Solution Proposal: Exam WS 22/23

Name, Vorname:

Matrikelnummer:

1.
a) Import der Daten:
> [Link] <- [Link]("exam_data.csv") %>% as_tibble()
> [Link]
# A tibble: 1,137 x 4
X exam attempt score
<int> <chr> <int> <int>
1 1 516647/Mathematics 1 75
2 2 193841/Data Bases 2 49
3 3 326761/Computer Networks 2 75
4 4 230275/Data Bases 1 73
5 5 911920/OOP 1 55
6 6 840873/Formal Languages 1 33
7 7 473291/Mathematics 2 56
8 8 412501/Mathematics 2 102
9 9 320446/Formal Languages 2 26
10 10 332749/Formal Languages 1 67
# ... with 1,127 more rows
# i Use `print(n = ...)` to see more rows

b) Determine the type and the scale of all variables:

[Link]: qualitative, nominal
gender: qualitative, nominal
semester: quantitative, ratio
course: qualitative, nominal
exam: qualitative, nominal
attempt: quantitative, ratio
score: quantitative, ordinal

c) Add a variable grade that indicates the grade of the exam.

> [Link] %>%
+ mutate(
+ grade = case_when(
+ score < 50 ~ 5,
+ score >= 50 & score < 65 ~ 4,
+ score >= 65 & score < 80 ~ 3,
+ score >= 80 & score < 90 ~ 2,
+ score >= 90 ~ 1
+ )
+ ) -> [Link]
> [Link]
# A tibble: 1,137 x 5
X exam attempt score grade
<int> <chr> <int> <int> <dbl>
1 1 516647/Mathematics 1 75 3
2 2 193841/Data Bases 2 49 5
3 3 326761/Computer Networks 2 75 3
4 4 230275/Data Bases 1 73 3
5 5 911920/OOP 1 55 4
6 6 840873/Formal Languages 1 33 5
 7 7 473291/Mathematics 2 56 4
8 8 412501/Mathematics 2 102 1
9 9 320446/Formal Languages 2 26 5
10 10 332749/Formal Languages 1 67 3
# ... with 1,127 more rows
# i Use `print(n = ...)` to see more rows

d) Split the variable exam into 2 columns

> [Link]
# A tibble: 1,137 x 6
X [Link] exam attempt score grade
<int> <chr> <chr> <int> <int> <dbl>
1 1 516647 Mathematics 1 75 3
2 2 193841 Data Bases 2 49 5
3 3 326761 Computer Networks 2 75 3
4 4 230275 Data Bases 1 73 3
5 5 911920 OOP 1 55 4
6 6 840873 Formal Languages 1 33 5
7 7 473291 Mathematics 2 56 4
8 8 412501 Mathematics 2 102 1
9 9 320446 Formal Languages 2 26 5
10 10 332749 Formal Languages 1 67 3
# ... with 1,127 more rows
# i Use `print(n = ...)` to see more rows

e) Determine the total number of tests in each exam and the number of students participating
Number of Tests:
> [Link] %>%
+ select([Link]) %>%
+ unique() %>%
+ summarise([Link] = n())
# A tibble: 1 x 1
[Link]
<int>
1 243

Number of students in each exam

> [Link] %>%
+ group_by(exam) %>%
+ summarise(n = n())
# A tibble: 6 x 2
exam n
<chr> <int>
1 Computer Networks 186
2 Data Bases 190
3 Formal Languages 190
4 Mathematics 185
5 OOP 196
6 Software Engineering 190

f) For each subject, determine the absolute frequencies of the grades and store the result in a tibble
with the variables grade, Computer Networks`, `Data Bases`, `Formal Languages`, Mathematics, OOP
and `Software Engineering`.
> [Link] %>%
+ count(exam,grade) %>% # 2 Punkte
+ spread(key = exam, value = n) # 3 Punkte
# A tibble: 5 x 7
grade `Computer Networks` `Data Bases` `Formal Languages` Mathematics O
OP `Software Engineering`
<dbl> <int> <int> <int> <int> <in
t> <int>
1 1 6 10 4 4
4 1
2 2 8 12 6 16
15 19
3 3 58 46 57 43
62 57
4 4 53 69 75 72
69 76
5 5 61 53 48 50
46 37
f) For each subject, determine the minimum, maximum, the three quartiles, the mean of the variable
score, the number of participants and the dropout rates.
> [Link] %>%
+ mutate(
+ fail = if_else(grade <= 4,0,1)
+ ) %>%
+ group_by(exam) %>%
+ summarise(
+ Min = min(score),
+ Max = max(score),
+ Q1 = quantile(score, probs = 0.25),
+ Q2 = quantile(score, probs = 0.5),
+ Q3 = quantile(score, probs = 0.75),
+ Mean = mean(score),
+ [Link] = n(),
+ [Link] = sum(fail)/[Link]
+ )
# A tibble: 6 x 9
exam Min Max Q1 Q2 Q3 Mean [Link]
[Link]
<chr> <int> <int> <dbl> <dbl> <dbl> <dbl> <int>
<dbl>
1 Computer Networks 29 108 48 60 70.8 59.4 186
0.328
2 Data Bases 9 95 49 59.5 71 60.3 190
0.279
3 Formal Languages 18 97 49.2 59 68 58.9 190
0.253
4 Mathematics 19 102 49 57 70 58.6 185
0.270
5 OOP 22 102 50 61 69 60.3 196
0.235
6 Software Engineering 24 94 52 60 71.8 61.2 190
0.195

h) Create side by side boxplots of the score for each subject and interpret the results.
> boxplot([Link]$score ~ [Link]$exam)
# no differences between the subjects

I ) Determine the contingency table of the variables attempt and grade and evaluate the indifference
table and chi-square value.
> [Link]([Link]$attempt, [Link]$grade)$observed %>% addmargins()
[Link]$grade
[Link]$attempt 1 2 3 4 5 Sum
1 14 49 206 226 184 679
2 14 21 84 142 84 345
3 1 6 33 46 27 113
Sum 29 76 323 414 295 1137
> [Link]([Link]$attempt, [Link]$grade)$expected %>% addmargins()
[Link]$grade
[Link]$attempt 1 2 3 4 5 Sum
1 17.318382 45.38610 192.89094 247.23483 176.16974 679
2 8.799472 23.06069 98.00792 125.62005 89.51187 345
3 2.882146 7.55321 32.10114 41.14512 29.31838 113
Sum 29.000000 76.00000 323.00000 414.00000 295.00000 1137
> [Link]([Link]$attempt, [Link]$grade)$statistic
X-squared
14.05124

2) A biased coin (head with probability 1/3) is tossed. If the coins shows tail a fair die is rolled 5 times
and if the coin shows head a biased die (6 with probability 0.4) is rolled 5 times. The number of sixes
are counted.

a) Determine the density of the random X which counts the number of sixes.

In case of head X ~ B(n=5, p=0.4) and in case of a tail X~B(n=5, p=1/6). Weg et

P( X=i ) = P( X=i | head) / 3 + P(X=i | tail) * 2/3

> [Link] <- tibble(
+ no = 0:5,
+ [Link] = dbinom(no, size = 5, prob = 0.4),
+ [Link] = dbinom(no, size = 5, prob = 1/6),
+ dens = [Link]/3 + 2*[Link]/3,
+ )
> [Link]
# A tibble: 6 x 4
no [Link] [Link] dens
<int> <dbl> <dbl> <dbl>
1 0 0.0778 0.402 0.294
2 1 0.259 0.402 0.354
3 2 0.346 0.161 0.222
4 3 0.230 0.0322 0.0982
5 4 0.0768 0.00322 0.0277
6 5 0.0102 0.000129 0.00350

b) Evaluate the expected value and the variance of the random variable X.
> EX <- sum([Link]$no * [Link]$dens)
> EX2 <- sum([Link]$no^2 * [Link]$dens)
> VarX <- EX2 - EX^2
> EX; VarX
[1] 1.222222
[1] 1.165432
c) What is the probability that the coin had shown a head if 3 sixes has been in the 5 rolls?

From the Bayes Theoren weg et

P( head | X=3 ) = P( head and X=3 ) / P(X=3) = P( X=3 | head) * P(head) / P( X=3 ) =

dbinom(3,size=5,prob=0.4) * (1/3) / 0.0982 = 0.782

3) The weight of bags of grain can be assumed to be random variable with expected value 50 kg and
standard deviation 2 kg. The price for one kilogram grain, which a farmer achieves, is 0.53 Euro per
kg. 300 bags of grain fit into a truck. Let X be the price a farmer can obtain for a fully loaded truck.

a) Determine an approximate distribution of the random variable X.

Applying the central limit theorem we get X approximately

N( mu = 0.53*300*50, sigma^2 = 300*0.53^2*2^2)

b) Find the probability that X is bigger than 8000 Euro.

1-pnorm(8000, mean = mu, sd = sigma) = 0.003231175

c) What are the lower and upper bounds of the interval containing the middle 80% of X?

qnorm(c(0.1,0.9), mean = mu, sd = sigma) = c(7926.471, 7973.529)

d) What is the minimum number of bags a farmer must sell to earn at least 20000 euros with a
probability of 95%?

Analytic solution:

P( X_1+…+X_n >= 20000 ) = 0.95, i.e. 20000 is approximately the 5% quantil of

N(n*0.53*50, n*0.53^2*2^2). From the relationship between quantile of an arbrtrary and standard
normal distribution we get

20000 = n * 0.53 * 50 + 2*0.53* sqrt(n) * qnorm(0.05)

With u = sqrt(n) we have the equation

20000 = u^2 * 0.53 * 50 + 2*0.53* u * qnorm(0.05). The solutions are

c(-qnorm(0.05)/50 - sqrt(-(qnorm(0.05)/50)^2 + 20000/(50*0.53)),
+ -qnorm(0.05)/50 + sqrt(-(qnorm(0.05)/50)^2 + 20000/(50*0.53)))^2
[1] 752.9095 756.5245
Since u must be positive we get n = 27.50499^2 = 756.5245

> tibble(
+ [Link] = 300:900,
+ [Link].20000 = 1-pnorm(20000,
+ mean = [Link]*0.53*50,
+ sd = (0.53^2*2^2*[Link])^0.5)
+ ) %>%
+ filter([Link].20000 > 0.95) %>%
+ summarise([Link] = min([Link]))
# A tibble: 1 x 1
[Link]
<int>
1 757
4) In a representative sample 100 people were asked if they prefer candidate A or not if the election
were held next Sunday.

a) Determine the relative frequency of voter prefering A.

survey <- c(1, 0, …, 1, 0, 1, 0, 0, 0)
> n <- length(survey)
> [Link] <- sum(survey)/n
> [Link]
[1] 0.32

b) Show that the relative frequency is an unbiased point estimators for the proportion of voters
preferring A in the whole population.

Let p be the probability that a randomly chosen voter supports A. Then X = # number of voters
preferring A in a sample of size n follows a B(n, p)-distribution. The expected value of the relative
frequency

[Link] = X / n is n*p_i / n = p_i, i.e. [Link] is an unbiased estimator of p.

c) Determine a normal approximation and the exact two-sided 95% confidence interval of the
unknown proportions p.
> alpha <- 0.05
> X <- sum(survey)
> n <- length(survey)
> [Link] <- [Link]-qnorm(1-alpha/2,mean=0, sd=1)*sqrt([Link]*([Link])/n)
> [Link] <- [Link]+qnorm(1-alpha/2,mean=0, sd=1)*sqrt([Link]*([Link])/n)
> # approx. CI
> c([Link], [Link])
[1] 0.2285724 0.4114276

Approximation auch mittels t-test möglich

> [Link](survey, alternative = "[Link]", [Link] = 1-alpha)

One Sample t-test

data: survey
t = 6.8256, df = 99, p-value = 7.095e-10
alternative hypothesis: true mean is not equal to 0
95 percent confidence interval:
0.2269747 0.4130253
sample estimates:
mean of x
0.32

d) Find an approximate value of the minimal sample size so that the length of the approximate
confidence interval for p is less the 0.1.

Length of the confidence interval

L = 2*sqrt([Link]*([Link])/n)*qnorm(1-alpha/2) <=

<= 2*sqrt(0.5*0.5/n)*qnorm(1-alpha/2) = sqrt(1/n)*qnorm(1-alpha/2) <= 0.1

ceiling((qnorm(1-alpha/2, mean = 0, sd = 1)/0.1)^2)
[1] 385

e) Determine the confidence level so that the width of the approximate confidence interval of p is
equal to 0.1.

From L = 2*sqrt([Link]*([Link])/n)*qnorm(1-alpha/2) = 0.1 ->

0.1 / (2*sqrt([Link]*([Link])/n)) = qnorm(1-alpha/2) = 0.1 -> 1-alpha =
2*pnorm(0.1/(2*sqrt([Link]*([Link])/n)))-1 # 0.7162198
[1] 0.7162198

5)

a) What are the null hypothesis and alternative for an appropriate statistical test?

H0: mu <= 5, H1: mu > 5

b) Perform an appropriate statistical test to verify the null hypothesis at the 5% level. What is the test
decision and what is the p-value?

one sample t-test

> [Link](x=X, alternative = "greater", mu = 5, [Link] = 0.95)

One Sample t-test

data: X
t = 3.8881, df = 19, p-value = 0.0004947
alternative hypothesis: true mean is greater than 5
95 percent confidence interval:
5.095785 Inf
sample estimates:
mean of x
5.1725
Since p-value <= 0.05 H0 can be rejected.

c) In making the above conclusion, which type of error are you risking, type I or type II?

type I error

d) Perform an appropriate statistical test to verify the conjecture at the 5% level and determine the
pvalue.
> Y <- c(5.11, 5.20, 4.91, 4.94, 5.22, 4.59, 5.45, 4.80, 5.13, 4.79,
+ 4.69, 4.14, 4.64, 5.20, 5.16, 5.49, 4.94, 5.75, 5.36, 4.69)

Since each car is used in each sample, we can use a paired samples t-test to determine if the mean co
nsumption has been changed by the fuel treatment.

> [Link](x=X, y=Y,alternative = "[Link]", paired = TRUE, [Link] = TRU

E,
+ [Link] = 0.95)

Paired t-test

data: X and Y
t = 2.2711, df = 19, p-value = 0.03496
alternative hypothesis: true difference in means is not equal to 0
95 percent confidence interval:
0.01274405 0.31225595
sample estimates:
mean of the differences
0.1625

Since the the p-value is lower than alpha the Null-Hypothesis that there is no difference can be
rejected.