9 Hypothesis testing using the

normal distribution
Skills check iii) Accept H0 if
⎛ 40 40 ⎞
x ∈ ⎜ 38.3 − 1.96 × , 38.3 + 1.96 × ⎟
1. P( X > x ) = P ⎛⎜ z > x − 57 ⎞⎟ = 0.025 ⎝ 60 60 ⎠
⎝ 12 ⎠
x − 57
= (36.7, 39.9) so reject H0
⇒ = 1.96 ⇒ x = 63.8
12 19
2. i) Accept H0 if x > 81.3 − 1.645 × = 79.7 so
20
reject H0
Exercise 9.1
34
Solutions for questions 1 and 2 are shown by both methods ii) Accept H0 if x > 81.3 − 1.645 × = 79.2 so
20
below. (cv: critical value) accept H0
34
Method 1: iii) Accept H0 if x > 81.3 − 1.645 × = 80.1 so
60
reject H0
1. i) Two-tail 5% test so cvs are ±1.96,
40.6 − 38.3 3. i) Comparing the first and second tests in each
z= = 2.57 > 1.96 so reject H0.
4 question: the population has a larger variance
20 in the second one and the acceptance region
ii) Two-tail 5% test so cvs are ±1.96, includes more values. It is harder to be sure
40.6 − 38.3 that an observed difference would not occur
z= = 1.626 < 1.96 so accept H0.
40 because of the randomness in the sampling
20 process. The same observed mean results in
iii) Two-tail 5% test so cvs are ±1.96, ii being rejected where i was accepted.
40.6 − 38.3 ii) Comparing the second and third tests in each
z= = 2.817 > 1.96 so reject H0. question: the sample size in the third one
40
60 is larger and the acceptance region shrinks.
With more data it is easier to be sure that the
2. i) One-tail 5% test so cv is −1.645,
observed difference would not occur because
79.5 − 81.3
z= = −1.847 < −1.645 so reject H0. of the randomness in the sampling process
19
– and the test goes back to accepting the null
20
hypothesis.
ii) One-tail 5% test so cv is −1.645,
79.5 − 81.3 4. H0: μ = 351.2 vs H1: μ ≠ 351.2 are the hypotheses
z= = −1.381 > −1.645 so accept H0.
34 for the test, with σ = 4.1; n = 40. Accept H0 if
20 ⎛ 4 .1 4 .1 ⎞
x ∈ ⎜ 351.2 − 1.96 × , 351.2 + 1.96 × ⎟
iii) One-tail 5% test so cv is −1.645, ⎝ 40 40 ⎠
79.5 − 81.3 = (349.9, 352.5)
z= = −2.391 < −1.645 so reject H0.
34 The sample mean is 349.9, so accept H0 and
60 conclude there is not sufficient evidence at the 5%
level to suggest the mean has changed.
Method 2:
Note that 349.9 is right on the border of the
1. i) Accept H0 if acceptance region. If you took the calculation to
⎛ 4 4 ⎞ more than 1 d.p. it is 349.93 and the sample mean
x ∈ ⎜ 38.3 − 1.96 × , 38.3 + 1.96 × ⎟
⎝ 20 20 ⎠ would be outside (and if you calculate the z-score
= (36.5, 40.1) so reject H0 it would be marginally outside the –1.96 limit, and
ii) Accept H0 if the test would be rejected). However, the data does
not have the degree of accuracy attached to it to
⎛ 40 40 ⎞
x ∈ ⎜⎜ 38.3 − 1.96 × , 38.3 + 1.96 × ⎟ justify making the call for rejecting the null
⎝ 20 20 ⎟⎠ hypothesis – which is a much stronger statement
= (35.5, 41.1) so accept H0 than accepting it.

© Oxford University Press 2018: this may be reproduced for class use solely for the purchaser’s institute Hypothesis testing using the normal distribution 1
5. H0: μ = 105 vs H1: μ > 105 are the hypotheses for the n = 40 P(Type II error) = 0.038
test, with σ = 7; n = 10; 2% test: Accept H0 n = 45 P(Type II error) = 0.029
7
if x < 105 + 2.054 × = 109.54… n = 50 P(Type II error) = 0.0234
10
For the data given x = 112.5 so reject H0 and n = 48 P(Type II error) = 0.0256
conclude that there is evidence at the 2% level to n = 49 P(Type II error) = 0.0245
suggest that this group has higher blood pressure
than usual. So a sample size of 49 is the smallest which
meets the MD’s request.
6. H0: μ = 85.3 vs H1: μ ≠ 85.3 are the hypotheses for
the test, with σ = 3.2. Exercise 9.2
Plan A:
i) n = 5, so 5% test is: Accept H0 if 1. i) Σx = 1846; Σx 2 = 86 835; n = 40
40  86 835
1846
−x 
2
= 46.15; s = 
2
⎛ 3 .2 3.2 ⎞ ⇒ x=
x ∈ ⎜ 85.3 − 1.96 × , 85.3 + 1.96 × ⎟ 40 39 40 
⎝ 5 5⎠
= 42.11
= (82.5, 88.1)
Two-tail 5% test is ‘accept H0 if
⎧ ⎛ 3.22 ⎞ ⎫
ii) P ⎨N ⎜ 87.3, ⎟ ∈ (82.5, 88.1) ⎬ ⎛ 42.11 42.11 ⎞
⎩ ⎝ 5 ⎠ ⎭ x ∈ ⎜ 45 − 1.96 × , 45 − 1.96 × ⎟
⎧ ⎫ ⎝ 40 40 ⎠
⎪⎪ 82.5 − 87.3 88.1 − 87.3 ⎪⎪ = (43.0, 47.0)’. Since x = 46.15, you accept H0
= P⎨ =− 3.354 < z < =0.559 ⎬
⎪ 3 . 2 3 . 2 ⎪ and conclude there is not sufficient evidence at
⎩⎪ 5 5 ⎪⎭ the 5% level to suggest that the mean is not 45.
= Φ(0.559) − Φ(−3.354) = 0.713 – 0 = 0.713
ii) Σx = 3514; Σx 2 = 199 281; n = 63
Note that using a downward shift in the mean of 3514
⇒ x = = 55.78;
2 hours would give the same answer because of 63
the symmetry of the situation (the top / bottom 63  199 281 2 
s =
2
−x
tail probabilities would be reversed). 62  63  = 52.87 one-tail 5%
test is ‘accept H0 if
Plan B:
52.87
i) n = 30, so 5% test is: accept H0 if x > 57.5 − 1.645 × = 56.0’
63
⎛ 3 .2 3 .2 ⎞
x ∈ ⎜ 85.3 − 1.96 × , 85.3 + 1.96 × ⎟ 
Since x = 55.78, you reject H0 and conclude
⎝ 30 30 ⎠
there is evidence at the 5% level to suggest that
= (84.2, 86.4)
the mean is less than 57.5.
⎪⎧ ⎛ 3.22 ⎞ ⎫⎪
ii) P ⎨N ⎜ 87.3, ⎟ ∈ ( 84.2,86.4 ) ⎬ iii) Σx = 24.3; Σx 2 = 994.27; n = 51
⎩⎪ ⎝ 30 ⎠ ⎭⎪
24.3 51 ⎛ 994.27 ⎞
⇒ x= = 0.476 =; s2 ⎜ −=
x 2 ⎟ 19.65
 84.2 − 84.3 = −5.31 < z < 86.4 − 87.3 = −1.463  51 50 ⎝ 51 ⎠
= P 3.2 3.2  one-tail 5% test is ‘accept H if
 30 30  0

19.54
x < − 0.5 + 1.645 × = 0.52’
= Φ(–1.463) – Φ(–5.36) = 0.072 – 0.000 = 0.072 51

Note: this illustrates why taking large samples 

Since x = 0.476, you accept H0 and conclude
is so much more effective in discriminating there is not sufficient evidence at the 5% level to
when there has been an important shift in the suggest that the mean has increased from −0.5.
mean. The probability of accepting the null 2. i) Using the normal approximation in a two-
hypothesis incorrectly has been reduced from tail 5% test (cvs ±1.96) gives the test as
71% to just 7%. ‘Accept H0 if the proportion observed is
iii) Seeing the shift from 71% to 6% as the sample ⎛ 0 .4 × 0 .6 0 .4 × 0 .6 ⎞
size increased from 5 to 30 gives a basis for ∈ ⎜ 0.4 − 1.96 × , 0.4 + 1.96 × ⎟
⎝ 145 145 ⎠
trial and improvement. (A spreadsheet could
= (0.320, 0.480)’
be used to reduce the effort required – this is
52
not something you would be asked to do in an Since the proportion observed is = 0.359,
145
examination.)

© Oxford University Press 2018: this may be reproduced for class use solely for the purchaser’s institute Hypothesis testing using the normal distribution 2
 you accept H0 and conclude there is not suff  icient ii) For a 5% one-tail test the cv is 1.645, so reject
evidence at the 5% level to suggest that the H0 and conclude that there is evidence at the
proportion is not 0.4. 5% level to suggest that the mean height has
increased.
ii) Using the normal approximation in a one-tail
1% test (cv is −2.326) gives the test as 6. a) i) 0.864 ii) 0.824
‘Accept H0 if the observed value is b) Around holiday periods and birthdays, there are
> 27.1 – 2.326 × 27.1 = 15.0’ likely to be more letters than usual.
Since the observed value is 19, you accept H0
c) Accept H0 and conclude there is not suff  icient
and conclude there is not sufficient evidence at
evidence to suggest the mean has increased.
the 1% level to suggest that the parameter is less
than 27.1. Exercise 9.3
iii) Using the normal approximation in a one-tail 257
1. a) n = 50; Σt = 257; Σt2 = 2433 ⇒
= t = 5.14;
2% test (cv is 2.054) gives the test as 50
50 ⎛ 2433 ⎞
‘Accept H0 if the proportion observed is =s2 ⎜ − x 2 ⎟ = 22.69
49 ⎝ 50 ⎠
0 .8 × 0 . 2
< 0.8 + 2.054 × = 0.8857’ 90% CI is
92
65 ⎛ 22.69 22.69 ⎞
Since the proportion observed is = 0.706, ⎜ 5.14 − 1.645 × , 5.14 + 1.645 × ⎟
92 ⎝ 50 50 ⎠
you accept H0 and conclude there is not = (4.03, 6.25)
sufficient evidence at the 5% level to suggest
that the proportion has increased from 0.8. 4.5 lies in this CI so a test at 10% level would
accept H0.
Note – if you calculated the observed
proportion was < 0.8, you could have drawn b) n = 117; Σx = 8596; Σx 2 = 2 186 591
this conclusion without doing the full ⇒
= x
8596
= 73.5;
calculations. 117
117 ⎛ 2186 591
3. X ~ B(85, p) H0 : p = 0.2 vs H1 : p > 0.2 = s2 ⎜ − x 2 ⎞⎟ = 13 405.5
116 ⎝ 117 ⎠
Using the normal approximation in a
one-tail 5% test (cv is 1.645) gives the test as 95% CI is
‘Accept H0 if the proportion observed is ⎛ 13 291 13 291 ⎞
⎜⎜ 73.5 − 1.96 × , 73.5 + 1.96 × ⎟
117 117 ⎟⎠
< 0.2 + 1.645 × 0.2 × 0.8 = 0.271’ ⎝
85 = (52.5, 91.5)
22
Since the proportion observed is = 0.259, you 90 lies in this CI so a test at 5% level would
85
accept H0 and conclude there is not suff  icient accept H0.
evidence at the 5% level to suggest that the c) n = 82; Σv = −459; Σv2 = 5109
proportion is greater than 0.2. −459
⇒v = = − 5.60
82
4. If X is the number of accidents in the six month
82 ⎛ 5109 2⎞
period, then test H0 : λ = 1.2 vs H1 : λ < 1.2 =s2 ⎜ − x= ⎟ 31.35
81 ⎝ 82 ⎠
where λ is the average number per week,
approx
X ~ Po(31.2) ~ N(31.2, 31.2), so the 1% 99% CI is
one-tail test (cv is 2.326) is ‘Accept H0 if ⎛ 31.35 31.35 ⎞
⎜ −5.60 − 2.576 × , − 5.60 + 2.576 × ⎟
⎝ 82 82 ⎠
X > 31.2 − 2.326 × 31.2 =
18.2’.
= (−7.19, −4.00)
Since 20 is > 18.2, you accept H0 and conclude
there is not sufficient evidence at the 1% level to −6 lies in this CI so a test at 1% level would
suggest that the average rate of accidents has accept H0.
decreased. 2. i) 95% CI is
x −μ 75.2 − 74.0 ⎛ 26.8 26.8 ⎞
5. i) z = ⇒ 1.867 = ⎜ 600.8 − 1.96 × , 600.8 + 1.96 × ⎟
s 5 .3 150 150 ⎠
⎝
n n = (599.97, 601.63)
1.867 × 5.3 
ii) 602 lies above this 95% CI so it does not support
⇒ n = = 8.246 ⇒ n = 68
75.2 − 74.0 the claim that the mean contents are 602 ml.

© Oxford University Press 2018: this may be reproduced for class use solely for the purchaser’s institute Hypothesis testing using the normal distribution 3
3. i) 95% CI is 4. H0: λ = 2.3 vs H1: λ < 2.3; where λ is the rate per
⎛ 16.8 16.8 ⎞ week. If X is the number of accidents in a 13-week
⎜ 248.5 − 1.96 × , 248.5 + 1.96 × ⎟ approx
⎝ 80 80 ⎠ period then X ~ Po(29.9) ~ N(29.9, 29.9) so the
= (247.6, 249.4) 5% one-tail test (cv is 1.645) is: ‘accept H0 if
ii) 250 lies above this 95% CI so it does not support X < 29.9 – 1.645 × 29.9 = 20.9’.
the claim that the mean contents are 250 g. Since 18 < 20.9, you reject H0 and conclude there is
evidence at the 5% level to suggest that the average
Summary exercise 9
rate of accidents has decreased.
1. i) A decrease might get them into legal diff  iculties,
and an increase in the mean contents will cost 5. H0: μ = 64.0 vs H1: μ > 64.0; s = 5.6, n = 60; so the
them money. 1% one-tail test (cv is 2.326) is: ‘accept H0 if
5.6
ii) H0: μ = 452.7 vs H1 : μ ≠ 452.7 are the hypotheses X < 64.0 + 2.326 × = 65.7’.
60
for the test, with σ = 6.1; n = 70 the test is: Accept H0
Since 65.2 < 65.7, you accept H0 and conclude
⎛ 6 .1 6 .1 ⎞
if x ∈ ⎜ 452.7 − 1.96 × , 452.7 + 1.96 × ⎟ there is not sufficient evidence at the 1% level to
⎝ 70 70 ⎠
suggest that the average height has increased.
= (451.3, 454.1). Reject H0 and conclude that
there is evidence at the 5% level to suggest that
6. a) This is a one-tail test of H0: µ = 14 vs H1: µ > 14,
the mean has changed.
with critical region z > 1.645 for a 5% test.
2. H0: μ = 80 vs H1 : μ < 80 are the hypotheses 14.12 − 14
z= = 1.61 , so accept H0 and conclude
for the test, with σ = 9; n = 12 the test is: Accept 0.5
9 45
H0 if x < 80 − 1.645 × = 75.7. that there is not sufficient evidence to suggest the
12
From the data, x = 52.3 so reject H0 and conclude mean length of pencils is greater than 14 cm.
b) Since this is a large sample (> 30) the Central
that there is evidence at the 5% level to suggest that
the pulse rates of trained athletes is lower than that Limit Theorem allows use of the normal to
of healthy young adults. approximate the sampling distribution even if
the distribution of the length of pencils is not
3. i) To take measurements of a complete population normal.
can be very expensive and take a long time.
A sample is cheaper and offers quicker access to 7. a) Hypotheses are H0: µ = 34.2 vs H1: µ < 34.2.
a reasonable level of information. The critical z-score for a one-tail 5% test is −1.645.
ii) X ~ B(65, p). H0 : p = 0.4 vs H1 : p < 0.4 33.5 − 34.2
z = = −1.90 < −1.645, so reject H0 and
2.6
Using the normal approximation in a
50
one-tailed 5% test (cv is 1.645) gives the
conclude that there is evidence at the 5% level
test as: ‘Accept H0 if the proportion observed
of significance to suggest the mean time for the
0 .4 × 0 .6 procedure has reduced.
is > 0.4 – 1.645 × = 0.30’. The observed
65 b) Note: there is no single ‘right answer’ to
15 this – the following is an illustration of a
proportion is = 0.23 so the claim would be
65 possible answer.
rejected at the 5% level of significance. The test was significant, suggesting that the mean

iii) What the spokesperson says may be true – and time is lower, but on average it has reduced by
you can construct scenarios which give plausible 0.7 minutes, which is 2%. The hospital should
explanations why this may not be reflected in consider cost, both the cost of the extra training
the customers’ experiences. For instance if they and the potential savings of a 2% reduction in
reduced very slightly the 40% most expensive operating time, and consider how much other
holidays that very few people actually bought last change the staff are experiencing – whether this
year then they could make the claim accurately is a reasonable burden if there are worthwhile
even if it presents a misleading picture to the savings to be made.
customer.

© Oxford University Press 2018: this may be reproduced for class use solely for the purchaser’s institute Hypothesis testing using the normal distribution 4
8. If X is the number of accidents in 365 days at Using the continuity correction,
that bend previously, then 243.5 − 215
z= = 2.57 > 2.326, so reject H0 and
4 2 approx 2 2 122.55
X ∼ Po  365 × = 48  ∼ N  48 ,48  .
 30 3  3 3 conclude that there is evidence at the 1% level
2 2 of significance to suggest that there has been
Test will be H0: µ = 48 vs H1: µ < 48 . an increase in the number of people taking
3 3
The critical z-score for a one-tail 5% test is −1.645. regular exercise in that area.
Using a continuity correction, b) A Type I error in this context would be to
2 conclude that there was an increase in people
34.5 − 48
z= 3 = −2.03 < −1.645, so reject H and taking regular exercise when there had been
0
2 no increase. Probability of Type I error is 5% –
48
3 since it is a test using a continuous distribution,
conclude that there is evidence at the 5% level it is the same as the significance level.
of significance to suggest that the number of c) If the aim of the advertising campaign is
accidents at that bend has reduced. to have a long-term effect on the exercise
behaviour of people it would be better to
9. a) Using a normal approximation, since n is large
conduct the survey after a reasonable gap in
and p is close to 0.5, time. A random sample of 500 people is large
approx
B ( 500,0.43) ∼ N ( 215,122.55) so enough that the results should be fairly reliable
H0: µ = 215 vs H1: µ > 215. The one-tail 1% if you can assume people answer truthfully.
critical value is 2.326.

© Oxford University Press 2018: this may be reproduced for class use solely for the purchaser’s institute Hypothesis testing using the normal distribution 5