Solutions Exercises Chapter 3 Statistics for Engineers

Exercise 1
a. An estimate of the expectation (µ) is the sample mean: 𝑥̅ = 28.25.
An estimate of the variance (𝜎 2 ) is the sample variance: 𝑠 2 = 14.37
(Remark: the notations µ = 28.25 and 𝜎 2 = 14.37 are false is, since µ and σ remain unknown!)
𝑠
b. Requested: a 95%-CI for μ with unknown σ (directly from the formula sheet) has bounds: 𝑥 ± 𝑐 ,
√𝑛
𝛼
where 𝑐 = 2.131, such that 𝑃(𝑇15 ≥ 𝑐) = = 0.025, from the 𝑡15 -distribution (→ brief notation for the
2
𝑡-distribution with 𝑑𝑓 = 15)
√14.37 √14.37
Substitute into the formula: 95% − 𝐶𝐼(µ) = (28.25 − 2.131 ∙ , 28.25 + 2.131 ∙ )
√16 √16
= (28.25 − 2.02, 28.25 − 2.02) = (26.23, 30.27)
(Note: in the formula we did not use the capitals 𝑋 and S , since 𝑥 and 𝑠 are the numerical values in a. after
observing the sample variables).
c. Requested is a 95%-CI for 𝜎 2 , so we will use (formula sheet):
(𝑛−1)𝑆 2 (𝑛−1)𝑆 2 2 2 𝛼
( , ) , with 𝑃(𝜒𝑛−1 ≤ 𝑐1 ) = 𝑃(𝜒𝑛−1 ≥ 𝑐2 ) = = 0.025
𝑐2 𝑐1 2
2
We have 𝑠 2 = 14.37 and 𝑛 = 16 (a.) and in the 𝜒16−1 -table we find 𝑐1 = 6.26 and 𝑐2 = 27.49.
(𝑛−1)𝑆 2 (𝑛−1)𝑆 2 15 ∙14.37 15 ∙14.37
So 95%-CI(𝜎 2 ) = ( , )=( , ) ≈ (7.84, 34.4).
𝑐2 𝑐1 27.49 6.26

Exercise 2
𝑠
a. A 99%-CI for μ has bounds: 𝑥 ± 𝑐 , where 𝑐 = 2.861 can be found in the 𝑡-table with 19 degrees of
√𝑛
𝛼 𝑠 39.50
freedom, such that 𝑃(𝑇19 ≥ 𝑐) = 2 = 0.005, so 𝑐 𝑛 = 2.861 ∙ ≈ 25.27
√ √20
99%-CI(μ) = (189.74 − 25.27, 189.74 + 25.27) = (164.47, 215.01)
b. At a confidence level 99% the mean (expected) number of hours of sunshine in the month of July in De Bilt
between 164.47 and 215.01. We are quite sure about this statement: this method will produce an interval
containing 𝜇 in 99% of the cases that we will apply the method in equivalent situations, that is, if we repeat
the random samples over and over again with equally many observations.
c. The value 164.1 of the observation 𝑥 in the year 1984 should not be compared to confidence interval in a.
since the interval concerns 𝜇, the mean number of hours of sunshine.
(Note that more observations result in a smaller interval and, of course, a relatively smaller proportion of
observations within the interval).
If we want to check how exceptional an observation such as 𝑥 = 164.1 is, we can compute its z-score:
𝑥−𝑥 164.1−189.74
𝑧= 𝑠 = ≈ −0.65. This value does not at all indicate an exceptionally low observation: it is
39.50
less than one standard deviation away from the mean.
d. Requested is the 95%-CI for the standard deviation σ, so we use (formula sheet):
(𝑛−1)𝑆 2 (𝑛−1)𝑆 2 2 2 𝛼
( , ) , where 𝑃(𝜒𝑛−1 ≤ 𝑐1 ) = 𝑃(𝜒𝑛−1 ≥ 𝑐2 ) = = 0.025
𝑐2 𝑐1 2
2
In this case 𝑛 = 20 and 𝑠 = 39.50 are given and in the 𝜒20−1 -table we find 𝑐1 = 8.91 and 𝑐2 = 32.85
19𝑠2 19𝑠2 19 ∙39.502 19 ∙39.502
So a 95%-CI(σ) = (√ ,√ ) = (√ ,√ ) ≈ (30.0, 57.7)
𝑐2 𝑐1 32.85 8.91

Exercise 3
a. Using the calculator we find: 𝑥̅ = 60 and 𝑠 2 = 51.25 (𝑠 ≈ 7.159).
( 𝑋̅ and 𝑆 2 are unbiased estimators of μ and 𝜎 2 )
b. We are interested in a confidence interval for μ, where in this case the other parameter 𝜎 2 is unknown.
𝑆
Assuming a normal distribution for the search times we will use: 𝑋 ± 𝑐 𝑛 with 𝑃(𝑇𝑛−1 ≥ 𝑐) = 12𝛼 ,
√

1
 where 𝑛 = 9, 𝑥̅ = 60, 𝑠 ≈ 7.159 and in the 𝑡8 -table we find 𝑐 = 2.306, such that 𝑃(𝑇8 ≥ 𝑐) = 0.025.
𝑠 𝑠
95%-CI(𝜇) = (𝑥 − 𝑐 ,𝑥 +𝑐 ) ≈ (54.5, 65.5)
√𝑛 √𝑛
c. In this part a 95%-BI for 𝜎 2 is requested, so we will use (formula sheet):
(𝑛−1)𝑆 2 (𝑛−1)𝑆 2 𝛼
( , ) , with 𝑃(𝜒82 ≤ 𝑐1 ) = 𝑃(𝜒82 ≥ 𝑐2 ) = = 0.025 , such that 𝑐1 = 2.18 and 𝑐2 = 17.53
𝑐2 𝑐1 2
(𝑛−1)𝑆 2 (𝑛−1)𝑆 2 8 ∙ 51.25 8 ∙ 51.25
So 95%-CI(𝜎 2 ) = ( , )=( , ) ≈ (23.4, 188.1)
𝑐2 𝑐1 17.53 2.18

Exercise 4
𝑝̂(1−𝑝̂)
a. 95%-CI for 𝑝: 𝑝̂ ± 𝑐√ with Φ(𝑐) = 1 − 12𝛼, (formula sheet)
𝑛
This a large-sample-interval since 𝑛 = 100 is large enough.
22
Furthermore 𝑝̂ = 100 = 0.22 and 𝑐 = 1.96 from the 𝑁(0, 1)-table such that Φ(𝑐) = 0.975
(never use the 𝑡-distribution if the binomial model applies to the observations!).
0.22 ∙ 0.78 0.22 ∙ 0.78
Substitute: (0.22 − 1.96√ , 0.22 + 1.96√ ) ≈ (0.22 − 0.08,0.22 + 0.08) = (0.14,0.30)
100 100

𝑝̂(1−𝑝̂)
b. The 95%-CI should not be wider than 0.02, so: 2 ∙ 1.96√ ≤ 0.02.
𝑛
1.96 1.96 2
This implies: √𝑛 ≥ 0.01 √𝑝̂ (1 − 𝑝̂ ), so 𝑛 ≥ (0.01) ∙ 𝑝̂ (1 − 𝑝̂ ).
Using the observed value in the small sample in a., so 𝑝̂ = 0.22, we find 𝑛 ≥ 6593.
(Remark 1: if the indication of 𝑝̂ in part a. is not given, one could choose a “safe” value ½ for 𝑝̂ or
1
𝑝̂ (1 − 𝑝̂ ) = 4: in that case we find: 𝑛 ≥ 9604
1
Remark 2: The inequality 𝑝̂ (1 − 𝑝̂ ) ≤ 4 can be shown by considering the
function 𝑓(𝑝) = 𝑝(1 − 𝑝) = 𝑝 – 𝑝2 for 0 ≤ 𝑝 ≤ 1: de graph of 𝑓 is a
parabola that intersects the X-axis at 𝑝 = 0 and 𝑝 = 1 and the maximum
value of f is attained at 𝑝 = ½ , for which 𝑓(𝑝) = 𝑝(1 − 𝑝) = ¼ )

Exercise 5
𝑝̂(1−𝑝̂)
a. This is another binomial situation: Use 𝑝̂ ± 𝑐 √ with Φ(𝑐) = 1 − 12𝛼, (formula sheet)
𝑛
73
In this case 1 − 𝛼 = 0.95 and Φ(𝑐) = 0.975, so 𝑐 = 1.96 , 𝑝̂ = 400 = 0.1825 and 𝑛 = 400.
Substitution in the formula results in:
0.1825 ∙0.8175 0.1825 ∙0.8175
95%-CI(𝑝) = (0.1825 − 1.96 ∙ √ , 0.1825 + 1.96 ∙ √ )
400 400

= (0.1825 − 0.0379, 0.1825 + 0.0379) ≈ (0.145, 0.220).

Interpretation: “We are 95% confident that the real population proportion of cars that do not meet the
conditions is between 14.5 and 22%.”
The more precise (frequency-)interpretation: “If we would repeat the sample many times (independently),
95% of all computed confidence intervals will contain the population proportion 𝑝 of substandard cars.”
𝑝̂(1−𝑝̂) 𝑐 2
b. Estimation error = half of the width (!), so: 𝑐 √ ≤ 0.02, it follows that 𝑛 ≥ (0.02) 𝑝̂ (1 − 𝑝̂ ).
𝑛
73
For a confidence level 99% is Φ(𝑐) = 0.995, so 𝑐 = 2.575 and from a. we can use 𝑝̂ = 400.
1
We find: 𝑛 ≥ 2473.1, or: 𝒏 = 𝟐𝟒𝟕𝟒. (if you used 𝑝̂ (1 − 𝑝̂ ) ≤ 4 you should find: 𝑛 ≈ 4145).