2.7 Measures of the Spread of the Data

The measure of variation the rest of the book runs on. The standard deviation does two jobs: it gives a single number for the overall spread of a data set, so that two sets with the same mean can be told apart, and it locates an individual value through the relation value equals mean plus a number of standard deviations times the standard deviation — which chapter 6 will rename the z-score. Then the computation itself: deviations from the mean, why they must be squared before averaging, the variance as the average squared deviation, the standard deviation as its square root, and the n minus 1 that distinguishes a sample from a population. Closes by choosing among the range, the interquartile range and the standard deviation.

Subject: Statistics · 71 slides · symbolic lesson

Open the interactive version of this deck

What this lesson covers

The lesson, slide by slide

1. Measures of the Spread of the Data

Title

Statistics · Chapter 2 — Descriptive Statistics

Standard deviation: how far the data stray from their mean, and how far one value strays

2. What this lesson gives you

Objectives

Sections 2.5 and 2.6 gave you the centre of a data set and its shape. A centre on its own can hide almost everything: two checkout lines can share an average wait and still feel completely different to stand in. This section measures how far the data stray from their centre.

The arithmetic in this section is long but never hard. What earns marks is the reasoning around it: why the squares, why one less than n, and what the final number says about the data in their own units. Each of those has its own slide.

OpenStax Introductory Statistics 2e, §2.7 Measures of the Spread of the Data §2.7, pp. 107-118 — the section this deck covers

3. Before anything new: two checkout lines

Warm-up

Section 2.5 gave you the mean, and Section 2.3 the range. Use both on five timed waits, in minutes, at each of two checkout lines.

Discussion prompt

Line A's waits are 2, 4, 5, 6 and 8 minutes. Line B's are 0, 2, 4, 8 and 11 minutes. Find the mean and the range of each line. Which line would you rather join if you have a train to catch?

The range already hints at the difference, but it looks at only two of the five waits, the smallest and the largest. This deck builds a measure that uses every wait, and these two lines will follow you through it.

OpenStax Introductory Statistics 2e, §2.7 Measures of the Spread of the Data §2.5 and §2.3 — the mean and the range

4. What the standard deviation measures

Section

Part 1

5. Same mean, different spread

Concept

Here are the two lines from the warm-up, drawn on one scale. Look at how the dots sit around the dashed line at the mean before reading anything else.

Figure (svg): Two dot plots on the same scale from 0 to 12 minutes. Line A has dots at 2, 4, 5, 6 and 8, bunched near a dashed line at the mean, 5. Line B has dots at 0, 2, 4, 8 and 11, spread much further from the same dashed line at 5.

Same centre, same number of waits, very different spread. A centre alone cannot tell these two lines apart.

Line A's waits huddle near five minutes. Line B's are flung out on both sides: one customer walked straight through and another waited eleven minutes. Yet the two means are identical, so any summary built only from the centre would call these lines the same.

The book puts it this way: in some data sets the values are concentrated closely near the mean, and in others they are more widely spread out from the mean. The most common measure of that variation is the standard deviation.

standard deviation — A number that measures how far data values are from their mean. It is zero when every value is the same, small when the data are concentrated near the mean, and larger when they are spread out.

OpenStax Introductory Statistics 2e, §2.7 Measures of the Spread of the Data §2.7, p. 107 — the opening paragraph

6. Supermarkets A and B: one number for overall spread

Concept

The book's own example: at supermarket A and at supermarket B the average wait is five minutes. The standard deviation at A is two minutes and at B it is four. You have not been shown a single wait, yet those two numbers already tell you a lot.

\[ \mu_A = \mu_B = 5 \text{ min}, \qquad \sigma_A = 2 \text{ min}, \qquad \sigma_B = 4 \text{ min} \]

Figure (svg): Two number lines from 0 to 12 minutes. On the upper one, for supermarket A, a shaded band runs from 3 to 7, one standard deviation of 2 minutes either side of the mean 5. On the lower one, for supermarket B, the band runs from 1 to 9, one standard deviation of 4 minutes either side of 5.

The band is the mean plus or minus one standard deviation. B's band is twice as wide: its waits stray twice as far from five minutes.

A standard deviation is measured in the same units as the data, minutes here. A useful first picture is the band from one deviation below the mean to one deviation above. At A that band runs from 3 to 7 minutes; at B it runs from 1 to 9.

Because B has the larger standard deviation, its waits vary more: they are more spread out from the average, while A's waits are more concentrated near it. If you hate surprises at the checkout, A is your supermarket, even though the average wait is the same.

It is no accident that this matches the warm-up. Line A's five waits have a standard deviation of exactly 2 and line B's of exactly 4, as Part 3 will compute, so the two lines are a small-scale picture of the two supermarkets.

OpenStax Introductory Statistics 2e, §2.7 Measures of the Spread of the Data §2.7, p. 107 — supermarkets A and B

7. Which set has the larger standard deviation?

Prediction

Each set below has a mean of 50. You do not need a formula: judge how far the values sit from 50, taken as a whole.

Look at more than the extremes. Two sets can have the same smallest and largest value and still differ in how many values sit far out.

Predict first

All four sets have mean 50. Which one has the largest standard deviation?

  • 10, 10, 90, 90
  • 10, 50, 50, 90
  • 40, 45, 55, 60
  • 50, 50, 50, 50

Correct: 10, 10, 90, 90

Why: Every value in that set sits 40 away from the mean. The set with 10, 50, 50 and 90 has the same range but two values right on the mean, so its typical distance is smaller. The set of all fifties has no spread at all.

\[ 10, 10, 90, 90: \quad \sigma = 40; \qquad 10, 50, 50, 90: \quad \sigma \approx 28.3 \]

The two sets with the same range have different standard deviations, which is the whole point of using every value. The range sees only 10 and 90 in both; the standard deviation also notices whether the middle values are far out or sitting on the mean.

Keep the set of four fifties in mind as well. Its standard deviation is zero, the smallest any data set can have.

8. Zero, small, large: reading the size of a standard deviation

Concept

The book lists the facts that let you read a standard deviation without computing one. Each follows from the idea of a typical distance from the mean.

Small and large only mean something against the data's own scale. A standard deviation of 2 minutes is large for checkout waits that average 5 minutes, and tiny for flight times that average 300 minutes. Always read it beside the mean and the units.

The book adds a caution worth taking seriously. In a symmetric distribution the standard deviation is very helpful, but in a skewed one it may not be, because the two sides of a skewed distribution have different spreads, and one number cannot describe both. Part 6 comes back to this.

And the book's last piece of advice is simple: because numbers can be confusing, always graph your data, in a histogram or a box plot, before you lean on any single summary of them.

OpenStax Introductory Statistics 2e, §2.7 Measures of the Spread of the Data §2.7, p. 111 — the facts about s and σ, and the caution for skewed data

9. Spread of zero, small, or large?

Sorting

For each small data set, judge its standard deviation against its own mean. You are not computing; you are reading how far the values sit from their centre compared with the size of the values themselves.

Sort into buckets

Sort each data set by its standard deviation.

exactly zero
7, 7, 7, 7, 7; 12.0, 12.0, 12.0
small next to the values
98, 99, 100, 101, 102; 499, 500, 500, 501
large next to the values
1, 50, 100, 150, 199; 2, 3, 20, 37, 38
z
Every value is identical, so every deviation from the mean is zero, and so is the standard deviation.
s
The values stray only one or two units from a mean of 100 or 500: a standard deviation near 1 is tiny beside them.
l
The values stray tens of units from a mean of 20 or 100, almost as far as the mean itself.

Notice that 98 to 102 and 499 to 501 are both small spreads, even though the numbers are very different sizes. What matters is how far the values move relative to where they sit.

The set 2, 3, 20, 37, 38 is worth a second look. Only one value is near the mean of 20; the other four are far out on both sides, so its standard deviation, about 17, is nearly as big as the mean.

10. Locating a value in standard deviations

Section

Part 2

11. Rosa and Binh: a distance counted in standard deviations

Concept

The standard deviation's second job is to say whether one particular value is close to the mean or far from it. Rosa and Binh both shop at supermarket A, where the mean wait is five minutes and the standard deviation two minutes. Rosa waits seven minutes; Binh waits one.

Figure (svg): A number line from 0 to 10 minutes. The mean, 5, is marked, and below the line the values 1, 3, 7 and 9 are labelled minus 2, minus 1, plus 1 and plus 2 standard deviations. One arc hops from 5 to 7, where Rosa's dot sits. Two arcs hop from 5 to 3 and from 3 to 1, where Binh's dot sits.

Count the hops, each the length of one standard deviation. Rosa is one hop above the mean; Binh is two hops below.

Rosa's seven minutes is two minutes above the average of five, and two minutes is exactly one standard deviation, so Rosa's wait is one standard deviation above the mean.

\[ 7 = 5 + (1)(2) \]

Binh's one minute is four minutes below the average, and four minutes is two standard deviations, so Binh's wait is two standard deviations below the mean.

\[ 1 = 5 + (-2)(2) \]

The book adds a caution to Binh's case. Two standard deviations from the average is just on the borderline of what many statisticians would call far, and that cut-off is a rule of thumb rather than a rigid rule.

Think of the standard deviation as the length of one hop along a ruler. Counting hops turns a distance in minutes into a distance in units that mean the same thing for any data set.

OpenStax Introductory Statistics 2e, §2.7 Measures of the Spread of the Data §2.7, p. 108 — Rosa and Binh, and Figure 2.24

12. Reading the value formula piece by piece

Notation

Rosa's and Binh's lines are two instances of one general relationship. The book writes it in words; here it is with each part named.

Annotate

On: \( \text{value} = \text{mean} + (\#\text{ of STDEVs})(\text{standard deviation}) \)

  • The one data value you are locating: Rosa's 7, Binh's 1.
  • Where you start counting from. x̄ for a sample, μ for a population.
  • How many hops, with a sign: positive above the mean, negative below. It need not be a whole number.
  • The length of one hop, in the data's units: s for a sample, σ for a population.

The book gives the sample and population versions separately. They are the same sentence with different letters, and the letters tell you where the numbers came from.

\[ \text{sample: } x = \bar{x} + (\#\text{ of STDEVs})(s) \]

\[ \text{population: } x = \mu + (\#\text{ of STDEVs})(\sigma) \]

Read the number of deviations as a signed count. It does not need to be an integer: a value can sit 1.5 or 0.21 deviations from its mean. In Part 5 this count gets its own name, the z-score.

OpenStax Introductory Statistics 2e, §2.7 Measures of the Spread of the Data §2.7, p. 108 — value = mean + (#ofSTDEV)(standard deviation)

13. Rosa and Binh: solving for the number of deviations

Worked example

The ruler made the hops easy to count because the numbers were tidy. Solve the formula for the number of deviations instead, so the same method works when they are not.

The plan: subtract the mean from both sides to get the distance, then divide by the standard deviation to turn that distance into hops. Do it for Rosa first, then Binh.

Start from the formula with Rosa's numbers

Why: the unknown is the number of deviations, k.

\[ 7 = 5 + k(2) \]

Subtract the mean from both sides

Why: this leaves the distance from the mean.

\[ 7 - 5 = 2k \quad\Rightarrow\quad 2 = 2k \]

Divide by the standard deviation

Why: the distance divided by one hop's length is the number of hops.

\[ k = \frac{2}{2} = 1 \]

Same two moves for Binh

Why: subtract the mean, then divide by 2.

\[ k = \frac{1 - 5}{2} = \frac{-4}{2} = -2 \]

In context: Rosa waited one standard deviation longer than the average shopper at A, and Binh two standard deviations less. The minus sign carries real information, below the mean, so keep it.

Check by rebuilding each value

Why: the mean plus the hops must give back the wait.

\[ 5 + (1)(2) = 7 \checkmark, \qquad 5 + (-2)(2) = 1 \checkmark \]

OpenStax Introductory Statistics 2e, §2.7 Measures of the Spread of the Data §2.7, p. 108 — Rosa and Binh

14. Line A and line B: two different waits, the same 1.5 deviations

Worked example

Line A's longest wait was 8 minutes and line B's longest was 11. Which wait is more out of the ordinary for its own line? Use each line's standard deviation, 2 and 4 minutes, with the shared mean of 5.

Raw distances say B's wait is further out, 6 minutes against 3. But a distance only means something next to that line's typical spread, so count hops on each line's own ruler.

Line A: distance from the mean

Why: subtract 5.

\[ 8 - 5 = 3 \text{ min} \]

Line A: divide by its standard deviation

Why: one hop is 2 minutes.

\[ k_A = \frac{3}{2} = 1.5 \]

Line B: distance from the mean

Why: subtract 5.

\[ 11 - 5 = 6 \text{ min} \]

Line B: divide by its standard deviation

Why: one hop is 4 minutes.

\[ k_B = \frac{6}{4} = 1.5 \]

In context: both longest waits sit one and a half standard deviations above the mean. Measured against what is normal for its own line, the 11-minute wait at B is no more unusual than the 8-minute wait at A; line B is simply a more variable line.

Check with the formula

Why: mean plus 1.5 hops must rebuild each wait.

\[ 5 + (1.5)(2) = 8 \checkmark, \qquad 5 + (1.5)(4) = 11 \checkmark \]

15. Four minutes below the mean: far, or not?

Counterexample

A claim you will hear: a wait four minutes shorter than the average is far from the average. Test it on the two supermarkets, which share a mean of five minutes.

Discussion prompt

Find a setting where a wait four minutes below the mean is far from the mean, and one where the same four minutes is not. What does this say about judging distance in minutes alone?

This is why the standard deviation earns its place. Without it, the claim sounds reasonable and is unanswerable; with it, you can say exactly how unusual the wait was, and the answer changes from one store to the next.

The same idea will carry the rest of the course. A test statistic in chapter 9 is nothing more than a distance from a mean counted in standard deviations.

OpenStax Introductory Statistics 2e, §2.7 Measures of the Spread of the Data §2.7, p. 108 — the standard deviation decides close or far

16. Locate 72 when the mean is 60 and s is 5

Fill the middle

A sample has mean 60 and standard deviation 5. Fill in the distance, the number of deviations, and the check. Keep the sign.

Fill in the blanks

72 - 60 = 12, \qquad k = \frac12}2.4 = 2.4, \qquad 60 + (___)(5) = 72

Why: The value is twelve above the mean, and twelve divided by a standard deviation of five is 2.4. So 72 sits 2.4 standard deviations above the mean, and rebuilding gives 60 plus 2.4 times 5, which is 72.

In context: 72 is 2.4 standard deviations above the mean. That is past two, so by the book's rule of thumb it counts as far from the mean, an unusually high value for this data set.

Notice the number of deviations came out as a decimal. The book stresses this: the count does not need to be an integer, and most real values sit a fractional number of deviations away.

17. The two-deviation rule of thumb

Estimation

The book offers a working rule for far. A value more than two standard deviations from the mean is usually considered far from it; exactly two is on the borderline. Before reading the book's caution, estimate.

Predict first

Is the two-deviation cut-off a strict law that holds the same way for every data set?

  • No, it is a rough guide, and how much data lies beyond two deviations depends on the shape
  • Yes, exactly 5 percent of any data set lies beyond two deviations
  • Yes, no value can ever lie more than two deviations from its mean
  • No, the cut-off should always be one deviation

Correct: No, it is a rough guide, and how much data lies beyond two deviations depends on the shape

Why: The book calls it a rule of thumb rather than a rigid rule. The shape of the distribution affects how much of the data lies further than two standard deviations away, which Part 5 makes precise with Chebyshev's rule and the empirical rule.

You have already seen values that clear the bar. Binh's one-minute wait is exactly two deviations below, the borderline case the book uses. The value 72 in the last slide, at 2.4, is past it.

Values far beyond two deviations certainly exist. In a skewed set, a long tail can carry values three or four deviations out, which is one reason the book insists you graph the data before judging.

OpenStax Introductory Statistics 2e, §2.7 Measures of the Spread of the Data §2.7, p. 108 — the rule of thumb for far from the mean

18. Computing it: deviations, squares, a root

Section

Part 3

19. Deviations: each value's signed distance from the mean

Concept

To build the standard deviation, start where the book does. For each value x, the difference x minus the mean is called its deviation. A data set has as many deviations as it has values.

\[ \text{deviation} = x - \bar{x} \;\text{(sample)} \qquad x - \mu \;\text{(population)} \]

Figure (svg): Line A's five waits as dots on a number line from 0 to 10, with a dashed line at the mean, 5. Arrows run from the mean to each wait: orange arrows left to 2 and 4, labelled minus 3 and minus 1; blue arrows right to 6 and 8, labelled plus 1 and plus 3. The wait at 5 has deviation 0. A note reads below, minus 4; above, plus 4.

The arrows on the left total 4 minutes of shortfall, the arrows on the right 4 minutes of overshoot. They cancel exactly.

For line A, with mean 5, the deviations are minus 3, minus 1, 0, plus 1 and plus 3. A positive deviation means the value is above the mean; a negative one, below it. The size says how far.

The obvious next move is to average the deviations and call that the spread. Add them up first and see what happens:

\[ (-3) + (-1) + 0 + 1 + 3 = 0 \]

Zero. Line A obviously has spread, yet its deviations add to nothing. The shortfall below the mean is exactly matched by the overshoot above it.

OpenStax Introductory Statistics 2e, §2.7 Measures of the Spread of the Data §2.7, pp. 108 and 111 — deviations, and why they sum to zero

20. Why the deviations always add to zero

Intuition

Line A's zero is not a coincidence. Line B's deviations, minus 5, minus 3, minus 1, plus 3 and plus 6, also add to zero. The book states it as a fact: if you add the deviations, the sum is always zero. Here is why, in three lines.

\[ \sum (x - \bar{x}) = \sum x - \sum \bar{x} \]

Split the sum: adding up the differences is the same as adding the values and then taking away n copies of the mean.

\[ \sum (x - \bar{x}) = \sum x - n\bar{x} \]

But the mean was defined as the total shared equally among n values, so n times the mean is the total itself:

\[ n\bar{x} = n \cdot \frac{\sum x}{n} = \sum x \quad\Rightarrow\quad \sum (x - \bar{x}) = 0 \]

So the mean is exactly the balance point Section 2.6 described. The deviations on one side always cancel those on the other, for every data set, which means the plain average of the deviations is always zero and can never measure spread.

A useful side effect: the zero total is a free check. If the deviations in your table do not add to zero, up to rounding, the mean or a subtraction is wrong.

OpenStax Introductory Statistics 2e, §2.7 Measures of the Spread of the Data §2.7, p. 111 — If you add the deviations, the sum is always zero

21. Squaring: deviations that cannot cancel

Concept

The cancelling came from the signs. Squaring removes them: a negative number squared is positive, so every squared deviation is zero or more, and their total can only grow as the data spread out.

Figure (svg): Each deviation drawn as a square whose side is the deviation's size. Line A's squares have sides 3, 1, 0, 1 and 3, with areas 9, 1, 0, 1 and 9, totalling 20. Line B's squares have sides 5, 3, 1, 3 and 6, with areas 25, 9, 1, 9 and 36, totalling 80.

Squares have no sign, so nothing cancels. Notice how the one 6-minute deviation in line B owns almost half of its total area.

Draw each squared deviation as an actual square on the deviation's length and the idea becomes visible. Line A's squares total 20 square minutes; line B's total 80. Nothing cancels now, and the more spread-out line has four times the area.

\[ A: 9 + 1 + 0 + 1 + 9 = 20, \qquad B: 25 + 9 + 1 + 9 + 36 = 80 \]

Squaring also weighs far values heavily. Line B's single 6-minute deviation contributes 36, nearly half of B's total, even though it is only one of five deviations. That is why the book warns that outliers can make the standard deviation very large.

Why squares and not simply the sizes, the absolute values? Both stop the cancelling. The squares win because they fit the mean exactly, as the next slide shows, and because the squared measure has the clean algebra the rest of statistics is built on.

OpenStax Introductory Statistics 2e, §2.7 Measures of the Spread of the Data §2.7, p. 111 — By squaring the deviations, you make them positive numbers

22. The mean is the centre that makes the squares smallest

Concept

Measure line A's waits from some centre other than the mean, say 4 minutes, and total the squares again. The distances become minus 2, 0, 1, 2 and 4:

\[ (-2)^2 + 0^2 + 1^2 + 2^2 + 4^2 = 4 + 0 + 1 + 4 + 16 = 25 \]

That is 25, more than the 20 you get from the mean. From 6 minutes the total is 25 again. Do this for every possible centre and plot the result:

Figure (svg): A parabola showing the total of the squared distances of line A's waits from a centre c, for c from 2 to 8 minutes. Its lowest point is at c equals 5, the mean, with a total of 20. At c equals 4 and c equals 6 the total is 25.

Measure from any centre other than the mean and the squares add to more. The mean is the centre that makes the squared distances as small as they can be.

The total of squares is a parabola with its lowest point exactly at the mean. Of all the places you could stand to measure distances, the mean makes the squared distances smallest. That is the real reason the squared deviations and the mean belong together.

Keep this fact; it explains the n minus 1 in Part 4. A sample's own mean sits exactly where that sample's squares are smallest, so squares measured from the sample mean always come out a little smaller than squares measured from the true population mean would.

23. Line A as a whole population: σ = 2

Worked example

Treat line A's five waits as the entire population of interest, say every customer in one quiet hour. Compute the population standard deviation step by step.

The recipe is the one the book describes: deviations, square them, average the squares to get the variance, then take the square root. Watch the units at every line, because the last step exists to repair them.

Find the mean

Why: the total shared among N = 5 values.

\[ \mu = \frac{2 + 4 + 5 + 6 + 8}{5} = \frac{25}{5} = 5 \text{ min} \]

Find each deviation

Why: each wait minus 5.

\[ -3, \; -1, \; 0, \; 1, \; 3 \]

Square each deviation

Why: the signs disappear.

\[ 9, \; 1, \; 0, \; 1, \; 9 \]

Total the squares

Why: add the five squares.

\[ \sum (x - \mu)^2 = 9 + 1 + 0 + 1 + 9 = 20 \]

Average the squares: the variance

Why: a population divides by N.

\[ \sigma^2 = \frac{20}{5} = 4 \text{ min}^2 \]

Take the square root

Why: back to minutes.

\[ \sigma = \sqrt{4} = 2 \text{ min} \]

In context: in that hour, a typical wait at line A strays about two minutes from the five-minute average, exactly supermarket A's standard deviation. Notice that 2 is not any single deviation; it is a kind of root-average of all five.

Check by undoing both steps

Why: square the answer and multiply by N to recover the total of squares.

\[ 2^2 \times 5 = 4 \times 5 = 20 \checkmark \]

OpenStax Introductory Statistics 2e, §2.7 Measures of the Spread of the Data §2.7, pp. 108-109 — the population variance and standard deviation

24. Line B as a whole population: σ = 4

Worked example

Now the more variable line, again treated as a whole population. Same recipe, bigger deviations.

Before computing, predict: line B's squares total four times line A's, and the root of four is two, so σ should be twice line A's 2 minutes.

Find the mean

Why: the total over N = 5.

\[ \mu = \frac{0 + 2 + 4 + 8 + 11}{5} = \frac{25}{5} = 5 \text{ min} \]

Find each deviation

Why: each wait minus 5.

\[ -5, \; -3, \; -1, \; 3, \; 6 \]

Square each deviation

Why: a 6 becomes 36.

\[ 25, \; 9, \; 1, \; 9, \; 36 \]

Total the squares

Why: add the five squares.

\[ \sum (x - \mu)^2 = 25 + 9 + 1 + 9 + 36 = 80 \]

Divide by N for the variance

Why: five values in the population.

\[ \sigma^2 = \frac{80}{5} = 16 \text{ min}^2 \]

Take the square root

Why: back to minutes.

\[ \sigma = \sqrt{16} = 4 \text{ min} \]

In context: a typical wait at line B strays about four minutes from the average, twice as far as at line A, matching supermarket B. The prediction held: four times the squared area gives twice the standard deviation, because of the square root.

Check the deviations first

Why: they must total zero, or the mean is wrong.

\[ -5 - 3 - 1 + 3 + 6 = 0 \checkmark, \qquad 4^2 \times 5 = 80 \checkmark \]

25. Reading the standard deviation formulas

Notation

The book gives four formulas: sample and population, each in a plain form and a frequency form. Here is the sample's frequency form, the one you will use most, decoded.

Annotate

On: \( s = \sqrt{\frac{\sum f(x - \bar{x})^2}{n - 1}} \)

  • Each value's deviation from the sample mean: how far it sits, with a sign.
  • Squared, so that no deviation can cancel another. Units become squared too.
  • The frequency: a value that appears three times contributes its squared deviation three times.
  • The sample's denominator, one less than the number of values. Part 4 explains why.
  • The root undoes the squaring of the units, returning the answer to the data's own units.

The population versions differ in exactly two places: the mean is μ instead of x-bar, and the denominator is N, the number of items in the population, instead of n minus 1.

\[ \sigma = \sqrt{\frac{\sum f(x - \mu)^2}{N}} \]

Everything under the root, before the square root is taken, is the variance: s squared for a sample, sigma squared for a population. You can think of the standard deviation, as the book suggests, as a special average of the deviations.

OpenStax Introductory Statistics 2e, §2.7 Measures of the Spread of the Data §2.7, p. 109 — Formulas for the sample and population standard deviation

26. Variance in square minutes; the root brings back minutes

Concept

The variance is the average squared deviation, and it is a perfectly good number, but it lives in squared units. Line A's variance is 4 square minutes, which is not a length of time anyone waited.

\[ \sigma^2 = 4 \text{ min}^2 \quad\longrightarrow\quad \sigma = \sqrt{4 \text{ min}^2} = 2 \text{ min} \]

The book's sentence is exact: the variance is a squared measure and does not have the same units as the data; taking the square root solves the problem. The standard deviation measures the spread in the same units as the data, so it can sit next to the mean and be added to it.

That is what made Rosa's line work. Five minutes plus one standard deviation of two minutes is seven minutes. Five minutes plus a variance of four square minutes is meaningless, like adding a length to an area.

The squaring also explains why doubling every deviation quadruples the variance but only doubles the standard deviation. Line B's deviations are roughly twice line A's; its variance is four times as large, 16 against 4, while its standard deviation is twice, 4 against 2.

OpenStax Introductory Statistics 2e, §2.7 Measures of the Spread of the Data §2.7, p. 111 — The variance is a squared measure

27. Trap: squaring the total instead of totalling the squares

Trap

The trap

Asked for line B's total of squared deviations, a first attempt adds the deviations, then squares:

\[ (-5 - 3 - 1 + 3 + 6)^2 \]

\[ = 0^2 = 0 \]

Wrong. It brings the cancelling straight back.

The fix

Square each deviation first, then add. The order matters because the signs must be gone before anything is added.

\[ 25 + 9 + 1 + 9 + 36 = 80 \]

The symbol says the same thing: in the sum of (x minus the mean) squared, the square sits inside the sum, so it acts on each term. A total of zero, or a standard deviation of zero for data that clearly vary, is the alarm bell.

OpenStax Introductory Statistics 2e, §2.7 Measures of the Spread of the Data §2.7, p. 111 — squaring the deviations

28. Order the steps of the computation

Ranking

Put the computation of a sample standard deviation in order. Each step needs the result of the one before it, so the order is not a matter of taste.

Put in order

Arrange the steps for computing s from a sample, first to last.

  1. Find the sample mean
  2. Subtract the mean from each value to get the deviations
  3. Square each deviation, times its frequency
  4. Add the squared deviations
  5. Divide by n − 1 to get the variance
  6. Take the square root

Why: The deviations need the mean, the squares need the deviations, the total needs the squares, the variance divides the total, and the root of the variance is the standard deviation. Swapping the squaring and the adding gives zero.

Two places in this chain attract mistakes. Adding before squaring gives zero, as the trap showed. Forgetting the last step leaves you with the variance, a number in squared units that is usually far too big or too small to be a sensible spread.

A quick sense check at the end: the standard deviation should be no larger than the biggest deviation in the table, and is usually somewhat smaller.

29. Example 2.32: the ages, their mean and the deviation table

Worked example

A fifth-grade teacher records the ages of a sample of n = 20 students, rounded to the nearest half year: 9; 9.5 twice; 10 four times; 10.5 four times; 11 six times; 11.5 three times. Find the mean and build the deviation table.

Twenty ages but only six different values, so work with frequencies: each value times how often it occurs. That is what the f in the formula is for, and it saves writing the same row many times.

Total the ages using frequencies

Why: each value times its count.

\[ \sum fx = 9(1) + 9.5(2) + 10(4) + 10.5(4) + 11(6) + 11.5(3) \]

Evaluate the products and add

Why: 9 + 19 + 40 + 42 + 66 + 34.5.

\[ \sum fx = 210.5 \]

Divide by n

Why: twenty students.

\[ \bar{x} = \frac{210.5}{20} = 10.525 \text{ years} \]

First row: deviation of 9

Why: subtract the unrounded mean.

\[ 9 - 10.525 = -1.525 \]

Square it and weight by f

Why: 9 occurs once.

\[ 1 \times (-1.525)^2 = 1 \times 2.325625 = 2.325625 \]

xfx − x̄(x − x̄)²f(x − x̄)²
91−1.5252.3256252.325625
9.52−1.0251.0506252.101250
104−0.5250.2756251.102500
10.54−0.0250.0006250.002500
1160.4750.2256251.353750
11.530.9750.9506252.851875

The book rounds the mean to 10.53 when it reports it but keeps 10.525 in the table, and so should you: intermediate results are not rounded. Notice also that the book's text later calls the last two deviations 0.97 and 0.47; those are 0.975 and 0.475 cut short, not rounded.

Check the deviations total zero

Why: weight each by its frequency.

\[ -1.525 - 2.05 - 2.1 - 0.1 + 2.85 + 2.925 = 0 \checkmark \]

OpenStax Introductory Statistics 2e, §2.7 Measures of the Spread of the Data §2.7, pp. 109-110 — Example 2.32 and Table 2.30

30. Example 2.32: from 9.7375 to s = 0.72 years

Worked example

Finish the computation from the table on the last slide. These are a sample of fifth graders, so the denominator is n minus 1.

Three moves remain: total the last column, divide by one less than the number of students, and take the root to get back to years.

Total the last column

Why: the sum of f times the squared deviation.

\[ 2.325625 + 2.10125 + 1.1025 + 0.0025 + 1.35375 + 2.851875 = 9.7375 \]

Find the denominator

Why: a sample, so one less than n.

\[ n - 1 = 20 - 1 = 19 \]

Divide: the sample variance

Why: the average squared deviation, sample style.

\[ s^2 = \frac{9.7375}{19} = 0.5125 \text{ years}^2 \]

Take the square root

Why: back to years.

\[ s = \sqrt{0.5125} = 0.7159 \approx 0.72 \text{ years} \]

Figure (svg): A dot plot of the twenty fifth graders' ages, from 9 to 11.5 years in half-year steps, with stacks of 1, 2, 4, 4, 6 and 3 dots. A dashed line marks the mean, 10.525. A shaded band runs one standard deviation either side, from 9.81 to 11.24, and holds 14 of the 20 ages.

Fourteen of the twenty ages sit within one standard deviation of the mean. The single 9-year-old sits more than two deviations below it.

In context: a typical fifth grader's age differs from the class average of about 10.5 years by roughly 0.72 years, about eight and a half months. The dot plot shows fourteen of the twenty ages inside one standard deviation of the mean.

Check against a calculator's 1-Var Stats

Why: the book's TI-83 run gives x-bar = 10.525 and Sx = 0.715891; squaring and multiplying back also recovers the table total.

\[ 0.715891^2 \times 19 = 0.5125 \times 19 = 9.7375 \checkmark \]

OpenStax Introductory Statistics 2e, §2.7 Measures of the Spread of the Data §2.7, p. 110 — Example 2.32(a)

31. Example 2.32(b)-(d): ages a set number of deviations out

Worked example

With x-bar = 10.525 and s = 0.7159, find the value one standard deviation above the mean, two below it, and 1.5 either side. Every part uses the same formula, value equals mean plus the number of deviations times s.

Here the number of deviations is given and the value is unknown, the reverse of Rosa and Binh. Multiply first, then add or subtract.

One deviation above

Why: k = +1.

\[ 10.525 + (1)(0.7159) = 11.24 \text{ years} \]

Two deviations below

Why: k = −2: the product is negative.

\[ 10.525 + (-2)(0.7159) = 10.525 - 1.4318 = 9.09 \text{ years} \]

1.5 deviations below

Why: k = −1.5.

\[ 10.525 - (1.5)(0.7159) = 10.525 - 1.0738 = 9.45 \text{ years} \]

1.5 deviations above

Why: k = +1.5.

\[ 10.525 + 1.0738 = 11.60 \text{ years} \]

Figure (svg): A number line of ages from 8.5 to 12 years. Marks show the mean 10.525, one standard deviation above at 11.24, two below at 9.09, and one and a half either side at 9.45 and 11.60.

Every mark is the mean plus a whole or fractional number of standard deviations. The two 1.5 marks sit symmetrically around the mean.

The book prints 11.25 for part (b) and 11.61 for the upper value in part (d). It worked from the rounded 10.53 and 0.72; from the unrounded values the answers are 11.24 and 11.60. The difference is small, but it is exactly why you keep full precision until the end.

In context: a fifth grader older than about 11.6 years or younger than about 9.45 would be more than one and a half standard deviations from the class mean. Only the single 9-year-old falls outside that range, and he is past two deviations below.

Check the symmetry

Why: the two 1.5 marks must average to the mean.

\[ \frac{9.45 + 11.60}{2} = \frac{21.05}{2} = 10.525 \checkmark \]

OpenStax Introductory Statistics 2e, §2.7 Measures of the Spread of the Data §2.7, pp. 110-111 — Example 2.32(b)-(d)

32. Try It 2.32: the baseball team's ages

Worked example

The ages of a baseball team's 25 players are 21, 21, 22, 23, 24, 24, 25, 25, 28, 29, 29, 31, 32, 33, 33, 34, 35, 36, 36, 36, 36, 38, 38, 38, 40. Find the mean and standard deviation, then the age two standard deviations above the mean.

The book says to use a calculator; here is the arithmetic behind its output so you can see where each number comes from. Treat the roster as a sample of ballplayers, so divide by n minus 1.

Total and divide for the mean

Why: 25 players.

\[ \bar{x} = \frac{767}{25} = 30.68 \text{ years} \]

Total the squared deviations

Why: each age minus 30.68, squared, summed over all 25.

\[ \sum (x - 30.68)^2 = 891.44 \]

Divide by n minus 1

Why: 24.

\[ s^2 = \frac{891.44}{24} = 37.143 \]

Take the root

Why: back to years.

\[ s = \sqrt{37.143} = 6.09 \text{ years} \]

Two deviations above the mean

Why: k = +2.

\[ 30.68 + (2)(6.0945) = 30.68 + 12.19 = 42.87 \text{ years} \]

Figure (svg): A dot plot of the 25 baseball players' ages from 21 to 40, with a stack of four at 36 and three at 38. A dashed line marks the mean, 30.68. A second dashed line at 42.87, two standard deviations above the mean, lies beyond the oldest player at 40.

Two standard deviations above the mean is older than anyone on the team. By the rule of thumb, no player's age is unusually high.

In context: the players' ages stray about six years from the team average of about 31, and an age of almost 43 would be needed to sit two deviations above it. The oldest player, at 40, is only about 1.5 deviations above, so no player is unusually old for this team.

Check by counting hops back

Why: the distance from the mean over s must be 2.

\[ \frac{42.87 - 30.68}{6.0945} = \frac{12.19}{6.0945} = 2.00 \checkmark \]

OpenStax Introductory Statistics 2e, §2.7 Measures of the Spread of the Data §2.7, p. 111 — Try It 2.32

33. Try It 2.33: pet food in the stores

Fill the middle

The numbers of different types of pet food carried by 29 stores are tallied below, with mean 269 over 29, about 9.28. The table's last column has already been filled in. Finish the computation.

xfx − x̄f(x − x̄)²
64−3.27642.925
75−2.27625.898
81−1.2761.628
94−0.2760.304
1050.7242.622
1141.72411.891
1262.72444.526
total29129.793

Fill in the blanks

s^2 = \frac284.635} = 4.635, \qquad s = \sqrt2.15} = ___

Why: These 29 stores are a sample, so divide the total of squared deviations by 28, giving a variance of about 4.635. Its square root is about 2.15, which the book's one-decimal answer reports as 2.2.

In context: stores carry about 9.3 types of pet food on average, give or take about 2.2 types. Look at where the squares come from: the 6s and the 12s, at the two ends, supply 87 of the 130, two thirds of the total.

OpenStax Introductory Statistics 2e, §2.7 Measures of the Spread of the Data §2.7, p. 113 — Try It 2.33

34. Find the error: a standard deviation of 20 minutes

Error analysis

A student treats line B's five waits as a sample and reports the standard deviation like this. The number is wildly wrong, and the error is a missing step.

Annotate

On: \( s = \frac{25 + 9 + 1 + 9 + 36}{5 - 1} = \frac{80}{4} = 20 \text{ min} \)

  • Correct: the squared deviations of 0, 2, 4, 8, 11 from 5 do add to 80.
  • Correct: a sample of five divides by four.
  • This is the variance, s², in square minutes. The square root was never taken.
  • s = √20 ≈ 4.47 min. A typical deviation of 20 minutes is impossible when no wait is more than 6 minutes from the mean.

\[ s = \sqrt{20} \approx 4.47 \text{ min} \]

The sense check that catches it: the standard deviation is a typical distance from the mean, so it can never exceed the largest distance in the data. Line B's largest deviation is 6 minutes; an answer of 20 minutes fails at a glance.

Notice the correct answer, 4.47, is a little more than the population value of 4 from Part 3. Same data, different denominator: Part 4 is about exactly that difference.

35. Sample against population

Section

Part 4

36. Two symbols, two denominators

Concept

The procedure depends on whether your numbers are the entire population or a sample from it. The calculations are similar but not identical, so the symbol records which one you did.

samplepopulation
meanx̄μ
variances² = Σ f(x − x̄)² ÷ (n − 1)σ² = Σ f(x − μ)² ÷ N
standard deviationsσ (lower-case sigma)
denominatorn − 1, one less than the sample sizeN, the population size

A census, where you have every member of the population, divides by N. A sample, where your data stand in for a larger group you did not measure, divides by n minus 1. If the sample has the same characteristics as the population, s should be a good estimate of σ.

The decision is about the question, not about the numbers. The same 25 baseball ages divide by N if you only want to describe that team, and by n minus 1 if you want to learn about ballplayers in general.

OpenStax Introductory Statistics 2e, §2.7 Measures of the Spread of the Data §2.7, pp. 108-109 — the sample and population standard deviations

37. A tiny population tests dividing by n

Intuition

The book says only that dividing by n minus 1 gives a better estimate of the population variance, for theoretical reasons. You can see it happen with a population small enough to list every sample. Take three waits, 2, 4 and 6 minutes, as the whole population.

\[ \mu = 4, \qquad \sigma^2 = \frac{(-2)^2 + 0^2 + 2^2}{3} = \frac{8}{3} \approx 2.67 \]

Now draw every possible sample of two, with replacement: nine samples, from 2 and 2 up to 6 and 6. For each one, measure the squares from that sample's own mean, then divide by n, which is 2, and separately by n minus 1, which is 1.

Figure (svg): A bar chart of nine samples of two drawn with replacement from the population 2, 4, 6. For each sample there is a blue bar, its squared deviations divided by n equals 2, and an orange bar, the same total divided by n minus 1 equals 1. The blue bars are 0, 1, 4, 1, 0, 1, 4, 1, 0 and average 1.33. The orange bars are 0, 2, 8, 2, 0, 2, 8, 2, 0 and average 2.67. A yellow line marks the population variance, 2.67.

Divided by n, the nine estimates average only half the true variance. Divided by n minus 1, they average exactly the true variance.

The blue estimates, dividing by n, average 12 over 9, about 1.33: only half the true 2.67. They are systematically too small. The orange estimates, dividing by n minus 1, average 24 over 9, which is exactly 8 over 3.

\[ \div n: \quad \frac{0 + 1 + 4 + 1 + 0 + 1 + 4 + 1 + 0}{9} = \frac{12}{9} = \frac{4}{3} \]

\[ \div (n - 1): \quad \frac{0 + 2 + 8 + 2 + 0 + 2 + 8 + 2 + 0}{9} = \frac{24}{9} = \frac{8}{3} \]

Drawing with replacement matters here: each draw is independent of the last, which is how a random sample from a large population behaves. Each of the nine samples is equally likely, so the plain average of the nine estimates is the long-run average of the method.

No single sample hits 2.67; a sample of two is a crude instrument. But on average the n minus 1 version is right and the n version is too low. That is what the book's better estimate means.

38. Why the sample's own mean makes the squares too small

Concept

Where does the shortfall come from? From the fact on the parabola slide: a set of values has the smallest total of squares when measured from its own mean. A sample is always measured from its own mean, not from μ, which you do not know.

Take the sample 2 and 4 from the tiny population. Measured from its own mean, 3, the squares total 2. Measured from the true mean, 4, they total 4:

\[ (2 - 3)^2 + (4 - 3)^2 = 2, \qquad (2 - 4)^2 + (4 - 4)^2 = 4 \]

The sample mean sits right in the middle of its own data, so the sample looks less spread out than the population really is. Dividing by a slightly smaller number, n minus 1 instead of n, scales the result back up by exactly enough to correct that on average.

There is also a counting way to say it. Once you know the sample mean, the deviations must total zero, so only n minus 1 of them are free; the last one is forced. The squares carry information from n minus 1 independent deviations, not n.

\[ -3 + (-1) + 0 + 1 + d = 0 \quad\Rightarrow\quad d = 3 \]

OpenStax Introductory Statistics 2e, §2.7 Measures of the Spread of the Data §2.7, p. 111 — dividing by n − 1 gives a better estimate of the population variance

39. Line A as a sample: s = 2.24, not 2

Worked example

Suppose line A's five waits are not the whole hour but a sample of five customers, used to learn about all customers at that line. Compute s and compare it with the population value of 2 minutes.

Only the denominator changes. Everything up to the total of squares is identical, so reuse it rather than recompute.

The total of squared deviations

Why: from Part 3: the mean is 5 either way.

\[ \sum (x - \bar{x})^2 = 20 \]

Divide by n minus 1

Why: a sample of five.

\[ s^2 = \frac{20}{5 - 1} = \frac{20}{4} = 5 \text{ min}^2 \]

Take the square root

Why: back to minutes.

\[ s = \sqrt{5} \approx 2.24 \text{ min} \]

Compare with the population version

Why: the same data with N in the denominator.

\[ \sigma = \sqrt{20/5} = 2 \text{ min} \]

In context: as an estimate for all of line A's customers, the typical wait strays about 2.24 minutes from the average. The sample version is larger because it corrects for the sample mean fitting its own five values too snugly.

Check the ratio

Why: s over σ must be the square root of n over n minus 1.

\[ \frac{2.236}{2} = 1.118 = \sqrt{5/4} \checkmark \]

40. N or n minus 1?

Discrimination

For each situation, decide whether the data are the entire group the question is about, or a sample standing in for a larger group. The numbers alone never settle it; the question does.

Sort into buckets

Sort each situation by the denominator its standard deviation should use.

population: divide by N
the heights of all 12 players on a team, to describe that team; every employee's salary at a 40-person firm, for the firm's own report; the final scores of every student in your class, to describe your class
sample: divide by n − 1
the heights of 12 players, to learn about all college basketball players; 20 fifth graders' ages, to describe fifth graders in general; 50 light bulbs tested from a production run of 10,000
N
The data are every member of the group the question asks about, so there is nothing larger to estimate.
n
The data are a part of a larger group, and the standard deviation is meant to estimate that group's spread.

The two basketball items use the very same twelve heights, yet they belong in different buckets. Sample or population is a property of the question you are asking, not of the numbers on the page.

When a problem does not say, look for the word sample. The book's examples nearly always say it: Example 2.32 is a sample of fifth graders, so it divides by 19.

41. How much the correction matters as n grows

Picture it

Dividing by n minus 1 instead of n always makes s a little larger than the divide-by-n version. How much larger depends only on the sample size: the ratio of the two is the square root of n over n minus 1.

\[ \frac{s}{\sqrt{\sum (x - \bar{x})^2 / n}} = \sqrt{\frac{n}{n - 1}} \]

Figure (svg): A falling curve of the square root of n over n minus 1, the factor by which s exceeds the divide-by-n version, for n from 2 to 40. It is 1.41 at n equals 2, 1.12 at n equals 5, 1.05 at n equals 10, 1.03 at n equals 20 and 1.02 at n equals 30.

With two values the correction raises s by 41 percent. By twenty values it is under 3 percent, and it keeps fading as n grows.

For line A's five waits the factor is 1.118, which is why s came out at 2.24 instead of 2. For the twenty fifth graders it is only 1.026: their s of 0.7159 against 0.6978 dividing by twenty.

So the correction matters most when samples are small, which is exactly when a sample mean is most likely to sit far from μ and make the squares too small. With hundreds of values the two versions agree to the second decimal place, but the reason for choosing correctly does not go away.

42. Trap: the calculator's population key

Trap

The trap

A TI-83 run on Example 2.32's twenty sampled ages shows two standard deviations. A first attempt reports:

\[ \sigma_x = 0.6978 \]

\[ \text{so } s \approx 0.70 \]

Wrong. That is the divide-by-N value.

The fix

These ages are a sample, so the right output is Sx, the line that divides by n minus 1.

\[ S_x = 0.7159 \approx 0.72 \]

\[ \sqrt{9.7375/19} = 0.7159 \]

The calculator cannot know whether your list is a sample or a population, so it shows both and you must choose. The book's note says it directly: select σx or sx from the summary statistics according to your data.

OpenStax Introductory Statistics 2e, §2.7 Measures of the Spread of the Data §2.7, pp. 109-110 — the calculator note and Example 2.32(a)

43. Symbol to what it names

Matching

This section uses eight closely related symbols. Match each to its meaning; the pairs differ in exactly one respect, sample against population or squared against not.

Match the pairs

  • s. s
  • sig. σ
  • s2. s²
  • sig2. σ²
  • xb. x̄
  • mu. μ
  • n1. n − 1
  • N. N
  • a. sample standard deviation
  • b. population standard deviation
  • c. sample variance
  • d. population variance
  • e. sample mean
  • f. population mean
  • g. denominator for a sample's variance
  • h. denominator for a population's variance

Why: Latin letters belong to samples and Greek letters to populations. A squared symbol is a variance, in squared units; its square root is the standard deviation, in the data's units. A sample's variance divides by n minus 1, a population's by N.

Two pairs are easy to swap under pressure: the sample variance and the sample standard deviation differ only by a square, and the two means differ only in where the data came from.

A rule of thumb that rarely fails in this course: Latin letters are statistics, computed from a sample; Greek letters are parameters, describing a population. Chapter 1 drew that same line between statistic and parameter.

When you write up an answer, the symbol is part of the answer. Writing σ for a value you computed from a sample tells your reader you divided by N and had the whole population, which is usually false.

44. Sampling variability: a preview of the standard error

Concept

The book closes its formulas with a look ahead. A statistic such as the sample mean changes from one sample to the next; how much it varies is called the sampling variability of the statistic, and it is measured by a standard error.

\[ \text{standard error of the mean} = \frac{\sigma}{\sqrt{n}} \]

Read it as a standard deviation of sample means. The tiny population showed this already: the nine samples of two had means of 2, 3, 4, 3, 4, 5, 4, 5 and 6, scattered around the true 4. Larger samples would scatter less, and the square root of n in the denominator says by how much.

You will not compute this until chapter 7, the Central Limit Theorem. For now the point is a distinction worth holding on to: σ describes how individual values spread, and the standard error describes how a sample mean would spread if you sampled again and again.

The standard error is always smaller than σ when n is more than one, because averages are steadier than single values. Ten waits averaged together vary far less than a single wait does.

OpenStax Introductory Statistics 2e, §2.7 Measures of the Spread of the Data §2.7, p. 109 — Sampling Variability of a Statistic

45. Comparing values and bounding the data

Section

Part 5

46. Values from different data sets: compare in standard deviations

Concept

Comparing two values directly can mislead when they come from data sets with different means and standard deviations. A 77 on one scale and a 2.85 on another cannot be compared at all as raw numbers.

The book's method: for each value, count how many standard deviations it sits from its own mean, then compare those counts. Solving the value formula for the count gives the z-score.

\[ z = \frac{\text{value} - \text{mean}}{\text{standard deviation}} \]

\[ \text{sample: } x = \bar{x} + zs, \qquad \text{population: } x = \mu + z\sigma \]

A z-score has no units: minutes divided by minutes, points divided by points. That is exactly what lets it compare across data sets measured in different units or on different scales. Its sign says above or below the mean; its size says how far, in that set's own typical distances.

One caution travels with every z comparison. Higher z means further above the mean, which is better only when higher values are better. For a race time, the more negative z wins.

OpenStax Introductory Statistics 2e, §2.7 Measures of the Spread of the Data §2.7, p. 116 — Comparing Values from Different Data Sets

47. Example 2.35: John and Ali, whose GPA is higher for his school?

Worked example

John's GPA is 2.85 at a school with mean 3.0 and standard deviation 0.7. Ali's GPA is 77 at a school with mean 80 and standard deviation 10. Which student has the higher GPA when compared to his school?

The two scales are incompatible, so do not compare 2.85 with 77, or even the raw shortfalls 0.15 and 3. Convert each to standard deviations from his own school's mean, paying attention to signs.

John: distance from his school's mean

Why: his GPA minus 3.0.

\[ 2.85 - 3.0 = -0.15 \]

John: divide by his school's SD

Why: 0.7 GPA points per deviation.

\[ z_J = \frac{-0.15}{0.7} = -0.214 \approx -0.21 \]

Ali: distance from his school's mean

Why: his GPA minus 80.

\[ 77 - 80 = -3 \]

Ali: divide by his school's SD

Why: 10 points per deviation.

\[ z_A = \frac{-3}{10} = -0.3 \]

Figure (svg): Two rulers, one per school, each scaled so that one standard deviation has the same length. John's ruler runs from 1.6 to 4.4 in steps of 0.7 with his GPA 2.85 just left of the mean 3.0. Ali's ruler runs from 60 to 100 in steps of 10 with his 77 just left of the mean 80. Dashed lines drop to a shared z scale, landing at minus 0.21 for John and minus 0.30 for Ali.

Rescale each school's ruler so a standard deviation is the same length, and the two GPAs can be read on one scale. John sits a little closer to his mean.

In context: both students are a little below their school's average, but John is only 0.21 deviations below while Ali is 0.30 below. Since a higher GPA is better, John's z of minus 0.21 is the higher one: John has the better GPA relative to his school.

Check by rebuilding each GPA

Why: mean plus z times the standard deviation.

\[ 3.0 + (-0.214)(0.7) = 2.85 \checkmark, \qquad 80 + (-0.3)(10) = 77 \checkmark \]

OpenStax Introductory Statistics 2e, §2.7 Measures of the Spread of the Data §2.7, p. 117 — Example 2.35 and Table 2.35

48. Try It 2.35: Angie and Beth, who swam faster for her team?

Worked example

In the 50-metre freestyle, Angie swam 26.2 seconds on a team with mean 27.2 and standard deviation 0.8. Beth swam 27.3 seconds on a team with mean 30.1 and standard deviation 1.4. Which swimmer had the faster time compared to her team?

Beth's raw time is slower, but her team is slower too. Standardise each time against its own team. This time a lower value is better, so watch which way the comparison goes.

Angie: distance from her team's mean

Why: 26.2 minus 27.2.

\[ 26.2 - 27.2 = -1.0 \text{ s} \]

Angie: divide by her team's SD

Why: 0.8 seconds per deviation.

\[ z_{\text{Angie}} = \frac{-1.0}{0.8} = -1.25 \]

Beth: distance from her team's mean

Why: 27.3 minus 30.1.

\[ 27.3 - 30.1 = -2.8 \text{ s} \]

Beth: divide by her team's SD

Why: 1.4 seconds per deviation.

\[ z_{\text{Beth}} = \frac{-2.8}{1.4} = -2.00 \]

Figure (svg): A z scale from minus 3 to plus 1. Angie's dot is at minus 1.25 and Beth's at minus 2.00; a dashed line marks the team mean at 0. A green arrow pointing left is labelled faster than her team.

For a swimming time, left is good: a more negative z means further below the team's average time. Beth is further left.

In context: Angie swam 1.25 standard deviations faster than her team's average, Beth a full 2 standard deviations faster than hers. For times, more negative is better, so Beth had the faster swim relative to her team, even though Angie touched the wall first.

Check by rebuilding Beth's time

Why: her team's mean plus z times its SD.

\[ 30.1 + (-2)(1.4) = 30.1 - 2.8 = 27.3 \checkmark \]

OpenStax Introductory Statistics 2e, §2.7 Measures of the Spread of the Data §2.7, p. 117 — Try It 2.35 and Table 2.36

49. Higher z, better result?

Prediction

Three people compare results using z-scores. Before choosing, ask of each measurement whether a larger raw value is good or bad.

Predict first

In which comparison does the person with the lower z-score have the better result?

  • Two golfers' scores, each compared with their own tour's average
  • Two students' exam marks, each compared with their own class
  • Two salespeople's monthly sales, each compared with their own region
  • Two high jumpers' heights, each compared with their own team

Correct: Two golfers' scores, each compared with their own tour's average

Why: In golf, fewer strokes is better, so a score further below the average, a more negative z, is the better round. For exam marks, sales and jump heights, larger values are better, so the higher z wins.

The z-score itself is neutral: it only measures where a value sits. Better or worse comes from the context, which is why the book's solution to Example 2.35 adds the sentence for GPA, higher values are better, before drawing its conclusion.

Make that sentence a habit. Write down which direction is good before you compare, and the swimming and golf versions stop being traps.

50. Chebyshev's rule: a guarantee for any data set

Concept

How much of a data set can lie far from the mean? The book gives a guarantee that holds for any data, whatever its shape. It is known as Chebyshev's rule.

All three come from one formula: within k standard deviations of the mean lies at least one minus one over k squared of the data.

\[ 1 - \frac{1}{2^2} = 0.75, \quad 1 - \frac{1}{3^2} = 0.889, \quad 1 - \frac{1}{4.5^2} = 0.951 \]

Figure (svg): A rising curve of 1 minus 1 over k squared, the fraction Chebyshev guarantees within k standard deviations, for k from 1 to 5. Yellow dots mark 0.75 at k equals 2, 0.89 at k equals 3 and 0.95 at k equals 4.5. Hollow pink dots mark the empirical rule's 0.68, 0.95 and 0.997 at k equals 1, 2 and 3, all above the curve.

The curve is a floor that every data set clears. Bell-shaped data clear it by a wide margin, which is why the empirical rule's numbers are larger.

Why can there not be more far-out values? Each value at least k deviations out contributes at least k squared times the variance to the total of squares, and the total is fixed. Too many far values would need more squared area than the data have, so the fraction beyond k is capped at one over k squared.

The guarantee is weak because it must cover every possible shape, including strange ones. Within one deviation it promises nothing at all, since one minus one is zero.

Use it in reverse, too. If a manager says a third of the deliveries took more than two standard deviations longer than average, Chebyshev says that cannot happen for any data set: at most one quarter can lie two or more deviations from the mean, on both sides together.

OpenStax Introductory Statistics 2e, §2.7 Measures of the Spread of the Data §2.7, p. 117 — Chebyshev's Rule

51. The empirical rule: for bell-shaped, symmetric data

Concept

When the data are bell-shaped and symmetric, the book gives much stronger statements, called the empirical rule.

Figure (svg): A bell-shaped curve with the regions within one, two and three standard deviations of the mean shaded in deepening layers. Double-headed arrows above the curve are labelled 68 percent for one standard deviation, 95 percent for two and 99.7 percent for three.

Only for a bell shape: about two thirds of the data within one deviation, nearly all within two, essentially all within three.

Compare it with Chebyshev at two deviations: at least 75 percent for any shape, about 95 percent for a bell. The bell does so much better because its values crowd near the centre and thin out quickly in both tails.

The book is firm about the condition: this rule only applies when the shape of the distribution is bell-shaped and symmetric. Chapter 6, the normal distribution, is where it comes from. Applied to skewed data it can be badly wrong, as the next slides show on real scores.

OpenStax Introductory Statistics 2e, §2.7 Measures of the Spread of the Data §2.7, pp. 117-118 — the Empirical Rule

52. Slide σ and watch the bell's bands

Tweak it

Here is a bell-shaped curve centred on 5 minutes with its standard deviation on a slider. Predict first: if σ doubles, what happens to the interval that holds the middle 68 percent?

Parameter explorer

Slide σ from 0.5 to 3. Read off where the curve has fallen to about 60 percent of its peak height; that point is one σ from the centre. How does the interval from 5 minus σ to 5 plus σ change, and what happens to the peak?

\[ \mu = 5, \quad \sigma = {s} \]

  • s — from 0.5 to 3: standard deviation σ (minutes)

What to notice: the middle 68 percent always lies between 5 minus σ and 5 plus σ, whatever σ is, so the band widens in proportion. At σ equal to 0.5 the curve is tall and narrow; at σ equal to 3 it is low and wide, and waits of 11 minutes become ordinary.

This is supermarkets A and B again in smooth form. A standard deviation of 2 puts the middle two thirds of waits between 3 and 7 minutes; a standard deviation of 4 spreads them from 1 to 9, if the waits are bell-shaped.

53. Example 2.33's scores against both rules

Worked example

Test the two rules on real data: the 31 first-exam scores from Professor Dean's pre-calculus class, with mean 73.52 and standard deviation 17.92. Count how many scores lie within one and within two standard deviations of the mean.

The plan is to turn each band into an interval of scores with the value formula, then count the dots inside it. Then compare with what each rule predicts.

The band one deviation either side

Why: mean minus and plus s.

\[ 73.52 - 17.92 = 55.59, \qquad 73.52 + 17.92 = 91.44 \]

Count the scores inside it

Why: 61 up to 90 are inside; 33 to 55 and 92 to 100 are outside.

\[ \frac{17}{31} = 0.548 \approx 55\% \]

The band two deviations either side

Why: mean minus and plus 2s.

\[ 73.52 - 35.85 = 37.67, \qquad 73.52 + 35.85 = 109.36 \]

Count the scores inside it

Why: every score except the 33.

\[ \frac{30}{31} = 0.968 \approx 97\% \]

Figure (svg): A dot plot of the 31 exam scores from 33 to 100, stacked where scores repeat. A dashed line marks the mean, 73.5. A darker band from 55.6 to 91.4, one standard deviation either side, holds 17 scores. A fainter band from 37.7 upward, two standard deviations either side, holds 30 of the 31; only the 33 lies outside it.

Only 55 percent of these scores lie within one deviation, well short of the bell's 68 percent: the scores are skewed left, not bell-shaped.

In context: Chebyshev promised at least 75 percent within two deviations, and the class delivers 97 percent, as it must. The empirical rule would predict about 68 percent within one deviation, but only 55 percent of these scores are there, because the scores are skewed left, with a cluster of 88s and 94s and a long tail down to 33.

Check the guarantee holds

Why: Chebyshev's floor at k = 2 must be cleared by every data set.

\[ \frac{30}{31} = 0.968 \ge 0.75 \checkmark \]

OpenStax Introductory Statistics 2e, §2.7 Measures of the Spread of the Data §2.7, pp. 111-112, 117 — Example 2.33's data, and the two rules

54. Choosing a measure of spread

Section

Part 6

55. Example 2.33(b): six numbers from 31 exam scores

Worked example

Using the same 31 exam scores, 33 up to 100, find the sample mean, the sample standard deviation, the median, the first and third quartiles, and the IQR, each to one decimal place. The book uses a TI-83; here is what the calculator is doing.

The scores are already in order, which the median and quartiles need. With 31 values, the median is the 16th, and each quartile is the median of the 15 values on its side.

The sample mean

Why: the total over 31.

\[ \bar{x} = \frac{2279}{31} = 73.52 \approx 73.5 \]

The sample variance

Why: the total of squared deviations over n minus 1, 30.

\[ s^2 = \frac{9637.74}{30} = 321.26 \]

The sample standard deviation

Why: the root.

\[ s = \sqrt{321.26} = 17.92 \approx 17.9 \]

The median

Why: the 16th of 31 ordered scores.

\[ M = 73 \]

The quartiles

Why: the 8th score of each half of 15.

\[ Q_1 = 61, \qquad Q_3 = 90 \]

The interquartile range

Why: the spread of the middle half.

\[ \text{IQR} = 90 - 61 = 29 \]

In context: the class averaged about 73.5 with a standard deviation of about 18 points, while the middle half of the class spanned 29 points, from 61 to 90. The book's answers agree with every one of these.

A calculator gives all six numbers in one screen, but knowing where they come from lets you catch a mistyped list: if your mean came out near 80 or your median at 74, one score was entered wrongly.

Check the quartile positions

Why: 15 scores lie below the median and 15 above, and each quartile is the middle, 8th, of its 15.

\[ 15 + 1 + 15 = 31 \checkmark, \qquad 7 + 1 + 7 = 15 \checkmark \]

OpenStax Introductory Statistics 2e, §2.7 Measures of the Spread of the Data §2.7, pp. 111-112 — Example 2.33(b)

56. Example 2.33(c): histogram and box plot on one axis

Picture it

Part (c) asks for a box plot and a histogram on the same axes. The book uses five intervals from 32.5 to 100.5, each 13.6 wide, so that no score falls on a boundary.

\[ \frac{100.5 - 32.5}{5} = 13.6: \quad 32.5, \; 46.1, \; 59.7, \; 73.3, \; 86.9, \; 100.5 \]

Figure (svg): A histogram of the 31 exam scores in five bins of width 13.6 from 32.5 to 100.5, with heights 2, 5, 9, 4 and 11. Above it, on the same horizontal scale, a box plot: whiskers from 33 to 61 and from 90 to 100, a box from 61 to 90, and a median line at 73.

The long left whisker and the short bars on the left tell the same story: the lower half of the scores is spread over 40 points, the upper half over only 27.

Read the two pictures together. The box plot's long left whisker matches the short, spread-out bars on the left of the histogram, and the tallest bar, 11 scores from 86.9 to 100.5, sits under the right end of the box.

The book's comment puts numbers on it: the spread of the lower 50 percent of scores is 73 minus 33, which is 40 points, while the upper 50 percent spans only 100 minus 73, which is 27. One standard deviation of 17.9 cannot describe two halves with such different spreads; the five-number summary can.

The book also notes that the middle 50 percent of scores, an IQR of 29, are Ds, Cs and Bs, and that the lowest quarter are Ds and Fs. Its Table 2.31 of cumulative relative frequencies ends at 0.998 rather than 1 only because each relative frequency was rounded to three places first.

OpenStax Introductory Statistics 2e, §2.7 Measures of the Spread of the Data §2.7, pp. 112-113 — Example 2.33(c), Figure 2.25 and Table 2.31

57. Range, IQR and standard deviation: what each one uses

Concept

You now have three measures of spread, and each looks at a different part of the data. The choice between them follows from what each one uses.

measureformulausesan outlier
rangemax − minonly the two extremeschanges it completely
IQRQ₃ − Q₁the middle 50% of the ordered datadoes not touch it
standard deviation√(Σ f(x − mean)² ÷ (n − 1))every value, weighted by its squared distanceinflates it, through the square

The standard deviation uses the most information, which is its strength for symmetric data without outliers. But the squaring that makes it work also makes it sensitive: one value far out contributes a huge square.

So the measures come in pairs with the centres from Section 2.5. The mean and the standard deviation both use every value and both are pulled by outliers, so they go together for symmetric data. The median and the IQR both ignore the extremes, so they go together for skewed data or data with outliers.

The book's advice for skewed data is to report the five-number summary, the smallest value, the first quartile, the median, the third quartile and the largest value, and to graph the data.

OpenStax Introductory Statistics 2e, §2.7 Measures of the Spread of the Data §2.7, p. 111 — in a skewed distribution, it is better to look at the quartiles

58. One typing slip: the score 100 entered as 1000

Worked example

Suppose the top exam score in Example 2.33 is mistyped as 1000 instead of 100. Recompute the three measures of spread and see which ones notice.

Only one value changes, and it is the largest, so the median and the quartiles keep their positions. The range and the standard deviation are another matter.

The range

Why: the new largest minus the smallest.

\[ 1000 - 33 = 967 \quad (\text{was } 67) \]

The quartiles and IQR

Why: the 8th score of each half: neither moves.

\[ Q_1 = 61, \; Q_3 = 90, \quad \text{IQR} = 29 \quad (\text{unchanged}) \]

The new mean

Why: the total rises by 900.

\[ \bar{x} = \frac{2279 + 900}{31} = \frac{3179}{31} = 102.5 \]

The new standard deviation

Why: squared deviations from the new mean, over 30.

\[ s = \sqrt{\frac{841\,179.7}{30}} = \sqrt{28\,039.3} = 167.4 \]

Figure (svg): A grouped bar chart of three measures of spread for the exam scores, before and after the score 100 is mistyped as 1000. The range goes from 67 to 967. The interquartile range stays at 29. The standard deviation goes from 17.9 to 167.4.

One bad value multiplies the range by fourteen and the standard deviation by nine. The interquartile range does not move at all.

In context: one slip makes the scores look as if they typically stray 167 points from an average above 100, on a test marked out of 100. The IQR still correctly reports that the middle half of the class spans 29 points.

The practical lesson is twofold. Look at the data before summarising them, because a graph would show the 1000 at once. And when you cannot rule out wild values, the median and IQR give a summary that one slip cannot wreck.

Check that the IQR could not move

Why: the two halves each still hold 15 scores, and 1000 sits in the upper half above Q3, where 100 sat.

\[ Q_3 = 24\text{th score} = 90, \quad 1000 > 90 \checkmark \]

59. Which measure of spread would you report?

Discrimination

For each data set, decide whether the standard deviation or the interquartile range is the better summary of spread. Ask first about shape and about outliers.

Sort into buckets

Sort each data set by the measure of spread you would report with it.

mean and standard deviation
heights of adult women in a large city, bell-shaped; weights of cereal boxes from a well-run filling machine; scores on a well-designed exam, roughly symmetric
median and IQR
house prices in a county, with a few mansions; household incomes, strongly skewed to the right; times to finish a marathon, with a long tail of walkers
sd
Symmetric data without outliers: every value counts and nothing extreme distorts the squares, so the standard deviation is the most informative choice.
iqr
Skewed data or extreme values: the squares of the far values would inflate the standard deviation, while the IQR describes the middle half untouched.

Each choice brings its partner. Reporting a standard deviation alongside a median, or an IQR alongside a mean, mixes a resistant measure with a sensitive one and invites confusion.

The house prices are the classic case. A handful of mansions can double the standard deviation of a county's prices while the IQR, which describes the ordinary homes, barely changes.

60. Example 2.34: the standard deviation of a grouped table

Worked example

Example 2.34 gives only a grouped frequency table: classes 0 to 2, 3 to 5, 6 to 8, 9 to 11, 12 to 14 and 15 to 17 with frequencies 1, 6, 10, 7, 0 and 2. The individual values are lost, so neither the mean nor the standard deviation can be found exactly. Estimate both.

The fix is to let each class's midpoint stand in for every value in that class. Then the table becomes a frequency table of midpoints, and the usual formula applies with m in place of x.

Be clear about what that costs. Using midpoints assumes the values in each class are spread evenly around the middle of the class. If the two values in the 15 to 17 class were really 15 and 17, their squared deviations would total about 143.8, not 141.8, so every grouped answer is an estimate.

Estimate the mean from the midpoints

Why: the sum of f times m, over the sum of f.

\[ \bar{x} = \frac{\sum fm}{\sum f} = \frac{1 + 24 + 70 + 70 + 0 + 32}{26} = \frac{197}{26} \approx 7.58 \]

Deviation of each midpoint

Why: for the first class, 1 minus 7.58.

\[ 1 - 7.58 = -6.58, \quad (-6.58)^2 = 43.2964 \]

Weight by frequency and total

Why: the book's last column.

\[ 43.2964 + 76.8984 + 3.364 + 40.9948 + 0 + 141.7928 = 306.3464 \]

Divide by n minus 1

Why: 26 values treated as a sample.

\[ s^2 = \frac{306.3464}{26 - 1} = \frac{306.3464}{25} = 12.254 \]

Take the root

Why: back to the data's units.

\[ s = \sqrt{12.254} \approx 3.50 \]

Figure (svg): A histogram of the grouped table: classes 0 to 2, 3 to 5, 6 to 8, 9 to 11, 12 to 14 and 15 to 17 with frequencies 1, 6, 10, 7, 0 and 2. A dot marks each class midpoint on top of its bar, and a dashed line marks the grouped mean, 7.58. Above each class its midpoint's deviation is written: minus 6.58, minus 3.58, minus 0.58, plus 2.42, plus 5.42 and plus 8.42.

Every value in a class is treated as if it sat at the class midpoint. The two values in the 15 to 17 class are the furthest out, and they dominate the squares.

A misprint to correct: the book's formula line shows 43.2964 over 26 minus 1 under the root, which is only the first row of the table. With that numerator the answer would be 1.32. The total, 306.3464, is what gives the 3.50 the book prints.

Check against the book's answer

Why: square 3.50 and multiply back by 25; the small gap is the rounding of 3.5006.

\[ 3.5006^2 \times 25 = 306.35 \approx 306.3464 \checkmark \]

OpenStax Introductory Statistics 2e, §2.7 Measures of the Spread of the Data §2.7, pp. 113-114 — Example 2.34 and Table 2.32

61. Try It 2.34: σx and sx from the same grouped table

Worked example

Try It 2.34 asks you to find the standard deviation of the same grouped table on a calculator: midpoints into L1, frequencies into L2, then 1-Var Stats. The screen shows two standard deviations. Compute both by hand and say which the example used.

The book's last step says you will see both a population standard deviation, σx, and a sample one, sx. They share the total of squares; only the denominator differs.

The shared total of squares

Why: from Example 2.34, unrounded mean 197 over 26.

\[ \sum f(m - \bar{x})^2 = 306.346 \]

Sample version, sx

Why: divide by n minus 1, 25.

\[ s_x = \sqrt{\frac{306.346}{25}} = \sqrt{12.254} = 3.50 \]

Population version, σx

Why: divide by N, 26.

\[ \sigma_x = \sqrt{\frac{306.346}{26}} = \sqrt{11.783} = 3.43 \]

Example 2.34 divided by 26 minus 1, so it treated the 26 values as a sample and its answer is sx. Had the table described a whole population, the right report would be 3.43. Either way, both are estimates, because midpoints replaced the real values.

Which one to report is the same question as ever. A table describing every student in a small club is a population and σx is right; a table from a survey standing in for a whole school is a sample and sx is right. The calculator shows both because it cannot read the question.

Check the ratio

Why: sx over σx must be the square root of 26 over 25.

\[ \frac{3.5006}{3.4326} = 1.0198 = \sqrt{26/25} \checkmark \]

OpenStax Introductory Statistics 2e, §2.7 Measures of the Spread of the Data §2.7, pp. 114-116 — Try It 2.34 and the calculator steps

62. The three measures side by side

Comparison

Fill in the blanks to finish the comparison. Each row is a measure of spread; each column is a property you would weigh when choosing one.

Comparison matrix

measurepairs withunitsresistant to outliers?Example 2.33 value
range(rarely reported alone)the data's unitsno67
IQRmedianthe data's unitsyes29
standard deviationmeanthe data's unitsno17.9
variancemeansquared unitsno321.3

Read down the units column: every measure except the variance is in the data's own units, which is why the variance is a stepping stone and the standard deviation is what gets reported.

The variance row earns its place for a different reason. It is the quantity the algebra works with, so later chapters add and compare variances, but whenever a number is reported to a reader, it is reported as a standard deviation.

Read down the resistance column next: only the IQR survived the 1000 typo. That column, together with the shape of the data, decides which pair, mean with standard deviation or median with IQR, belongs in your summary.

63. Where spread is the whole story

Real world

In many real settings the average is fixed by design and the spread is what people actually manage. A cereal company sets its filling machine to an average of 500 grams; what decides whether boxes are legally underweight is the standard deviation.

Discussion prompt

A filling machine averages 500 grams per box with standard deviation 4 grams, and the weights are bell-shaped. The label promises 492 grams. How unusual is an underweight box, and what should the company change to make them rarer: the mean, the standard deviation, or either?

The same logic runs through investment risk, where the standard deviation of returns is called volatility, and medicine, where a test result is flagged when it lies more than two standard deviations from the healthy mean. In each case the question is the one from Part 2: how many deviations out is this value?

OpenStax Introductory Business Statistics 2e, §2.7 Measures of the Spread of the Data §2.7 Measures of the Spread of the Data — the same ideas in business settings

64. Putting it together

Section

Part 7

65. Pattern: computing and using a standard deviation

Pattern

Every problem in this section ran through the same short sequence. The arithmetic is mechanical; the decisions at the start and the end are where the marks are.

Figure (svg): A flow diagram in two rows. Top row: sample or population, then the mean, then the deviations, then square and total, then divide by n minus 1 or N, then take the square root. An arrow drops to a bottom row with three uses: locate a value with z, compare values from different sets, and bound the data with Chebyshev or the empirical rule. A side box says choose the IQR instead when the data are skewed or have outliers.

The top row computes the number; the bottom row uses it. The red box is the question to ask before trusting either.
  1. Sample or population? Decide from the question, not from the numbers. It fixes the symbol and the denominator.
  2. Mean, deviations, squares, total. Keep the unrounded mean, square before adding, and check the deviations total zero.
  3. Divide and take the root. n − 1 for a sample, N for a population. Forgetting the root leaves the variance, in squared units.
  4. Locate or compare. z = (value − mean) ÷ SD; decide which direction is good before comparing.
  5. Bound. Chebyshev for any data; 68-95-99.7 only for bell-shaped data.
  6. Choose. Skewed or outliers: median and IQR. Symmetric: mean and standard deviation.

\[ s = \sqrt{\frac{\sum f(x - \bar{x})^2}{n - 1}}, \qquad z = \frac{x - \bar{x}}{s} \]

Work the top row in order and the checks come for free: deviations that total zero confirm the mean, and a root that is smaller than the biggest deviation confirms the last two steps. The bottom row is where the question's context comes back in.

Most wrong answers in this section are right arithmetic on the wrong choice: N where n minus 1 belonged, a variance reported as a standard deviation, a high z read as better for a race time, or a standard deviation quoted for data so skewed that it describes neither tail.

66. Check: the denominator

Check

Decide sample or population before you touch the arithmetic.

The squared deviations are already totalled for you. Your only decisions are the denominator and the last step.

Check your understanding

A researcher records the commute times of 10 randomly chosen workers in a city, to learn about all the city's workers. The squared deviations from the sample mean total 90. What standard deviation should she report?

  • A. √(90 ÷ 9) ≈ 3.16 minutes (correct)
  • B. √(90 ÷ 10) = 3 minutes
  • C. 90 ÷ 9 = 10 minutes
  • D. 90 ÷ 10 = 9 minutes

Answer: A

Why: Ten workers chosen to learn about all workers are a sample, so divide the total of squares by n minus 1, which is 9, giving a variance of 10 square minutes. The standard deviation is its square root, about 3.16 minutes.

Why B tempts people
That divides by N, treating the ten workers as the whole population. They are a sample standing in for the city.
Why C tempts people
That is the sample variance, in square minutes. The standard deviation needs the square root.
Why D tempts people
That both divides by N and skips the root: two separate slips.

Notice the four options are built from the same 90. Each wrong one makes exactly one or two of the slips this deck has named, so working out why each is wrong is as useful as finding the right one.

In context: a typical commute in this city differs from the average by roughly three minutes, as estimated from this sample.

Two quick checks rule out wrong options at a glance. A standard deviation cannot be larger than every deviation in the data, so a large number from a small total of squares suggests a missing root; and the word randomly chosen signals a sample.

67. Check: compare in standard deviations

Check

Decide which direction is better before you compare.

Convert each result to a z-score against its own group, then remember what a high value means for this measurement.

Check your understanding

Two sprinters run the 100 metres. Kai runs 11.2 seconds in a league with mean 11.8 and standard deviation 0.4. Lee runs 10.9 seconds in a league with mean 11.2 and standard deviation 0.3. Who ran faster relative to their own league?

  • A. Kai, with z = −1.5 (correct)
  • B. Lee, with z = −1.0
  • C. Lee, because 10.9 < 11.2
  • D. They are equal: both ran 0.3 to 0.6 s under the mean

Answer: A

Why: Kai's time is 0.6 seconds below his league's mean, which is 0.6 over 0.4, or 1.5 standard deviations below. Lee's is 0.3 below, which is 0.3 over 0.3, one standard deviation below. For a race time lower is better, so Kai's more negative z is the stronger run.

Why B tempts people
Lee's z is minus 1, but for times the more negative z is better, and Kai's is minus 1.5.
Why C tempts people
That compares raw times from different leagues, which the z-score exists to avoid.
Why D tempts people
The raw gaps differ, 0.6 against 0.3, and the leagues' standard deviations differ too, so the z-scores differ.

In context: Kai beat his league's average by one and a half typical deviations, Lee by one, so Kai's run was the more exceptional, even though Lee's clock time was faster.

This is Try It 2.35 in a new setting. The trap answer, the faster raw time, is exactly the comparison the book warns can be misleading when means and standard deviations differ.

68. Check: what the rules promise

Check

Ask about the shape before choosing a rule.

Nothing is said about the shape of these data, so think about which rule is allowed.

Check your understanding

A data set of delivery times has mean 30 minutes and standard deviation 5 minutes. Nothing is known about its shape. What can you say for certain about the deliveries between 20 and 40 minutes?

  • A. At least 75% of deliveries (correct)
  • B. About 95% of deliveries
  • C. About 68% of deliveries
  • D. Nothing at all can be said

Answer: A

Why: Twenty to forty minutes is the mean plus or minus two standard deviations. With no information about shape only Chebyshev's rule applies, and it guarantees at least one minus one quarter, which is 75 percent, within two deviations.

Why B tempts people
That is the empirical rule's figure, which needs bell-shaped, symmetric data. The shape here is unknown.
Why C tempts people
That is the empirical rule's figure for one standard deviation, 25 to 35 minutes, and it also needs a bell shape.
Why D tempts people
Chebyshev's rule holds for any data set, so a guarantee is available whatever the shape.

Chebyshev also caps the far side. At most one delivery in four can take longer than 40 minutes or less than 20 minutes combined, however strange the shape of the delivery times turns out to be.

In context: at least three deliveries in four arrive within ten minutes of the half-hour average, and if the times turn out to be bell-shaped the true share is closer to 95 percent.

The empirical rule is a stronger statement with a condition attached; Chebyshev is a weaker one with no condition. When the condition is unknown, report the weaker statement and say why.

69. Explain why the deviations are squared

Explain it to yourself

The squaring is the step students most often perform without understanding. Explaining it in your own words is a good test of whether Part 3 has landed.

Discussion prompt

Explain in two or three sentences why the standard deviation squares the deviations instead of simply averaging them, and why it then takes a square root.

A strong answer names both problems and both fixes: cancelling, solved by squaring; units, solved by the root. A very strong one adds the cost of squaring: far values are weighted heavily, which is why outliers inflate the standard deviation and why skewed data are better summarised by the IQR.

If your explanation said squaring is just what the formula does, revisit the line A arrows and squares: the zero total and the 20 square minutes are the whole argument in a picture.

70. Exit ticket: two quizzes, one student

Exit ticket

Finish with a problem that uses the whole section: compute a standard deviation from a small sample, then use it to compare.

Discussion prompt

Maya's five quiz scores in history were 6, 7, 8, 8 and 11, a sample of her work. Find the mean and the sample standard deviation. On her next history quiz she scores 10; on a maths quiz where her mean is 15 with s = 4 she scores 18. Which score is better relative to her usual performance?

The five scores are a sample of her work, so the denominator is four. The deviations, minus 2, minus 1, 0, 0 and plus 3, total zero, which checks the mean.

If you divided by five you would get s equal to about 1.67 and a z of 1.20: the same conclusion here, but the wrong number. And comparing 10 with 18 directly is the mistake the whole of Part 5 exists to prevent.

71. Recap

Recap

The standard deviation measures how far data values are from their mean. It does two jobs: one number for the overall spread of a data set, and a ruler for locating a single value.

ideaformularead it as
deviationx − x̄ or x − μsigned distance from the mean; they total zero
sample variances² = Σ f(x − x̄)² ÷ (n − 1)average squared deviation, sample style
population varianceσ² = Σ f(x − μ)² ÷ Naverage squared deviation over everyone
standard deviations = √s², σ = √σ²the root brings back the data's units
locating a valuevalue = mean + (z)(SD)count the hops of one SD
z-scorez = (value − mean) ÷ SDunitless: compares across data sets
Chebyshevwithin k SD: at least 1 − 1/k²any shape: 75% within 2, 89% within 3
empirical rule68%, 95%, 99.7%only for bell-shaped, symmetric data

The work is mostly in the decisions: sample or population, which direction is better, and whether the data are symmetric enough for a standard deviation at all. Skewed data and outliers call for the median and the IQR instead.

The Introductory Business Statistics text covers the same section with business data if you want more practice. Chapter 3 turns to probability; when chapter 6 arrives, the z-score and the empirical rule from this deck become the normal distribution.

OpenStax Introductory Statistics 2e, §2.7 Measures of the Spread of the Data §2.7, pp. 107-118 — the whole section

OpenStax Introductory Business Statistics 2e, §2.7 Measures of the Spread of the Data §2.7 Measures of the Spread of the Data — the same measures in business contexts

Sources

  1. OpenStax Introductory Statistics 2e, §2.7 Measures of the Spread of the Data — Illowsky & Dean, OpenStax / Rice University, CC BY 4.0, pp. 106-118
  2. OpenStax Introductory Business Statistics 2e, §2.7 Measures of the Spread of the 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