Technical University of Denmark Page 1 of 36 pages.

Written examination: 27 May 2018

Course name and number: Introduction to Statistics (02402)

Aids and facilities allowed: All

The questions were answered by

(student number) (signature) (table number)

This exam consists of 30 questions of the “multiple choice” type, which are divided between 11
exercises. To answer the questions, you need to fill in the “multiple choice” form (6 seperate
pages) on CampusNet with the numbers of the answers that you believe to be correct.

5 points are given for a correct “multiple choice” answer, and −1 point is given for a wrong
answer. ONLY the following 5 answer options are valid: 1, 2, 3, 4, or 5. If a question is left
blank or an invalid answer is entered, 0 points are given for the question. Furthermore, if more
than one answer option is selected for a single question, which is in fact technically possible in
the online system, 0 points are given for the question. The number of points needed to obtain
a specific mark or to pass the exam is ultimately determined during censoring.

The final answers should be given by filling in and submitting the form
online via CampusNet. The table provided here is ONLY an emergency alter-
native. Remember to provide your student number if you do hand in on paper.

Exercise I.1 I.2 I.3 I.4 II.1 II.2 II.3 III.1 IV.1 IV.2
Question (1) (2) (3) (4) (5) (6) (7) (8) (9) (10)
Answer
4 5 2 3 1 3 4 2 4 2

Exercise IV.3 IV.4 IV.5 V.1 V.2 VI.1 VI.2 VI.3 VII.1 VII.2
Question (11) (12) (13) (14) (15) (16) (17) (18) (19) (20)
Answer
3 2 5 1 4 2 2 3 3 4

Exercise VII.3 VIII.1 IX.1 IX.2 X.1 X.2 X.3 XI.1 XI.2 XI.3
Question (21) (22) (23) (24) (25) (26) (27) (28) (29) (30)
Answer
5 4 1 3 2 2 5 2 4 3

The exam paper contains 36 pages.

Continue on page 2

1
Multiple choice questions: Note that in each question, one and only one of the answer
options is correct. Furthermore, not all the suggested answers are necessarily meaningful. Al-
ways remember to round your own result to the number of decimals given in the answer options
before you choose your answer.

Exercise I

In order to investigate the download speed at a work station, download times (in seconds) were
recorded for 53 files of different sizes (measured in MB). The download times are saved in the
vector time in R, while the corresponding file sizes are saved in the vector size. Furthermore,
the following code was executed in R:

logtime <- log(time)

model1 <- lm(logtime ~ size)
summary(model1)

##
## Call:
## lm(formula = logtime ~ size)
##
## Residuals:
## Min 1Q Median 3Q Max
## -2.571 -0.673 0.132 0.745 2.190
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) -0.109 0.482 -0.23 0.82
## size 0.127 0.023 5.50 1.2e-06 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.979 on 51 degrees of freedom
## Multiple R-squared: 0.372,Adjusted R-squared: 0.36
## F-statistic: 30.2 on 1 and 51 DF, p-value: 1.25e-06

The statistical model given by model1 is a linear regression model of the form
Yi = β0 + β1 xi + εi ,
where εi , i = 1, . . . , 53, are assumed to be independent and identically N (0, σ 2 ) distributed.

Question I.1 (1)

Identify the dependent variable and the explanatory variable in the model given by model1.

1  Download time is the dependent variable. File size is the explanatory variable.

2
 2  File size is the dependent variable. The logarithm of download time is the explanatory
variable.

3  Download time is the explanatory variable. File size is the dependent variable.

4*  File size is the explanatory variable. The logarithm of download time is the dependent
variable.

5  As (the logarithm of) download time depends on file size, both file size and the logarithm
of download size are dependent variables. There is no explanatory variable in the model.

----------------------------------- FACIT-BEGIN -----------------------------------

The variable to the left of the tilde (∼) in the R code corresponds to the dependent variable
Yi , while the variable to the right of the tilde corresponds to the explanatory variable xi .

------------------------------------ FACIT-END ------------------------------------

Question I.2 (2)

Give estimates for the parameters of the model given by model1.

1  β̂0 = 0.127, β̂1 = −0.109, σ̂ = 0.979

2  β̂0 = −0.109, β̂1 = 0.127, σ̂ = 0.9792

3  β̂0 = −0.109, β̂1 = 0.127, σ̂ = 0.372

4  β̂0 = 0.127, β̂1 = −0.109, σ̂ = 0.9792

5*  β̂0 = −0.109, β̂1 = 0.127, σ̂ = 0.979

----------------------------------- FACIT-BEGIN -----------------------------------

The estimates of the model intercept, β̂0 , and slope, β̂1 , are given in the column Estimate
in the R output (in the rows (Intercept) and size, respectively). The estimated standard
deviation of the error, σ̂, is termed Residual standard error in the R output.

------------------------------------ FACIT-END ------------------------------------

Question I.3 (3)

The following code was also executed in R:

3
 mean(size)

## [1] 20.088

(53-1)*var(size)

## [1] 1807.6

Using the model given by model1 as a starting point, give a 90% prediction interval for the
logarithm of the download time for a file of size 17 MB.

q
2
1 −0.109 + 0.127 · 17 ± 2.0076 · 0.979 · 1 + 531
+ (17−20.088)
1807.6
q
2
2*  1
17 · 0.127 − 0.109 ± 0.979 · 1.6753 · 1 + 53 + (20.088−17)
1807.6

√ q
2
3 −0.109 + 0.127 · 17 ± 1.6753 · 0.979 · 1 + 53 1
+ (17−20.088)
1807.6
q
2
4 1
−0.109 + 17 · 0.127 ± 1.6753 · 0.979 · 53 + (17−20.088)
1807.6
q
2
5 1
17 · 0.127 + 0.109 ± 0.979 · 2.0076 · 1 + 53 + (20.088−17)
1807.6

----------------------------------- FACIT-BEGIN -----------------------------------

Use Method 5.18, equation (5-60). Note that 1.6753 is the 0.95 quantile of the t-distribution
with 51 degrees of freedom, qt(0.95, df = 51) in R.

------------------------------------ FACIT-END ------------------------------------

Question I.4 (4)

Using the model given by model1 as a starting point, one would like to investigate whether
there is a significant linear relationship between the logarithm of download time and file size.
Formulate the corresponding statistical null hypothesis that there is no association between the
two variables.

1  H0 : β̂1 = 0
2  H0 : β1 6= β0
3*  H0 : β1 = 0
4  H0 : β1 6= 0

4
5  H0 : β̂1 6= 0

----------------------------------- FACIT-BEGIN -----------------------------------

Under the null hypothesis H0 : β1 = 0, the model Yi = β0 +β1 xi +εi reduces to Yi = β0 +ε. The
latter corresponds to a model for one sample (in which the logarithm of the download time, Yi ,
does not depend on file size, xi ). Answer 1 is wrong since we are making a null hypothesis for
the true mean and not our estimate.

------------------------------------ FACIT-END ------------------------------------

Continue on page 6

5
 Exercise II

Train carriages in a mine are to be loaded with large pieces of blasted rock. Initially, a machine
has sorted the pieces into two piles based on their size and weight. The weights of the pieces
of rock are assumed to be independent and normally distributed, such that the weight of a
randomly extracted piece from pile 1 can be represented by X1 ∼ N (20, 52 ) kg and from pile 2
by X2 ∼ N (50, 102 ) kg.

Question II.1 (5)

What is the probability that a randomly selected piece from pile 1 weighs more than 25 kg?

1*  15.9 %

2  84.1 %

3  42.1 %

4  57.9 %

5  None of the values listed.

----------------------------------- FACIT-BEGIN -----------------------------------

Here,
P (X1 > 25) = 1 − P (X1 ≤ 25) = 1 − FX1 (25)
is to be computed, where FX1 denotes the cumulative distribution function of the normal
distribution with mean 20 and standard deviation 5. In R, the result may be found as:

1 - pnorm(q = 25, mean = 20, sd = 5)

## [1] 0.1586553

------------------------------------ FACIT-END ------------------------------------

Question II.2 (6)

Choose a correct statement:

There is a 20% probability of a randomly selected piece of rock from pile 2 being heavier than

1  41.6 kg.

6
 2  52.5 kg.

3*  58.4 kg.

4  67.4 kg.

5  134 kg.

----------------------------------- FACIT-BEGIN -----------------------------------

Let FX2 be the cumulative distribution function of the normal distribution with mean 50 and
standard deviation 10. Here, we need to find x such that

P (X2 > x) = 0.2 ,

which corresponds to finding x such that

P (X2 ≤ x) = FX2 (x) = 0.8.

In R, x may be found as:

qnorm(0.8, mean = 50, sd = 10)

## [1] 58.41621

------------------------------------ FACIT-END ------------------------------------

Question II.3 (7)

If the train carriages are loaded too heavily, operations must stop. A manual crane must be
used to remove pieces of rock from the overloaded carriage, and this procedure is very costly.

Two robotic cranes load the train carriages. One crane takes pieces from pile 1, and the other
takes pieces from pile 2. If each crane takes 10 pieces from its pile, and all 20 pieces are loaded
into the same (empty) carriage, what is the probability that the total load of this train carriage
exceeds 800 kg?

1  The total weight is Y ∼ N (700, 15625), so P (Y > 800) = 21.2%.

2  The total weight is Y ∼ N (700, 12500), so P (Y > 800) = 18.6%.

3  The total weight is Y ∼ N (700, 2500), so P (Y > 800) = 2.28%.

4*  The total weight is Y ∼ N (700, 1250), so P (Y > 800) = 0.234%.

5  The total weight is Y ∼ N (700, 225), so P (Y > 800) = 1.31 · 10−9 %.

7
----------------------------------- FACIT-BEGIN -----------------------------------

Let X1i and X2i , i = 1, . . . , 10, with X1i ∼ N (20, 52 ) and X2i ∼ N (50, 102 ), be independent
random variables which represent the weights of 10 randomly selected pieces of rock from pile
1 and pile 2, respectively. Then, the total load may be represented by
10
X 10
X
Y = X1i + X2i .
i=1 i=1

According to Theorem 2.40 and Theorem 2.56, Y is normally distributed with

10
X 10
X
E(Y ) = E(X1i ) + E(X2i ) = 10 · 20 + 10 · 50 = 700
i=1 i=1
10
X 10
X
V(Y ) = V(X1i ) + V(X2i ) = 10 · 52 + 10 · 102 = 1250
i=1 i=1

so
P (Y > 800) = 1 − P (Y ≤ 800) = 1 − FY (800)
where FY is the cumulative distribution function for the N (700, 1250) distribution. In R, the
result may then be computed as:

1 - pnorm(800, mean = 700, sd = sqrt(1250))

## [1] 0.002338867

------------------------------------ FACIT-END ------------------------------------

Continue on page 9

8
 Exercise III

The resistances in the electrical circuit

R1

R2

are estimated to be R̂1 = 2 Ω and R̂2 = 3 Ω, with an estimated standard deviation for the
estimators (standard error ) of σ̂R̂1 = 0.2 and σ̂R̂2 = 0.5, respectively. R1 and R2 can be assumed
independent and normally distributed.

The total resistance through the circuit is given by

1
R= 1 1
R1
+ R2

Question III.1 (8)

Using simulation, determine a 95% confidence interval for the total resistance R. The following
R code must be used to get the specified result (and after this code has been executed, only
one additional function call in R is needed to get the result):

set.seed(7643)
k <- 10000
R1 <- rnorm(k, mean = 2, sd = 0.2)
R2 <- rnorm(k, mean = 3, sd = 0.5)
R <- 1/(1/R1 + 1/R2)

1  [1.11, 1.29]
2*  [0.96, 1.40]
3  [0.92, 1.47]
4  [0.82, 1.58]
5  [0.72, 1.68]

----------------------------------- FACIT-BEGIN -----------------------------------

We have used the parametric bootstrap simulation approach to error propagation as described
in Method 4.7:

9
set.seed(7643)
k <- 10000
R1 <- rnorm(k, mean = 2, sd = 0.2)
R2 <- rnorm(k, mean = 3, sd = 0.5)
R <- 1/(1/R1 + 1/R2)
quantile(R, c(0.025,0.975))

## 2.5% 97.5%
## 0.9647361 1.4016874

------------------------------------ FACIT-END ------------------------------------

Continue on page 11

10
 Exercise IV

Students’ choice of higher education has great political awareness. The table below is based
on numbers from ”Universiteternes Statistiske Beredskab”, and contains the number of newly
enrolled students by discipline for selected years (row and column sums are included in italics)

Y2012 Y2016 Y2017 Sum

Hum 7966 7297 6691 21954
Soc 10173 10253 10006 30432
Hlth 2789 3137 3157 908 3
TechNat 8551 10130 10339 29020
Sum 29479 30817 30193 90489

Question IV.1 (9)

Based on the numbers for 2017, one wants to test the hypothesis that the proportion of students
newly enrolled in a technical or natural science bachelor education (TechNat) is 32.0%. The
relevant test statistic for this hypothesis, which is assumed to be well approximated by a
standard normal distribution, is

√
1  (10253 − 0.68 · 30817)/ 30817 · 0.32 · 0.68 = −130.70
√
2  (10339 − 0.32 · 30193)/ 30193 · 0.32 · 0.32 = 12.18
√
3  (10130 − 0.32 · 30817)/ 30817 · 0.32 · 0.68 = 3.28
√
4*  (10339 − 0.32 · 30193)/ 30193 · 0.32 · 0.68 = 8.36
√
5  (10339 − 0.68 · 30193)/ 30193 · 0.32 · 0.32 = −183.30

----------------------------------- FACIT-BEGIN -----------------------------------

Using equation (7-16) with x = 10339, n = 30193, and p0 = 0.32:

(10339 - 0.32 * 30193) / sqrt(30193 * 0.32 * 0.68)

## [1] 8.355261

------------------------------------ FACIT-END ------------------------------------

Question IV.2 (10)

It is also of interest, whether there was a change from 2016 to 2017 in the proportion of students
who were newly enrolled in the humanities (Hum).

11
Four different tests are performed in R using the following code:

prop.test(x = 6691, n = 30193, p = 7297/30817, correct = FALSE)

##
## 1-sample proportions test without continuity correction
##
## data: 6691 out of 30193, null probability 7297/30817
## X-squared = 38.5, df = 1, p-value = 5.5e-10
## alternative hypothesis: true p is not equal to 0.23678
## 95 percent confidence interval:
## 0.21696 0.22633
## sample estimates:
## p
## 0.22161

prop.test(x = c(7297, 6691), c(30817, 30193), correct = FALSE)

##
## 2-sample test for equality of proportions without continuity
## correction
##
## data: c(7297, 6691) out of c(30817, 30193)
## X-squared = 19.9, df = 1, p-value = 8.2e-06
## alternative hypothesis: two.sided
## 95 percent confidence interval:
## 0.0085083 0.0218461
## sample estimates:
## prop 1 prop 2
## 0.23678 0.22161

binom.test(x = 6691, n = 30193, p = 7297/30817)

##
## Exact binomial test
##
## data: 6691 and 30193
## number of successes = 6691, number of trials = 30193, p-value = 4.3e-10
## alternative hypothesis: true probability of success is not equal to 0.23678
## 95 percent confidence interval:
## 0.21693 0.22634
## sample estimates:
## probability of success
## 0.22161

12
 binom.test(x = 6691, n = 30193, p = (6691+7297)/(30193+30817))

##
## Exact binomial test
##
## data: 6691 and 30193
## number of successes = 6691, number of trials = 30193, p-value = 0.0015
## alternative hypothesis: true probability of success is not equal to 0.22927
## 95 percent confidence interval:
## 0.21693 0.22634
## sample estimates:
## probability of success
## 0.22161

Which line of code performs the desired test?

1  binom.test(x = 6691, n = 30193, p = 7297/30817)

2*  prop.test(x = c(7297, 6691), c(30817, 30193), correct = FALSE)

3  binom.test(x = 6691, n = 30193, p = (6691+7297)/(30193+30817))

4  prop.test(x = 6691, n = 30193, p = 7297/30817, correct = FALSE)

5  None of the four lines of code tests the relevant hypothesis

----------------------------------- FACIT-BEGIN -----------------------------------

The task is to compare proportions between two populations (Section 7.3) (and not, e.g., to
test whether a proportion has a specific value).

------------------------------------ FACIT-END ------------------------------------

Question IV.3 (11)

Using a hypothesis test, it is also to be investigated whether the distribution of newly enrolled
students across disciplines has changed over the three years for which data is given. The number
of degrees of freedom in the relevant distribution of the test statistic is:

1 3

2 4

3*  6

13
 4  12

5  20

----------------------------------- FACIT-BEGIN -----------------------------------

Comparison of distributions in different groups (Method 7.22). With four disciplines and three
years, (r − 1) · (c − 1) = 3 · 2 = 6.

------------------------------------ FACIT-END ------------------------------------

Question IV.4 (12)

Assuming independence between year and discipline, the expected number of newly enrolled
students in TechNat in the year 2017 is estimated to be

1  10339

2*  29020 · 30193/90489 = 9683

3  (8551 + 10130 + 10339)/3 = 9673

4  10339 · 29020/30193 = 9937

5  10339 · 30193/29020 = 10757

----------------------------------- FACIT-BEGIN -----------------------------------

As read in chapter 7.5.1 the expected number in a cell is calculated as:

column total × row total 30193 × 29020
= = 9683
grand total 90489

------------------------------------ FACIT-END ------------------------------------

Question IV.5 (13)

As the next step in testing whether the distribution across disciplines has changed over the
years, the following is given: The test statistic is calculated to be 314.5. The significance level
is set to α = 0.05. In the distribution used to assess the test statistic, the 0.95 and 0.975
quantiles are, respectively, 12.59 and 14.45. What may be concluded? (Both the conclusion
and reasoning must be correct).

14
 1  The numbers provided above cannot be used to argue statistically, whether the distribu-
tion across disciplines has changed.

2  The distribution across disciplines has not changed significantly, as the test statistic is
greater than the given 0.95 quantile.

3  The distribution across disciplines has not changed significantly, as the test statistic is
greater than the given 0.975 quantile.

4  The distribution across disciplines has changed significantly, as, under the null hypothesis,
there is a 95% probability of observing a test statistic greater than 12.59.

5*  The distribution across disciplines has changed significantly, as the test statistic is greater
than the given 0.95 quantile.

----------------------------------- FACIT-BEGIN -----------------------------------

Method 7.22. The test statistic is greater than the given 0.95 quantile, so the null hypothesis of
no difference between groups (no change across years) is rejected, and the change is concluded
to be significant.

------------------------------------ FACIT-END ------------------------------------

Continue on page 16

15
 Exercise V

An experiment was conducted with the purpose of investigating the shelf life of a certain type
of medicine. Altogether 26 identical, unopened bottles of medicine with the same production
date were used for the experiment. Half of the bottles were stored at room temperature (21 ◦ C),
the other half at fridge temperature (5 ◦ C). After 90 days the bottles were opened, and the
content of active substance (in mg/ml) in each bottle was measured. The results were read
into R in two vectors, hightemp (measurements from bottles stored at 21 ◦ C) and lowtemp
(measurements from bottles stored at 5 ◦ C).

Furthermore, the following code was executed in R:

t.test(lowtemp, hightemp)

##
## Welch Two Sample t-test
##
## data: lowtemp and hightemp
## t = 5.3, df = 24, p-value = 2e-05
## alternative hypothesis: true difference in means is not equal to 0
## 95 percent confidence interval:
## 1.4316 3.2592
## sample estimates:
## mean of x mean of y
## 7.4397 5.0943

Question V.1 (14)

What may be concluded when the significance level is set to α = 0.05?

1*  The mean content of the active substance is significantly larger after storage at fridge
temperature than at room temperature. The difference is estimated to be 2.35 mg/ml.

2  It can be seen from the p-value that there is no significant difference between the mean
content of the active substance in bottles stored at room temperature and at fridge
temperature, respectively.

3  The mean content of the active substance is significantly larger after storage at fridge
temperature than at room temperature. The difference is estimated to be 5.30 mg/ml.

4  The mean content of the active substance is significantly less after storage at fridge
temperature than at room temperature. The difference is estimated to be 2.35 mg/ml.

5  The 95% confidence interval contains 2.35. Therefore, there is no significant difference
between the mean content of the active substance in bottles which were stored at room
temperature and fridge temperature, respectively.

16
 ----------------------------------- FACIT-BEGIN -----------------------------------

The p-value 2 · 10−5 is much smaller than the significance level α = 0.05 (or: 0 is not contained
in the 95% confidence interval), so the difference is significant. The mean content of the active
substance is estimated to be 7.4397 − 5.0943 = 2.35 mg/ml larger in bottles which were stored
at fridge temperature than in those which were stored at room temperature.

------------------------------------ FACIT-END ------------------------------------

Question V.2 (15)

A 99% confidence interval for the difference in the mean content of the active substance between
bottles stored at fridge temperature and room temperature may be determined as follows:

3.2592−2.3454
1  2.3454 ± 2.7969
· 2.0639 = [1.67, 3.02]

2  2.3454 ± 3.2592−2.3454
2.4922
· 2.7969 = [1.32, 3.37]

3  1.4316 · 2.7969 2.7969

 
2.0639
, 3.2592 · 2.0639 = [1.94, 4.42]
3.2592−2.3454
4*  2.3454 ± 2.0639
· 2.7969 = [1.11, 3.58]
3.2592−2.3454
5  2.3454 ± 2.0639
· 2.4922 = [1.24, 3.45]

----------------------------------- FACIT-BEGIN -----------------------------------

Using the notation from Method 3.47, it may be concluded from the R output that x̄ − ȳ =
7.4397 − 5.0943 = 2.3454, and that the degrees of freedom ν = 24, so that t0.975 = 2.0639 and
t0.995 = 2.7969 (qt(0.975, df = 24) and qt(0.995, df = 24), respectively, in R).

Furthermore, according to the R output, [1.4316, 3.2592] is a 95% confidence interval

q 2 for the
s1 s22
difference in mean content. The only thing in the equation we do not know is n1 + n2 , but
since we have one confidence interval given (95%), we can isolate this term from that equation.
Thus, it follows from Method 3.47 that
s s
s21 s22 s21 s2
x̄ − ȳ ± t0.975 · + = 2.3454 ± 2.0639 · + 2 = [1.4316, 3.2592]
n1 n2 n1 n2

which may be used to conclude that

s s
2 2
s1 s2 s21 s22 3.2592 − 2.3454
2.3454 + 2.0639 · + = 3.2592 ⇔ + = .
n1 n2 n1 n2 2.0639

17
Now, the 99% confidence interval can be computed using Method 3.47, as well:
s
s21 s2 3.2592 − 2.3454
x̄ − ȳ ± t0.995 · + 2 = 2.3454 ± 2.7969 · = [1.11, 3.58] .
n1 n2 2.0639

------------------------------------ FACIT-END ------------------------------------

Continue on page 19

18
 Exercise VI

In a production process things sometimes go wrong, and a component must be discarded after an
inspection. From experience, it is known that there is a 20% probability of a component needing
to be discarded. The assessment (“keep” or “discard”) for a given component is independent of
the assessments for the other components. One can assume that 20 components are produced
per day.

Question VI.1 (16)

What is the probability that, on a randomly selected day, no components need to be discarded?

1  The number of components which need to be discarded is hypergeometrically distributed,

and the probability is 0.
2*  The number of components which need to be discarded is binomial distributed, and the
probability is 0.0115.
3  The number of components which need to be discarded is binomial distributed, and the
probability is 0.0576.
4  The number of components which need to be discarded is hypergeometrically distributed,
and the probability is 0.0692.
5  The number of components which need to be discarded is binomial distributed, and the
probability is 0.630.

----------------------------------- FACIT-BEGIN -----------------------------------

The number of components that get the assessment discard (“number of successes”) out of 20
independently assessed components (“number of independent trials”) is binomial distributed
with probability p = 0.2 and size n = 20.

Let X be binomial distributed with probability p = 0.2 and size n = 20. Then P (X = 0) (or,
equivalently, P (X ≤ 0)) may be computed in R as follows

dbinom(0, size = 20, p = 0.2)

## [1] 0.01152922

pbinom(0, size = 20, p = 0.2)

## [1] 0.01152922

------------------------------------ FACIT-END ------------------------------------

19
 Question VI.2 (17)

A simulation of the number of discarded components per day has been carried out. Let Xi
denote the number of discarded components on a randomly selected day i. A sample of n =
20 values has been simulated, and is denoted by xi , i = 1, . . . , 20. Which of the following
statements is the only one that can be correct?

1  The expected number of components to be discarded during one day is µX = 4.5. The
sample mean of the simulated values was µ̂X = 4.3.

2*  The expected number of components to be discarded during one day is µX = 4. The

sample mean of the simulated values was µ̂X = 3.7.

3  The expected number of components to be discarded during one day is µX = 5. The

sample mean of the simulated values was µ̂X = 4.9.

4  The expected number of components to be discarded during one day is µX = 2. The

sample mean of the simulated values was µ̂X = 2.2.

5  The expected number of components to be discarded during one day is µX = 4.25. The
sample mean of the simulated values was µ̂X = 4.25.

----------------------------------- FACIT-BEGIN -----------------------------------

Again, let X be binomial distributed with probability p = 0.2 and size n = 20. See theorem
2.21. The expected number of components to be discarded during one day, µX , may then be
computed as
µX = E(X) = np = 20 · 0.2 = 4 ,
so the answer option with µX = 4 is the only one that can be correct.

------------------------------------ FACIT-END ------------------------------------

Question VI.3 (18)

An experiment is planned in order to estimate the proportion of components that need to be

discarded.

How many days does the experiment need to run in order to obtain a 95% confidence interval
for the proportion of components to be discarded, which has an expected width of 5 percentage
points?

1  n = (1.96 · 0.2/0.05)2 = 61.5, i.e. 4 days.

2  n = 0.16 · (1.96/0.05)2 = 246, i.e. 13 days.

20
3*  n = 0.16 · (1.96/0.025)2 = 983, i.e. 50 days.

4  n = 0.2 · (1.96/0.025)2 = 1229, i.e 62 days.

5  n = (1.96 · 0.2/0.025)2 = 3934, i.e. 197 days.

----------------------------------- FACIT-BEGIN -----------------------------------

Use Method 7.13 with p = 0.2, remembering that the ME is half the expected width of the
confidence interval, and using the information that 20 components are produced per day.

------------------------------------ FACIT-END ------------------------------------

Continue on page 22

21
 Exercise VII

A factory which produces percussion instruments and accessories produces drumsticks as well.
For the purpose of quality control, the lengths (in cm) of 20 drumsticks selected at random were
measured. These lengths were read into the vector length in R. The lengths can be assumed
to be independent and normally distributed with mean µ and variance σ 2 .

Question VII.1 (19)

The following commands were executed in R:

sum(length)

## [1] 793.1

var(length)

## [1] 0.01099

Give an expression for a 95% confidence interval for the mean length of the drumsticks.

793.1 0.01099
1 20
± 2.093 · √
20

0.01099
2  793.1 ± 2.086 · √
20
√
793.1 0.01099
3*  20
± 2.093 · √
20
√
793.1 0.01099
4 20
± 2.086 · √
20

793.1 0.01099
5 20
± 2.093 · √
19

----------------------------------- FACIT-BEGIN -----------------------------------

√
Use Method 3.9 / (3-11) with n = 20, x̄ = 793.1/20, s = 0.01099. The 0.975 quantile for the
t-distribution with 19 degrees of freedom is:

qt(0.975, df = 19)

## [1] 2.093024

------------------------------------ FACIT-END ------------------------------------

Question VII.2 (20)

22
 A t-test is performed in order to investigate whether the mean length of the drumsticks may
be assumed to be 39.60 cm. The observed t-test statistic is calculated to be tobs = 2.38. Which
of the following is the only correct conclusion?

1  The mean length is significantly different from 39.60 cm when the significance level α =
0.01 is used.

2  As the p-value is greater than 0.05, the null hypothesis that the mean length is 39.60 cm
is rejected, no matter whether the significance level is set to α = 0.05 or α = 0.01.

3  As the p-value is less than 0.05, the null hypothesis that the mean length is 39.60 cm is
accepted when the significance level is set to α = 0.05.

4*  The mean length is significantly different from 39.60 cm when the significance level is set
to α = 0.05.

5  As the p-value is greater than 0.05, the null hypothesis that the mean length is 39.60 cm
is accepted, no matter whether the significance level α = 0.05 or α = 0.01 is used.

----------------------------------- FACIT-BEGIN -----------------------------------

The p-value is

p = 2 · P (T > |tobs |) = 2 · P (T > 2.38) = 2 · (1 − P (T ≤ 2.38)) = 0.028

where P (T ≤ 2.38) is computed in R as pt(2.38, df = 19).

As the p-value lies in between 0.01 and 0.05, the null hypothesis is rejected at significance level
α = 0.05 (but not at significance level α = 0.01).

------------------------------------ FACIT-END ------------------------------------

Question VII.3 (21)

In relation to subsequent quality control, plans are made to select a new sample. The sample
size must be sufficiently large to detect a difference of 0.5 mm between the mean length of
the drumsticks and the desired mean length 39.60 cm with power 90%, at significance level
α = 0.01. Here, 0.11 is to be used as a guess for the population standard deviation. Use the
power.t.test function in R to determine the necessary sample size.

1  n = 20

2  n = 22

3  n = 146

23
4  n = 53

5*  n = 76

----------------------------------- FACIT-BEGIN -----------------------------------

power.t.test(power = 0.9, delta = 0.05, sd = 0.11,

sig.level = 0.01, type = "one.sample")

##
## One-sample t test power calculation
##
## n = 75.36328
## delta = 0.05
## sd = 0.11
## sig.level = 0.01
## power = 0.9
## alternative = two.sided

See example 3.67 for examples of power.t.test.

------------------------------------ FACIT-END ------------------------------------

Continue on page 25

24
 Exercise VIII

Two quantitative variables have been read into R as x and y. One would like to describe the
relationship between these two variables using a linear regression model. To this end, two
different linear regression models have been estimated in R, see the R code below.

logx = log(x)
logy = log(y)
model1 <- lm(y ~ x)
model2 <- lm(logy ~ logx)

Question VIII.1 (22)

Plots of the residuals against the fitted values for model1 (left) and model2 (right), respectively,
are shown below. On the basis of these plots, would one prefer to analyse the data using the
statistical model given by model1 or the one given by model2? (Both the conclusion and
reasoning must be valid).
6

6
4

4
2

2
Residuals

Residuals
0

0
−2

−2
−4

−4
−6

−6

0 1 2 3 4 −4 −3 −2 −1 0 1

Fitted values Fitted values

1  There are clear linear associations between the residuals and the fitted values in the plot
to the left, while no linear association is seen in the plot to the right. Thus, one would
prefer to use model1.

2  The assumption of variance homogeneity is clearly not satisfied for model2, while the
assumption seems reasonable for model1. Thus, one would prefer to use model1.

3  In the plot to the left, a lot of the fitted values lie in the interval [0,1], while they are
better spread out over the whole x axis in the plot to the right. Thus, one would prefer
to use model2.

25
4*  The assumption of variance homogeneity is clearly not satisfied for model1, while the
assumption seems reasonable for model2. Thus, one would prefer to use model2.

5  The residuals in the figure to the right are obviously uniformly distributed in an interval
around 0 (thus not normally distributed), while the residuals in the figure to the left
might well be normally distributed. Thus, one would prefer to use model1.

----------------------------------- FACIT-BEGIN -----------------------------------

In the plot to the left, it is clear that the variance of the residuals increases with the fitted
values, revealing that the assumption of variance homogeneity does not hold for model1. In
the plot to the right, the variance of the residuals seems to be quite constant across the fitted
values, indicating that the assumption of variance homogenity should be ok for model2.

------------------------------------ FACIT-END ------------------------------------

Continue on page 27

26
 Exercise IX

Many factors affect the indoor climate of a building. One of the most common measures for
the quality of the indoor climate is the level of CO2 . If there is insufficient ventilation, the CO2
level becomes too high, which, among other things, decreases peoples’ ability to concentrate.
In new buildings with classrooms, the CO2 level may not exceed 1000 ppm - in outdoor air
there is around 400 ppm (before the industrial revolution it was around 280 ppm!).

In a study of the indoor climate in classrooms, samples of the CO2 level were taken from two
different classrooms. Both samples consist of one-hour average values measured over a period
of 2 months. Only values where people were present in the classroom have been included in
the samples. The observations for room 1 and room 2, respectively, were loaded into R in the
vectors room1CO2 and room2CO2.

Question IX.1 (23)

The following code was run in R:

length(room1CO2)

## [1] 304

length(room2CO2)

## [1] 252

sum((room1CO2 - mean(room1CO2))^2)

## [1] 131606104

(length(room2CO2)-1)*var(room2CO2)

## [1] 12775276

Determine the sample standard deviation for room 1 (s1 ) and room 2 (s2 ), respectively.

1*  s1 = 659.0475 and s2 = 225.6048

2  s1 = 434343.6 and s2 = 50897.51

3  s1 = 657.9626 and s2 = 225.1567

4  s1 = 11471.97 and s2 = 3574.252

5  None of the four answers above can be correct.

27
----------------------------------- FACIT-BEGIN -----------------------------------

See definition 1.11.

r
131606104
s1 = = 659.0475
304 − 1
r
12775276
s2 = = 225.6048
252 − 1

------------------------------------ FACIT-END ------------------------------------

Question IX.2 (24)

In addition, the following has been run in R:

Q3 <- function(x){ quantile(x, 0.75) }

simSamples1 <- replicate(10000, sample(room1CO2, replace = TRUE))

simSamples2 <- replicate(10000, sample(room2CO2, replace = TRUE))

simQ3s1 <- apply(simSamples1, 2, Q3)

simQ3s2 <- apply(simSamples2, 2, Q3)
simQ3sdiff <- simQ3s1 - simQ3s2

quantile(simQ3s1, c(0, 0.025, 0.05, 0.95, 0.975, 1))

## 0% 2.5% 5% 95% 97.5% 100%

## 1417.896 1562.332 1583.146 1833.104 1838.021 1953.792

quantile(simQ3s2, c(0, 0.025, 0.05, 0.95, 0.975, 1))

## 0% 2.5% 5% 95% 97.5% 100%

## 772.0833 827.5000 831.1042 916.4583 920.5000 966.6667

quantile(simQ3sdiff, c(0, 0.025, 0.05, 0.95, 0.975, 1))

## 0% 2.5% 5% 95% 97.5% 100%

## 534.6458 685.8297 712.4562 976.1042 991.6250 1093.4167

Use this R output to determine a 95% confidence interval for the difference between the 0.75
quantiles for the CO2 level in room 1 and 2.

1  [916, 1833]

28
 2  [828, 921]

3*  [686, 992]

4  [1556 − 828, 1838 − 921] = [728, 917]

5  [828, 1838]

----------------------------------- FACIT-BEGIN -----------------------------------

The 95% bootstrap confidence interval is determined by the 0.025 and 0.975 quantiles of the
simulated differences (simQ3sdiff).

------------------------------------ FACIT-END ------------------------------------

Continue on page 30

29
 Exercise X

The table below shows the average yield (measured in hkg/acres) for 5 crops (Crop 1-5) in
Denmark in the years 2014-2017.

2014 2015 2016 2017 Average

Crop 1 79 80 73 83 78.75
Crop 2 46 48 47 52 48.25
Crop 3 64 63 57 66 62.50
Crop 4 66 68 62 68 66.00
Crop 5 57 60 55 58 57.50
Average 62.40 63.80 58.80 65.40 62.60

In addition to the row and column averages given in the table (in italics), it is given that
SS (Year) = 118.8 and SST = 2172.8.

In this exercise, the 20 average crop yields in the table are considered to be observations from
20 different randomly selected fields. It is assumed that within each year, there is no difference
between the expected yields from the five crops. The analysis must therefore be carried out as
if the same crop was sown on all the fields (and the information about crop type should not be
used in the exercise). We use a model of the form

Yij = µ + αi + εij

αi = 0, where εij ∼ N (0, σ 2 ) and independent.

P
with

Question X.1 (25)

Let α1 describe the effect of year 2014 on the expected yield. Give the estimate of α1 .

1  α̂1 = 78.75 − 62.40 = 16.35

2*  α̂1 = 62.40 − 62.60 = −0.20

3  α̂1 = 78.75

4  α̂1 = 62.60

5  α̂1 = 62.40

----------------------------------- FACIT-BEGIN -----------------------------------

As stated in equation 8-4 α̂1 is computed as the average yield in the year 2014 (62.40) minus
the overall average yield (62.60).

------------------------------------ FACIT-END ------------------------------------

30
 Question X.2 (26)

Set the significance level to α = 0.05. Give the critical value for the usual test used to investigate
whether the expected crop yield differs between years.

1  3.73

2*  3.24

3  26.30

4  3.49

5  2.96

----------------------------------- FACIT-BEGIN -----------------------------------

Theorem 8.6. The F -test statistic is evaluated using the F distribution with

(k − 1, n − k) = (4 − 1, 20 − 4) = (3, 16)

degrees of freedom, and the critical value is the 0.95 quantile of this distribution:

qf(0.95, df1 = 3, df2 = 16)

## [1] 3.238872

------------------------------------ FACIT-END ------------------------------------

Question X.3 (27)

Give the estimate of σ 2 .

1  σ̂ 2 = 2054

2  σ̂ 2 = 3.058

3  σ̂ 2 = 36.7

4  σ̂ 2 = 29.7

5*  σ̂ 2 = 128.375

31
----------------------------------- FACIT-BEGIN -----------------------------------

As can be read in chapter 8.2.2, MSE and MST are both central estimators for the the variance,
but that MSE holds true both if the null-hypothesis is rejected or not, so in this case we will
use this as the estimate.

SSE = SST − SS (Year) = 2172.8 − 118.8 = 2054

and
SSE 2054
σ̂ 2 = MSE = = = 128.375
n−k 16

------------------------------------ FACIT-END ------------------------------------

Continue on page 33

32
 Exercise XI

The data set used in this exercise is the same as in the previous exercise. However, here, it is
to be used that data describes the average yield (measured in hkg/acres) for 5 different crops
(Crop 1-5) in Denmark in the years 2014-2017.

2014 2015 2016 2017 Average

Crop 1 79 80 73 83 78.75
Crop 2 46 48 47 52 48.25
Crop 3 64 63 57 66 62.50
Crop 4 66 68 62 68 66.00
Crop 5 57 60 55 58 57.50
Average 62.40 63.80 58.80 65.40 62.60

In addition to the row and column averages given in the table (in italics), it is given that
SS (Crop) = 2017.3, SS (Year) = 118.8 and SST = 2172.8. The average crop yields are assumed
to be realizations of normally distributed random variables.

Question XI.1 (28)

A key question is whether there has been a development in the yield over time. To investigate
this, it is tested whether there is a statistically significant difference between the four years,
when the variation between the different crops is taken into account. Give the p-value and the
conclusion at significance level α = 0.05 for the usual test.

1  There is no significant difference in yield between the years, since p = 0.82 > 0.05.
2*  There is a significant difference in yield between the years, since p = 0.00045 < 0.05.
3  There is no significant difference in yield between the years, since p = 2 · 10−10 < 0.05.
4  There is a significant difference in yield between the years, since p = 0.82 > 0.05.
5  There is a significant difference in yield between the years, since p = 8 · 10−7 < 0.05.

----------------------------------- FACIT-BEGIN -----------------------------------

Use Theorem 8.22. First, find SSE:

SSE = SST − SS (Crop) − SS (Year) = 2172.8 − 2017.3 − 118.8 = 36.7 .

Then, the F -test statistic may be computed by

SS(Year)/(k − 1) 118.8/(4 − 1) 118.8/3
FYear = = = ,
SSE /((k − 1)(l − 1)) 36.7/((4 − 1)(5 − 1)) 36.7/12
and the p-value by

33
 1-pf((118.8/3)/(36.7/12), df1 = 3, df2 = 12)

## [1] 0.0004526046

------------------------------------ FACIT-END ------------------------------------

Question XI.2 (29)

It was decided in advance that the difference between the yields in 2014 and 2017 should be
seen as an indicator of the overall development. Set the significance level to α = 0.05 and
test the hypothesis that there is no difference between the two years. What is the conclusion?
(Both the conclusion and reasoning must be correct).

65.4−62.4
1  There is a significant difference in yield between the two years, since 62.6
< 0.05.
118.8
2  There is no significant difference in yield between the two years, since 2017.3
> 0.05.

3  There is no significant difference in yield between the two years, since √3 < 2.18.
36.7

−3
4*  There is a significant difference in yield between the two years, since 1.11
< −2.18.
3
5  There is a significant difference in yield between the two years, since 36.7
> 0.05.

----------------------------------- FACIT-BEGIN -----------------------------------

Use Method 8.10 modified for two-way ANOVA. With 5 observations in each of the two years,
and MSE = SSE /12, the relevant t-test statistic becomes
62.40 − 65.40 −3
tobs = q = = −2.712 .
36.7 2
· 1.11
12 5

The negative/smaller of the two critical values for the test is:

qt(0.025, df = 12)

## [1] -2.178813

Since the test-statistic is more extreme than the critical value there is a significant different.

------------------------------------ FACIT-END ------------------------------------

As part of the model validation, different plots are given in the figure below.

34
 Plot 1: Yield vs. Year Plot 2: Yield vs. Crop
80

80
70

70
Yield

Yield
60

60
50

50
2014 2015 2016 2017 1 2 3 4 5

Year Crop

Plot 3: Residuals vs. Fitted values Plot 4: Normal Q−Q Plot

2

2
Sample Quantiles
1

1
Residuals

0
−1

−1
−2

−2

50 60 70 80 −2 −1 0 1 2

Fitted values Theoretical Quantiles

Question XI.3 (30)

Based on the figure, which of the following statements is correct?

1  As the variation within years is much greater than the variation within crops (Plot 1 and
Plot 2), the assumption of variance homogeneity is clearly not fulfilled.

2  The normal distribution assumption is clearly not fulfilled (Plot 3), while the assumption
of variance homogeneity is clearly fulfilled (Plot 4).

3*  The normal distribution assumption appears to be fulfilled (Plot 4), and the same applies
to the assumption of variance homogeneity (Plot 3).

35
4  The yield cannot be different for the five crops, as the residuals vary randomly around
the horizontal line through y = 0 (Plot 2 and Plot 3).

5  There cannot be any difference between the yields in the different years (Plot 1), but the
normal distribution assumption is clearly fulfilled (Plot 3).

----------------------------------- FACIT-BEGIN -----------------------------------

A plot of the residuals against the fitted values may be used to investigate the assumption of
variance homogeneity for a two-way ANOVA model, while a normal QQ-plot may be used to
investigate the assumption of normality.

------------------------------------ FACIT-END ------------------------------------

The exam is finished. Have a great summer!

36