Chp-07 124-144.

qxd 26/5/04 4:26 pm Page 124

CHAPTER 7

Correlation

Learning objectives
In earlier chapters, only single variables have been considered. Now you will be working with
pairs of variables.
After studying this chapter, you should be able to:
■ investigate the strength of a linear relationship between two variables by using suitable
statistical analysis
■ evaluate and interpret the product moment correlation coefficient.

7.1 Scatter diagrams

Scatter diagrams are used where we are examining
possible relationships between two variables.

An athlete, recovering from injury, had her pulse rate measured

after performing a predetermined number of step-ups in a
gymnasium. The measurements were made at weekly intervals.
The table below shows the number of step-ups, x, the pulse rate,
y beats per minute, and the week in which the measurement was
made.

Week 1 2 3 4 5 6 7 8
x 15 50 35 25 20 30 10 45
y 114 155 132 112 96 105 78 113

To construct a scatter diagram you simply plot points with

coordinates (x, y) for each of the 8 weeks.
Chp-07 124-144.qxd 26/5/04 4:26 pm Page 125

Correlation 125

y
160

140

120
Pulse rate
(beats per
minute)

100

80

0 10 20 30 40 50 x 7
Number of step-ups

In this case, as you would expect, there appears to be a clear

tendency for the pulse rate to increase with the number of
step-ups.

7.2 Interpreting scatter diagrams

Interpreting a scatter diagram is often the easiest way for
you to decide whether correlation exists. Correlation means
that there is a linear relationship between the two
variables. This could mean that the points lie on a straight
line, but it is much more likely to mean that they are
scattered about a straight line.
Chp-07 124-144.qxd 26/5/04 4:26 pm Page 126

126 Correlation

The main types of scatter diagram

Positive correlation Negative correlation No correlation
y y y

x x x

● positive or ● negative or ● little or no

direct inverse correlation,
correlation correlation no linear
relationship
● x increases as ● x decreases as ● x and y are not
y increases y increases linked
● clear linear ● clear linear ● x and y appear
relationship relationship to be
exists. exists. independent.
y y

x x

● x and y are clearly linked by a non-linear relationship.

7.3 Studying results

The table below gives the marks obtained by ten pupils taking
maths and physics tests.
Pupil A B C D E F G H I J
Maths mark
(out of 30)
x 20 23 8 29 14 11 11 20 17 17
Physics mark
(out of 40) Physics
y 30 35 21 33 33 26 22 31 33 36
40
Is there a connection between the marks obtained by the ten
35
pupils in the maths and physics tests?
The starting point would be to plot the marks on a scatter diagram. 30

The areas in the bottom-right and top-left of the graph are 25

almost empty so there is a clear tendency for the points to run
20
from bottom-left to top-right. This indicates that positive 5 10 15 20 25 30
correlation exists between x and y. Maths
Chp-07 124-144.qxd 26/5/04 4:26 pm Page 127

Correlation 127

Calculating the means: Note: importance of scale.

170

x
 
17
10
Consider this change,
Physics
and 100
300

y

10

30. 75
Using these lines, the graph can be divided into four regions to 50
show this tendency very clearly.
25
Physics
x
17
40 0
5 10 15 20 25 30
35 Maths

30
The appearance of the scatter
y
30 diagram is now very different.
25 The existence of correlation is
much more difficult to identify.
20 Scales should cover the range of
5 10 15 20 25 30
Maths
the given data.

The table below gives the marks obtained by the ten pupils
taking maths and history tests.
Pupil A B C D E F G H I J
Maths mark
(out of 30) 7
x 20 23 8 29 14 11 11 20 17 17
History mark History
x
17
(out of 60) 60
z 28 21 42 32 44 56 36 24 51 26
50
Calculating the mean for z:
40 z
36
360

z

10

36
30
The scatter diagram for maths and history shows a clear
20
tendency for points to run from top-left to bottom-right. This 5 10 15 20 25 30
indicates that negative correlation exists between x and z. Maths

7.4 Product moment correlation

coefficient (PMCC)
(This is often known as Pearson’s correlation coefficient
after Karl Pearson, an applied mathematician who worked on
the application of statistics to genetics and evolution.)

How can the strength of correlation be quantified?

There are two main points to consider.
● How close to a straight line are the points?
● Is the correlation positive or negative?
Chp-07 124-144.qxd 26/5/04 4:26 pm Page 128

128 Correlation

The product moment correlation coefficient, r, gives a

standardised measure of correlation which can be used for
comparisons between different sets of data.
Sxx, Syy and Sxy are used to evaluate r where:
Sxx
(xi 
x)2, Syy
(yi 
y)2 and These formulae are given in the
AQA formulae book.
Sxy
(xi 
x)(yi 
y)
Sxy
r is given by 
S
xxSyy

Formula r is obtainable directly from all

The computational form of this equation which is most calculators with regression
commonly used is: facility. This is recommended in
the exam.
xy
xy  
n
r

{ (x)2
x2  
n }{ (y)2
y2  
n } Sketches to illustrate examples
of possible values of r.

Values of r y

Some worked examples

Returning to the maths and physics marks in section 8.2.
r
1
To illustrate the calculation involved in evaluating r, the
following additional summations are needed:
x2
3250, y2
9250, xy
5313. x
Exact positive correlation
You can then see that
y
1702
Sxx
3250  
360
10
and
3002
Syy
9250  
250
10 r
0.6
170  300
Then, Sxy
5313  
213 x
10 Weak positive correlation
So y

213
r

0.71
3

6

0



52
0


This, of course, can be found directly from your calculator.
The interpretation of the value of r is very important. The value r
0
of r tells you how close the points are to lying on a straight x
line. No correlation
Chp-07 124-144.qxd 26/5/04 4:26 pm Page 129

Correlation 129

It is always true that: y

r
0.65

1  r  1

r
1 indicates ALL the points lie on a line with

positive gradient

r
1 indicates ALL the points lie on a line with x

negative gradient Weak negative correlation

r
0 indicates that there is no linear connection at y

r
0.9
all between the two sets of data.

The value obtained in this example, r
0.71, would indicate a

fairly strong positive correlation between the test score in maths
and the test score in physics.
x
Strong negative correlation
Worked example 7.1
A group of 12 children participated in a psychological study
designed to assess the relationship, if any between age, x years,
and average total sleep time (ATST), y minutes. To obtain a
measure for ATST, recordings were taken on each child on five
consecutive nights and then averaged. The results are below.
x
9
7
Child Age ATST y
x (years) y (minutes)
A 4.4 586
B 6.7 565
C 10.5 515
D 9.6 532 550
E 12.4 478
F 5.5 560 y
531
G 11.1 493 ATST
H 8.6 533
I 14.0 575 500
J 10.1 490
K 7.2 530
L 7.9 515

Calculate the product moment correlation

coefficient between x and y and interpret 450
4 6 8 10 12 14 x
your result. Age

Solution
x
108 and y
6372
x2
1060.1, y2
3 396 942
xy
56 825.4
Chp-07 124-144.qxd 26/5/04 4:26 pm Page 130

130 Correlation

1082
Sxx
1060.1  
88.1
12
63722
Syy
3 396 942  
13 410
12
Then
108  6372
Sxy
56 825.4  
522.6
12
So Note that it would be worth
522.6
r

0.481 (to 3 s.f.) investigating child I who seems to
8

8

.1



3
1

1
4
0

have an abnormally high ATST.
Considering the value of r and the scatter diagram, there is Perhaps the child was ill during
evidence of weak negative correlation between age and ATST. the experiment or perhaps there
is some other reason for the
This would indicate that older children have less ATST than
excessive amount of sleep.
younger children. However, the relationship is fairly weak.

Worked example 7.2

The following data indicate the level of sales for ten models of
pen sold by a particular company. The sales, together with the
selling price of the pen, are given in the table below.
y
Model Price, Sales, 35
x (£) y (00s)
A 2.5 30 30
B 5 35
C 15 25 25
D 20 15
Sales
E 7.5 25 20
F 17.5 10
G 12 15 15
H 6 20
I 25 8 10
J 30 10
5
x
140.5 y
193 0 10 20 x
Price
x2
2723.75 y2
4489 xy
2087.5
Note that care must be taken
Plot these data on a scatter diagram. Evaluate the product
not to approximate prematurely
moment correlation coefficient and interpret your answers with
in calculations or else r may be
reference to the data supplied. inaccurate.

Solution
140.52
Sxx
2723.75  
749.725
10
1932
Syy
4489  
764.1
10
Chp-07 124-144.qxd 26/5/04 4:26 pm Page 131

Correlation 131

Then, Using prematurely rounded

140.5  193 figures:
Sxy
2087.5  
624.15 Sxx
750 and Syy
764
10
Sxy
624
So
624
624.15 r

r

0.825 (to 3 s.f.) 7

5
0



6
7
4

7

4

9

.7

2
5



6
7
4

.1


624
Considering the value of r and the scatter diagram, there is

0.824 (to 3 s.f.)
757
evidence of quite strong negative correlation between x and y.
An error has now occurred.
This would indicate that there are fewer sales of the more It is only the final answer which
expensive pens and this trend follows a linear relationship. should be rounded to three
significant figures.

7.5 Limitations of correlation Remember it is recommended

that r is obtained directly using a
It is very important to remember a few key points about calculator.
correlation.

Non-linear relationships
As illustrated in section 7.2, r measures linear relationships Note that clear non-linear
only. It is of no use at all when a non-linear relationship is relationships identified on scatter
evident. There may well be a very clear relationship diagrams should always be
between the variables being considered but if that commented upon but the 7
evaluation of r is not appropriate.
relationship is not linear then r will not help at all.

The scatter diagram should reveal this.

Cause and effect

A student does some research in a primary school and discovers Note that any suggestion that
a very strong direct correlation between length of left foot and correlation may indicate cause
score in a mental maths test. Does this mean that stretching a and effect in the relationship
child’s foot will make them perform better in maths? between two variables should be
considered very carefully!
Clearly this is ridiculous and the probable hidden factor is age:
older children have bigger feet and a better ability at maths. Foot length

The correlation found between foot length and score in Age

maths is often called spurious and should be treated with Mental maths score
caution.

Freak results
An unusual result can drastically alter the value of r.
Unexpected results should always be commented upon
and investigated further as their inclusion or exclusion
in any calculations can completely change the final
result.
Chp-07 124-144.qxd 26/5/04 4:26 pm Page 132

132 Correlation

Imagine the effect on r if the point P is to be removed from

correlation calculations using the data below.
x
12 r
0.706 x
10
y y
r
0
30 P 30
25 25
20 20
15 15
y
12
10 10 y
10
5 5

5 10 15 20 25 30 x 5 10 15 20 25 30 x

Worked example 7.3

Plot scatter diagrams on separate axes for the following data
sets:
(a) x 15 10 5 20 25 10 25 10
y 3 2.5 5 5 4 5 5 3

(b) x 2.5 2.8 3 3.2 4.5 5 6 8

y 20 14 10 8 6 4 3 2

It has been suggested that the product moment correlation

coefficient should be evaluated for both sets of data. By careful
examination of your scatter diagrams, comment on this
suggestion in each case.

Solution
(a) The scatter diagram indicates little or no correlation
between the two variables. r could be evaluated but would
clearly be close to zero.
y
6

2
0 5 10 15 20 25 x

(b) r may well indicate fairly strong negative correlation

between the two variables but the scatter diagram clearly
shows that the relationship is non-linear and hence r is an
inappropriate measure.
Chp-07 124-144.qxd 26/5/04 4:26 pm Page 133

Correlation 133

y
20

10

0
2 4 6 8 x

Worked example 7.4

A tasting panel was asked to assess biscuits baked from a new
recipe. Each member was asked to assign a score from 0 to 100
for texture (x), for flavour (y) and for sweetness (z).
The scores assigned by the ten tasters were as follows:
Taster 1 2 3 4 5 6 7 8 9 10
x 43 59 76 28 53 55 81 49 38 47
y 67 82 75 48 91 63 67 51 44 54
(a) Draw a scatter diagram to illustrate the data.
(b) Calculate the value of the product moment correlation 7
coefficient between x and y.
(c) State, briefly, how you would expect the scatter diagram to
alter if the tasters were given training in how to assign
scores before the tasting took place.
(d) Given that (zi 
z)2
2516.9 and (yi 
y)(zi 
z)
1979.8,
calculate the product moment correlation coefficient
between y and z.

Solution
(a) y

80

60

40
20 40 60 80 x
Chp-07 124-144.qxd 26/5/04 4:26 pm Page 134

134 Correlation

(b)  x
529,
x
52.9, y
642,
y
64.2

 xy
35 187,  x2
30 339, y2
43 334
5292 Note that a calculator can be
Sxx
30 339  
2354.9
10 used to obtain r directly.
6422
and Syy
43 334  
2117.6
10
529  642
and Sxy
35 187  
1225.2
10
1225.2
therefore, r

0.549 (to 3 s.f.)
2

3

5

4

.9



1
2
1

7

.6

(c) The scores would be less variable. Training would lead to a more
The scatter diagram would be more compact but the overall consistent scale for x and y.
shape would be similar. Without training, people’s views
on texture or flavour would vary
(yi 
y)(zi 
z)
(d) r
 widely.
(y

i 

y
)2(z

i
z)2
Sz A calculator cannot be used to

y, obtain r directly in this case – the
S

yySzz
formula must be used.
therefore,
1979.8
r

0.858
2

1

1

7

.6



5
2
1

6

.9

Worked example 7.5

The following data show the annual income per head, x ($US),
and the infant mortality, y (per thousand live births), for a
sample of 11 countries.

Country x y
A 130 150
B 5950 43
C 560 121
D 2010 53
E 1870 41
F 170 169
G 390 143
H 580 59
I 820 75
J 6620 20
K 3800 39

x
22 900, x2
102 724 200,

y
913, y2
103 517, xy
987 130.
Chp-07 124-144.qxd 26/5/04 4:26 pm Page 135

Correlation 135

(a) Draw a scatter diagram of the data. Describe the

relationship between income per head and infant mortality
suggested by the diagram.
(b) An economist asks you to calculate the product moment
correlation coefficient.
(i) Carry out this calculation.
(ii) Explain briefly to the economist why this calculation
may not be appropriate.

Solution
(a) y
160

140

120

100
Infant
mortality 80

7
60

40

20

0 2000 4000 6000 x

Income
Infant mortality appears to decline as income per head
increases.
The decrease is not uniform but is much more marked
for the very low incomes than for the higher income
countries.
22 9002
(b) (i) Sxx
102 724 200  
55 050 563.64
11
9132
and Syy
103 517  
27 738
11
22 900  913
and Sxy
987 130  
913 570
11
913 570
therefore, r

0.739 Note that this can be found
5

5


5
0
0


6
5
3
.6

4



7
2

3
7
8

directly from a calculator.
(to 3 s.f.)
Chp-07 124-144.qxd 26/5/04 4:26 pm Page 136

136 Correlation

(ii) PMCC measures the strength of a linear relationship. It Look back to the beginning of
is not a suitable measure for data which clearly shows section 7.5. A clear curve is seen.
a non-linear relationship, as in this case.

EXERCISE 7A
1 (a) For each of the following scatter diagrams, state whether
or not the product moment correlation coefficient is an
appropriate measure to use.
(i) y (ii) y

r
1 x r
0.3 x

(iii) y

r
1.2 x
(b) State, giving a reason, whether or not the value
underneath each diagram might be a possible value of
this correlation coefficient.

2 Estimate, without undertaking any calculations, the

product moment correlation coefficient between the variables
in each of the scatter diagrams given:
(i) y (ii) y

x x
(iii) y (iv) y

x x
Chp-07 124-144.qxd 26/5/04 4:26 pm Page 137

Correlation 137

3 For each of the following sets of data:

(a) draw a scatter diagram,
(b) calculate the product moment correlation coefficient
between x and y.
(i) x 1 3 6 10 12
y 5 13 25 41 49
(ii) x 1 3 6 10 12
y 44 34 24 14 4
(iii) x 1 1 3 5 5
y 5 1 3 1 5
(iv) x 1 3 6 9 11
y 12 28 37 28 12
4 The diameters of the largest lichens growing on gravestones
were measured.
Age of gravestone Diameter of lichen
(x years) (y mm)
9 2
18 3
20 4
31 20
44 22
52 41 7
53 35
61 22
63 28
63 32
64 35
64 41
114 51
141 52

(a) Plot a scatter diagram to show the data.

(b) Calculate the values of
x and y
and show these as vertical
and horizontal lines.
(c) Find the values of Sxx, Syy and r.
5 A metal rod was gradually heated and its length, L, and
temperature, T, were measured several times.
Temperature, Length,
T (˚C) L (cm)
14 100.0
21 103.8
24 100.6
29 111.0
37 116.1
40 119.9
Chp-07 124-144.qxd 26/5/04 4:26 pm Page 138

138 Correlation

(a) Draw a scatter diagram to show the data, plotting L

against T.
(b) Find the value of r, the product moment correlation
coefficient.
(c) It is suspected that a major inaccuracy may have occurred
in one or more of the recorded values. Discard any
readings which you consider may be untrustworthy and
find the new value for r.
Comment on your results.

6 In a workshop producing hand-made goods a score is

assigned to each finished item on the basis of its quality
(the better the quality the higher the score). The number of
items produced by each of 15 craftsmen on a particular day
and their average quality score are given below.

No. of items Average quality

Craftsman produced, x score, y
1 14 6.2
2 23 7.3
3 17 4.9
4 32 7.1
5 16 5.2
6 19 5.7
7 17 5.9
8 25 6.4
9 27 7.3
10 31 6.1
11 17 5.4
12 18 5.7
13 26 6.9
14 24 7.2
15 22 4.8

(a) Draw a scatter diagram to show the data.

(b) Calculate the product moment correlation coefficient
between x and y.
(c) The owner of the firm believes that the quality of the
output is suffering because some of the craftsmen are
working too fast in order to increase bonus payments.
Explain to him the meaning of your results, and state
what evidence, if any, they provide for or against his
belief.
Chp-07 124-144.qxd 26/5/04 4:26 pm Page 139

Correlation 139

7 During the summer of 1982 the National Leisure Council, on

behalf of the Government, conducted a survey into all aspects
of the nation’s leisure time. The table shows the amount
spent per month on sporting pastimes and the total amount
spent per month on all leisure activities for a random sample
of 13 young married men.
Man Amount on sport, x Total amount, y
A 9.0 50.1
B 4.2 46.6
C 12.9 52.4
D 6.1 45.1
E 14.0 56.3
F 1.5 46.6
G 17.4 52.0
H 10.2 48.7
I 18.1 56.0
J 2.9 48.0
K 11.6 54.1
L 15.2 53.3
M 7.3 51.7

(a) Draw a scatter diagram for this data.

(b) Calculate the product moment correlation coefficient for 7
the data.
(c) Comment, with reasons, upon the usefulness, or
otherwise, of the above correlation analysis.
8 A clothing manufacturer collected the following data on the
age, x months, and the maintenance cost, y (£), of his sewing
machines.
Machine Age, x Cost, y
A 13 24
B 75 144
C 64 110
D 52 63
E 90 240
F 15 20
G 35 40
H 82 180
I 25 42
J 46 50
K 50 92

(a) Plot a scatter diagram of the data.

(b) Calculate the product moment correlation coefficient.
(c) Comment on your result in (b) by making reference to
the scatter diagram drawn in (a).
Chp-07 124-144.qxd 26/5/04 4:26 pm Page 140

140 Correlation

9 The following data relate to a random sample of 15 males,

all aged between 40 and 60 years. The measurements given
are the level of heart function (out of 100), the percentage of
baldness and the average number of hours spent watching
television each day.

Male Heart function Baldness (%) Hours of TV

1 42 83 6.2
2 65 66 2.2
3 86 32 1.8
4 32 74 8.3
5 56 69 7.6
6 48 74 6.5
7 92 25 0.8
8 78 30 5.9
9 68 32 2.2
10 52 54 4.4
11 53 58 4.6
12 69 76 2.7
13 57 63 5.8
14 89 38 0.2
15 65 41 4.6

(a) Calculate the value of the product moment correlation

coefficient between heart function and percentage
baldness.
(b) Calculate the value of the product moment correlation
coefficient between heart function and average number
of hours of television watched per day.
(c) Comment on the values of the correlation coefficients
found in (a) and (b) and interpret your results.
(d) Do you consider that males aged between 40 and 60
should be advised to reduce the number of hours that
they spend watching television in order to ensure a
better heart function? Explain your answer.
Chp-07 124-144.qxd 26/5/04 4:26 pm Page 141

Correlation 141

Key point summary

1 A scatter diagram should be drawn to judge p124
whether correlation is present.
2 The product moment correlation coefficient, p128, 129, 131
xy
xy  
n Sy
r
or x
S
xxSyy

{ (x)2
}{
x   y  
2
n
2 (y)2
n }
Remember, this can be found directly using a
calculator.
r is a measure of linear relationship only and
1  r  1
Do not refer to r if a scatter diagram clearly shows a
non-linear connection.
3 r
1 or r
1 implies that the points all exactly p129
lie on a straight line.
r
0 implies no linear relationship is present.
But … no linear relationship between the variables
does not necessarily mean that r
0.
4 Even if r is close to 1 or 1, no causal link should p131 7
be assumed between the variables without thinking
very carefully about the nature of the data involved.
Remember the feet stretching! Will it really help you
to get better at maths?
Chp-07 124-144.qxd 26/5/04 4:26 pm Page 142

142 Correlation

Test yourself What to review

1 Which of the following could not be a value for a product Section 7.4
moment correlation coefficient?
(a) r
0.98,
(b) r
0.666,
(c) r
1.2,
(d) r
0.003.
2 Which of the following scatter diagrams has a corresponding Section 7.4
product moment correlation coefficient given which is not
appropriate?
(a) r
0.86
y

(b) r
0.784
y

(c) r
0.145
y

x
Chp-07 124-144.qxd 26/5/04 4:26 pm Page 143

Correlation 143

Test yourself (continued) What to review

3 For the following data, plot a scatter diagram and evaluate the Section 7.4
product moment correlation coefficient.
x 8 6 5 2 1 3 6
y 9 8 8 0 5 2 7

4 Explain the meaning of spurious correlation with reference to Section 7.4

the following statement:
‘Between 1988 and 1998, the product moment correlation
coefficient between the number of incidents of violent juvenile
offences taken to court each year and the average number of
hours per week which 16- to 19-year-olds spent watching
television was found to be 0.874, indicating a high level of
correlation.’
5 The weight losses for ten females enrolled on the same Watch Section 7.4
and Weight course at a local Sports Centre are given below.

Weeks on course Weight loss (kg)

5 7.6
15 23
12 19.6 7
3 1.2
10 17.4
8 15.2
20 25.5
10 14
5 2.4
8 9.5

(a) Plot a scatter diagram.

(b) Evaluate the product moment correlation coefficient and comment on its value, referring
also to the scatter diagram.
1 (c).
2 (c).
3 r
0.904.
Graph of y against x
y
8
4
4 0 4 8 x
4
8
4 Spurious refers to the fact that the link between the two variables may
not be causal. They may be two effects from a different cause.
5 (a) Graph of weight loss against time on course
y
25
20
Weight
15
loss
(kg)
10
5
0
0 4 8 12 16 20 x
Time on course (weeks)
(b) r
0.936. Strong positive correlation. Some suggestion from
scatter diagram that weight loss is reaching a peak.
ANSWERS Test yourself
Correlation 144
Page 144 4:26 pm 26/5/04 Chp-07 124-144.qxd