 almost as easy as one over 100.

The cumulative blanks each add one row's share to the entry above: 0.26 and 0.30 make 0.56, 0.72 and 0.18 make 0.90, and the final 0.10 brings the total to exactly 1.00. The frequencies add to 50, so no rounding interferes.

Keep this table in view. The next two slides answer the book's four Try It questions from it.

OpenStax Introductory Statistics 2e, §1.3 Frequency, Frequency Tables, and Levels of Measurement §1.3, p. 30 — Table 1.14

54. Try It 1.14 and 1.15: rainfall below and between

Worked example

Two questions on Table 1.14, the same two shapes as Examples 1.14 and 1.15. Try It 1.14 asks for the percentage of rainfall less than 9.01 inches; Try It 1.15 asks for the percentage between 6.99 and 13.05 inches.

Figure (svg): A histogram of Table 1.14, annual rainfall in 50 towns: six bars from 2.95 to 15.07 inches with heights 6, 7, 15, 8, 9 and 5. The first three bars, below 9.01 inches, are highlighted.

Everything left of 9.01 inches: 6, 7 and 15 towns, which is 28 of the 50.

Try It 1.14: less than 9.01 is the first three classes

Why: 9.01 is the top of the third class, so read that row's cumulative entry.

\[ \text{cum}(6.99\text{--}9.01) = 0.56 = 56\% \]

Try It 1.15: between 6.99 and 13.05

Why: the classes from 6.99 to 9.01 up to 11.03 to 13.05, three rows.

\[ 0.30 + 0.16 + 0.18 = 0.64 = 64\% \]

Both questions are about towns, not about rain. The 56 percent is a share of the fifty towns in the sample, and the individual being counted, a town, is what every percentage in this table is a percentage of.

Say each answer in context: 56 percent of the sampled towns get less than 9.01 inches of rain a year, and 64 percent get between 6.99 and 13.05 inches.

Check both from the cumulative column and the counts

Why: below 13.05 minus below 6.99 for the second; six, seven and fifteen towns for the first.

\[ 0.90 - 0.26 = 0.64 \;\checkmark \qquad \dfrac{6 + 7 + 15}{50} = \dfrac{28}{50} = 0.56 \;\checkmark \]

Both answers are confirmed by an independent route, which is the standard you should hold every range answer to.

OpenStax Introductory Statistics 2e, §1.3 Frequency, Frequency Tables, and Levels of Measurement §1.3, p. 30 — Try It 1.14 and Try It 1.15

55. Try It 1.16 and 1.17: a count and a fraction of towns

Worked example

Two more questions on Table 1.14. Try It 1.16 asks for the number of towns with rainfall between 2.95 and 9.01 inches; Try It 1.17 asks what fraction of towns get between 11.03 and 13.05 inches.

Try It 1.16: a number of towns means adding frequencies

Why: 2.95 to 9.01 covers the first three classes.

\[ 6 + 7 + 15 = 28 \text{ towns} \]

Figure (svg): The rainfall histogram with only the fifth bar, from 11.03 to 13.05 inches, highlighted in orange; it has height 9.

One class, one row of the table: nine towns out of fifty.

Try It 1.17: one class, so one relative frequency

Why: 11.03 to 13.05 is the fifth row, nine towns.

\[ \dfrac{9}{50} = 0.18 \]

Notice the different forms the two questions want. The number of towns is a count, 28, not a percentage; the fraction of towns is nine fiftieths, which you may also write as 0.18. Answer the question in the form it asks.

Check the count against the cumulative column

Why: the cumulative entry below 9.01 times the sample size must give the same count.

\[ 0.56 \times 50 = 28 \;\checkmark \]

Multiplying a cumulative share back by n recovers the count, which is the reverse of how the share was made and a reliable check whenever n is known.

OpenStax Introductory Statistics 2e, §1.3 Frequency, Frequency Tables, and Levels of Measurement §1.3, pp. 31-32 — Try It 1.16 and Try It 1.17

56. Example 1.18: years of service for 70 federal employees

Worked example

Table 1.16 gives the number of years of service for 70 federal employees, from 24 to 35 years. Note that 27 years has a frequency of zero: no one in the sample had exactly 27 years, but the row still belongs in the table.

Figure (svg): A bar chart of years of service for 70 federal employees, from 24 to 35 years. The bars read 2, 1, 3, 0, 4, 6, 11, 12, 7, 8, 6 and 10; the bars from 30 to 35 years are highlighted, and 27 years has no bar at all.

The highlighted block is part (a). The empty slot at 27 years is a real row with frequency zero, and it still belongs in the table.

Check the frequencies against n

Why: seventy employees.

\[ 2+1+3+0+4+6+11+12+7+8+6+10 = 70 \]

(a) Frequency for 30 to 35 years, inclusive

Why: the six highlighted bars.

\[ 11 + 12 + 7 + 8 + 6 + 10 = 54 \]

(b) Relative frequency for exactly 30 years

Why: one row.

\[ \dfrac{11}{70} \approx 0.157 \]

Eleven seventieths does not terminate, so it has to be rounded. The book gives 0.157, or 15.7 percent. Keep the fraction until the last line and round once, exactly as the rounding rule asks.

(c) Relative frequency for 30 years or less

Why: every row from 24 to 30.

\[ \dfrac{2+1+3+0+4+6+11}{70} = \dfrac{27}{70} \approx 0.386 \]

(d) Relative frequency for 25 years or more

Why: everyone except the two with 24 years.

\[ \dfrac{70 - 2}{70} = \dfrac{68}{70} \approx 0.971 \]

Check (a) and (c) together

Why: 30 or less and 31 or more must cover everyone, and 31 or more is part (a) minus the 11 at 30.

\[ 27 + (54 - 11) = 27 + 43 = 70 \;\checkmark \]

The book calls part (a) a cumulative frequency; it is really the frequency of a range, the count of employees with 30 to 35 years. Answers are to three decimals, as the book gives them, because the data are whole numbers and a count over 70 does not terminate.

OpenStax Introductory Statistics 2e, §1.3 Frequency, Frequency Tables, and Levels of Measurement §1.3, pp. 32-33 — Example 1.18, Table 1.16

57. Try It 1.18: eighteen years of fatal crashes

Worked example

Table 1.17 gives the total number of fatal motor vehicle traffic crashes in the United States in each of 18 years, 653,782 in all. The book's questions say deaths, but the table counts crashes, so these answers are about crashes.

Figure (svg): A bar chart of fatal motor vehicle crashes by year for Years 1 to 18, in thousands. The bars stay between about 36 and 39 thousand up to Year 14, then fall to about 34, 31, 30 and 30 thousand; Years 14 to 18 are highlighted.

Thirteen fairly level years, then five falling ones. The highlighted bars are the ones part (b) asks about.

(a) Frequency from Year 7 through Year 11

Why: five years added.

\[ 37{,}526 + 37{,}862 + 38{,}491 + 38{,}477 + 38{,}444 = 190{,}800 \]

(b) Percentage after Year 13

Why: Years 14 to 18, divided by the total.

\[ \dfrac{37{,}435 + 34{,}172 + 30{,}862 + 30{,}296 + 29{,}757}{653{,}782} = \dfrac{162{,}522}{653{,}782} \approx 0.249 \]

(c) Relative frequency in Year 7 or before

Why: Years 1 to 7.

\[ \dfrac{260{,}086}{653{,}782} \approx 0.398 \]

(d) Percentage in Year 18

Why: one year.

\[ \dfrac{29{,}757}{653{,}782} \approx 0.046 = 4.6\% \]

(e) Cumulative relative frequency for Year 13

Why: Years 1 to 13 together.

\[ \dfrac{491{,}260}{653{,}782} \approx 0.751 \]

Part (e) asks what that number tells you: about 75 percent of the crashes in the 18-year period happened in the first 13 years. Since 13 years is only 72 percent of the period, crashes were more concentrated early on, which matches the falling bars at the right of the chart.

Check (b) and (e) against each other

Why: after Year 13 and up to Year 13 are complements, so the two must add to 1.

\[ 162{,}522 + 491{,}260 = 653{,}782 \;\checkmark \]

The counts add to the printed total exactly, which also confirms the book's total of 653,782. The book gives no answers for this Try It; all five were computed from the table.

OpenStax Introductory Statistics 2e, §1.3 Frequency, Frequency Tables, and Levels of Measurement §1.3, pp. 33-34 — Try It 1.18, Table 1.17

58. Your own class: the sibling survey

Real world

The book suggests a survey you can run in any room: ask everyone how many siblings they have, build the four-column table, and answer three questions from it. Here is what that looks like with a class of 16 whose answers happen to be these.

siblingsfrequencyrelative frequencycumulative
030.18750.1875
160.3750.5625
240.250.8125
320.1250.9375
410.06251.0000

Discussion prompt

From the table, what percentage of this class have no siblings? What percentage have from one to three siblings? What percentage have fewer than three? Say which column you read each answer from, and state the level of measurement of the variable.

The phrase fewer than three is where people slip, by reading the cumulative entry at three. Fewer than three stops at two; three or fewer would include the row for three.

Try it with a real group. Your table will not match this one, and the die experiment already told you why that is not a problem: a different sample gives a different table, and both are honest records.

OpenStax Introductory Statistics 2e, §1.3 Frequency, Frequency Tables, and Levels of Measurement §1.3, p. 31 — the collaborative exercise on siblings

59. Putting it together

Section

Part 6

60. From raw data to answers, with a check at every stage

Pattern

The whole lesson is one procedure with a question in front of it. Look at the flow once and notice that every computed column comes with its own check.

Figure (svg): A flow diagram with five boxes: raw data, sort and count, frequency column that must add to n, divide each by n to get relative frequency that must add to 1, and a running total giving cumulative relative frequency that must end at 1. A check is written under each of the last three.

Three columns, three checks. The checks cost seconds and they catch every tallying slip in this lesson.
  1. Name the level of measurement first. Nominal and ordinal data are counted and ordered, never averaged; interval and ratio data support arithmetic
  2. Sort and count. One row per value, or per class for continuous data, with boundaries one decimal place finer than the data
  3. Frequencies add to n. Check before using n for anything
  4. Relative frequency is f over n. The column adds to 1, or very close to it after rounding
  5. Cumulative is the running total. It never falls and it ends at 1
  6. Answer range questions from the cumulative column. At most is a lookup, more than is a complement, between is a difference
  7. Round only the final answer, to one more decimal place than the data

If you remember one thing, remember the three checks. Each costs one addition, and between them they catch almost every slip this section makes possible, including the one the book deliberately planted in Example 1.17.

The level of measurement comes first in the list for a reason. Every column of a frequency table can be built for nominal data, but the moment you compute a mean or a ratio from the values you have assumed a level, and deciding it first stops a tidy table from leading to a meaningless summary.

61. Check: which variable is interval data?

Check

One question on Part 1. Run the three questions on each option, and remember that a zero printed on a scale is not necessarily a true zero.

Check your understanding

Which of these variables is measured at the interval level?

  • A. temperature in degrees Fahrenheit (correct)
  • B. score on a statistics exam out of 100
  • C. cruise rating from unsatisfactory to excellent
  • D. brand of smartphone

Answer: A

Why: Fahrenheit temperatures have an order and meaningful differences, but zero degrees is not the absence of heat, so ratios mean nothing: that is interval data. The exam score has a true zero, the rating has order without measurable gaps, and the brand has no order at all.

Why B tempts people
An exam score has a true zero, no points earned, so 80 really is four times 20. That makes it ratio data, one level above interval.
Why C tempts people
The ratings come in a definite order, but the gap between satisfactory and good is not a measured amount, so this is ordinal data.
Why D tempts people
Smartphone brands have no agreed order, however strong people's preferences are, so this is nominal data, the lowest level.

Temperature is the textbook example of interval data precisely because it looks like ratio data. It is numerical, it can be subtracted, and it has a zero printed on the thermometer. The test that exposes it is asking what that zero means.

If you chose the exam score, the difference to hold on to is that zero points really is nothing, while zero degrees Fahrenheit is merely a cold day.

62. Check: at least four hours

Check

One question on Part 4, using Table 1.12. Decide first whether the question is a lookup, a complement or a difference.

hours234567
cumulative0.150.400.550.850.951.00

Check your understanding

What share of the twenty students work at least four hours per day?

  • A. 0.60 (correct)
  • B. 0.55
  • C. 0.45
  • D. 0.15

Answer: A

Why: At least four hours means four or more, which is everyone except those who work three hours or fewer. The cumulative entry at three hours is 0.40, so the answer is one minus 0.40, which is 0.60. Counting confirms it: 3, 6, 2 and 1 students make 12 of 20.

Why B tempts people
This is the cumulative entry at four hours, which means four hours or fewer. It is the share at the other end of the table.
Why C tempts people
This is one minus the cumulative entry at four hours, which is more than four hours. It leaves out the three students who work exactly four.
Why D tempts people
This is the relative frequency of exactly four hours, one row, not the range the question asks about.

The three wrong options are the three ways a range question goes wrong: reading the wrong end, dropping the boundary row, and answering for one value instead of a range. Saying at least four as four or more before you start avoids all three.

The complement has to start from the row just below the boundary. Four or more is the complement of three or fewer, so the entry to subtract sits in the three-hour row.

63. Check: towns with at least 9.01 inches of rain

Check

One question on Part 5, using the rainfall table. The boundary 9.01 is the top of the third class, so no town sits exactly on it.

Check your understanding

From Table 1.14, what proportion of the 50 towns get 9.01 inches of rain or more per year?

  • A. 0.44 (correct)
  • B. 0.56
  • C. 0.34
  • D. 0.16

Answer: A

Why: Towns with 9.01 inches or more are those in the last three classes. The cumulative entry below 9.01 is 0.56, so the proportion is one minus 0.56, which is 0.44. Adding the last three relative frequencies, 0.16 and 0.18 and 0.10, gives the same 0.44, which is 22 of the 50 towns.

Why B tempts people
This is the cumulative entry for the third class, the proportion below 9.01 inches, which is the complement of what was asked.
Why C tempts people
This adds the fourth and fifth classes but leaves out the last class, from 13.05 to 15.07 inches, which is also above 9.01.
Why D tempts people
This is only the class immediately above the boundary, from 9.01 to 11.03 inches, not everything above it.

Because the boundaries carry more decimal places than the data, or at least than any reading that lands on them, at least 9.01 and more than 9.01 give the same answer here. With whole-number data like the commuting distances, that distinction would change the rows you use.

Checking with the counts is worth the few seconds: 8, 9 and 5 towns make 22, and 22 of 50 is 0.44.

OpenStax Introductory Statistics 2e, §1.3 Frequency, Frequency Tables, and Levels of Measurement §1.3, p. 30 — Table 1.14

64. Why must the columns add to one?

Explain it to yourself

The checks in this lesson only help if you know why they must pass. Put the reasons into your own words; being able to say them is what lets you trust a table you built yourself.

Discussion prompt

Explain in three or four sentences why the relative frequency column must add to 1 and the cumulative column must end at 1. Then explain what it means if a relative frequency column adds to 1.01, and what it means if it adds to 0.95.

A good explanation names n as the reason. Everything in the check depends on the same n being the total of the frequencies and the divisor of every row, and if either fails, the check fails with it.

If your explanation used the word percent, test it: a cumulative column that ends at 100 percent is the same statement, since one hundred percent of the data have been accumulated by the last row.

65. Exit ticket: a table from scratch

Exit ticket

Sixteen students were asked how many pets they have. Build the complete four-column table, then answer two questions from it. Run all three checks.

\[ 0;\,1;\,2;\,1;\,0;\,3;\,1;\,2;\,0;\,1;\,1;\,4;\,2;\,0;\,1;\,1 \]

Discussion prompt

Build the frequency, relative frequency and cumulative relative frequency columns. Then give the share of students with at most one pet and the share with more than one pet, and state the level of measurement of the variable.

The two places this goes wrong are the tally and the phrase at most. If your frequency column added to 15 or 17, recount before anything else. If you gave 0.4375 for at most one pet, you answered for exactly one pet; at most one includes the students with none.

If everything checked, you have done the whole section on a data set you had never seen, which is the skill the rest of the course assumes.

66. What you can do now

Recap

This section gave you two tools that every later chapter uses without comment: a way to decide what arithmetic your data allow, and a way to organise data so that questions can be answered from a table rather than a raw list.

the questionwhat to dothe check
what can I compute from this variable?climb the ladder: ordered, measurable gaps, true zerodoes zero mean none of it?
how often does each value occur?sort, tally, write the frequency columnfrequencies add to n
what share of the sample is that?divide each frequency by nthe column adds to 1, give or take rounding
what share is at most, more than, or between?read, complement or subtract the cumulative columnthe running total ends at 1
continuous data?group into classes whose boundaries have one extra decimalevery observation lands in exactly one class
how do I report the answer?one more decimal place than the data, rounding only at the endno intermediate rounding

\[ \text{relative frequency} = \dfrac{f}{n}, \qquad \text{cumulative} = \text{running total, ending at } 1 \]

The thread to carry forward is that every conclusion rests on evidence you can point to. A percentage answer should name the rows it came from, and a claim that one value is so many times another should rest on a true zero.

Chapter 2 turns these tables into pictures, histograms, frequency polygons and box plots, and explains how the heights intervals were chosen. The cumulative column you built here becomes the tool for percentiles and quartiles there.

OpenStax Introductory Business Statistics 2e, §1.3 Levels of Measurement §1.3 — Levels of Measurement, the same material in the business statistics text

Sources

  1. OpenStax Introductory Statistics 2e, §1.3 Frequency, Frequency Tables, and Levels of Measurement — Illowsky & Dean, OpenStax / Rice University, CC BY 4.0, pp. 25-34
  2. OpenStax Introductory Business Statistics 2e, §1.3 Levels of Measurement — Illowsky & Dean, OpenStax / Rice University, CC BY 4.0
  3. 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