Unit 7

Factors affecting reading

 Introduction

Introduction
In Unit 6, we met some statistical techniques which enabled us to compare
the truancy rate in large secondary schools in the East of England with the
general truancy rate experienced by all secondary schools in the East of
England. The method employed was to analyse sample data and use the
results of the analysis to make inferences about the population from which
the sample was drawn. In particular, we saw how the sign test enabled us
to decide whether or not to reject, at the 5% significance level, the
hypothesis that the population median takes a particular value.
Medians are just one measure of location. In this unit we return to
hypotheses about location, and you will meet another hypothesis test, the
‘z-test’, which concerns means, rather than medians. For much of the unit
we will be dealing with situations where we have a sample from a single
population, as in Unit 6. We will then develop the ideas of hypothesis
testing so as to compare two populations in terms of their locations. This
involves setting up a hypothesis about the locations of the two populations
(means, here, rather than medians) – the most common hypothesis is that
the locations of the two populations are equal. A random sample of data is
taken from each population, and these data are analysed to see whether or
not to reject the hypothesis. Such tests are called ‘two-sample tests’, in
contrast to ‘one-sample tests’ in the case of one population.
The emphasis will be on the development of statistical techniques, and, as
in Unit 6, we shall explore many of the ideas in the context of a question
taken from the general area of education. This time we shall be looking at
the achievement of 7- and 8-year-old children in reading:
What factors affect a child’s reading ability?
Section 1 starts with a brief discussion of this question. We shall then look
at an available source of data and identify what aspects of the general
question we can consider.
The next step will be to define specific questions of interest and use them
to set up appropriate hypotheses. We will then begin to develop an
appropriate sample statistic – a test statistic – with which to perform our
hypothesis tests, by revisiting the idea of sampling distributions (as
introduced in Unit 4) in Section 2. This notion will lead us to consider a
particular distribution known as the ‘normal distribution’. In Section 3, we
look closely at this distribution, which is of great importance in statistics.
In Section 4, we go on to consider how the normal distribution helps us to
define a usable test statistic, along with its sampling distribution.
Section 5 is concerned with the application of the resulting z-test to the
analysis of a sample of data from one population. Section 6 extends these
ideas to investigate the difference between the means of two populations.
One important aspect of these z-tests is that they are suitable only for
dealing with (quite) large samples of data.

95
Unit 7 Factors affecting reading

In Section 7, you will use Minitab to perform z-tests and learn to interpret
the resulting p-values. Section 8 draws some conclusions about the
educational question raised in the first section, and makes some general
points about z-tests.
Section 7 directs you to the Computer Book. You are also guided to the
Computer Book at the end of Subsections 3.1 and 3.3.

1 Clarifying the question

In Unit 6, the modified modelling diagram was introduced.

1. 2.
CLARIFY COLLECT
question data from
about population random sample

4. 3.
INTERPRET ANALYSE
results data
for population from sample

The modified modelling diagram (Figure 2 from Unit 6)

In this section, you are going to consider the first two stages of the
modified modelling process: clarify question and collect data.

1.1 The question to be clarified

The question What factors affect a child’s reading ability? is rather too
general for us to attempt to answer it straight away. We need to make it
more explicit. The first step is to understand what is meant by reading
ability.

Activity 1 Measuring reading ability

How would you measure reading ability? Write down two or three
measures of reading ability of 7- and 8-year-old children that you might
use.

96
 1 Clarifying the question

There are various different reading tests available to teachers, and they
normally combine several of the measures mentioned in the solution to
Activity 1. We shall be using data that have already been collected for us,
so the measures used have already been defined.
The next step is to consider the factors that might affect a child’s reading
ability.

Activity 2 Factors affecting a child’s reading ability

Write down some factors that you think might affect a child’s ability to
read.

The data that we shall use to explore this area will not allow exploration
of all these factors. Therefore the data have to be examined before a A reading class
decision can be made as to which factors can be explored and what
questions can be asked.

1.2 The data to be used

We shall be looking at the population of British children aged 7 and 8.
The sample we shall use consists of 7- and 8-year-old children of a certain
group of parents defined as follows. At least one of the child’s parents – a
‘cohort member’ – was born in a particular week in April 1970, resides in
Great Britain, and has been part of a long-term study known as the
British Cohort Study (BCS). There were more than 17 000 such people.
The BCS had its origins in what was called the British Births Survey,
which was originally designed to examine the social and biological
characteristics of the cohort member’s mother. That study looked at
neonatal morbidity, and its results were compared with those of a similar,
earlier study, the National Child Development Study, carried out in 1958.
‘Neonatal morbidity’ refers to disease of the child (the cohort member) in
its first month of life.
Since 1970, the aims of the BCS have broadened considerably. There have
been eight follow-up surveys, or ‘sweeps’, carried out in 1975, 1980, 1986,
1996, 1999–2000, 2004–2005, 2008–2009 and 2012. The follow-up surveys
attempted to trace the original sample and, in the case of the first two
follow-ups, to include immigrants born in the same week as the original
sample. Each follow-up survey looked at different areas of the original
group’s development into adulthood. The one which included reading skills A book based on the results of
of the cohort member’s children, and from which the data we shall be using the BCS and similar studies
have been taken, was the one carried out by the Centre for Longitudinal
Studies at the Institute of Education, University of London, in 2004–2005.
At that time, the age of the people under study was 34 years.

97
Unit 7 Factors affecting reading

We are therefore going to concentrate on data relating to the reading

ability of children who were 7 and/or 8 years old in 2004–2005 and whose
parents were part of the BCS. The 2004–2005 sweep of the BCS provided
data on 745 children aged 7 or 8 in total. Of these, only 679 were tested
for their reading ability. It is this sample of 679 that we shall concentrate
on in this unit. As we progress through the analysis, you will find that we
shall be using sample sizes smaller than this, because not all the additional
information needed was provided in the answers to the questionnaire.
However, the sample sizes concerned will remain pretty large.

Activity 3 Is it a random sample?

Write down some reasons why this sample of children can or cannot be
considered a random sample of the population of 7- and 8-year-old children
in Great Britain in 2004–2005.

We shall return to the issue of randomness of the sample in Section 8, but

for most of the unit, despite our doubts, we shall assume that it is
acceptable to treat the sample as if it were a random sample.
We next need to consider what data we have available. Table 1 shows data
on the first few children in the sample. In the first column is the child’s
reading ability as scored using a standard reading test called the BAS II
Word Reading Ability Score, where BAS stands for ‘British Ability Scales’.
(BAS II was, in fact, updated to BAS 3 in 2011.) This value will be
referred to simply as the child’s ‘reading score’ in this unit. The second
column gives the child’s age in months.
The remaining columns of Table 1 are in coded form; that is, they use
simple numerically coded values to represent attributes of the child in
place of more complicated wordings, ranges of numbers or exact numerical
values. For example, the third column shows the gender of the child, coded
as 1 for a boy, 2 for a girl. The fourth column again relates to age; this
time whether the child is aged 7 or 8 is recorded. The fifth column, headed
‘Parental education’, actually shows whether the cohort member’s
partner/spouse finished full-time education by the age of 16 or at some age
over 16; it is used here as a measure of the level of education of the child’s
parents. The sixth column shows the occupation of the child’s father. The
codes for the values in columns three to six are given beneath the main
body of the table.

98
 1 Clarifying the question

Table 1 Part of the dataset on reading from BCS 2004–2005

Reading Age Gender Coded Parental Father’s

score (months) age education occupation
106 91 1 1 1 –
123 95 2 1 1 1
123 86 2 1 – 1
110 92 1 1 1 1
92 90 2 1 2 –
129 93 2 1 1 –
118 97 1 2 – 2
115 107 2 2 1 2
117 93 2 1 2 –
134 89 1 1 1 1
25 85 1 1 – 2
110 93 2 1 1 1
172 94 1 1 1 1
138 90 2 1 – 2
56 105 1 2 – 1
136 100 2 2 1 1
115 90 2 1 1 –
160 94 1 1 1 1
.. .. .. .. .. ..
. . . . . .

(This data is copyright and owned by the Economic and Social Data Service.)

‘Gender’, 1: boy; 2: girl. ‘Coded age’, 1: 7 years old; 2: 8 years old.

‘Parental education’, 1: finished aged 16 or less; 2: finished aged over 16.
‘Father’s occupation’, 1: managerial, technical, professional and skilled
non-manual occupations; 2: skilled manual, partly skilled and unskilled
occupations.
You will notice that in some cases information is missing in the sample
data. This is to be expected, because some people either cannot or do not
wish to answer specific questions in the questionnaire. The missing data
will just be ignored for now, but we will return to a brief consideration of
its possible effects in Section 8.
A number of factors may have an effect on a child’s reading ability. With
our choice of data, the factors we can consider are child’s age, child’s
gender, parental education and father’s occupation.
Consider ‘child’s gender’ first. What precise question can we ask? It must,
as usual, be about the appropriate population (that of British children
aged 7–8 in 2004–2005) and not merely the sample. We might ask, Within
this population, do boys and girls differ in their reading ability? But we
should be more precise. As in Unit 6, we shall be looking for a difference in
location, in this case between reading scores of boys and girls.

99
Unit 7 Factors affecting reading

In the next subsection we shall be more precise about the particular

measure of location to use, but for now a reasonably precise question is:
For British children aged 7–8 in 2004–2005, did boys’ and girls’
ability scores differ in location?
Similar questions can be asked about the other factors.
For British children aged 7–8 in 2004–2005, did reading scores differ
in location according to their father’s occupation?
For British children aged 7–8 in 2004–2005, did reading scores differ
in location according to the level of parental education?
The questions can also be made more focused. For instance, consider the
first question again. Perhaps there is a difference between boys and girls
aged 7, but no such difference for 8-year-olds. Because of possibilities like
this, it may be more appropriate to consider the two age groups separately,
asking
For British children aged 7 in 2004–2005, did boys’ and girls’ reading
scores differ in location?
as well as
For British children aged 8 in 2004–2005, did boys’ and girls’ reading
scores differ in location?
The questions on education and occupation could also be split up
according to age, in a similar way.

1.3 Setting up the hypotheses

We shall try to answer most of these very specific questions by means of
hypothesis tests. Let us then remind ourselves what is involved by
referring back to the procedure for the sign test discussed in Unit 6, but
setting it up in a more formal way.
We began by making a statement about the population of interest that we
wished to test. In particular, this was the hypothesis that the population
median was equal to a specified value, M . The hypothesis that the
population median is equal to M is known as the null hypothesis. This
hypothesis is usually denoted by the symbol H0 . Thus the null hypothesis
in the case of the sign test can be stated precisely in the form
H0 : Population median = M.
We then looked at the data to see if there was any evidence that the
population median did not, in fact, equal M . If there is evidence against
the truth of the null hypothesis, H0 , then we reject this hypothesis and we
conclude that there is evidence that the population median is not equal to
M . That the population median is not equal to M is called the
alternative hypothesis. An alternative hypothesis is usually denoted by

100
 1 Clarifying the question

the symbol H1 . Thus, if we reject the null hypothesis

H0 : Population median = M,
then we are left with the alternative hypothesis
H1 : Population median =
% M,
where ‘%=’ is the symbol for ‘is not equal to’, and we say we are rejecting
the null hypothesis in favour of the alternative hypothesis.
In Unit 6, a trial in a law court was used as an analogy to hypothesis
testing. In that context, the null hypothesis is that ‘the defendant is not
guilty’, while the alternative hypothesis is that ‘the defendant is guilty’. If
the evidence against the null hypothesis is sufficiently great, then the jury
should reject that hypothesis in favour of the alternative hypothesis, and
conclude that the defendant is guilty.
Returning to the questions concerning children’s reading ability, the first
step therefore is to set up the appropriate null and alternative hypotheses.
As you might expect, these correspond to whether or not there is a
difference in location in reading ability between two groups of children. We
shall start with one of the questions on gender.
For British children aged 7 in 2004–2005, did boys’ and girls’ reading
scores differ in location?
However, before defining the hypotheses, it is worth thinking again about
the data. We actually have data on reading scores and gender for
396 children who are aged 7 (of these, 206 are boys and 190 are girls).
Printing all 396 scores here would clearly be cumbersome and waste space.
We can summarise the data as shown in Table 2.
Table 2 Summary statistics for data on reading scores of 7-year-old children

Sample size Sample mean Sample standard deviation

Boys 206 109.31 27.671
Girls 190 113.42 25.464
(This data is copyright and owned by the Economic and Social Data Service.)

You may well be wondering why the summary measures are the mean, x,
and standard deviation, s, and not some other measures of location and
spread, such as the median, M , and interquartile range, IQR. A minor
reason is that the mean and standard deviation are commonly used in
practice, so more people are familiar with them than with other measures.
The main reason, though, is that x and s can be used to construct a
reasonably simple test, in a way that M and IQR cannot.

101
Unit 7 Factors affecting reading

Activity 4 Calculating a mean and standard deviation

Because it is some time now since you worked with the sample mean (x)
and the sample standard deviation (s), here is a reminder of how to
calculate these summary measures:
!
x
x= ,
n
!
where x is the sum of all the sample values and n is the sample size; and
√
s = variance,
!
(x − x)2
where the variance is .
n−1
(a) Data on the first eighteen 7- and 8-year-olds taken from the BCS
2004–2005 results were given in Table 1 (in Subsection 1.2). Extract
from that table the values of the reading scores for all the 7-year-old
boys. What is the value of n for this small sample?
(b) Calculate x and s for the reading scores for 7-year-old boys that you
extracted in part (a).

Having paused briefly to examine the sample data, we now move on. We
still need to state the null and alternative hypotheses associated with the
question
For British children aged 7 in 2004–2005, did boys’ and girls’ reading
scores differ in location?
in their precise forms. The null hypothesis will be
H0 : For British children aged 7 in 2004–2005, the mean reading score
for girls was equal to the mean reading score for boys.
As you will have noticed, H0 is phrased in terms of the population means
and not, for example, the population medians. The alternative hypothesis
is naturally taken to be
H1 : For British children aged 7 in 2004–2005, the mean reading score
for girls was not equal to the mean reading score for boys.
The null and alternative hypotheses for other questions listed at the end of
the previous subsection are similar.
The British Ability Scales reading score system gives overall mean test
scores for different age groups in Great Britain. These overall mean test
scores are given for quite finely defined age groups, from which the authors
of this unit have come up with the following means for 7- and 8-year-olds:
the population mean for 7-year-old children is 96, and for 8-year-olds it is
116. (Actually, these means come from very large samples of children and
not the whole population, but in practice we can treat them as population
means.) So a further appropriate question to ask about the data on

102
 1 Clarifying the question

reading scores for 7-year-old children, for example, is whether they are
consistent with a population mean of 96. In other words, we could test the
following hypotheses:
H0 : For British children aged 7 in 2004–2005, the mean reading score
was equal to 96
H1 : For British children aged 7 in 2004–2005, the mean reading score
was not equal to 96.

In testing hypotheses about population medians in Unit 6, the next step

was to define a quantity that we could calculate from the data that would
help us to evaluate the truth or otherwise of the null hypothesis. In the
law-trial analogy, this is the evidence. In the sign test, this quantity was
the smaller of the numbers of [+]s and [−]s that the sample contains. (See
Section 4 of Unit 6.) In general, in hypothesis testing, this quantity is
called the test statistic. So now we need to find suitable test statistics to
assess the hypotheses about children’s reading abilities. Since these
hypotheses are about population means or differences between population
means, the obvious test statistics would involve sample means or the
differences between sample means. But, as with the sign test in Unit 6, the
awkward part involves finding what is called the sampling distribution of
the test statistic; so in the next section we look again at sampling
distributions.

Exercises on Section 1
Exercise 1 Mean and standard deviation for 8-year-olds
(a) Extract from Table 1 the values of the reading scores for all the
8-year-old children in the table. What is the value of n for this small
sample?
(b) Calculate x and s for the reading scores for 8-year-old children that
you obtained in part (a).

Exercise 2 Parental education and occupation

(a) Extract the values of the reading scores for all the children in Table 1
whose parent’s age on finishing full-time education was 16 or less and
whose father’s occupation is managerial, technical, professional or
skilled non-manual. What is the value of n for this small sample?
(b) Calculate x and s for the reading scores for the sample of children
that you obtained in part (a).

Exercise 3 Null and alternative hypotheses?

Suggest null and alternative hypotheses for comparing the reading abilities
of the 8-year-old children according to their gender.

103
Unit 7 Factors affecting reading

2 Sampling distributions revisited

In Subsection 3.3 of Unit 4, you saw what is meant by the sampling
distribution of the median of a sample, and what happened to such
sampling distributions as the sample size increased. We now review these
ideas, but rather than just repeating exactly what was done before, we
look at the sampling distribution of the mean as opposed to that of
the median.
As in Unit 4, in order to look at these sampling distributions precisely, we
really need to know all the relevant information about the whole
population. Nobody has information about the reading ability of all 7- and
8-year-old children in Great Britain, so we cannot work with data exactly
like those from the BCS. Instead let us look at a population where we
do have data on everyone, and investigate sampling distributions using
that. The population is that of all students taking the examination for the
Open University module Exploring mathematics (MS221) in a particular
presentation. There were 1234 students in the presentation chosen, and
their marks in the examination are displayed in Figure 1. This plot is very
like a histogram with lines instead of bars. (Histograms were introduced in
Subsection 1.5 of the Computer Book.) The numbers of students achieving
each mark from 0 to 100 are given by the heights of the lines drawn at
each mark. These heights are the same as the areas of the bars that would
have been used on the histogram. But, in addition, the top ends of the
lines have been joined together.
This representation gives a good picture of the shape of the population
distribution of examination marks of students on MS221 in one
presentation.
Number of students obtaining each mark

30

25

20

15

10

0
0 10 20 30 40 50 60 70 80 90 100
Mark
Figure 1 Numbers of students obtaining each examination mark in MS221

104
 2 Sampling distributions revisited

Now, there is a modification that we need to make. What will be

important later are the proportions of students in the population gaining
each mark. Thus instead of using the vertical axis to measure the actual
number of students who obtained each mark in the examination, we want
the population distribution to be described in terms of the proportion of
students in the population who obtained each mark. We can do this
simply by dividing each of the actual numbers represented in Figure 1 by
the total number of students in the population (1234). Hence,
1
1 becomes " 0.0008,
1234

2
2 becomes " 0.0016,
1234

3
3 becomes " 0.0024,
1234
and so on.

Activity 5 From number to proportion

The actual number of students scoring 75 marks in Figure 1 is 21. What

proportion of students on MS221 in the presentation in question achieved
75 marks?

The result of changing from actual numbers to proportions is shown in

Figure 2. Notice that Figure 2 looks just the same as Figure 1; only the
scale on the vertical axis has changed.
Proportion of students obtaining each mark

0.025

0.02

0.015

0.01

0.005

0
0 10 20 30 40 50 60 70 80 90 100
Mark
Figure 2 Proportions of students obtaining each examination mark in MS221

105
Unit 7 Factors affecting reading

However, it is not the characteristics of the population distribution of

exam marks, above, that we are interested in as such. Our focus is going
to be on the sampling distributions of means of random samples of exam
marks taken from this population. This is because we will be interested in
testing hypotheses about the mean examination mark, such as
H0 : For students on MS221, the mean examination score
is equal to 65
H1 : For students on MS221, the mean examination score
is not equal to 65
or (using data from other years)
H0 : For students on MS221, the mean examination score for the
current presentation is equal to the mean examination score
for the previous presentation
H1 : For students on MS221, the mean examination score for the
current presentation is not equal to the mean examination score
score for the previous presentation.
Now we begin our investigation of the sampling distribution of the mean.
Consider first all possible random samples of size 2 that we might select
from the population data of 1234 examination marks. There is a great
number of possibilities (760 761, to be precise!), and we cannot concisely
picture all the sample values in every one of these possible samples.
However, as in Unit 4, we can summarise each sample using a summary
measure, and then picture these in the form of the sampling distribution of
that summary measure. This time, as suggested above, we use the sample
mean as our summary measure.

Activity 6 Sample means of samples of size 2

(a) Find the sample means of each of the following samples of size 2:
(i) 15, 35 (ii) 65, 77 (iii) 65, 52 (iv) 37, 80.
(b) The exam marks in the population, and hence in any sample, are all
integers (whole numbers). Are the sample means of samples of size 2
necessarily integers? If not, what other kinds of value can these
sample means take?

In this way it would be possible to calculate the sample mean for every one
of the 760 761 possible samples of size 2. Different samples can give the
same sample mean, as (iii) and (iv) in part (a) of Activity 6 illustrate. The
sampling distribution records the proportions of all these samples with
each value of the sample mean. A picture of this is shown in Figure 3.
This represents the sampling distribution of the mean for samples of size 2
from the population of exam marks. Here all the possible values of the
sample mean x are indicated on the horizontal axis, and the vertical lines
represent the heights of the bars that would be used for a histogram of the

106
 2 Sampling distributions revisited

proportion of samples (out of 760 761 possibilities) which have each of

these values as the sample mean. Notice that there are many more lines in
this diagram than there are in Figure 2. That is because this sample mean
can take about twice as many values, integers and half-integers, as you saw
in Activity 6, so the histogram can have twice as many bars. Joining the
tops of the lines again provides us with a good picture of the shape of the
distribution.
0.028

0.024

0.02
Proportion

0.016

0.012

0.008

0.004

0
0 10 20 30 40 50 60 70 80 90 100
x
Figure 3 Sampling distribution of the mean for samples of size 2 from the
population of MS221 exam marks

Activity 7 Distribution of sample means of size 2

What are the main features of the distribution of sample means of size 2
shown in Figure 3?

Let us now find out, as in Unit 4, what happens to the sampling

distribution as the sample size increases. Let’s first look at the sampling
distribution of the mean for samples of size 3.

Activity 8 Sample means of samples of size 3

Find the sample mean of each of the following samples of size 3:

(a) 10, 20, 45 (b) 82, 24, 33 (c) 52, 61, 73 (d) 78, 64, 46.

Activity 8 indicates that there are even more possible values of the sample
mean for samples of size 3 than there are for samples of size 2. This means
that the vertical lines in the sampling distribution will be even closer
together. For this reason, we stop plotting the lines and just concentrate
on the shape of the distribution as indicated by the tops of the lines; we

107
Unit 7 Factors affecting reading

obtain the picture of the sampling distribution shown in Figure 4. In fact,

the ‘joining’ line shown in Figure 4 is made up of lots of very short lines,
each one joining two adjacent vertical lines.
0.036
0.03
0.024

Proportion
0.018
0.012
0.006
0
0 10 20 30 40 50 60 70 80 90 100
x
Figure 4 Sampling distribution of the mean for samples of size 3 from the
population of MS221 exam marks

Activity 9 Distributions of sample means of sizes 2 and 3

How does the distribution of sample means of size 3 shown in Figure 4

compare with the distribution of sample means of size 2 shown in Figure 3?

Activity 10 Distributions of sample means of sizes 3 and 5

Figure 5 shows the distribution of sample means of size 5 from the

population of MS221 examination marks. How does the distribution shown
in Figure 5 compare with the distribution of sample means of size 3 shown
in Figure 4?

0.05

0.04
Proportion

0.03

0.02

0.01

0
0 10 20 30 40 50 60 70 80 90 100
x
Figure 5 Sampling distribution of the mean for samples of size 5

108
 2 Sampling distributions revisited

Activity 11 Distributions of sample means of larger sample

sizes
Figure 6 contains pictures of the sampling distributions of the mean for
larger sample sizes. Notice that we have not indicated the scale on the
vertical axes in Figure 6, but it is the same in each case. Describe the
changes in shape of these distributions, as the sample size n increases.
Proportion

0 10 20 30 40 50 60 70 80 90 100 x
(a) n = 10
Proportion

0 10 20 30 40 50 60 70 80 90 100 x
(b) n = 20
Proportion

0 10 20 30 40 50 60 70 80 90 100 x
(c) n = 50
Figure 6 Sampling distributions of the mean for samples of size n

109
Unit 7 Factors affecting reading

The common shape of the distributions in Figures 4, 5 and 6 is sometimes

called a ‘bell shape’. You can use the following picture of Big Ben to
decide whether or not you agree that these distributions are bell-shaped!

Big Ben: bell-shaped?

Now the interesting thing about the sampling distribution of the mean is
that it will nearly always be approximately bell-shaped (looking something
like the above figures), no matter what population distribution is taken as
the starting point. (The sampling distributions of some other quantities,
such as the sample median, show similar features.)

Example 1 Sampling distributions of means based on

earnings data
Figure 7 provides a rough picture of the population distribution of
earnings of all full-time employees in the UK in 2011.
Proportion

0 500 1000 1500 x

Weekly gross earnings (£)
Figure 7 Population distribution of earnings of full-time employees

110
 2 Sampling distributions revisited

This population distribution is very smooth. The smoothness results

from the fact that the population is extremely large and there are so
many possible earnings that we can record. This means that the
vertical lines representing the various adjacent proportions would be
so close together that we could not distinguish between them and so,
effectively, the line joining the tops is a smooth curve. The
distribution is, however, clearly right-skew since it has a long tail to
the right. This reflects the fact that while most employees earn a
moderate to ‘medium’ wage, some employees earn considerably more,
and a few earn very considerably more again.
Figure 8 contains pictures of the sampling distributions of the mean
for samples of various sizes from this population distribution.
Proportion

Proportion

Proportion
0 500 1000 1500 x 0 500 1000 1500 x 0 500 1000 1500 x
n=2 n=5 n = 10
Proportion

Proportion

0 500 1000 1500 x 0 500 1000 1500 x

n = 20 n = 30
Figure 8 Sampling distributions of means based on earnings data

Activity 12 Distributions of sample means of earnings data

Describe the main changes in shape of the sampling distributions in

Figure 8, as the sample size n increases.

111
Unit 7 Factors affecting reading

So, again, we see from Example 1 and Activity 12 that even though the
population distribution is skew, as the sample size n increases, the
sampling distribution of the mean becomes more and more symmetric and
bell-shaped.
What is surprising, though, is that if the sample size is large enough, the
sampling distribution of the mean will nearly always be this sort of shape,
no matter what shape the population distribution is.

The shape of sampling distributions of the mean

For most practical purposes, whatever the shape of the original
population distribution, the sampling distributions of the mean for
large enough sample sizes are always symmetric and bell-shaped.

These symmetric bell-shaped distributions that we obtain as sampling

distributions for large enough values of n are called normal
distributions.
As you will see in Section 3, these distributions have some very interesting
properties which help us to develop the test statistic that we are working
towards.

What is a large enough sample?

As a rough guide, you can assume that, whatever the population
distribution, for sample sizes greater than 25, the sampling
distribution of the mean will always be approximately normal, and in
practice, we generally assume that it is normal.

In fact, the sampling distribution of the mean will actually be

approximately normal for sample sizes (much) smaller than n = 25 for
many population distributions. On the other hand, there are atypical
population distributions for which the sampling distribution of the mean is
not (approximately) normal. You will not deal with samples from such
populations in M140. This allows us to rephrase a previously highlighted
statement.

112
 2 Sampling distributions revisited

The shape of sampling distributions of the mean,

rephrased
For most practical purposes, whatever the shape of the original
population distribution, the sampling distributions of the mean for
large enough sample sizes are always approximately normal.

Exercises on Section 2

Exercise 4 Means of samples of size 2 from two small

populations
Consider the following two small populations of values:
Population A: 10 20 30 40 and Population B: 10 38 39 40
(a) Find the sample mean of each of the six different samples of size 2
that you can obtain from Population A. Make a very rough plot of
the positions of the six sample means along the horizontal axis.
(b) Repeat what you did in part (a) for Population B.
(c) Compare the graphs you obtained in parts (a) and (b). Which of the
two displays a more bell-shaped distribution of sample means? Can
you think of a reason why this should be so?

Exercise 5 Change in shape as sample size changes?

The BCS sample with which we are concerned in this unit comprises a
total of 679 reading scores (of 7- and 8-year-old children in 2004–2005). We
will now pretend that this large sample of reading score values is actually
the entire population of reading score values. Figure 9 contains pictures of
the sampling distributions of the mean for samples of various sizes from the
(pseudo-)population distribution of reading scores. Describe the changes in
shape of these sampling distributions, as the sample size n increases.

113
Unit 7 Factors affecting reading

0.04
0.035
0.03

Proportion
0.025
0.02
0.015
0.01
0.005
0
(a) 10 60 110 160 210 x

0.028
0.024
0.02
Proportion

0.016
0.012
0.008
0.004
0
(b) 10 60 110 160 210 x
n=2
0.03

0.024
Proportion

0.018

0.012

0.006

0
(c) 10 60 110 160 210 x
n=3
0.05
0.04
Proportion

0.03

0.02

0.01
0
(d) 10 60 110 160 210 x
n = 10
Figure 9 Sampling distributions of the mean as sample size changes

114
 3 Normal distributions

3 Normal distributions
In Section 2 we saw that the sampling distribution of the mean is nearly
always approximately normal, provided the sample size is sufficiently large.
In this section we examine some of the properties of normal distributions
and begin to discover just how important sampling distributions really are.
But first, we need to introduce some important new terminology. You are
already familiar with the idea of a sample mean, x:
!
x sum of sample values
x= = .
n sample size
In this section we shall also need to refer to the population mean. For a
population of finite – but very large – size, N , this is calculated in exactly
the same way, but using all the data values in the population. By
convention it is labelled µ, so that
!
x sum of population values
µ= = .
N population size
(The symbol µ is the lower-case Greek letter ‘mu’, pronounced to rhyme
with ‘new’.) Because N is often very large indeed, the population is often
actually assumed to be of infinite size. For an infinite population, the
population mean value, µ, is the mean of a truly enormous sample – the
sample size must approach infinity.
There is a similar distinction between the sample standard deviation, s,
and the population standard deviation, which is denoted by the
symbol σ. (This is the!lower-case version of the Greek letter ‘sigma’, which
in upper-case form is , but there is no connection between the ways that
these two symbols are used here.) The formulas are
&! '
! 2 !
(x − x)2 x − ( x)2 /n
s= = ,
n−1 n−1
where the summations are over sample values, and
&! '
! 2 !
(x − µ) 2 x − ( x)2 /N
σ= = ,
N N
where the summations are over population values.
An important property to note is that σ is always a positive number.

115
Unit 7 Factors affecting reading

Off duty from their work in statistics

class µ and σ take a much needed
Spring Break in their native Greece

3.1 Normal distributions: location and

spread
Normal distributions are important in statistics for two different reasons.
You met the first of these in Section 2: many sampling distributions of
summary statistics are approximately normal for large enough samples.
The other reason is that the distributions of many populations are
approximately normal. One example is the population of men’s heights
that you will look at below. In Unit 10 you will see further examples of
population distributions that are approximately normal.

Importance of the normal distribution

Normal distributions are important both as (approximate) population
distributions, in some cases, and as (approximate) sampling
distributions, in many more cases.

The distribution is called the normal distribution because it arises so

commonly. Normal distributions are also called Gaussian distributions
after the great German mathematician and scientist C.F. Gauss, who was
instrumental in their development. They also appear in popular literature
as the ‘bell curve’.

116
 3 Normal distributions

Carl Friedrich Gauss

Gauss (1777–1855) was a phenomenal mathematician – one of the
most productive mathematicians ever. He made exceptional
contributions in many fields, perhaps number theory most notably,
but also astronomy, geometry, algebra, geophysics and, amongst
others, statistics. During the early part of his career he took up the
challenge of predicting where Ceres would be found. Ceres was a dwarf
planet that had been observed in 1801 but which then disappeared
behind the Sun and could not be found when it first reappeared.
Gauss developed new methods of estimation and approximation to
locate its position. He later published a monograph on the theory of
the motion of small planets disturbed by large planets, and in this he
introduced several important statistical concepts, including the
normal distribution. It is for this reason that the normal distribution
is also called the Gaussian distribution, though Gauss did not
contribute most to the development of its properties. (The
contribution of Laplace (1749–1827), for example, is greater.) Carl Friedrich Gauss
(1777–1855)

In this unit, we want to explore certain characteristics of normal

distributions in order to apply them to sampling distributions. It would be
possible to do this exploration using the sort of sampling distributions we
met in the last section. However, the descriptions of what is going on tend
to look rather complicated, because they involve means of sampling
distributions of means. To make things clearer, the exploration is therefore
done in the context of a normally distributed population.
Each normal distribution is a precise distribution defined by a
mathematical formula involving the mean and standard deviation. We
shall not need to use this formula in this module. But despite this
mathematical precision, in practice the word ‘approximate’ is very
important above. Real-world populations never have exact normal
distributions in terms of the mathematical formula; but many are close
enough to a normal distribution so that it makes sense to treat them as
having normal distributions, in which case we say they are approximately
normally distributed.
Figure 10 provides a picture of the population distribution of the heights of
all men in Scotland in 2008, based on information given by the Scottish
Health Survey, 2008.

117
Unit 7 Factors affecting reading

Proportion

1.50 1.55 1.60 1.65 1.70 1.75 1.80 1.85 1.90 1.95 2.00
Height (metres)
Figure 10 Population distribution of Scottish men’s heights (in metres)

This population distribution is very smooth, symmetric and bell-shaped.

For the rest of this section, we shall assume that the distribution is indeed
normal.
The symmetry of the distribution means that the population mean height
is the value corresponding to the mode (peak) of the distribution: about
1.75 metres. (In fact, as well as being the mode and the mean, this value is
also the population median!) This characteristic applies more generally
so that any normal distribution is symmetric about its mean µ.
Figure 11 shows normal distributions for different values of the mean µ
and Figure 12 shows normal distributions for different values of the
standard deviation σ.

118
 3 Normal distributions

−3 −2 −1 0 1 2 3 −2 −1 0 1 2 3 4 0 1 2 3 4 5 6 7 8
µ=0 µ=1 µ=5
(a) (b) (c)

0 17 18 19 20 21 22 23 −4 −3 −2 −1 0 1 2 3
µ = 20 µ = −1
(d) (e)
Figure 11 Normal distributions with different locations

µ=0

µ=0

−2 −1 0 1 2 −4 −3 −2 −1 0 1 2 3 4
(a) σ = 0.5 (b) σ=1

µ=0

−6 −4 −2 0 2 4 6
(c) σ=2
Figure 12 Normal distributions with different spreads

119
Unit 7 Factors affecting reading

The location of a normal distribution on the horizontal axis depends on

the value of its mean µ, as demonstrated by Figure 11.
As with any distribution, the spread of a normal distribution can be
measured by the standard deviation of the population, σ. Thus a small
value of σ means that the distribution is tightly clustered about the mean;
the larger the value of σ, the more spread out the distribution will be – as
demonstrated by Figure 12.

Activity 13 What are µ and σ for this normal distribution?

Figure 13 shows another normal distribution. By comparing it with

Figures 11 and 12, can you identify the values of µ and σ for this normal
distribution?

−5 −3 −1 1 3 5 7
Figure 13 A normal distribution related to those in Figures 11 and 12

Location and spread of the normal distribution: 1

The normal distribution has location specified by the population
mean, µ, and spread specified by the population standard deviation, σ.

You have now covered the material needed for Subsection 7.1 of
the Computer Book.
You have also now covered the material related to Screencast 1
for Unit 7 (see the M140 website).

120
 3 Normal distributions

3.2 Normal distributions: relating means,

standard deviations and plots
For a normal distribution, almost the whole of the distribution
(about 99.7%) is contained within plus or minus three standard deviations
of the mean. For example, the population distribution of Scottish men’s
heights (in metres) is normal with mean µ " 1.75 and standard deviation
σ " 0.07. Thus 3σ " 0.21, and so almost the whole of the distribution is
contained within plus or minus 0.21 metres of the mean 1.75 metres
(i.e. between 1.75 − 0.21 = 1.54 metres and 1.75 + 0.21 = 1.96 metres).
You can check for yourself in Figure 14 – which is an annotated copy of
Figure 10 – that this is indeed the case.

1.50 1.55 1.60 1.65 1.70 1.75 1.80 1.85 1.90 1.95 2.00
Height (metres)

1.54 1.75 1.96

0.21 0.21
Figure 14 Annotated population distribution of Scottish men’s heights

(Similar percentages are known for all other numbers of standard

deviations; for instance, 95.4% of the distribution is contained within plus
or minus two standard deviations, and 68.3% within plus or minus one
standard deviation.)

Location and spread of the normal distribution: 2

The normal distribution has its mode at µ, and almost the whole of
the normal distribution is contained between µ − 3σ and µ + 3σ.

The links between the graph of a normal distribution and its mean and
standard deviation suggest that a picture of the distribution can be used
to obtain approximate values for its mean and standard deviation.

121
Unit 7 Factors affecting reading

Example 2 Approximate values for the mean and

standard deviation
The population distribution of a certain variable x is known to be
normal. This distribution is pictured in Figure 15.

3 4 5 6 7 x
Figure 15 A normal distribution

The mode of this normal distribution occurs at about x = 5. This

means that the population mean must be approximately equal to 5.
So, µ " 5. We say approximately equal because µ may not be exactly
equal to 5. It could be 5.1 or 4.9; it is impossible to give an exact
value here.

3 4 5 6 7 x
3.5 6.5
1.5 1.5
Figure 16 Investigating the spread of a normal distribution

The dashed lines in Figure 16 indicate that almost all of the

distribution is contained between x = 3.5 and x = 6.5 (i.e. within

122
 3 Normal distributions

5 ± 1.5). This means that 3σ " 1.5, so σ " 0.5.

In summary, the normal distribution plotted in Figure 15 is
approximately the normal distribution with mean µ = 5 and standard
deviation σ = 0.5.

Activity 14 Approximate values for the mean and standard

deviation
For each of the normal distributions shown in the parts of this activity,
find approximate values for the mean and standard deviation, using the
method described above.
(a)

4 5 6 7 8 9 10 11 12 13 14 15 16 x
Figure 17 Another normal distribution
(b)

40 50 60 70 80 90 100 110 120 130 140 150 160 x

Figure 18 Yet another normal distribution

123
Unit 7 Factors affecting reading

(c)

0.5 0.6 0.7 0.8 0.9 1 1.1 1.2 1.3 1.4 1.5 x
Figure 19 And yet one more normal distribution

Conversely, knowing the mean and standard deviation of a normal

distribution enables us to make a rough sketch of the distribution. Any
sketch of a normal distribution will show a symmetric and bell-shaped
curve. More specifically, the distribution must be symmetric about the
mean. In addition, almost the whole of the distribution must be contained
within plus or minus three standard deviations of the mean.

Example 3 Sketching a normal distribution

The normal distribution of a variable x has mean µ = 15 and
standard deviation σ = 3. To sketch this distribution, draw a
The 1997 music CD
‘abnormally normal or symmetric, bell-shaped curve centred on the value of µ, which in this
normally abnormal?’ by the case is 15. The standard deviation is σ = 3, so that 3σ = 9. We
band ‘standard deviation’. therefore know that just about all the distribution is contained within
15 ± 9 (i.e. lies between 15 − 9 = 6 and 15 + 9 = 24). A sketch of the
distribution can therefore be drawn and should resemble Figure 20.

124
 3 Normal distributions

6 15 24 x
Figure 20 The normal distribution with µ = 15 and σ = 3

The scale that is used for the horizontal axis certainly affects the shape of
the normal distribution, as demonstrated by Figure 21. The important
thing, though, is that the information conveyed by the sketch remains
exactly the same.

6 15 24 x 6 15 24 x
Figure 21 The same normal distribution plotted on different horizontal
scales
Also, for the aspects we are investigating, the height of the distribution
does not really matter; all the information we require about the
relationship between the distribution and its mean and standard deviation
is provided by the scale on the horizontal axis. For this reason there is no
need to bother with a vertical scale at all.

Activity 15 Sketching a normal distribution

Sketch the following distributions:
• The normal distribution of a variable x with mean 1000 and standard
deviation 100.
• The normal distribution of a variable x with mean 2 and standard
deviation 0.25.

Activity 15 demonstrates that it always makes sense to think of the

horizontal axis of a normal distribution in terms of the number of standard

125
Unit 7 Factors affecting reading

deviations of the variable away from the mean. This is illustrated in

Figure 22, which is an important picture in understanding the normal
distribution. Notice how the horizontal scale is marked off using µ and σ.

µ − 3σ µ − 2σ µ−σ µ µ + σ µ + 2σ µ + 3σ
x
Figure 22 The normal distribution with its scale marked in terms of µ and σ

You have now covered the material related to Screencast 2 for

Unit 7 (see the M140 website).

3.3 The standard normal distribution

We can go one step further than that represented by Figure 22
(Subsection 3.2) and think of all normal distributions in terms of one
special normal distribution. This special normal distribution has mean
zero and standard deviation one, and is called the standard normal
distribution. It looks like Figure 23. Figure 23, in turn, looks like
Figure 22 with µ and σ in the labels on the horizontal axis replaced by 0
and 1, respectively: so, µ − 3σ has become 0 − (3 × 1) = −3, µ − 2σ has
become 0 − (2 × 1) = −2, and so on.

-3 -2 -1 0 1 2 3
x
Figure 23 The standard normal distribution

126
 3 Normal distributions

The standard normal distribution

The standard normal distribution is the particular normal distribution
that has mean µ = 0 and standard deviation σ = 1.

It turns out that we can transform all normal distributions to the standard
normal distribution.

Example 4 Transforming to the standard normal

distribution
The normal distribution of a variable x with mean µ = 10 and
standard deviation σ = 2 is illustrated in Figure 24.

5 10 15 x

6 6
Figure 24 The normal distribution with µ = 10 and σ = 2

First, we can shift the whole of the distribution to the left so that the
mode occurs at zero just by subtracting 10 from each value of x. This
is shown in Figure 25. It changes the location of the distribution but
leaves the spread unchanged.

127
Unit 7 Factors affecting reading

Subtract 10 from
each value of x

−5 0 5 10 15 x
Figure 25 Shifting the distribution of x

The dashed curve in Figure 25 is now a new normal distribution with

mean zero and standard deviation 2. This new distribution is the
distribution of the variable v, say, where v = x − 10. The normal
distribution of v differs from the standard normal distribution only by
having standard deviation 2 rather than 1. However, if we now think
of the horizontal axis in terms of the number of standard deviations of
v away from the mean, then we obtain Figure 26.

−6 −4 −2 0 2 4 6 v
(−3σ) (−2σ) (−σ) 0 (σ) (2σ) (3σ)
Figure 26 The normal distribution of v = x − 10 with mean 0 and
standard deviation 2

Then, dividing every value of v by the standard deviation 2 gives the

distribution of v/2. This distribution, shown in Figure 27, is the
standard normal distribution, as required.

128
 3 Normal distributions

−3 −2 −1 0 1 2 3 v/2
(−3σ) (−2σ) (−σ) 0 (σ) (2σ) (3σ)
Figure 27 The normal distribution of v/2 with mean 0 and standard
deviation 1

We have shown that if the variable x has a normal distribution with

mean 10 and standard deviation 2, then the variable v/2 = (x − 10)/2
has the standard normal distribution.

Example 4 is a specific example of the following general result. If we start

with a normal distribution for x, with mean µ and standard deviation σ,
then:
• By subtracting µ from each value of x we obtain the distribution of
v = x − µ. This distribution is normal with mean zero and standard
deviation σ.
• By then dividing each value of v by σ we obtain the variable z = v/σ,
which has the standard normal distribution.
Combining the formulas for z and v, we find that
x−µ
z= .
σ

Transforming a normal distribution to the standard normal

distribution
If a variable x has a normal distribution with mean µ and standard
deviation σ, then the variable
x−µ
z=
σ
has the standard normal distribution.

129
Unit 7 Factors affecting reading

Activity 16 Transforming some particular normal

distributions
For the normal distributions with the following values of µ and σ, write
down the appropriate formula to transform the variable x to the variable z
that follows the standard normal distribution.
(a) µ = 10, σ = 2 (b) µ = 100, σ = 20 (c) µ = 1, σ = 0.1

Activity 17 Transforming the distribution of Scottish men’s

heights
(a) Assume that the population distribution of Scottish men’s heights h
(in metres) is normal with mean µ = 1.75 and standard deviation
σ = 0.07. Write down the formula for z which transforms each value
of the variable h to the number of standard deviations from its mean.
(b) Calculate the value of z corresponding to each of the following values
of h (in metres). In each case, interpret your answer by completing a
sentence of the form ‘So a height of *** metres is *** standard
deviations *** the mean height of *** metres’.
h = 1.96; h = 1.61; h = 1.785.

Importance of the standard normal distribution

The development in this subsection implies that by describing every
normal distribution in terms of z, the number of standard deviations
by which the variable differs from its mean, we can think of all normal
distributions in terms of just one distribution: the standard normal
distribution.

130
 3 Normal distributions

Wall space increased at the Statistics Art Gallery

when it became clear that only one picture was
required in the ‘normal distribution’ collection.

You have now covered the material needed for Subsection 7.2 of
the Computer Book.
You have also now covered the material related to Screencast 3
for Unit 7 (see the M140 website).

131
Unit 7 Factors affecting reading

Exercises on Section 3

Exercise 6 Approximating the mean and standard deviation

Find approximate values for the mean and standard deviation of the
normal distribution shown below.

0 10 20 x
Figure 28 Yet again, a normal distribution

Exercise 7 Approximating another mean and standard

deviation
Find approximate values for the mean and standard deviation of the
normal distribution shown below.

−12 −5 2 x
Figure 29 And one more time, another normal distribution

Exercise 8 Sketching a normal distribution

The normal distribution of a variable x has mean −1 and standard
deviation 1. Sketch the distribution.

132
 3 Normal distributions

Exercise 9 Sketching another normal distribution

The normal distribution of a variable x has mean 4 and standard
deviation 4. Sketch the distribution.

Exercise 10 Obtaining z for a normal distribution

Write down the appropriate formula to transform the variable x to the
variable z that follows the standard normal distribution when
(a) x has the normal distribution with mean 6 and standard deviation 3.3;
(b) x has the normal distribution with mean −6 and standard deviation 2.

Exercise 11 Calculating z from x

Assume that x follows the normal distribution with mean µ = 2 and
standard deviation σ = 10. Write down the appropriate formula for z
which transforms the variable x to the number of standard deviations from
its mean. Calculate the value of z corresponding to x = 3.

Exercise 12 Calculating z from x for another normal

distribution
Assume that x follows the normal distribution with mean µ = −1 and
standard deviation σ = 0.5. Write down the appropriate formula for z
which transforms the variable x to the number of standard deviations from
its mean. Calculate the value of z corresponding to x = 0.

133
Unit 7 Factors affecting reading

4 Sampling distributions re-revisited

We now take a closer look at the sampling distributions of the sample
mean that you met in Section 2. As we said there, provided the sample
size is sufficiently large (roughly speaking, greater than 25), these sampling
distributions are approximately normal. Thus the ideas discussed in
Section 3, which apply to all normal distributions, apply (approximately)
to these sampling distributions as well. These ideas will enable us to find a
suitable test statistic to use for testing some of the hypotheses we are
interested in for the BCS survey.

We begin by examining the relationship between sampling distributions of

the mean and the original population distribution in a little more detail.

Activity 18 Means of distributions of sample means

(a) Consider again the population distribution of MS221 examination

marks which you met in Section 2. In fact, this population
distribution has mean µ = 66 and standard deviation σ = 22.
Figure 30 shows the sampling distributions of the mean for various
sample sizes. (Figure 30 is similar to Figure 6 but for some different
values of n.)
What do you notice about the means of these sampling distributions
compared with the population mean?

134
 4 Sampling distributions re-revisited

Proportion

0 10 20 30 40 50 60 70 80 90 100 x
n = 25
Proportion

0 10 20 30 40 50 60 70 80 90 100 x
n = 50
Proportion

0 10 20 30 40 50 60 70 80 90 100 x
n = 100
Figure 30 Sampling distributions of the mean for samples of size n from the
population of exam marks

(b) Consider again the population distribution of full-time employees’

earnings which you met in Example 1, in Section 2. This population
distribution has mean µ = 491 and standard deviation σ = 283 (in £).
Figure 31 shows again the sampling distributions of the mean for
various sample sizes. (Figure 31 is similar to Figure 8 but for different
values of n.)
What do you notice about the means of these sampling distributions
compared with the population mean?

135
Unit 7 Factors affecting reading

Proportion
0 500 1000 1500 x
n = 25

Proportion

0 500 1000 1500 x

n = 50
Proportion

0 500 1000 1500 x

n = 100
Figure 31 Sampling distributions of the mean for samples of various sizes
from the population of employees’ earnings

The conclusions of Activity 18 hold more generally so that whatever the

population distribution (no matter what shape) and whatever the sample
size (no matter how small), the mean of the sampling distribution is
always equal to the population mean µ.
Now let us take a closer look at the spread of the sampling distributions.

Activity 19 Standard deviations of distributions of sample

means (1)
(a) Consider again the sampling distributions of the mean for MS221
examination marks that are shown in Figure 30. What do you notice
about the standard deviations of these sampling distributions?
(b) Consider again the sampling distributions of the mean for full-time
employees’ earnings that are shown in Figure 31. What do you notice
about the standard deviations of these sampling distributions?

136
 4 Sampling distributions re-revisited

In fact it can be shown that for population standard deviation σ, the

standard deviation of the sampling distribution of the mean for samples of
√
size n is σ/ n.

Activity 20 Standard deviations of distributions of sample

means (2)
The population distribution of examination marks has standard deviation
σ = 22. Use the formula to find the standard deviation of the sampling
distribution of the mean for samples of size
(a) 25; (b) 50; (c) 100.

√
Both the formula σ/ n and the calculations in Activity 20 confirm that
the standard deviation of the sampling distribution of the mean does
decrease as n increases, as was suggested in Activity 19. What is not so
clear, and is perhaps unexpected, is the precise way in which the standard
deviation of the sampling distribution of the mean depends on n – through
its square root.
The expression ‘standard deviation of the sampling distribution of the
mean’ is a bit of a mouthful. It is often referred to as the standard error
of the mean for samples of size n, or sometimes just the standard error
for short, in which case it can be abbreviated to the symbol SE. Using this
√
abbreviation, we obtain the formula SE = σ/ n, which is easier to
remember.
The above result holds generally for all sampling distributions, no matter
what the population distribution and no matter what sample size is
involved. So there is a very precise relationship between sampling
distributions and the population distribution. It can be summarised as
follows.

Mean and standard deviation of the sampling distribution

of the mean
• The mean of the sampling distribution is equal to µ, the population
mean.
• The standard deviation of the sampling distribution is called the
standard error of the mean. It is given by
σ
SE = √ ,
n
where n is the sample size and σ is the population standard
deviation.

137
Unit 7 Factors affecting reading

It seems you’ve calculated

the standard deviation of the mean
when the question asked for the population
standard deviation. We see this a lot.
It’s a standard error.

The relationship between sampling distributions and the population

distribution is particularly useful when the sample size is large and the
sampling distribution is approximately normal. In practice, we usually
have very little information about the population distribution itself.
Indeed we often have only a sample of data on which to base our analysis;
there is no other information about the population. Yet many techniques
of statistical inference require us to make some assumptions about the
population distribution.
The advantage of working with large samples is that, no matter what
shape the population distribution is, the sampling distribution of the mean
for samples of size n will always be more or less normal. Moreover, we
know that the mean of this sampling distribution is equal to the
population mean, µ, and the standard deviation is the standard error,
√
given by SE = σ/ n, where σ is the population standard deviation. This
is summarised below.

138
 4 Sampling distributions re-revisited

Approximate normality of the sampling distribution of the

mean
If n is large, no matter what shape the population distribution is, the
sampling distribution of the mean for samples of size n will be
approximately normal with mean equal to the population mean, µ,
√
and standard deviation equal to the standard error, SE = σ/ n.
(This important result is often called the central limit theorem.)

Activity 21 Approximate distribution of ball bearing

diameters
The population distribution of the diameters of ball bearings produced by
a particular manufacturer has mean µ = 2 mm and standard deviation
σ = 0.01 mm. Find the standard deviation of the sampling distribution of
the mean for samples of 25 such ball bearings. Hence give the approximate
distribution of the mean diameter of ball bearings in samples of size 25.

What this implies is that we can base our analysis on the relationship
between the sample data and the sampling distribution of the mean. Thus
we infer back from the evidence provided by the sample data to the
sampling distribution. Then our knowledge of the links between this
sampling distribution and the population distribution allows us to draw
conclusions about the population mean. This is a very important strategy
in statistics.
The new hypothesis test, the z-test, is based on just this principle and will
be fully discussed in Section 5. As you now know, the sampling
distribution of the mean, x, for large samples of size n is approximately
√
normal with mean µ and standard deviation SE = σ/ n. As with any
normal distribution, we can transform this normal sampling distribution
into the standard normal distribution. This means that the distribution of
the variable
x−µ
z=
SE
is the standard normal distribution (with mean zero and standard
deviation one). There is a strong connection between this result and the
z-test to follow.
You have now covered the material related to Screencast 4 for
Unit 7 (see the M140 website).

139
Unit 7 Factors affecting reading

Exercises on Section 4

Exercise 13 Standard deviations of the mean as sample size

changes
The population distribution of full-time employees’ earnings has standard
deviation σ = 283. Find the standard deviation of the sampling
distribution of the mean for samples of size
(a) 9; (b) 25; (c) 100.

Exercise 14 Standard deviations of another mean

The population distribution of a certain quantity has standard deviation
σ = 3.6. Find the standard deviation of the sampling distribution of the
mean for samples of size
(a) 4; (b) 19; (c) 300.

Exercise 15 Standard deviation of the average content of

water bottles
The population distribution of the amount of water contained in a
nominally one-litre bottle from a certain manufacturer has mean
µ = 1.01 litres and standard deviation σ = 0.01 litres. Find the standard
deviation of the sampling distribution of the mean for samples of 40 such
bottles. Hence give the approximate distribution of the mean amount of
water contained in samples of 40 one-litre bottles from this manufacturer.

5 The one-sample z-test

In this section we shall develop a new hypothesis test, called the
one-sample z-test. However, we often omit the phrase ‘one-sample’
unless we need to distinguish it from the two-sample z-test (which will be
developed in Section 6). The hypotheses are concerned with the mean, µ,
of the population from which the sample is selected. We shall suppose that
a particular value, say A, is of special interest as a potential value for µ.
The null hypothesis is
H0 : µ = A,
and the alternative hypothesis is
H1 : µ %= A.
Alternative hypotheses of this form are often called two-sided
alternative hypotheses. This is because they include both µ < A and
µ > A.

140
 5 The one-sample z-test

The above is the first of the four stages of hypothesis testing that you were
introduced to at the start of Section 4 of Unit 6. In abbreviated form,
these are:
1. Set up the hypotheses that we wish to test.
2. Determine the sampling distribution of a test statistic under the
assumption that the null hypothesis is true.
3. Ascertain how unlikely the observed value of the test statistic is on the
basis of the sampling distribution.
4. If the test statistic turns out to have a very unlikely value, then either:
• a very unusual event has happened, or
• the sample has provided evidence against the correctness of the null
hypothesis.
To develop ideas in the current context, we first consider the simpler case
where the population standard deviation is assumed to be known, and in
Subsection 5.2 we consider the more realistic case where it is unknown.
The tests that are developed make use of the results presented in Section 4
about the sampling distribution of the sample mean.

5.1 The z-test with the standard deviation

assumed to be known
To describe the z-test we will use a simple (constructed) example.

Example 5 Has a new method of teaching made a

difference?
For many years a teacher has been using the same method of teaching
children to read. The scores the children obtain on a reading test have
a mean of 54.6 and a standard deviation of 8.3. These values will be
taken to be the population mean and the population standard
deviation under the old method of teaching. The teacher tries a new
method with her current class of 34 children, and their average score
on the reading test is 58.1. She wants to test whether random
variation underlies the difference between the average of this class
(58.1) and the long-term average of previous classes (54.6), or whether
there is a genuine difference.
The null and alternative hypotheses are:
H0 : The old method and new method of teaching children to read
are equally effective.
H1 : The old method and new method of teaching children to read
differ in their effectiveness.

141
Unit 7 Factors affecting reading

If µ denotes the mean reading score of children taught by the new

method, we can recast these hypotheses as
H0 : µ = 54.6
H1 : µ =
% 54.6.
The sample mean, x, is based on the performances of n = 34 children.
Hence, its sampling distribution is approximately normal, as a sample
size of 34 is quite large. Moreover, as shown in Section 4:
• the mean of the sampling distribution of x is equal to µ
• the standard deviation of the sampling distribution of x (i.e. the
√
standard error of x) is equal to σ/ n.
Now, to perform a hypothesis test based on x, the sampling
distribution under which we calculate probabilities is the sampling
distribution of x assuming that the null hypothesis, H0 , is true.
In the present case, if H0 is true, then µ = 54.6 and the distribution of
x is approximately normal with mean 54.6 and standard deviation
√
σ/ n. We know that n equals 34 but need to know the value of σ.
For this example, we shall assume that the population standard
deviation of scores with the new method is the same as with the old
method, so σ = 8.3. All told,
A = 54.6, x = 58.1, n = 34, σ = 8.3.
Now, from the end of Section 4, if the sampling distribution of x is
approximately normal with mean µ and standard deviation
√
SE = σ/ n, then the distribution of the variable
x−µ
z=
SE
is (approximately) the standard normal distribution (with mean zero
and standard deviation one). Thus, if H0 is true, so that
µ = A = 54.6, the distribution of the variable
x − 54.6
z=
SE
is (approximately) the standard normal distribution.
The variable z is the test statistic for the z-test. Its numerical value in
this example is
x−A 58.1 − 54.6
z= = √ " 2.46.
SE 8.3/ 34

The main result that we have obtained so far is summarised in the

following box.

142
 5 The one-sample z-test

Test statistic and its sampling distribution when H0 is true

and σ is assumed known
For a one-sample z-test, when H0 : µ = A is true, the test statistic,
x−A
z= ,
SE
follows (approximately) the standard normal distribution, where
√
SE = σ/ n.

Activity 22 Value of z

Calculate the value of the test statistic z for the test of

H0 : µ = 120
H1 : µ =
% 120,
when n = 100, x = 112, and σ = 15.

Critical values and critical regions

If the null hypothesis, H0 , is true, then z should follow the standard
normal distribution. This distribution has a mean of 0, so if the value of z
given by our data was very large in size (positive or negative), it would
suggest that H0 is false. The idea, then, is to reject H0 if the observed
value of z is ‘too extreme’ and therefore unlikely. Notice that ‘too extreme’
covers both large positive values and large negative values, in line with H1 ,
which specifies µ %= A, ‘in either direction’ away from A. If we cannot
believe that the observed z is an observation from a standard normal
distribution, then we cannot believe H0 .
We have calculated the value of the test statistic z that is given by our
data. Suppose now that the test is to be performed at the 5% significance
level. As discussed in Subsection 4.1 of Unit 6, H0 will be rejected at the
5% significance level if z is in the most extreme 5% of values under the
sampling distribution that applies if H0 is true. This ‘most extreme’ region
is the critical region of the test. (In this case it is the critical region at
the 5% significance level.) Because of the discussion in the previous
paragraph, the critical region consists of two parts: one part comprises the
most extremely high 2.5% of values under the standard normal
distribution, and the other part comprises the most extremely low 2.5% of
values under the standard normal distribution.
The values defining the ‘inner ends’ of the critical region are the critical
values. The critical values for the z-test at the 5% significance level are
1.96 and −1.96. Figure 32 shows the critical values and critical region
pictorially.

143
Unit 7 Factors affecting reading

Probability that z Probability that z

lies in here is 0.025 lies in here is 0.025

−1.96 1.96

−3 −2 −1 0 1 2 3 z

Critical region at the

5% significance level

Figure 32 The standard normal distribution with the critical region and
critical values (1.96 and −1.96) shown for a test at the 5% significance level

Instead of using the 5% significance level for the hypothesis test, we might
want to perform the test at the more stringent 1% significance level. To do
this, all that changes is the values of the critical values and hence the
critical region. The critical values become 2.58 and −2.58, and the critical
region is rather smaller: see Figure 33.

Probability that z Probability that z

lies in here is 0.005 lies in here is 0.005

−2.58 2.58
−3 −2 −1 0 1 2 3 z

Critical region at the

1% significance level

Figure 33 The standard normal distribution with the critical region and
critical values (2.58 and −2.58) shown for a test at the 1% significance level

The procedure to be followed to complete the z-test is as follows.

144
 5 The one-sample z-test

Completing the z-test

If z ≥ 2.58 or z ≤ −2.58, reject H0 at the 1% significance level.
If 1.96 ≤ z < 2.58 or −2.58 < z ≤ −1.96, reject H0 at the 5%
significance level but not at the 1% significance level.
If −1.96 < z < 1.96, do not reject H0 at the 5% significance level.

Activity 23 Where is z in marginal circumstances?

On a sketch of the standard normal distribution, show where the value of z
must lie in the marginal case where H0 is rejected at the 5% significance
level but not at the 1% significance level.

As noted in Subsection 5.1 of Unit 6, we conclude that there is strong

evidence against the null hypothesis if we reject H0 at the 1% significance
level. (The evidence might in fact be ‘very strong’, but we will not be
testing H0 at the 0.1% level.) If we reject it at the 5% significance level
but not the 1% level, we conclude that there is moderate (but not strong)
evidence against the null hypothesis. If we do not reject H0 at the 5%
significance level, we have, in the words of Subsection 5.1 of Unit 6, either
‘little’ or ‘weak’ evidence against H0 .

Example 6 Completing the z-test started in Example 5

In Example 5, the test statistic (the data z-value) takes the value
2.46. This value exceeds 1.96 but not 2.58. We conclude that there is
moderate evidence that the old and new methods are not equally
effective at teaching children to read.
As the new method gave an average score of 58.1, while the average
under the old method was 54.6, and a higher test score means better
reading ability, there is moderate evidence that the new method is
better than the old method.

Activity 24 A z-test in manufacturing

A firm is engaged in putting finishes on work surfaces for kitchen

manufacturers. Previously, the work was done in very large batches, so the
time spent setting up the machine did not affect production too much.
However, with a change in the pattern of demand the batch size has had to
be considerably reduced, so the time spent setting the machine to different
specifications is becoming more important.

145
Unit 7 Factors affecting reading

Last year the manufacturing manager found that the machine setting had
been changed very many times and the mean time taken for a change was
26.1 minutes. The operators suggested a way in which the set-up time
might be reduced, but the manager was unconvinced and feared that the
set-up time might actually be increased. Nevertheless, it was agreed to try
out this new method for two weeks. A z-test would then be performed to
examine whether or not the mean time for setting up under the new
method differs from the mean time taken last year.
In the two-week testing period, the machine was reset on 53 occasions,
taking a mean time of 20.9 minutes.
(a) What are the appropriate null and alternative hypotheses?
(b) Give the values of A, x and n.
(c) Assume that the standard deviation, σ, equals 12.3. Calculate the
value of the test statistic.
(d) Is the null hypothesis rejected at the 5% significance level? Is it
rejected at the 1% significance level?
(e) What do you conclude from the hypothesis test?

5.2 The z-test with unknown standard

deviation
In Subsection 5.1, we developed one-sample z-tests under the assumption
that σ is known. Now σ is the standard deviation of the population from
which the sample data are drawn. Typically its actual value will not be
known, but if we have a large sample then the sample standard deviation,
s, provides a good estimate of σ. Moreover, provided the sample size is
large, the one-sample z-test can be performed with σ replaced by s.
Specifically, we calculate the estimated standard error (ESE) of x,
s
ESE = √ ,
n
and put
x−A
z= .
ESE
You might be slightly disquieted by the bald assertion that, for large
samples, replacing SE by its estimated value ESE makes no difference to
the (approximate) standard normal distribution of (x − A)/SE. After all,
ESE is not the correct quantity to divide by; SE is. It is the assumption of
a large sample that saves the day. In Unit 10 we give tests for small
samples (t-tests) which take the difference between ESE and SE into
account. Differences between those tests and z-tests are small when the
sample size is above about 25.

146
 5 The one-sample z-test

At the end of Section 2, it was asserted that the sampling distribution of

the mean will always be approximately normal for sample sizes greater
than 25. It was also argued that the sampling distribution of the mean will
actually be approximately normal for sample sizes (much) smaller
than n = 25 for many population distributions. In that sense, the notion
of n = 25 being large enough errs on the ‘careful’ side. When SE is
replaced by its estimated value (ESE), however, a sample size of 25 is only
just enough for a z-test to be usable. We will continue to use this ‘rule of
thumb’, but n = 25 is no longer a ‘generous’ value – many would prefer to
use z-tests only for samples that are a bit larger than that.

What is a large enough sample for a z-test?

As a rough guide you can assume that, whatever the population
distribution, for sample sizes of at least 25, the z-test is applicable.

As this jolly logo shows, ESE also stands for ‘Exceptional Student Education’
. . . an educational program in schools in Citrus County, Florida, USA

The next two boxes lay out the full requirements and procedure for the
one-sample z-test. They cover both the cases where σ is known and where
it must be estimated. The first box gives the key pieces of information that
you should pick out for a z-test when you are reading details about a
survey or experiment.

Key values for a one-sample z-test

The information you need to know for a one-sample z-test is:
• the hypothesised population mean (A) under the null hypothesis
• the sample mean (x)
• the sample size (n)
• the population standard deviation (σ), or a good estimate of σ.

147
Unit 7 Factors affecting reading

Procedure: the one-sample z-test

1. Set up the null and alternative hypotheses,
H0 : µ = A
H1 : µ =
% A,
where µ is the population mean.
2. Calculate the test statistic, z:
• If the population standard deviation (σ) is known,
x−A σ
z= , where SE = √ .
SE n
• If σ is unknown but the sample size (n) is 25 or more,
x−A s
z= , where ESE = √ .
ESE n
Here x is the sample mean and s is the standard deviation of the
sample. SE is the standard error of the mean and ESE is the
estimated standard error.
3. Compare z with the appropriate critical values, which are 1.96 and
−1.96 at the 5% significance level and 2.58 and −2.58 at the 1%
significance level.
• If z ≥ 2.58 or z ≤ −2.58, then H0 is rejected at the 1%
significance level.
• If 1.96 ≤ z < 2.58 or −2.58 < z ≤ −1.96, then H0 is rejected at
the 5% significance level but not at the 1% significance level.
• If −1.96 < z < 1.96, then H0 is not rejected at the 5%
significance level.
4. State the conclusions that can be drawn from the test.

We are now in a position to start answering some of the questions we

asked about the BCS survey data in Subsection 1.3. The investigation
illustrates use of the one-sample z-test when σ is unknown.

Example 7 Reading scores of 7-year-old children in BCS

survey
In a question that was posed at the end of Subsection 1.3, we asked
whether the sample of children from the BCS survey in 2004–2005
could be considered to have come from the population of children for
whom the British Ability Scales reading score was developed. The
overall population mean reading scores for British children are taken
to be 96 for 7-year-old children. We wrote down the following null and

148
 5 The one-sample z-test

alternative hypotheses:
H0 : For British children aged 7 in 2004–2005, the mean reading score
is equal to 96
H1 : For British children aged 7 in 2004–2005, the mean reading score
is not equal to 96.
We can recast these hypotheses as
H0 : µ = 96
H1 : µ =
% 96,
where µ is the population mean of the reading scores of all British
7-year-old children in 2004–2005. The data from the BCS concerning
7-year-old children are summarised in Table 3.
Table 3 Further summary statistics for data on reading scores of 7-year-old
children

Sample size Sample mean Sample standard deviation

396 111.28 26.668
(This data is copyright and owned by the Economic and Social Data Service.)

Although σ is unknown, the sample size is considerably greater than

25, so the ESE may be used in calculating z. Thus the information
required for the z-test is:
A = 96, x = 111.28, n = 396, s = 26.668.
We can now calculate the test statistic:
x−A x−A 111.28 − 96
z= = √ = √ " 11.40.
ESE s/ n 26.668/ 396
Assuming that the null hypothesis, H0 , is true, 11.40 is a value from
the standard normal distribution. However, 11.40 is much bigger than
the 1% critical value of 2.58. Hence the z-test clearly rejects H0 at the
1% level.
We conclude that there is strong evidence that the mean reading score
for 7-year-old children in 2004–2005 is not equal to the overall mean
reading score for 7-year-old children. At face value, this is a little
surprising – there seems no obvious factor to cause a difference in
reading ability between the 7-year-old children in the 2004–2005 BCS
survey and the 7-year-olds in the population of British children for
whom the reading test was originally developed. Given that the mean
reading score for the 7-year-old children in the BCS survey is larger
than the overall mean reading score, for some reason the BCS children
seem to have performed rather better than expected (on average).

149
Unit 7 Factors affecting reading

The following two activities provide you with practice in applying the
z-test. The first one continues our investigation of the BCS data. It
concerns the reading scores of 8-year-old children. The second concerns
some data on earnings.

Activity 25 Reading scores of 8-year-old children in BCS

survey
In the BCS investigation, the following results were obtained for 8-year-old
children.
Table 4 Summary statistics for data on reading scores of 8-year-old children

Sample size Sample mean Sample standard deviation

283 126.92 27.711
(This data is copyright and owned by the Economic and Social Data Service.)

The overall mean reading score for 8-year-old children is 116.

Carry out a z-test to investigate whether the sample of 8-year-old children
was selected from a population whose mean reading score is equal to the
overall mean score for 8-year-old children. Comment on your result.

Activity 26 Wages of female employees

A random sample of 810 female local government clerical officers and

assistants had a mean wage of £373.40 per week in 2011 with a standard
deviation of £138.20. The overall mean weekly wage for female employees
in 2011 was £381.50. (Source: Annual Survey of Hours and Earnings,
2011.) Investigate whether the mean weekly wage of female local
government clerical officers and assistants differed from the overall mean
weekly wage for female employees in 2011. Comment on your result.

You have now covered the material related to Screencast 5 for

Unit 7 (see the M140 website).

Exercises on Section 5

Exercise 16 Reading scores of 7-year-old girls in BCS survey

In the BCS investigation, the following results were obtained for 7-year-old
girls. (These results have been extracted from Table 2 in Subsection 1.3.)
Table 5 Summary statistics for data on reading scores of 7-year-old girls

Sample size Sample mean Sample standard deviation

190 113.42 25.464
(This data is copyright and owned by the Economic and Social Data Service.)

150
 5 The one-sample z-test

The overall mean reading score for 7-year-old children is 96.

Carry out a z-test to investigate whether the sample of 7-year-old girls was
selected from a population whose mean reading score is equal to the overall
mean score for 7-year-old children. Comment on your result.

Exercise 17 Weight of pigs

A random sample of 533 pigs of a certain breed that had been fed a special
diet were weighed. They had a mean weight of 81.92 kg with a standard
deviation of 15.65 kg. The mean weight of this breed of pig when fed the
standard diet is 80 kg. Evaluate the evidence that the special diet changes
the mean weight of this breed of pig.

Exercise 18 An exciting exercise: paint drying

A consumer magazine, when comparing various brands of paint, stated
that the drying time of one particular brand was exactly four hours. The
manufacturers of that paint were not particularly pleased with this as they
believed the drying time for their paint was shorter. They organised a trial
in which the paint was tested by a random sample of 40 customers, all of
whom were decorating their living rooms. For this sample the mean drying
time was found to be 3.80 hours and the standard deviation was 0.55 hours.
(a) Analyse the sample data to test whether the drying time given by the
consumer magazine is correct.
(b) What reservations might there be about your conclusion?

In 2011/12, the internet (including one national newspaper) was abuzz

with news of the forthcoming inaugural World Watching Paint Dry
Championships to held in Stoke-on-Trent in July 2012. Competitors
were to each be given a one-metre square patch of freshly emulsioned
wall at which to stare as it slowly dried. There were said to be 42
entrants, from the UK, USA, India and Hungary. Unfortunately, there
is no evidence that the event actually took place.

151
Unit 7 Factors affecting reading

6 The two-sample z-test

In this section we develop the two-sample z-test, which is used to analyse
the difference in locations between two populations. There were plenty of
examples of this raised in the context of the BCS and its data on reading
scores in Subsections 1.2 and 1.3. One such question posed there was:
For British children aged 7 in 2004–2005, did boys’ and girls’ reading
scores differ in location?
Here, the two populations which we wish to compare in terms of their
reading abilities are the population of British boys aged 7 in 2004–2005
and the population of British girls aged 7 in 2004–2005. Another example
is the question
For British children aged 7–8 in 2004–2005, did reading scores differ
in location according to their father’s occupation?
Here, the two populations which we wish to compare in terms of their
reading abilities are the population of British children aged 7–8 in
2004–2005 whose father’s occupation was coded 1 in Table 1 (managerial,
technical, professional and skilled non-manual occupations) and the
population of British children aged 7–8 in 2004–2005 whose father’s
occupation was coded 2 (skilled manual, partly skilled and unskilled
occupations).
As in Section 5, comparisons will be made using hypothesis tests
comparing means, and the two-sample z-test will be appropriate when
both samples are large. To develop this test we use the reading scores from
the BCS sample. We examine the first of the above questions:
For British children aged 7 in 2004–2005, did boys’ and girls’ reading
scores differ in location?
The following are appropriate null and alternative hypotheses:
H0 : For British children aged 7 in 2004–2005, the mean reading score
for girls is equal to the mean reading score for boys
H1 : For British children aged 7 in 2004–2005, the mean reading score
for girls is not equal to the mean reading score for boys.
We shall now introduce some symbols that will enable us to express our
hypotheses more concisely and will also be helpful in explaining a
theoretical result that we need. We are investigating two populations of
values: the reading scores of all British 7-year-old girls in 2004–2005 and
the reading scores of all British 7-year-old boys in 2004–2005. We shall let
the means of these two populations be µg and µb , and the standard
deviations be σg and σb . (In this example, the subscripts ‘g’ and ‘b’ always
relate to girls and boys, respectively.) It is worth noting that the values of
these quantities cannot be known: not all British 7-year-old girls and boys
actually took this test in 2004–2005. So there is no way we could actually
calculate µg , µb , σg and σb , but they enable us to make precise statements.

152
 6 The two-sample z-test

For a start, we can use µg and µb to write the hypotheses concisely as

H 0 : µg = µ b
H 1 : µg =
% µb ,
or, equivalently, as
H 0 : µg − µ b = 0
H1 : µg − µb %= 0.
This last form is the one we shall actually use to derive the test statistic.
Although we do not know test values for all children, the values for the
samples of girls and boys in the BCS are known. We shall denote these
samples’ sizes by ng and nb , the sample means by xg and xb , and the
sample standard deviations by sg and sb . Their values were set out in
Table 2 (Subsection 1.3), but we do not need them at the moment.
As we have expressed our null hypothesis as µg − µb = 0, it seems
intuitively sensible to test the hypothesis by looking at the difference
between the sample means, xg − xb . Before we can develop our hypothesis
test, we need a theoretical result about the sampling distribution of the
difference between two sample means.
You already know, from Section 4, that, because ng and nb are large, the
sampling distribution of xg is approximately normal with mean µg and
√
standard error σg / ng , and similarly that the sampling distribution of xb
√
is approximately normal with mean µb and standard error σb / nb . We
may conceive the first of these sampling distributions by thinking of all the
possible samples of size ng that we could select from the population of
scores of all 7-year-old girls. We then imagine that we could calculate xg
for each of these samples and look at their distribution. Similar
considerations apply to the sampling distribution of xb .
Now, think of all the possible means xg of samples of size ng of girls and
also all the possible means xb of samples of size nb of boys. If we select
just one value of xg and one value of xb , we can calculate the difference
xg − xb . Now think of all the possible pairs of values xg and xb we could
select, and suppose we calculate xg − xb for each of them. Then the
distribution of all these differences is the sampling distribution of the
difference between two means.
We require three results that are known about this sampling distribution.
First, the mean of the sampling distribution of xg − xb is equal to µg − µb ,
as you might expect. The second result requires the two samples to be
independent of each other – here that is clearly the case, as the choice of
girls was completely separate from the choice of boys. As long as the
samples are independent, the standard deviation of the sampling
distribution is given by
'
σg2 σb2
SE = + ,
ng nb

153
Unit 7 Factors affecting reading

and this standard deviation is called the standard error of the

difference between two means. Notice that it is larger than the
√ √
standard errors of xg and xb , which are σg / ng and σb / nb , respectively.
This is because we are looking at the difference between two sample
means; both means can vary, so there is more variation in the difference
between them. Notice also that the standard error of the difference
between two means is neither the sum nor the difference of the standard
errors of the individual means. The next box summarises these results.

Mean and standard deviation of the sampling distribution

of the difference between two means
• The mean of the sampling distribution is equal to µg − µb , the
difference between the population means.
• The standard deviation of the sampling distribution is called the
standard error of the difference between two means, and is given by
'
σg2 σ2
SE = + b,
ng nb
where ng and nb are the sizes of the samples, and σg and σb are the
population standard deviations.

Furthermore, provided the sample sizes are sufficiently large, the sampling
distribution of the differences between two sample means is approximately
normal. This is the third result that we require.

Approximate normality of the sampling distribution of the

difference between two means
If ng and nb are large, no matter what shape the population
distributions, the sampling distribution of the difference between two
means based on samples of sizes ng and nb will in practice be
approximately normal.

From these results, xg − xb is approximately normally distributed with

mean µ = µg − µb and standard deviation σ = SE. Thus, the formula
given in Subsection 3.3 can be used to transform xg − xb to a quantity
which follows (approximately) the standard normal distribution:
x−µ (xg − xb ) − (µg − µb )
z= = .
σ SE

154
 6 The two-sample z-test

Now to obtain our test statistic, we assume that the null hypothesis H0 is
true, so µg − µb = 0. We still cannot calculate z, as we do not know σg and
σb . We deal with this problem exactly as we did in Subsection 5.1, by
replacing σg by sg and σb by sb . This leads to the estimated standard
error of xg − xb :
'
s2g s2
ESE = + b.
ng nb

Test statistic and its sampling distribution when H0 is true

For a two-sample z-test, when H0 : µg − µb = 0 is true, the test
statistic,
'
xg − x b s2g s2
z= , where ESE = + b,
ESE ng nb
follows (approximately) the standard normal distribution.

For the one-sample z-test, we used the rule of thumb that the sample size This logo suggests a more
had to be at least 25. To justify use of a two-sample z-test, we apply this relaxing form of ESE
rule of thumb to both samples and require that each sample size should be
at least 25.

Key values for a two-sample z-test

In general, call the two groups A and B. The information you need to
know for a two-sample z-test is:
• the sample means (xA and xB )
• the sample sizes (nA and nB )
• the population standard deviations (σA and σB ), or good estimates
of them (sA and sB ).

Since the test statistic above has the standard normal distribution
(approximately) when the null hypothesis is true, the critical values are
exactly the same as those in Subsection 5.1 for a one-sample hypothesis
test. We can reject H0 at the 1% significance level if z ≥ 2.58 or if
z ≤ −2.58, and we can reject H0 at the 5% significance level if z ≥ 1.96 or
z ≤ −1.96. Otherwise we cannot reject H0 .

155
Unit 7 Factors affecting reading

Example 8 Comparing the mean reading scores of girls

and boys
We are now able to perform the two-sample z-test with which the
current subsection was introduced. The hypotheses are:
H 0 : µg = µb
H 1 : µg =
% µb ,
where µg is the population mean reading score for 7-year-old girls in
2004–2005, and µb is the population mean reading score for 7-year-old
boys in 2004–2005. The data on which the test will be based were
given as Table 2 (Subsection 1.3) and are repeated in Table 6.
Table 6 Summary statistics for data on reading scores of 7-year-old
children

Sample size Sample mean Sample standard deviation

Boys 206 109.31 27.671
Girls 190 113.42 25.464
(This data is copyright and owned by the Economic and Social Data Service.)

In this example we are using ‘g’ and ‘b’ to distinguish the two groups,
rather than A and B. We have:
xg = 113.42, xb = 109.31, ng = 190, nb = 206,
sg = 25.464, sb = 27.671.
Both ng = 190 and nb = 206 are greater than 25, so we can assume
that the z-test is applicable.
In the two-sample case, it is easier to calculate the value of z in two
stages. We first calculate the value of ESE, the estimated standard
error of xg − xb :
' &
s2g s2b 25.4642 27.6712
ESE = + = + " 2.670.
ng nb 190 206
Hence the value of the test statistic is
xg − xb 113.42 − 109.31
z= " " 1.54.
ESE 2.670
The critical values are 1.96, −1.96 (5%) and 2.58, −2.58 (1%). Since
−1.96 < 1.54 < 1.96, we cannot reject H0 at the 5% significance level.
There is little evidence to suggest that the mean reading scores in
2004–2005 for 7-year-old boys and girls were different.

156
 6 The two-sample z-test

The procedure for the two-sample z-test is summarised in the following

box.

Procedure: two-sample z-test

1. Set up the null and alternative hypotheses,
H0 : µ A = µ B
H1 : µ A =
% µB ,
where µA and µB are the means of populations A and B,
respectively.
2. Calculate the test statistic
xA − x B
z= ,
ESE
where the estimated standard error of xA − xB is
'
s2A s2
ESE = + B.
nA nB
Here, nA and nB are the sample sizes of random samples from
populations A and B respectively, xA and xB are the sample
means, and sA and sB are the sample standard deviations.
3. Compare z with the appropriate critical values, which are 1.96 and
−1.96 at the 5% significance level, and 2.58 and −2.58 at the 1%
significance level.
• If z ≥ 2.58 or z ≤ −2.58, then H0 is rejected at the 1%
significance level.
• If 1.96 ≤ z < 2.58 or −2.58 < z ≤ −1.96, then H0 is rejected at
the 5% significance level but not at the 1% significance level.
• If −1.96 < z < 1.96, then H0 is not rejected at the 5%
significance level.
4. State the conclusions that can be drawn from the test.

In the two-sample z-test, it doesn’t actually matter which of the two

groups of interest you label A and which B. If you swapped the roles of A
and B over, you would change the sign of z but nothing else. In particular,
the conclusions of the test would be the same in either case.

157
Unit 7 Factors affecting reading

Activity 27 Mean reading scores of girls and boys at age 8

In the BCS investigation, the following results were obtained for 8-year-old
children.
Table 7 Summary statistics for data on reading scores of 8-year-old children

Sample size Sample mean Sample standard deviation

Boys 145 126.38 29.927
Girls 138 127.49 25.064
(This data is copyright and owned by the Economic and Social Data Service.)

Carry out a two-sample z-test to investigate whether the mean reading

score of 8-year-old girls in 2004–2005 was equal to the mean reading score
of 8-year-old boys in 2004–2005. Comment on your result.

You might have noticed something interesting about the results for
7-year-old and 8-year-old children. For the younger children, the girls’
sample mean score was 113.42 − 109.31 = 4.11 more than that for boys,
whereas for the older children the girls’ sample mean score was
127.49 − 126.38 = 1.11 higher. One might have thought at first glance that
there was an interesting effect here: at the younger age, girls are ahead of
boys in reading ability, but a year later boys seem to be catching up. Not
so, however: our hypothesis tests showed that in neither case was there
any evidence of a real difference, or therefore, of any such effect. The
differences in the samples that we observed can easily have arisen by
chance.
Does the level of education of parents have an affect on the reading scores
of their children? In the next activity you will investigate this in the
context of the BCS survey. This study classified parental education into
two categories: those who finished full-time education by age 16 and those
who continued after 16 (see Table 1, Subsection 1.2).

Activity 28 Mean reading scores according to parental

education
This activity addresses another of the questions raised in Subsection 1.2:
For British children aged 7–8 in 2004–2005, did reading scores differ
in location according to the level of their parents’ education?
Table 8 provides the relevant summary data from the BCS.
Table 8 Summary statistics for data on reading scores and parental education

Parental education Sample size Sample mean Sample standard deviation

Ended by age 16 389 116.12 28.775
Continued after age 16 199 123.15 24.603
(This data is copyright and owned by the Economic and Social Data Service.)

158
 6 The two-sample z-test

(Note that 679 children were tested for their reading ability, but no
information was available on when the parents of 91 of the children
completed their education.)
Carry out a hypothesis test to investigate whether children whose parents’
education continued beyond age 16 scored differently on average on the
reading test from those children whose parents’ education ended by age 16.

You might have expected the answer to Activity 28 before analysing the
data. That is, denoting µC as the mean reading score of children whose
parental education continued after age 16 and µE as the mean reading
score of children whose parental education ended by age 16, you might
have thought of doing the following: testing the null hypothesis that
µC = µE with the purpose of seeing whether, as you suspect, µC is actually
greater than µE , disregarding the possibility that µC could be less than µE .
Hypothesis tests undertaken when a particular type of inequality between
the two groups is of interest are the one-sided tests mentioned in a margin
note at the start of Section 5 and to be looked at briefly in Unit 10.
You have now covered the material related to Screencast 6 for
Unit 7 (see the M140 website).

Exercises on Section 6
Exercise 19 Mean reading scores according to fathers’
occupations
This exercise concerns another question posed in Subsection 1.2, namely:
For British children aged 7–8 in 2004–2005, did reading scores differ
in location according to their fathers’ occupations?
Table 9 provides the relevant summary data from the BCS. Note that, as
in Table 1 (Subsection 1.2), ‘1’ denotes ‘managerial, technical, professional
and skilled non-manual’ occupations while ‘2’ denotes ‘skilled manual,
partly skilled and unskilled’ occupations.
Table 9 Summary statistics for data on reading scores and father’s occupation

Father’s occupation Sample size Sample mean Sample standard deviation

1 316 120.55 24.221
2 203 117.17 30.085
(This data is copyright and owned by the Economic and Social Data Service.)

(No information was available on father’s occupation for 160 individuals.)

Carry out a two-sample hypothesis test to investigate whether the mean
score of children whose father had an occupation coded 1 differs from that
of children whose father had an occupation coded 2. Comment on your
result.

159
Unit 7 Factors affecting reading

Exercise 20 Calcium for babies

This exercise is related to an investigation of the effect of vitamin D
supplementation for the prevention of low levels of calcium in newborn
babies. The data given in Table 10 come from a clinical trial in which a
sample of babies who were breast-fed were compared with a sample of
babies who were bottle-fed: the measured quantity was the level of calcium
in the baby’s blood (‘serum calcium’) at 1 week of age.
Table 10 Summary statistics for data on serum calcium for week-old babies

Sample size Sample mean Sample standard deviation

Breast-fed 64 2.45 0.292
Bottle-fed 169 2.30 0.274
(Source: Cockburn et al. (1980) ‘Maternal vitamin D intake and mineral
metabolism in mothers and their newborn infants’, British Medical Journal,
vol. 281, pp. 11–14)

Carry out a two-sample z-test to investigate whether the mean serum

calcium level of week-old babies was the same whether they were
breast-fed or bottle-fed.

Exercise 21 Peak flow rate of lungs

The peak flow rate is a measure of how well a person’s lungs are
functioning. It is the maximum rate in litres per minute at which air can
be expelled through a peak flow meter. In an investigation of the
possibility that chronic bronchitis, although a disease of adult life, starts in
childhood, the peak flow rates of a large number of school children without
persistent coughs were measured. Amongst other details recorded were
whether the child lived in an urban or a rural area. Data for urban and
rural areas are summarised in Table 11. Use a two-sample z-test to
examine whether the average peak flow rate of children differs in these two
groups.
Table 11 Peak flow rates for children without persistent coughs

Sample size Sample mean Sample standard deviation

Urban 485 226 52
Rural 637 231 53
(Source: unpublished data collected by Professor J.R.T. Colley, University of
Bristol)

160
 8 Conclusions and reservations

7 Computer work: one-sample

z-tests
In this section you will use Minitab to perform one-sample z-tests. These
are similar to the tests you have performed earlier in this unit, except that
Minitab gives the results of hypothesis tests in terms of p-values, while in
earlier sections we have only considered specific significance levels (5% and
1% significance levels). The use of p-values with sign tests was explained in
Unit 6. Their use with z-tests is identical, but is described explicitly in the
Computer Book.
You should now turn to the Computer Book and work through Chapter 7.
The chapter starts with the interactive computer resources connected with
Section 3 of this unit; you should do them now if you have not already
done so. You should then do the Minitab work that is contained in the rest
of Chapter 7.

8 Conclusions and reservations

We have answered many of the questions raised in Section 1, and we have
learned a lot about children’s reading ability and factors affecting it, at
any rate for British children aged 7 and 8 in 2004–2005. We summarise our
conclusions below. As usual, though, after coming to such conclusions, we
should stop and look for reservations that might arise.
• Are there any problems with the data that might throw doubt on
conclusions drawn from them?
• Were appropriate statistical methods used in analysing the data?
We shall look at both these questions. To address the second question we
shall discuss when z-tests should be used. We then note limitations on the
way conclusions are stated and interpreted.

Conclusions
We began this unit by asking the general question:
What factors affect a child’s reading ability?
In Section 1, we refined this question to produce several more specific
questions that we could attempt to answer using BCS data. In Sections 5
and 6, we carried out hypothesis tests that related to these questions. All
these tests involved hypotheses about the population from which the BCS
sample was drawn, that of British children aged 7 and 8 in 2004–2005.
In Example 7 (Subsection 5.2), we found that we could reject the null
hypothesis that 7-year-old British children in 2004–2005 had the overall
population mean reading score for 7-year-olds. Similarly, in Activity 25
(Subsection 5.2), we found that we could reject the null hypothesis that
8-year-old British children in 2004–2005 had the overall population mean

161
Unit 7 Factors affecting reading

reading score for 8-year-olds. (A related result for 7-year-old girls was
obtained in Exercise 16 in Section 5.)
In Example 8 (Section 6), we found that, for 7-year-old children, the null
hypothesis that the population mean for boys was equal to that for girls
could not be rejected. In Activity 27 (Section 6), we also found that the
same was true for 8-year-old children.
In Activity 28 (Section 6), we found strong evidence that the mean reading
score was higher for children whose parents’ education had lasted longer.
(Something less expected happened with respect to father’s occupation in
Exercise 19.)

Reservations about the data

Probably the main reservation about the data is whether they can be
considered a random sample from the relevant population. As was
discussed in Activity 3 (Subsection 1.2), the data do not come from a
formal random sample of British children aged 7 and 8 in 2004–2005, of
the sort one might draw using a sampling frame and random numbers. But
it might still be the case that the data can be treated as if they had been
drawn in that way. How would a ‘real’ random sample differ from the
BCS 2004–2005 sample? The main difference was raised in Activity 3; all
the children in our sample have at least one parent who is in the BCS
survey, and therefore was born in a particular week in 1970. In a true
random sample of British 7- and 8-year-old children, not every child would
have a parent aged 34.
There are other features of the sampling process that might lead to the
sample of children being unrepresentative:
• Children could be included only if the BCS 2004–2005 investigators had
managed to trace their parents. People in the original BCS sample whose
lifestyles involve moving around a lot may have been harder to trace,
and therefore their children would be less likely to be in the sample.
• There are missing data. This data may not be missing completely ‘at
random’ – which might be OK, provided there is not too much of it –
but its very missingness might be connected to the things you are trying
to measure. (This is a common problem in real-world statistics.)
For example, parents with less education might be more reluctant to say
so in response to a survey, in which case children with such parents might
be under-represented; worse, such parents might be more likely to not
respond to the education question if they know their child is not reading
especially well, and they don’t want to be ‘blamed’ for this situation.
• The data on parental education simply give the age at which one of the
parents left full-time education, and say nothing about which parent it
was. Also, nothing is said about any qualifications he or she gained, or
about any part-time study.

162
 8 Conclusions and reservations

These reservations about the randomness/representativeness of the sample

are probably less important than the reservation about the parents’ age,
but they should not be forgotten.

Any other reservations?

8.1 When to use the z-test

The z-test can be applied in many situations, though it does have
limitations. In this subsection the characteristics of the test are described
so that you can recognise when it is appropriate.

The sample size must be large

It is unnecessary to know anything about the distribution of the
population from which the sample is selected, because the test is based on
the fact that the sampling distribution of the mean of a sample of size n is
approximately normal, provided n is sufficiently large.
As a simple rule of thumb, we assume that in the one-sample case, n
should be greater than 25, and in the two-sample case, nA and nB should
both be greater than 25. If the sample size is less than 25, you should not
apply the z-test. (If you believe that the population distribution is
extremely skew – which has not been the case for any distribution in this
unit – then it is safer only to use the z-test if the sample size is
considerably greater than 25.)
In Unit 10 you will meet another hypothesis test, the t-test, which you can
apply under some circumstances when the sample size is less than 25.

The sample values should consist of numerical measurements

The z-test should be applied only to data which consist of numerical
measurements. Length, weight, time, scores in a test and petrol
consumption are all examples of such data. The z-test cannot be applied,
for example, to data which might be coded, such as perhaps hair colour or
disease type – with such data the concept of a population mean or a
sample mean is not really meaningful.

163
Unit 7 Factors affecting reading

The samples should be unrelated

This restriction applies only to the two-sample z-test. The samples from
the two populations should be unrelated and so not consist of data
collected in pairs, each pair coming from the same individual.
All the hypothesis tests that we have performed in this unit were based on
data that met the requirements for the one- or two-sample z-tests. Sample
sizes were above 25 (substantially so for the BCS data), sample values were
numerical measurements (often scores on a reading test), and the
two-sample z-test was only ever applied to unrelated samples from two
separate populations.

8.2 Limitations in stating conclusions

In stating conclusions from any hypothesis test, the following factors must
be borne in mind.
• A sampling error may have occurred.
• The conclusions should match the population from which the sample
was drawn.
• The conclusions must not make causal statements which are not
supported by the way the data arose.
Let us look at each of these briefly.

Sampling errors
You should always bear in mind that a sampling error might have occurred;
that is, the result of any hypothesis test might be due to sampling
variation. Hypothesis tests do not provide proofs of the truth of either the
null or alternative hypotheses. They just attempt to assess the evidence
for or against the hypotheses. For example, if the null hypothesis is
rejected, that means that there is evidence against the null hypothesis, but
not that the null hypothesis is definitely wrong. However, with the BCS
data, in most of the hypothesis tests where we rejected the null hypothesis,
the test statistic came out much higher numerically than the critical values
(it easily gave ‘strong evidence’), so with those tests it is unlikely – but
still possible – that sampling error has led to erroneous conclusions.

What can we say about the populations?

A major difficulty with the BCS data is that it is not clear that these data
can be treated as a random sample from any population. But they are
clearly likely to be much more representative of the population of British
children than of, say, Ugandan children. The stated conclusions were
explicit in referring to British children and to the year, 2004–2005, in
which the data were collected, although we should perhaps have referred to
the population of British children, aged 7 or 8 in 2004–2005, who had at
least one parent aged 34, as all these characteristics are common to the
children in our sample. Nevertheless, it seems reasonably plausible that

164
 8 Conclusions and reservations

the data would still be representative of British children in some other year
close to 2004–2005, say 2003 or 2007, since reading skills are unlikely to
change very rapidly. But it would be a mistake to apply the conclusions
directly to the population of British children in 1970, say, or 2013.

What can we say about causal statements?

Can we make any conclusions about what might have caused any
differences for which we have evidence? For the BCS data, the answer is,
essentially, ‘no’ ! Our conclusions are not worded in causal terms; for
instance, we concluded (in Activity 28) that, for British children aged 7
and 8 in 2004–2005, those whose parental education was beyond the age of
16 had a higher mean reading score than did those whose parent left
education earlier. Worded like that, the conclusion says nothing about how
this difference arose; but there is a great temptation to suppose that the
parent’s level of education caused the difference in mean reading score.
This causal conclusion goes beyond what the data tell us. Instead, there
could well be one or more other factors that underly both a child’s reading
ability and whether a parent of the child was educated past the age of 16.
We just cannot tell about such things from these data, since they do not
give us the appropriate information.

165
Unit 7 Factors affecting reading

Summary
In terms of statistical methodology, you have been introduced to the most
important distribution in statistics – the normal distribution – and you
have learned to use the distribution in two hypothesis tests, the one-sample
and two-sample z-tests. In this unit, the normal distribution arose out of
consideration of the sampling distributions of the sample mean: regardless
of the distribution of the original data, such sampling distributions were
seen to become more and more normal-like as the sample size, n, increased.
You then learned about the normal distribution itself. You saw the way in
which it depends on two quantities, the population mean, µ – controlling
its location – and the population standard deviation, σ – controlling its
spread. You also learned how any normal distribution can be related to a
special normal distribution: the standard normal distribution with µ = 0
and σ = 1. You then found that the sampling distribution of the sample
mean can be approximated by a normal distribution with mean µ and
√
standard deviation σ/ n, which is called the standard error of the mean.
The z-test was first introduced in its one-sample form to address null and
alternative hypotheses concerning the value of µ. Its test statistic was
developed in two forms, for σ assumed to be known and, more usefully, for
σ unknown. You saw how the sampling distribution of the test statistic,
and hence the critical values associated with the test, arose from the above
results for the normal distribution. Having learned how to implement the
one-sample z-test, you went on to learn how to adapt those ideas to
produce the two-sample z-test; this is applicable to testing hypotheses
concerning whether or not the means of two unrelated populations are
equal. In each case, you applied what you learned about hypothesis testing
in Unit 6 to interpret results in terms of the amount of evidence the data
provide against the null hypothesis.
What you learned about children’s reading abilities from the BCS survey
has been summarised and discussed in Section 8.

166
 Learning outcomes

Learning outcomes
After you have worked through this unit, you should be able to:
• appreciate the steps taken to make the unit’s original question, which is
rather general, more specific
• recall that the null and alternative hypotheses required for the z-test are
expressed in terms of population means
• recognise a bell-shaped distribution
• appreciate that population distributions can have different shapes, some
of which are normal
• appreciate that, whatever the shape of the population distribution, for a
large enough sample size the sampling distribution of the mean is nearly
always approximately normal
• appreciate the relationship between the location and spread of a normal
distribution and its mean and standard deviation
• appreciate that it makes sense to think of normal distributions in terms
of the number of standard deviations of the variable away from its mean,
and that we can therefore think of all normal distributions in terms of
only one distribution: the standard normal distribution
• apply the formula that transforms any variable x with a given normal
distribution to the variable z with the standard normal distribution
• understand what is meant by the standard error (of the mean) and the
estimated standard error in one- and two-sample situations
• write down the mean and standard deviation of the sampling
distribution of the mean for samples of size n, given the population
mean, µ, and standard deviation, σ
• follow the reasoning behind the one-sample z-test and apply the test
when σ is assumed known
• adapt and apply the one-sample z-test when σ is unknown
• understand and apply the two-sample z-test to analyse the difference
between means
• use Minitab to perform the one-sample z-test
• be aware of questions to ask which might lead to reservations about the
conclusions of a hypothesis test
• be aware of some of the characteristics of the z-test, and recognise when
it is necessary to exercise some caution in its use.

167
Unit 7 Factors affecting reading

Solutions to activities
Solution to Activity 1
There are many possible answers to this question. You can test a child’s
reading ability by how well they read a coherent passage, recognise
separate words, name letters, or pronounce separate words. Perhaps you
have thought of other measures; or you may have thought in terms of a
standard reading test of some kind.

Solution to Activity 2
Some of the factors you may have thought of are pre-school education,
parents’ education, precise age of child, whether there are other children in
the family, mental or physical disability, social deprivation, quality of
teaching, method of teaching, school class size, and parent’s reading to the
child at an early age. You may have been able to think of a different set of
possibilities.

Solution to Activity 3
To be a random sample of exactly the sort you met in Unit 4, the sample
would have had to be chosen by using random numbers to select children
from a sampling frame of all 7- and 8-year-old children in the country.
Clearly this was not done, so in this sense the sample is not random.
However, you have previously met examples where a sample that was not
chosen in this way was nevertheless considered to be representative in the
same way that a formally selected random sample would be. In other
words, the key question is not ‘Was this sample chosen using a sampling
frame and random numbers?’, but ‘Was this sample chosen in such a way
that it has the same properties as one chosen using a sampling frame and
random numbers?’
The answer to the second question is not so clear in this case. It might
seem reasonable to treat the original BCS sample of people born in a
particular week in 1970 as being representative of the general population of
people born in Great Britain around that time, in the same way that a
random sample would be representative. It is perhaps less reasonable to
treat their 7- and 8-year-old children as if they were a random sample from
the population of all 7- and 8-year-old children in 2004–2005. This is
because in a true random sample of children, the ages of the children’s
parents would vary more – in this sample all the children have at least one
parent born in a particular week in 1970. This might be quite a problem
because the age and experience of their parents might well be linked to
how a child’s reading develops.

168
 Solutions to activities

Solution to Activity 4
(a) The 7-year-old boys are identified in Table 1 by having a value of 1 in
the third column (Gender – 1 denotes boy) and 1 in the fourth
column (Coded age – 1 denotes 7 years old). There are six individuals
in Table 1 that have 1 in each of the third and fourth columns. They
have reading scores
106 110 134 25 172 160.
The sample size is n = 6.
(b) To calculate x,
"
x = 106 + 110 + 134 + 25 + 172 + 160 = 707,
and so
!
x 707
x= = " 117.8.
n 6
Using Method 2 from Unit 3 (Subsection 3.1) to calculate s,
" " !
2 2 ( x)2
(x − x) = x −
n
(707)2
= 97 101 −
6
" 13 792.833.
This means that the variance is
!
(x − x)2 13 792.833
=
n−1 5
" 2758.5667.
So
√ √
s= variance = 2758.5667
" 52.5.

Solution to Activity 5
The proportion of students on MS221 in the presentation in question
achieving 75 marks is the actual number of students receiving 75 marks
(21) divided by the total number of students sitting the exam (1234).
That is,
21
" 0.0170.
1234

169
Unit 7 Factors affecting reading

Solution to Activity 6
15 + 35 50
(a) (i) Sample mean = = = 25.
2 2
65 + 77 142
(ii) Sample mean = = = 71.
2 2
65 + 52 117
(iii) Sample mean = = = 58.5.
2 2
37 + 80 117
(iv) Sample mean = = = 58.5.
2 2
(b) The sample means of samples of size 2 are either integers (as in (i)
and (ii) in part (a)) or else ‘half-integers’, that is, values of the form
‘an integer plus a half’ (as in (iii) and (iv) of part (a)).

Solution to Activity 7
The distribution of sample means of size 2 shown in Figure 3 is much
smoother and less jagged than the distribution of the population data
shown in Figure 2. The distribution of sample means of size 2 is also fairly
symmetric, about a maximal value at around 70. However, there are
slightly more sample means less than 70 than greater than 70, meaning
that the distribution is slightly left-skew (see Subsection 5.2 of Unit 1).
You might also note that the distribution fades away to almost nothing –
corresponding to very rare sample mean values – at about 10 or so.

Solution to Activity 8
10 + 20 + 45 75
(a) Sample mean = = = 25.
3 3
82 + 24 + 33 139
(b) Sample mean = = " 46.3.
3 3
52 + 61 + 73 186
(c) Sample mean = = = 62.
3 3
78 + 64 + 46 188
(d) Sample mean = = " 62.7.
3 3

Solution to Activity 9
The distribution of sample means of size 3 shown in Figure 4 is much
smoother than the distribution of sample means of size 2 shown in
Figure 3 – it is made up of many more very short lines whose overall effect
is closer to a smooth curve. The sampling distribution in Figure 4 is a
little more compressed from side to side than that in Figure 3; that is, it
has a smaller spread. The sampling distribution in Figure 4 is perhaps
even closer to symmetric than the one in Figure 3. The maximum value
about which the sampling distribution is approximately symmetric is,
however, at approximately the same place as the maximum in Figure 3 –
that is, at about, or a little under, 70. Finally, corresponding to its smaller
spread, the distribution in Figure 4 fades away to almost nothing at about
20 or so (and just below 100).

170
 Solutions to activities

Solution to Activity 10
The spread of the sampling distribution in Figure 5 is a little smaller again
than the spread of the sampling distribution in Figure 4. It is also the case
that any skewness apparent in Figure 4 is no longer apparent in Figure 5:
this time, the distribution is symmetric, falling away smoothly on either
side of a maximum value a little way below 70. But aside from the change
in spread, the sampling distribution in Figure 5 is rather similar to the
sampling distribution in Figure 4; in particular, the maximum is at
approximately the same place in the two figures, while in both cases the
sampling distributions fall away from the maximum, first more rapidly and
then more slowly as they ‘level out’ a long way from the maximum.

Solution to Activity 11
As the sample size n increases, the sampling distributions, which all have
the same symmetric shape, rise more and more sharply to a mode (at a
little below 70, it seems). Also, the distributions become more and more
compressed (i.e. the spread decreases as the sample size increases).

Solution to Activity 12
For n = 2, the sampling distribution of the mean is right-skew, but a little
less so than the population distribution. As the sample size n increases,
the sampling distributions again become more symmetric and bell-shaped.
The distributions also become more and more peaked and compressed
about the mode (at about £500).

Solution to Activity 13
The centre of this normal distribution is located at the value 1, so, as in
Figure 11(b), this means that µ = 1. The distribution also appears to have
the same spread as the normal distribution in Figure 12(c), so σ = 2. To
confirm these claims, notice that the x-axis labels on Figure 11(b) have 1
added to them (when µ = 1) compared with the corresponding labels on
Figure 11(a) (when µ = 0); similarly, the x-axis labels on Figure 13 have 1
added to them (when µ = 1) compared with the corresponding labels on
Figure 12(c) (when µ = 0).
Don’t worry if you didn’t get this activity right. There is much more on
changing both µ and σ in the normal distribution in the Computer Book
and Subsections 3.2 and 3.3 to follow.

171
Unit 7 Factors affecting reading

Solution to Activity 14
(a) The mode of this normal distribution occurs at about x = 10. So
µ " 10. Almost all the distribution is contained between x = 4 and
x = 16 (i.e. within 10 ± 6). So 3σ " 6 and σ " 2. That is, the normal
distribution plotted in Figure 17 is approximately the normal
distribution with mean µ = 10 and standard deviation σ = 2.
(b) The mode of this normal distribution occurs at about x = 100. So
µ " 100. Almost all the distribution is contained between x = 40 and
x = 160 (i.e. within 100 ± 60). So 3σ " 60 and σ " 20. That is, the
normal distribution plotted in Figure 18 is approximately the normal
distribution with mean µ = 100 and standard deviation σ = 20.
(c) The mode of this normal distribution occurs at about x = 1. So
µ " 1. Almost all the distribution is contained between x = 0.7 and
x = 1.3 (i.e. within 1 ± 0.3). So 3σ " 0.3 and σ " 0.1. That is, the
normal distribution plotted in Figure 19 is approximately the normal
distribution with mean µ = 1 and standard deviation σ = 0.1.

Solution to Activity 15
You should have obtained something like the sketches below, although,
since you may have used different scales, yours could look a bit different.
The important thing is that the information on your horizontal axes
should match those in the figures.
• The following normal distribution is centred at µ = 1000 and has just
about all the distribution contained within
1000 ± (3 × 100) = 1000 ± 300, i.e. between 700 and 1300.

500 600 700 800 900 1000 1100 1200 1300 1400 1500 x
The normal distribution with µ = 1000, σ = 100

• The following normal distribution is centred at µ = 2 and has almost all

the distribution contained within 2 ± (3 × 0.25) = 2 ± 0.75, i.e. between
1.25 and 2.75.

172
 Solutions to activities

1 1.5 2 2.5 3 x
The normal distribution with µ = 2, σ = 0.25

Solution to Activity 16
(a) Here µ = 10 and σ = 2, so
x − 10
z= .
2
(b) Here µ = 100 and σ = 20, so
x − 100
z= .
20
(c) Here µ = 1 and σ = 0.1, so
x−1
z= .
0.1
If you prefer, you could equivalently write this as
x−1
z= = 10(x − 1).
1/10

Solution to Activity 17
(a) The appropriate formula is
h−µ
z= ,
σ
where µ = 1.75 and σ = 0.07. Hence
h − 1.75
z= .
0.07
(b) When h = 1.96,
1.96 − 1.75 0.21
z= = = 3.
0.07 0.07
So a height of 1.96 metres is 3 standard deviations above the mean
height of 1.75 metres.
When h = 1.61,
1.61 − 1.75 −0.14
z= = = −2.
0.07 0.07

173
Unit 7 Factors affecting reading

So a height of 1.61 metres is 2 standard deviations below the mean

height of 1.75 metres.
When h = 1.785,
1.785 − 1.75 0.035
z= = = 0.5.
0.07 0.07
So a height of 1.785 metres is 0.5 standard deviations above the mean
height of 1.75 metres.
You can check the picture of the distribution in Figure 14
(Subsection 3.2) to see if each of the values of h in this activity is the
appropriate z standard deviations away from the mean.

Solution to Activity 18
(a) In each case the sampling distribution is symmetric about a mode at
about 66 marks. So the means of the sampling distributions appear to
be the same as the population mean µ = 66 marks.
(b) The sampling distributions all look symmetric with a mode at about
£491 or so. So again the mean of each of the sampling distributions
appears to be the same as the population mean µ = £491.

Solution to Activity 19
(a) The standard deviation of the sampling distribution of mean exam
marks decreases (i.e. the distributions become more compressed) as
the sample size n increases.
(b) The standard deviation of the sampling distribution of mean
employees’ earnings also decreases (i.e. the distributions become more
compressed) as the sample size n increases.

Solution to Activity 20
(a) When n = 25,
σ 22
√ = √ = 4.4,
n 25

(b) When n = 50,

σ 22
√ = √ " 3.11.
n 50

(c) When n = 100,

σ 22
√ =√ = 2.2.
n 100

174
 Solutions to activities

Solution to Activity 21
When n = 25 and σ = 0.01,
σ 0.01 0.01
√ =√ = = 0.002.
n 25 5
It follows that the sampling distribution of the mean for samples
of 25 ball bearings from this manufacturer is approximately normal with
√
mean µ = 2 mm and standard deviation σ/ n = 0.002 mm.

Solution to Activity 22
The value of z is
x−A 112 − 120
z= = √ " −5.33.
SE 15/ 100

Solution to Activity 23
If H0 is rejected at the 5% significance level but not at the 1% significance
level, then z lies in the critical region shown in Figure 32 but not in the
critical region shown in Figure 33, that is 1.96 ≤ z < 2.58 or
−2.58 < z ≤ −1.96. This is shown in the following figure.

−2.58 −1.96 1.96 2.58

−3 −2 −1 0 1 2 3 z

z must lie in one

of these intervals

Sketch of standard normal distribution with possible values of z indicated

Solution to Activity 24
(a) The null and alternative hypotheses are
H0 : µ = 26.1
H1 : µ =
% 26.1,
where µ is the population mean set-up time of the new method.
(b) A = 26.1, as this is the value of µ under H0 . The sample values are
x = 20.9 and n = 53.

175
Unit 7 Factors affecting reading

(c) The test statistic is

x−A 20.9 − 26.1
z= = √ " −3.08.
SE 12.3/ 53

(d) As −3.08 is less than −1.96 and −2.58, the null hypothesis is rejected
at both the 5% significance level and the 1% significance level.
(e) There is strong evidence against H0 . Thus there is strong evidence
that the mean set-up time under the new method differs from that
under the old method – there is strong evidence that the new method
is faster.

Solution to Activity 25
The appropriate null and alternative hypotheses are
H0 : µ = 116
H1 : µ =
% 116,
where µ is the population mean reading score of all British 8-year-old
children in 2004–2005.
As the sample size, 283, is much greater than 25, it is appropriate to apply
the z-test. We have
A = 116, x = 126.92, n = 283, s = 27.711.
The test statistic is
x−A x−A 126.92 − 116
z= = √ = √ " 6.63.
ESE s/ n 27.711/ 283
The critical values are 1.96, −1.96 (5%) and 2.58, −2.58 (1%).
Since 6.63 ≥ 2.58, we can reject the null hypothesis at the 1% significance
level and conclude that there is strong evidence that the mean reading
score of 8-year-olds in 2004–2005 was not equal to 116.
This might have been a little surprising if we had not seen a similar result
for 7-year-old children in Example 7.

Solution to Activity 26
The null and alternative hypotheses are
H0 : µ = 381.50
H1 : µ =
% 381.50,
where µ is the population mean weekly wage (in £) of female local
government clerical officers and assistants in 2011.
As the sample size, 810, is greater than 25, it is appropriate to apply the
z-test. We have
A = 381.5, x = 373.4, n = 810, s = 138.2.

176
 Solutions to activities

The test statistic is

x−A x−A 373.4 − 381.5
z= = √ = √ " −1.67.
ESE s/ n 138.2/ 810
The critical values are 1.96, −1.96 (5%) and 2.58, −2.58 (1%).
Since −1.96 < −1.67 < 1.96, the null hypothesis is not rejected at the 5%
significance level. There is little evidence that the mean weekly wage of
female local government clerical officers and assistants differed from the
overall mean weekly wage of female employees in 2011.

Solution to Activity 27
The null and alternative hypotheses are
H 0 : µg = µ b
H 1 : µg =
% µb ,
where µg and µb are the population mean reading scores for 8-year-old
girls and boys, respectively. We have:
xg = 127.49, xb = 126.38, ng = 138, nb = 145,
sg = 25.064, sb = 29.927.
Both ng = 138 and nb = 145 are greater than 25, so we can assume that
the z-test is applicable.
The estimated standard error is
' &
s2g s2b 25.0642 29.9272
ESE = + = + " 3.276,
ng nb 138 145
and the test statistic is
xg − x b 127.49 − 126.38
z= " " 0.34.
ESE 3.276
The critical values are 1.96, −1.96 (5%) and 2.58, −2.58 (1%).
Since −1.96 < 0.34 < 1.96, we cannot reject the null hypothesis at the 5%
significance level. There is no reason to doubt that the mean reading
scores of 8-year-old girls and boys were the same.

177
Unit 7 Factors affecting reading

Solution to Activity 28
Let ‘E’ denote quantities relating to children whose parental education
ended by age 16, and ‘C’ denote quantities relating to children whose
parental education continued after age 16. The null and alternative
hypotheses are
H 0 : µC = µ E
H 1 : µC =
% µE ,
where µC and µE are the population mean reading scores of interest.
We have:
xE = 116.12, xC = 123.15, nE = 389, nC = 199,
sE = 28.775, sC = 24.603.
Both nC = 199 and nE = 389 are greater than 25, so we can assume that
the z-test is applicable.
The estimated standard error is
' &
s2C s2E 24.6032 28.7752
ESE = + = + " 2.274,
nC nE 199 389
and the test statistic is
xC − xE 123.15 − 116.12
z= " " 3.09.
ESE 2.274
The critical values are 1.96, −1.96 (5%) and 2.58, −2.58 (1%).
Since 3.09 ≥ 2.58, we can reject H0 at the 1% significance level. There is
strong evidence that the mean reading score of children whose parental
education continued after age 16 differs from the mean reading score of
children whose parental education ended by age 16. There is strong
evidence that the children of parents who stayed longer in full-time
education did better than those of parents who left education earlier.

178
 Solutions to exercises

Solutions to exercises
Solution to Exercise 1
(a) The 8-year-old children are identified in Table 1 by having a value of 2
in the fourth column (Coded age ‘2’ denotes 8 years old). There are
four individuals in Table 1 that have 2 in the fourth column. They
have reading scores
118 115 56 136.
The sample size is n = 4.
(b) To calculate x,
"
x = 118 + 115 + 56 + 136 = 425,
and so
!
x 425
x= = = 106.25.
n 4
To calculate s,
" " !
( x)2
(x − x)2 = x2 −
n
4252
= 48 781 −
4
= 3624.75,
which means the variance is
!
(x − x)2 3624.75
= = 1208.25.
n−1 3
So
√ √
s= variance = 1208.25 " 34.8.

Solution to Exercise 2
(a) The children of interest in this exercise are identified in Table 1 by
having a value of 1 in the fifth column (Parental education ‘1’ denotes
finished aged 16 or less) and a value of 1 in the sixth column (Father’s
occupation ‘1’ denotes managerial, technical, professional and skilled
non-manual occupations). There are seven individuals in Table 1 that
have 1 in both the fifth and sixth columns. They have reading scores
123 110 134 110 172 136 160.
The sample size is n = 7.
(b) To calculate x,
"
x = 123 + 110 + 134 + 110 + 172 + 136 + 160 = 945,

179
Unit 7 Factors affecting reading

and so
!
x 945
x= = = 135.
n 7
To calculate s,
" " !
2 2 ( x)2
(x − x) = x −
n
9452
= 130 965 −
7
= 3390,
which means the variance is
!
(x − x)2 3390
= = 565.
n−1 6
So
√ √
s= variance = 565 " 23.8.

Solution to Exercise 3
Suitable null and alternative hypotheses are
H0 : For British children aged 8 in 2004–2005, the mean reading score
for girls was equal to the mean reading score for boys
H1 : For British children aged 8 in 2004–2005, the mean reading score
for girls was not equal to the mean reading score for boys.

Solution to Exercise 4
(a) For Population A, the six different samples of size 2 with their sample
means are listed below:
10 + 20 30
Sample: 10 20; sample mean = = = 15
2 2
10 + 30 40
Sample: 10 30; sample mean = = = 20
2 2
10 + 40 50
Sample: 10 40; sample mean = = = 25
2 2
20 + 30 50
Sample: 20 30; sample mean = = = 25
2 2
20 + 40 60
Sample: 20 40; sample mean = = = 30
2 2
30 + 40 70
Sample: 30 40; sample mean = = = 35
2 2
The sample means are plotted along the horizontal axis in the
following figure.

15 20 25 30 35
Plot of values of sample means from Population A

180
 Solutions to exercises

(b) For Population B, the six different samples of size 2 with their sample
means are listed below:
10 + 38 48
Sample: 10 38; sample mean = = = 24
2 2
10 + 39 49
Sample: 10 39; sample mean = = = 24.5
2 2
10 + 40 50
Sample: 10 40; sample mean = = = 25
2 2
38 + 39 77
Sample: 38 39; sample mean = = = 38.5
2 2
38 + 40 78
Sample: 38 40; sample mean = = = 39
2 2
39 + 40 79
Sample: 39 40; sample mean = = = 39.5
2 2
The sample means are plotted along the horizontal axis in the
following figure.

24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39
Plot of values of sample means from Population B

(c) The points in the graph in part (a) are symmetrically distributed
around a central mode, while the points in the graph in part (b) are
split into two groups some distance apart. Hence the graph in part (a)
seems more bell-shaped than the graph in part (b). This happens
because the points in Population A are more symmetric – and more
evenly spread out – than the points in Population B, which consist of
three points close together (38, 39 and 40) and another far away (10).

Solution to Exercise 5
The distribution of reading scores – ‘sample means’ when n = 1 – is very
jagged, but if you squint your eyes you get an impression of a fairly
symmetric distribution with perhaps a slight preponderance of low, as
opposed to high, values. The distribution of sample means of size n = 2 is
smoother, though still with some jaggedness towards its right-hand side,
fairly close to symmetric but with a little bit of left skewness. When n = 3
the distribution is smoother again, and any lack of symmetry is pretty
small. It is also clear that the vertical scale of the sampling distribution of
the mean when n = 3 is larger than the vertical scale of the distribution of
the data (n = 1). By the time n = 10, the distribution of sample means is
very smooth, symmetric, bell-shaped/normal-like and with a larger vertical
scale still.
So, again, we see that even though the population distribution is not
especially normal-like, as the sample size n increases, the sampling
distribution of the mean quite quickly becomes much more normal-like.

181
Unit 7 Factors affecting reading

Solution to Exercise 6
The mode of this normal distribution occurs at about x = 10. So µ " 10.
Almost all the distribution is contained between x = 0 and x = 20
(i.e. within 10 ± 10). So 3σ " 10 and σ " 3.33. That is, the normal
distribution plotted in Figure 28 is approximately the normal distribution
with mean µ = 10 and standard deviation σ = 3.33.

Solution to Exercise 7
The mode of this normal distribution occurs at about x = −5. So µ " −5.
Almost all the distribution is contained between x = −12 and x = 2
(i.e. within −5 ± 7). So 3σ " 7 and σ " 2.33. That is, the normal
distribution plotted in Figure 29 is approximately the normal distribution
with mean µ = −5 and standard deviation σ = 2.33.

Solution to Exercise 8
You should have obtained something like the sketch in the figure below,
although, since you may have used different scales, yours could look a bit
different. This normal distribution is centred at µ = −1 and has just about
all the distribution contained within −1 ± (3 × 1) = −1 ± 3,
i.e. between −4 and 2.

−4 −3 −2 −1 0 1 2 x
A normal distribution with µ = −1, σ = 1

182
 Solutions to exercises

Solution to Exercise 9
You should have obtained something like the sketch in the figure below,
although, since you may have used different scales, yours could look a bit
different. This normal distribution is centred at µ = 4 and has just about
all the distribution contained within 4 ± (3 × 4) = 4 ± 12, i.e. between −8
and 16.

−8 −4 0 4 8 12 16 x
A normal distribution with µ = 4, σ = 4

Solution to Exercise 10
(a) Here µ = 6 and σ = 3.3, so
x−6
z= .
3.3
(b) Here µ = −6 and σ = 2, so
x − (−6) x+6
z= = .
2 2

Solution to Exercise 11
The appropriate formula is
x−2
z= .
10
When x = 3,
3−2 1
z= = = 0.1.
10 10

183
Unit 7 Factors affecting reading

Solution to Exercise 12
The appropriate formula is
x − (−1) x+1
z= = = 2(x + 1).
0.5 1/2
When x = 0,
z = 2(0 + 1) = 2.

Solution to Exercise 13
(a) When n = 9,
σ 283 283
√ = √ = " 94.3.
n 9 3

(b) When n = 25,

σ 283 283
√ =√ = = 56.6.
n 25 5

(c) When n = 100,

σ 283 283
√ =√ = = 28.3.
n 100 10

Solution to Exercise 14
(a) When n = 4,
σ 3.6 3.6
√ =√ = = 1.8.
n 4 2

(b) When n = 19,

σ 3.6
√ = √ " 0.83.
n 19

(c) When n = 300,

σ 3.6
√ =√ " 0.21.
n 300

Solution to Exercise 15
When n = 40 and σ = 0.01,
σ 0.01
√ = √ " 0.0016.
n 40
It follows that the sampling distribution of the mean for samples of 40
one-litre bottles of water from this manufacturer is approximately normal
√
with mean µ = 1.01 litres and standard deviation σ/ n = 0.0016 litres.

184
 Solutions to exercises

Solution to Exercise 16
The appropriate null and alternative hypotheses are
H0 : µ = 96
H1 : µ =
% 96,
where µ is the population mean reading score of all British 7-year-old girls
in 2004–2005.
As the sample size, n = 190, is much greater than 25, it is appropriate to
apply the z-test. We have
A = 96, x = 113.42, n = 190, s = 25.464.
The test statistic is
x−A x−A 113.42 − 96
z= = √ = √ " 9.43.
ESE s/ n 25.464/ 190
The critical values are 1.96, −1.96 (5%) and 2.58, −2.58 (1%). Since
9.43 ≥ 2.58, we can reject the null hypothesis at the 1% significance level.
Hence there is strong evidence that the mean reading score of 7-year-old
girls in 2004–2005 is not equal to 96.
This result corresponds to the similar result observed for all 7-year-old
children (not just girls) in Example 7.

Solution to Exercise 17
The null and alternative hypotheses are
H0 : µ = 80
H1 : µ =
% 80,
where µ is the mean weight (in kg) of this breed of pig when fed the
special diet.
As the sample size, 533, is greater than 25, it is appropriate to apply the
z-test. We have
A = 80, x = 81.92, n = 533, s = 15.65.
The test statistic is
x−A x−A 81.92 − 80
z= = √ = √ " 2.83.
ESE s/ n 15.65/ 533
The critical values are 1.96, −1.96 (5%) and 2.58, −2.58 (1%). Since
2.83 ≥ 2.58, the null hypothesis is rejected at the 1% significance level.
There is strong evidence that the mean weight of this breed of pig when
fed the special diet is not equal to 80 kg. There is strong evidence that the
mean weight is higher for the special diet.

185
Unit 7 Factors affecting reading

Solution to Exercise 18
(a) The null and alternative hypotheses are
H0 : µ = 4
H1 : µ =
% 4,
where µ is the population mean drying time in hours of the
manufacturers’ paint. As the sample size, n = 40, is greater than 25,
it is appropriate to apply the z-test. We have
A = 4, x = 3.80, n = 40, s = 0.55.
The test statistic is
x−A x−A 3.80 − 4
z= = √ = √ " −2.30.
ESE s/ n 0.55/ 40
The critical values are 1.96, −1.96 (5%) and 2.58, −2.58 (1%). Since
−2.58 < −2.30 ≤ −1.96, we can reject the null hypothesis at the 5%,
though not at the 1%, significance level. We conclude that there is
moderate evidence that the drying time given by the consumer
magazine is incorrect. The manufacturers’ paint appears to dry more
quickly than the magazine claimed.
(b) You might think that such ‘marginal’ (moderate) evidence is not
enough to conclude that the manufacturers’ paint dries more quickly
than the consumer magazine claimed.
The time at which paint is declared ‘dry’ is not well-defined: different
customers might measure drying time differently or have different
ideas about what ‘dry’ means.
Even if the measures are reliable and the test result is correct, 0.20
hours or 12 minutes is not a very large reduction. Most customers
would not consider this an important difference.
You may have thought of other reservations.

Solution to Exercise 19
The null and alternative hypotheses are
H 0 : µ1 = µ2
H 1 : µ1 =
% µ2 ,
where µ1 and µ2 are the population mean reading scores of interest. Here
and below, ‘1’ denotes quantities relating to children with father’s
occupation coded 1 and ‘2’ denotes quantities relating to children with
father’s occupation coded 2. The summary statistics are:
x1 = 120.55, x2 = 117.17, n1 = 316, n2 = 203,
s1 = 24.221, s2 = 30.085.
Both n1 = 316 and n2 = 203 are greater than 25, so we can assume that
the z-test is applicable.

186
 Solutions to exercises

The estimated standard error is

' &
s21 s22 24.2212 30.0852
ESE = + = + " 2.513,
n1 n2 316 203
and the test statistic is
x1 − x 2 120.55 − 117.17
z= " " 1.35.
ESE 2.513
(You might have got 1.34, correct to two decimal places, if calculating z all
in one go. Such a difference doesn’t matter.)
The critical values are 1.96, −1.96 (5%) and 2.58, −2.58 (1%). Since
−1.96 < 1.35 < 1.96, we cannot reject H0 at the 5% significance level.
There is little evidence that the mean reading score of children whose
father had an occupation coded 1 differed from the mean reading score of
children whose father had an occupation coded 2. This goes against
conventional wisdom, at least from other contexts.

Solution to Exercise 20
Let ‘A’ denote quantities relating to breast-fed babies and ‘B’ denote
quantities relating to bottle-fed babies. The null and alternative
hypotheses are
H 0 : µA = µ B
H 1 : µA =
% µB ,
where µA and µB are the population mean serum calcium levels of interest.
The summary statistics are:

xA = 2.45, xB = 2.30, nA = 64, nB = 169,

sA = 0.292, sB = 0.274.
Both nA = 64 and nB = 169 are greater than 25, so we can assume that
the z-test is applicable.
The estimated standard error is
' &
s2A s2B 0.2922 0.2742
ESE = + = + " 0.042,
nA nB 64 169
and the test statistic is
xA − x B 2.45 − 2.30
z= " " 3.57.
ESE 0.042
(You might have got 3.56.)
The critical values are 1.96, −1.96 (5%) and 2.58, −2.58 (1%). Since
3.57 ≥ 2.58, we can reject the null hypothesis at the 1% significance level.
There is strong evidence that the mean serum calcium level of week-old
babies was different depending on whether they were breast-fed or
bottle-fed. The evidence is that it was higher in those who were breast-fed.

187
Unit 7 Factors affecting reading

Solution to Exercise 21
Let ‘R’ denote quantities relating to children from rural areas and ‘U’
denote quantities relating to children from urban areas. The null and
alternative hypotheses are
H 0 : µR = µ U
H 1 : µR =
% µU ,
where µR and µU are the population mean peak flow rates of interest (in
litres per minute). The summary statistics are:
xU = 226, xR = 231, nU = 485, nR = 637,
sU = 52, sR = 53.
Both nU = 485 and nR = 637 are greater than 25, so we can assume that
the z-test is applicable.
The estimated standard error is
' &
s2R s2U 532 522
ESE = + = + " 3.160,
nR nU 637 485
and the test statistic is
x R − xU 231 − 226
z= " " 1.58.
ESE 3.160
The critical values are 1.96, −1.96 (5%) and 2.58, −2.58 (1%).
Since −1.96 < 1.58 < 1.96, we cannot reject the null hypothesis at the 5%
significance level. Thus, there is little evidence to suggest that the mean
peak flow rate differs between children who live in rural areas and those
who live in urban areas.

188
 Acknowledgements

Acknowledgements
Grateful acknowledgement is made to the following sources:
Subsection 1.1 photo (a reading class): Christopher Futcher /
[Link]
Subsection 1.2 figure (Changing Britain, Changing Lives):
[Link]/Changing-Britain-Lives-Generations-Century/
dp/0854736506
Subsection 1.2 figure (BAS 3 documentation): [Link]
Section 2 photo (Big Ben), taken from: [Link]/
mybellringing/[Link]
Section 3 cartoon (statistics in Greece): [Link]
Subsection 3.2 figure (‘abnormally normal or normally abnormal’), taken
from: [Link]/[Link]?pid=7362503
Subsection 5.2 ESE logo: [Link].k12.fl.us/ese/[Link]
Section 6 ESE logo (‘a more relaxing form’): Easy Serving Espresso,
[Link]
c Otnaydur / [Link]
Section 8 photo (reserved sign) $
Every effort has been made to contact copyright holders. If any have been
inadvertently overlooked the publishers will be pleased to make the
necessary arrangements at the first opportunity.

189