Griffiths D. Head First Statistics

Подождите немного. Документ загружается.

you are here 4 101

measuring variability and spread

Here are box and whisker diagrams for two more basketball players. Compare the ranges of

their scores. If you had to choose between having player A or player B on the team, which would

you pick? Why?

Score

Player A

Player B

Score

4 6 8 10 12 14 16 18 20 22 24 26 28 30

Basketball Player A and B scores

I’m sure I’ve seen box and whisker diagrams that look a

bit different than this.

There are actually several versions of box and whisker

diagrams. Some have deliberately shorter whiskers and explicitly

show outliers as dots or stars extending beyond the whiskers. This

makes it easier to see how many outliers there are and how extreme

they really are. Other diagrams show the mean as a dot, so you can

see where it’s positioned in relation to the median. If you’re taking a

statistics course, it would be a good idea to check which version of

the box and whisker diagram is likely to be used.

So if you show the mean as a dot, is it to the left or right

of the median?

If the data is skewed to the right, then the mean will be to the

right of the median, and the whisker on the right will be longer than

that of the left. If the data is skewed to the left, the mean will be to

the left of the median, and the whisker on the left will be the longest.

102 Chapter 3

Player A

Player B

Here are box and whisker diagrams for each basketball player. Compare the ranges of their

scores. If you had to choose between having player A or player B on the team, which would you

pick? Why?

Player A has a relatively small range, and his median score is a bit higher

than Player B’s.

Player B has a very large range. Sometimes this player scores a lot

higher than Player A, but sometimes a lot lower.

Player A plays more consistently and usually scores higher than Player B

(compare the medians and interquartile range), so we’d pick Player A.

Percentiles split your data into percentages.

They’re useful for benchmarking.

The kth percentile is k% of the way through

your data. It’s denoted by P

An interpercentile range is like the

interquartile range but, this time, between two

percentiles.



Box and whisker diagrams, or box plots, are

a useful way of showing ranges and quartiles

on a chart. A box shows where the quartiles

and interquartile range are, and the whiskers

give the upper and lower bounds. More than

one set of data can be shown on the same

chart, so they’re useful for comparisons.



Score

4 6 8 10 12 14 16 18 20 22 24 26 28 30

Basketball Player A and B scores

exercise solution

you are here 4 103

measuring variability and spread

The coach doesn’t just need to compare the range of the

players’ scores; he needs some way of more accurately

measuring where most of the values lie to help him

determine which player he can truly rely on come game day.

In other words, he needs to find the player whose scores vary

the least.

The problem with the range and interquartile range is that

they only tell you the difference between high and low values.

What they don’t tell you is how often the players get these high

or low scores versus scores closer to the center—and that’s

important to the coach.

The coach needs a team of players he can rely on. The last

thing he wants is an erratic player who will play well one

week and score badly the next.

What can we do to help the coach make his decision?

How can we more accurately measure variability?

The interquartile range looks useful, but

what about players who sometimes get really

low scores? If a player messes up on game day,

it could cost us the league! I’m not sure that the

range or the interquartile range tell me which

player is really the most consistent.

104 Chapter 3

Variability is more than just spread

We don’t just want to measure the spread of each set of scores;

we want some way of using this to see how reliable the player is.

In other words, we want to be able to measure the variability of

the players’ scores.

One way of achieving this is to look at how far away each value

is from the mean. If we can work out some sort of average

distance from the mean for the values, we have a way of

measuring variation and spread. The smaller the result, the closer

values are to the mean. Let’s take a look at this.

The values here are spread out quite a long way from

the mean. If the coach picks this player for the team,

he’s unlikely to be able to predict how the player will

perform on game day. The player may achieve a

very high score if he’s having a good day. On a bad

day, however, he may not score highly at all, and that

means he’ll potentially lose the game for the team.

The values for this second set of data are much closer

to the mean and vary less. If the coach picks this

player, he’ll have a good idea of how well the player

is likely to perform in each game.

Let’s find out.

So does that mean we

just calculate the average

distance from the mean?

exploring variability

frequency

score

frequency

score

Player 1’s basketball scores

Player 2’s basketball scores

you are here 4 105

measuring variability and spread

Calculating average distances

Imagine you have three numbers: 1, 2, and 9. The mean

is 4. What happens if we find the average distance of

values from the mean?

Average distance = (1 to μ) + (2 to μ) + (9 to μ)

= 3 + 2 + (-5)

= 0

These distances cancel

each other out.

The average distance of values from the mean is always

0. The positive and negative distances cancel each other

out. So what can we do now?

Why do we have -5 in that

equation? I would have thought the

distance would be 5. Why is it negative?

A: The distance from 9 to μ is negative

because μ is less than 9. 1 and 2 are both

less than μ, so the distance is positive for

both of them. That’s why the distances

cancel each other out.

Can’t we just take the positive

distances and average those?

A: That sounds intuitive, but in practice,

statisticians rarely do this. There’s another

way of making sure that the distances don’t

cancel out, and you’ll see that very soon.

This other way of determining how close

typical values are to the mean is used a lot

in statistics, and you’ll see it through most of

the rest of the book.

Surely the distances don’t cancel

out for all values. Maybe we were just

unlucky.

A: No matter what values you choose, the

distances to the mean will always cancel out.

Here’s a challenge for you: take a group of

numbers, work out the mean, work out the

distance of each value from the mean, and

then add the distances together. The result

will be 0 every time.

Can’t I use the interquartile range

to see how reliable the scores are?

A: The interquartile range only uses part

of the data for measuring spread. If a player

has one bad score, this will be excluded

by the interquartile range. In order to truly

determine reliability and consistency, we

need to consider all the scores.

The range uses all the scores. Why

can’t we use that then?

A: The range is only really good for

describing the difference between the

highest and lowest number. As you saw

earlier, this doesn’t represent how the values

are actually distributed. We need another

measure to do this.

μ = 4

distance = 2

distance = 3

distance = -5

The positive and

negative distances

from the mean

cancel each other

out.

106 Chapter 3

We need a way of making all the

numbers positive. Maybe it’ll work if

we square the distances first. Then

each number is bound to be positive.

Let’s try this with the same three numbers.

Average (distance)

= (1 to μ)

+ (2 to μ)

+ (9 to μ)

= 3

+ 2

+ (-5)

= 9 + 4 + 25

= 12.67 (to 2 decimal places)

This time we get a meaningful number, as the distances don’t

cancel each other out. Every number we add together has to

be non-negative because we’re squaring the distance from the

mean. Adding these numbers together gives us a non-negative

result—every time.

This method of measuring spread is called the variance, and

it’s a very common way of describing the spread of a set of data.

Here’s a general form of the equation:

The variance is

the average of

the distances from

the mean squared.

We want a way to measure the average distance of values from

the mean in a way that stops the distances from cancelling each

other out.

Remember that μ = 4.

This time we’re

adding together three

positive numbers.

We can calculate variation with the variance…

variance and standard deviation measure variability

(x - )

Variance =

Variance

The variance is a way of

measuring spread, and it’s the

average of the distance of

values from the mean squared.

Vital StatisticsVital Statistics

(x - )

you are here 4 107

measuring variability and spread

But why should I have to think about

distances squared? I hardly call that

intuitive. Isn’t there another way?

Statisticians use the variance a lot as a means of measuring the

spread of data. It’s useful because it uses every value to come up

with the result, and it can be thought of as the average of the

distances from the mean squared.

Standard deviation know-how

We’ve seen that the standard deviation is a way of saying how far

typical values are from the mean. The smaller the standard deviation,

the closer values are to the mean. The smallest value the standard

deviation can take is 0.

Like the mean, the standard deviation has a special symbol, σ. This is

the Greek character lowercase Sigma. (We saw uppercase Sigma, Σ, in

Chapter 2 to represent summation.)

To find σ, start off by calculating the variance, and then

take the square root.

…but standard deviation is a more intuitive measure

= variance

What we really want is a number that gives the spread

in terms of the distance from the mean, not distance

squared.

The problem with the variance is that it can be quite difficult to think about

spread in terms of distances squared.

There’s an easy way to correct this. All we need to do is take the square root

of the variance. We call this the standard deviation.

Let’s work out the standard deviation for the set of numbers we had before.

The variance is 12.67, which means that

Standard deviation = √12.67

= 3.56 (to 2 decimal places)

In other words, typical values are a distance of 3.56 away from the mean.

σ = √variance

I’m the standard

deviation. If you need to

measure distances from

the mean, give me a call.

108 Chapter 3

Head First: Hey, Standard Deviation, great to see

you.

Standard Deviation: It’s a real pleasure, Head

First.

Head First: To start off, I was wondering if you

could tell me a bit more about yourself and what you

do.

Standard Deviation: I’m really all about measuring

the spread of data. Mean does a great job of telling

you what’s going on at the center, but quite often,

that’s not enough. Sometimes Mean needs support to

give a more complete picture. That’s where I come in.

Mean gives the average value, and I say how values

vary.

Head First: Without meaning to be rude, why

should I care about how values vary? Is it really all

that important? Surely it’s enough to know just the

average of a set of values.

Standard Deviation: Let me give you an example.

How would you feel if you ordered a meal from the

local diner, and when it arrived, you saw that half of it

was burnt and the other half raw?

Head First: I’d probably feel unhappy, hungry, and

ready to sue the diner. Why?

Standard Deviation: Well, according to Mean,

your meal would have been cooked at the perfect

temperature. Clearly, that’s not the full picture; what

you really need to know is the variation. That’s where

I come in. I look at what Mean thinks is a typical

value, and I say how you can expect values to vary

from that number.

Head First: I think I get it. Mean gives the average,

and you indicate spread. How do you do that, though?

Standard Deviation: That’s easy. I just say how

far values are from the mean, on average. Suppose

the standard deviation of a set of values is 3 cm. You

can think of that as saying values are, on average, 3

cm away from the mean. There’s a bit more to it than

that, but if you think along those lines, you’re on the

right track.

Head First: Speaking of numbers, Standard

Deviation, is it better if you’re large or small?

Standard Deviance: Well, that really all depends

what you’re using me for. If you’re manufacturing

machine parts, you want me to be small, so you can

be sure all the pieces are about the same. If you’re

looking at wages in a large company, I’ll naturally be

quite large.

Head First: I see. Tell me, do you have anything to

do with Variance?

Standard Deviation: It’s funny you should ask that.

Variance is just an alter ego of mine. Square me, and I

turn into Variance. Take the square root of Variance,

and there I am again. We’re a bit like Clark Kent and

Superman, but without the cape.

Head First: Just one more question. Do you ever feel

overshadowed by Mean? After all, he gets a lot more

attention than you.

Standard Deviation: Of course not. We’re great

friends, and we support each other. Besides, that

would make me sound negative. I’m never negative.

Head First: Standard Deviation, thank you for your

time.

Standard Deviation: It’s been a pleasure.

Standard Deviation Exposed

This week’s interview:

Getting the measure of Standard Deviation

interview with standard deviation

you are here 4 109

measuring variability and spread

It’s time for you to flex those standard deviation muscles. Calculate the mean and standard

deviation for the following sets of numbers.

1 2 3 4 5 6 7

1 2 3 4 5 6

110 Chapter 3

Variance = (1-3.5)

+ (2-3.5)

+ (3-3.5)

+ (4-3.5)

+ (5-3.5)

+ (6-3.5)

= 2.5

+ 1.5

+ 0.5

+ (-0.5)

+ (-1.5)

+ (-2.5)

= 6.25 + 2.25 + 0.25 + 0.25 + 2.25 + 6.25

= 17.5

= 2.92 (to 2 decimal places) = 2.92

= 1.71 (to 2 d.p.)

It’s time for you to flex those standard deviation muscles. Calculate the mean and standard

deviation for the following sets of numbers.

1 2 3 4 5 6 7

Let’s start off by calculating the mean.

μ = 1 + 2 + 3 + 4 + 5 + 6 + 7

= 28

= 4

1 2 3 4 5 6

μ = 1 + 2 + 3 + 4 + 5 + 6

= 21

= 3.5

The standard deviation calculation

can quickly become complicated.

To find the standard deviation, you first have to

calculate variance, finding (x - μ)

for every value of x.

But there is a much simpler formula for variance

that produces the same result.The equation’s on

the opposite page, but first you’ll need to rescue the

derivation from the pool.

exercise solution

Those calculations were tricky.

Isn’t there an easier way?

Variance = (1-4)

+ (2-4)

+ (3-4)

+ (4-4)

+ (5-4)

+ (6-4)

+ (7-4)

= 3

+ 2

+ 1

+ 0

+ (-1)

+ (-2)

+ (-3)

= 9 + 4 + 1 + 0 + 1 + 4 + 9

= 28

= 4 =

4 = 2