1.
On 19 June 2021, The Straits Times published the figure below, taken from a popu-
lation census of Singapore. Each household may only belong to a single category.
What can be said about the resident households, earning more than $6,999 from work?
From the following statements, select all that apply.
(A) A majority of resident households are earning more than $6,999 from work in
2020.
(B) A larger proportion of resident households are earning more than $6,999 from
work in 2020, as compared to 2010.
(C) rate(Income > $6,999 | 2020) > rate(Income > $6,999 | 2010). Here “Income”
represents Household monthly income from work.
(D) rate(Income > $6,999 | 2020) < rate(Income > $6,999 | 2010). Here “Income”
represents Household monthly income from work.
(A), (B) and (C) are correct. Proportions must add up to 100%. In 2020, the propor-
tion earning more than $6,999 is
100% − (13.3 + 2 + 10.2 + 10.6 + 10.4)% = 53.5%.
Thus a majority of resident households are earning more than $6,999 in 2020.
In 2010, the proportion earning more than $6,999 is
100% − (10.5 + 3.5 + 15.2 + 16.2 + 14.1)% = 40.5%.
This is smaller than in 2020.
Recall that proportions are the same as rates. Since 53.5% is greater than 40.5%, this
means that rate(Income > $6,999 | 2020) is greater than rate(Income > $6,999 | 2010),
showing that a larger proportion is earning more than $6,999 in 2020, as compared to
2010.
2. A researcher collected data on his study subjects. Unfortunately, he spilled coffee on
his table of values, resulting in some missing information. Based on the remaining
information in the table, calculate rate(Male) in the study.
X Y Row total
Female 300 100
Male 50
Column total 500
1
(A) rate(Male) = 0.2.
(B) rate(Male) = 1.
(C) rate(Male) = 0.7.
(D) rate(Male) = 0.1.
Answer is (A). We can fill in the missing information in the table completely using
100
the information given. Thus rate(Male) = 500 = 0.2.
X Y Row total
Female 300 100 400
Male 50 50 100
Column total 350 150 500
3. For the year 2020, the marginal death rate of country A is greater than the death
rate among the females of country A, or in other words, rate(Death) > rate(Death |
Female). Which of the following statements must be true in country A for the year
2020?
(A) rate(Male) < rate(Male | Death).
(B) rate(Male) > rate(Male | Death).
(C) rate(Male) = rate(Male | Death).
Answer is (A). By the basic rule of rates,
rate(Death | Male) > rate(Death) > rate(Death | Female),
so symmetry rule gives us
rate(Male | Death) > rate(Male | No death).
By the basic rule of rates again,
rate(Male | Death) > rate(Male) > rate(Male | No death).
4. Categorical variables A and B are associated with each other. This means that
rate(A | B) ̸= rate(A | not B).
Based on the given information, which of the following statements is/are always true?
(I) rate(B | A) ̸= rate(B | not A).
(II) rate(B | A) = rate(A | B).
(III) rate(A | B) ̸= rate(not A | B).
(A) Only (II).
(B) Only (II) and (III).
(C) Only (I).
(D) All the statements are true.
Answer is (C). Statement (I) is true. If A and B are associated, then so are B and A
by definition of association. Statement (II) need not be true. Consider the following
example:
2
Heart disease No heart disease Row total
Smokers 38 14962 15000
Non-smokers 44 84956 85000
Column total 82 99918 100000
Here, rate(Heart disease | Smokers) ̸= rate(Smokers | Heart disease). Statement (III)
need not be true as well. Consider the following example:
Heart disease No heart disease Row total
Smokers 100 100 200
Non-smokers 300 500 800
Column total 400 600 1000
Observe that smoking and heart disease are associated in this data set since rate(Heart
disease | Smokers) = 12 while rate(Heart disease | Non-smokers) = 83 . But note that
rate(Heart disease | Smokers) = rate(No heart disease | Smokers) = 0.5.
Here, we let A be heart disease and B be smokers.
5. The following data is coming from a survey done on the effectiveness of the coaching
sessions on job hunting for fresh graduates. In this survey, three questions were asked
of the participants:
Q1: How much is their salary or if they are unemployed?
Q2: If they have received coaching? (Answer YES or NO.)
Q3: If their job was a continuation of their internship? (Answer YES or NO.)
The table below summarises the answers received for the three questions.
Q2: NO Q2: YES
Monthly Salary Q3: YES Q3: NO Q3: YES Q3: NO Total
> $4000 26 85 8 28 147
Q1:
< $3000 5 31 0 7 43
$3000 − $4000 41 181 13 48 283
Unemployed 604 141 745
Based on the data provided above, which of the following statements is/are correct?
(I) There is an association between receiving coaching and having a salary above
$4000.
(II) There is an association between receiving coaching and landing a job in continu-
ation of an internship.
(A) Only (I).
(B) Only (II).
(C) Neither (I) nor (II).
(D) Both (I) and (II).
Answer is (D). Comparing the conditional rates:
rate(>$4000 | Coaching) = 8+28
8+28+7+0+13+48+141 = 0.15.
rate(>$4000 | No Coaching) = 26+85
26+85+5+31+41+181+604 = 0.11.
Since these two rates are different, there is an association between receiving coaching
and having a salary above $4000. Hence, statement (I) is correct. We also compare:
8+0+13
rate(Internship continuation | Coaching) = 8+0+13+28+7+48+141 = 0.086.
3
26+5+41
rate(Internship continuation | No Coaching) = 26+5+41+85+31+181+604 = 0.074.
Since these two rates are different, there is an association between receiving coaching
and landing a job continuation of an internship. Hence, statement (II) is correct.
6. The contingency table below shows the classification of hair descriptions of students
studying in an international school in Singapore.
Hair type
Straight Curly
Hair colour Male Female Male Female Total
Red 7 9 8 5 29
Brown 35 20 12 16 83
Blonde 51 55 38 27 171
Black 22 25 19 24 90
Total 115 109 77 72 373
The marginal rate, rate(Curly), is %; while the joint rate, rate(non-Black
and Female) is %. Give each answer as a percentage correct to 2 decimal
places.
To calculate the marginal rate, rate(Curly), we take the column totals of all Curly-
haired persons (both Male and Female) dvided by the grand total of everyone in the
data set, i.e.
(77 + 72)
≈ 39.95% (2 decimal places).
373
Then, to calculate the joint rate, rate(non-Black and Female), we take the count of
“Females with non-black hair” divided by once again the grand total of everyone in
the data set, i.e.
(9 + 20 + 55 + 5 + 16 + 27)
≈ 35.39% (2 decimal places).
373
7. A group of market researchers were commissioned to investigate the relationship be-
tween two food delivery companies (Grabfood and Foodpanda) and their punctuality
of deliveries (whether they are on time or late). The following chart was used by the
researchers to aid in the presentation of their findings.
4
Which of the following statements is/are true based on the information given above?
Select all that apply.
(A) Grabfood is positively associated with being on time for food deliveries.
(B) Grabfood is positively associated with being late for food deliveries.
(C) Foodpanda is positively associated with being on time for food deliveries.
(D) Foodpanda is positively associated with being late for food deliveries.
(B) and (C) are correct. Let A be “Grabfood”, so NA is “Foodpanda”. Let B represent
“late” and NB represent “On Time”. Since
20 30
rate(Late | Grabfood) = > = rate(Late | Foodpanda),
150 300
there is a positive association between A and B. Hence, Grabfood is positively associ-
ated with being late for food deliveries. This also means that Foodpanda is negatively
associated with being late for food deliveries, or that Foodpanda is positively associ-
ated with being on time for food deliveries.
8. A newspaper article had a headline “30% of local university students admitted last
year graduated from a polytechnic”. Assume there are only 2 universities (Uni A
and Uni B). In Uni A, 50% of its local students admitted last year graduated from
a polytechnic. In Uni B, the percentage of its local students admitted last year who
graduated from a polytechnic must be
(A) more than 50%..
(B) 40%.
(C) between 30% and 50%.
(D) less than 30%.
Answer is (D). By the basic rule on rates, since the overall rate is 30% and the rate
at Uni A is 50%, the rate at Uni B must be less than 30%.
5
9. By “elderly”, we mean a person who is more than 65 years old. In Singapore, the
percentage of elderlies among women is higher than the percentage of elderlies among
men. Which of the following statements must be true?
(I) In Singapore, the percentage of women among elderlies is higher than the per-
centage of women among the non-elderlies.
(II) In Singapore, the percentage of women is higher than the percentage of men
among elderlies.
(A) Only (I).
(B) Only (II).
(C) Both (I) and (II).
(D) Neither (I) nor (II).
Answer is (A). From the information given,
rate(Elderlies | Women) > rate(Elderlies | Men)
and thus women and elderlies are positively associated. This also means that
rate(Women | Elderlies) > rate(Women | Non-elderlies),
which is expressed by statement (I). We cannot determine the percentage of women
and men among the elderlies with the information given.
10. Su is investigating the association between blood pressure and workaholism in a cer-
tain population. Someone who works more than 75 hours per week is considered a
workaholic.
The income level and blood pressure (high or normal) for each subject and whether
or not they are classified as workaholic is recorded and summarised in the table be-
low. Here “HBP” denotes “high blood pressure” while “NBP” denotes “normal blood
pressure”.
Income Group
Low Middle High
HBP NBP HBP NBP HBP NBP
Workaholic 25 75 23 87 26 134
Non-workaholic 25 80 18 72 9 51
Consider the “Low” income level group.
rate(HBP | Workaholic) is (1) % while rate(HBP) is (2) %.
Fill in the blanks in the statement above, giving your answers as percentages correct
to the nearest whole numbers.
25
The answer for the first blank is 25 since (25+75) = 0.25 = 25%. The answer for the
50
second blank is 24 since (50+155) = 0.24 = 24%.
11. The Lord of the Rings: The Fellowship of the Ring was released in December 2001.
Suppose that
(I) Among the people in Singapore who were born before 2000, 10% watched the
film.
(II) Among the people in Singapore who were born during or after 2000, 20% watched
the film.
6
Choose the best option below. Among all the people in Singapore, the percentage who
watched the film .
(A) must be 15%.
(B) must be between 10% and 20%.
(C) can be less than 10%.
(D) can be more than 20%.
Answer is (B). By the basic rule on rates, since 10% and 20% are the respective rates
in the two groups, the overall rate must be between 10% and 20%.
12. Darren is planning a surprise Bubble Tea Party for his class of 30 students during the
last tutorial of GEA1000. Each student chooses either milk tea or fruit tea (but not
both). The following information is what he has gathered about his tutorial class:
Of the 30 students, 40% are males.
60% of the students who drink milk tea at this party are males.
70% of the students who drink fruit tea at this party are females.
Which of the following statements can be concluded from the above information about
bubble tea consumption in Darren’s tutorial class?
(I) There is positive association between being male and drinking milk tea.
(II) The majority of the 30 students are fruit tea drinkers.
(A) Only (I).
(B) Only (II).
(C) Both (I) and (II).
(D) Neither (I) nor (II).
Answer is (C). Since rate(Female | Fruit tea) = 70%, we have rate(Male | Fruit tea)
= 30%. We are also given that rate(Male | Milk tea) = 60%. Therefore rate(Male
| Milk tea) > rate(Male | Fruit tea) and there is positive association between being
male and drinking milk tea. Since rate(Male | Fruit tea) = 30%, rate(Male) = 40%,
and rate(Male | Milk tea) = 60%, rate(Male) = 40% is closer to the conditional rate
rate(Male | Fruit tea) as compared to rate(Male | Milk tea), we can deduce that
rate(Fruit tea) is greater than 50%. Hence, both statements (I) and (II) are correct.
13. A researcher was studying the effects of Paxlovid for COVID-19 treatment on a group
of patients. The researcher took notes on whether the treatment was successful or not
(Success/Failure), and the gender of each patient (Male/Female).
Which of the following statements is/are correct?
(I) rate(Success | Male) refers to the proportion of successful treatments that are
males.
(II) rate(Failure & Male) refers to the proportion of males that had a failed treatment.
(III) rate(Female) refers to the proportion of females among the patients.
(A) Only (III).
(B) Only (I) and (II).
(C) Only (II) and (III).
(D) All three statements are correct.
7
Answer is (A). Rate(Success | Male) refers to the proportion of males that had a
successful treatment. Rate(Failure & Male) refers to the proportion of patients that
were male and had a failed treatment.
14. Suppose that 70% of the male graduates in 2018 from university GER got married by
2022. In addition, 5000 of all 6250 graduates in 2018 from university GER got married
by 2022. The percentage of female graduates in 2018 from university GER who got
married by 2022 is
(A) lower than 70%.
(B) between 70% and 80%.
(C) higher than 80%.
(D) impossible to determine from the information given.
Answer is (C). By the basic rule on rates, since the overall rate of graduates who are
married is 5000/6250 = 80%, and the rate of male graduates who are married is 70%,
the rate of female graduates who are married must be higher than 80%.
15. An incomplete breakdown of students in University S by its four faculties and by sex
is as follows:
Male Female Row Total
Engineering 700
Arts 250 400 650
Architecture 50 50 100
Science 550
Column Total 1200 800 2000
Based on the above, which of the following deductions must be true? Select all that apply.
(A) rate(Arts) is closer to rate(Arts | Male) than to rate (Arts | Female).
(B) There is a negative association between Arts and Male.
(C) rate(Science) is closer to rate(Science | Female) than to rate (Science | Male).
(D) There is no association between Architecture and Female.
(A) and (B) are correct. Since rate(Male) > rate(Female), we would expect that
rate(X), where X can refer to any of the four faculties, namely Engineering, Arts,
Architecture or Science, to be closer to rate(X | Male), as compared to rate(X |
Female); unless rate(X | Male) = rate(X | Female), in which case rate(X) = rate(X
| Male) = rate(X | Female). Thus (A) is true but (C) is not true. Since
rate(Arts | Male) = 250/1200 < rate (Arts | Female) = 400/800,
There is a negative association between Arts and Male. That is, (B) is true as well.
Finally, rate(Architecture | Male) = 50/1200, which is not equal to rate(Architecture
| Female) = 50/800. Therefore, there is an association between Female and Architec-
ture, and (D) is not true.
16. A researcher conducted a study to investigate if the usage of ChatGPT is associated
with the passing of an exam. He found the rate(pass | usage of ChatGPT) to be equal
to 0.4 and rate(fail | no usage of ChatGPT) to be equal to 0.3. Which of the following
statements is/are true? Select all that apply.
(A) There is insufficient information to make a conclusion about the association be-
tween the two variables.
(B) The usage of ChatGPT is positively associated with failing the exam.
8
(C) Passing the exam is negatively associated with the usage of ChatGPT.
(D) Failing the exam is positively associated with no usage of ChatGPT.
(B) and (C) are correct.
Rate(pass | no usage of ChatGPT) = 1–rate(fail | no usage of ChatGPT) = 1–0.3 = 0.7.
Since rate(pass | usage of ChatGPT) < rate(pass | no usage of ChatGPT), there is
a negative association between passing the exam and the usage of ChatGPT. This
means (C) is correct and (A) is wrong. It is also equivalent to saying that there is
a positive association between the usage of ChatGPT and failing the exam, which
means (B) is correct. Lastly, (B) being correct implies that (D) is wrong.
17. Ron is studying the reading habits of a certain group of adults in relation to their
sex and whether they wear spectacles. Some of the data he collected are presented
in the incomplete contingency table below. In addition, you are given that rate(Wear
spectacles) = 0.5.
Males Females
Read books Wear Do not wear Wear Do not wear Total
weekly spectacles spectacles spectacles spectacles
Yes 72 65
No 20 34
Total 239 480
Fill up the missing cells in the table where possible and determine which of the fol-
lowing statements must be true. Select all that apply.
(A) rate(Yes and Wear spectacles) = 61/240.
(B) rate(Do not wear spectacles and Females) = 31/96.
(C) rate(No | Males) cannot be determined.
(D) rate(Females | Wear spectacles) = 42/120.
(A), (B) and (D) are correct. We can fill up the table in the order as shown below.
(i) 480–239 = 241.
(ii) 241–72–65–20 = 84.
Since rate(Wear spectacles) = 0.5, the total number who wear spectacles is 240. Thus
we have
(iii) 240–72–84–34 = 50.
To show that (A) is true, we find
rate(Yes and Wear spectacles) = (72 + 50)/480 = 61/240.
(B) is true since
rate(Do not wear spectacles and Females) = (239 − 50 − 34)/480 = 31/96.
To show that (C) is not true, we have
rate(No | Males) = (84 + 20)/241 = 104/241.
We can check that (D) is true since
rate(Females | Wear spectacles) = (50 + 34)/240 = 42/120.
9
Males Females
Read books Wear Do not wear Wear Do not wear Total
weekly spectacles spectacles spectacles spectacles
Yes 72 65 (iii) 50
No (ii) 84 20 34
Total (i) 241 239 480
18. To tackle the rise in dengue infections among residents in Compass Condominium, the
building management sent out a survey form to all of its 100 households to find out
water disposal habits among its residents. Households were randomly assigned into
only one of the two modes: hard-copy surveys placed in the residents’ mailboxes, and
email surveys sent to the residents’ registered email addresses. Based on the above
random assignment, 55 of the 100 households were given hard-copy surveys in their
mailboxes. Half of those who responded did so using email surveys, while 40% of those
who did not respond were sent email surveys. What can we conclude from the above?
Select all that apply.
(A) The response rate of the survey is 50%.
(B) There is a negative association between those who received hard-copy surveys
and those who responded.
(C) 50% of those who were asked to do email surveys responded.
(D) The rate of response among email surveys is greater than the overall rate of
response.
(A), (B) and (D) are correct. Since rate(email | response)= 50% and rate(email | no
response)= 40% and rate(email)= 45%, we can deduce that rate(response)= rate(no
response)= 50%. Therefore (A) is correct. Moreover, from the above we can deduce
that rate(email | response) > rate(email | no response), so there is a positive association
between email and response, which implies that there is a negative association between
hard-copy and response, by the symmetry rule. Therefore (B) is correct. (C) cannot
be concluded from the above, as rate(response | email) may not be equal to rate(email
| response). Generally, rate(A | B) = rate(A and B)/rate(B) is not equal to rate(B |
A)= rate(B and A)/rate(A), unless rate(A and B) = 0, or rate(A) = rate(B), which
is not the case here. (D) is correct, as rate(response | email) > rate(response) >
rate(response | no email) can be deduced using the symmetry rule, and the basic rule
on rates.
19. Consider the following partial contingency table that gives the breakdown of students
by gender in department A and department B of a local university. We are told that
there is no association between gender and department. The total number of students
in both departments is .
Male Female Row Total
A 30 90
B 140
Answer is 300. The table can be filled in easily.
(1) Number of female students in Department A is 60.
(2) Since there is no association between department and gender, the proportion of
Males to Females in both departments will be the same. Hence, there are 70 male
students in Department B. Thus the total number of students from departments A
and B is 90 + 210 = 300.
10
Male Female Row Total
A 30 60 90
B 70 140 210
20. The table below provides the number of all the teachers employed in the different
institutions in 2021 in Singapore, categorised by age and sex.
Using only the information in the table, fill in the blanks below (giving your answers
to 2 decimal places). Among all the teachers aged 30-39, (1) % of them
were male secondary school teachers. In addition, (2) % of female primary
school teachers were aged 45 years old and above.
Answer: (1) 14.00. (2) 37.59.
((2444−1728)+(2559−1660))
For blank 1, we have (2785+2832+2444+2559+402+513) = 14.00%.
(2228+1494+1166)
For blank 2, we have (68+1030+2341+2181+2495+2228+1494+1166) = 37.59%.
21. Suppose that in a population, it is known that within males and within females,
smoking and binge drinking are positively associated. However, Simpson’s Paradox
is observed when the male and female subgroups are combined. From the statements
below, select all that are true.
(A) Overall rate(Binge drinker | Smoker) ≤ overall rate(Binge drinker | Non-smoker).
(B) Overall rate(Smoker | Binge drinker) > overall rate(Smoker | Non-binge drinker).
(C) Overall rate(Smoker | Binge drinker) ≤ overall rate(Smoker | Non-binge drinker).
(D) Overall rate(Non-binge drinker | Smoker) ≥ overall rate(Non-binge drinker | Non-
smoker).
(A), (C) and (D) are correct. Since we know that smoking and binge drinking are
positively associated in the subgroups, and Simpson’s Paradox is observed when the
subgroups are combined, this means that overall, smoking and binge drinking are
negatively associated (or not associated). Hence, we can conclude that
overall rate(Binge drinker | Smoker) ≤ overall rate(Binge drinker | Non-smoker)
11
and
overall rate(Smoker | Binge drinker) ≤ overall rate(Smoker | Non-binge drinker).
Since smoking and binge drinking are negatively associated (or not associated), we
can also conclude that smoking and not binge drinking are positively associated (or
not associated). Hence we have
overall rate(Non-binge drinker | Smoker) ≥ overall rate(Non-binge drinker | Non-smoker).
22. A researcher wants to find out if drinking tea helps to reduce memory loss. He in-
terviewed 100 elderly citizens from an Elder Care Center and inquired if they were
tea drinkers. 60 of them were classified as tea drinkers, while the remaining 40 were
not. He then asked them to play a specific memory game to test their memory. The
researcher also noted that a potential confounding variable was “gender”. To control
for this potential confounder (gender), the researcher could perform
(A) double blinding.
(B) random assignment.
(C) slicing of the data.
Answer is (C). This is an observational study, and thus only slicing can be done
to control for this potential confounder. Random assignment is only suitable for
experimental studies. Double blinding also does not control for gender as a confounder.
23. The table below shows male and female patients undergoing two treatment types, X
or Y . The outcome of the treatment is designated as either successful or unsuccessful.
The success rates of the respective treatments across genders are also calculated.
Male Female
Patients Succ. # Suc. Rate Patients Succ. # Succ. Rate
X ? ? 50% 40 32 80%
Y ? ? ? ? ? 60%
Total 100 50 50% ? ? ?
Unfortunately, some of the data is missing. We know that all missing values are
non-zero. Which of the following statements must be true?
(I) Simpson’s Paradox is observed when the subgroups of Treatment X and Treat-
ment Y are combined, when considering the relationship between gender and
outcome.
(II) Treatment type is a confounder between the variables gender and outcome.
(A) Only (I).
(B) Only (II).
(C) Neither (I) nor (II).
(D) Both (I) and (II).
Answer is (C). By the basic rule of rates, among males, rate of success given Treatment
Y must be 50%. By the basic rule of rates, among females, overall rate of success must
be between 60% and 80%. Simpson’s Paradox is not observed when the subgroups
12
are combined, hence statement (I) is false. Depending on the data, treatment type
may or may not be a confounder. For the following given set of values in the table,
80 40
rate(X | Male) = = = rate(X | Female),
100 50
so treatment type is not associated with gender. Treatment type is not necessarily a
confounder, hence statement (II) is false.
Male Female
Patients Succ. # Suc. Rate Patients Succ. # Succ. Rate
X 80 40 50% 40 32 80%
Y 20 10 50% 10 6 60%
Total 100 50 50% 50 38 76%
24. A tuition agency is interested to see whether teaching English via a new online platform
is more effective compared to their current teaching methods for Primary 3 students.
Every child has a choice whether he/she wants to enrol in the class that teaches using
the online platform (denoted as Class A) or in the class without the online platform
(denoted as Class B) and parental consent is also obtained. At the end, all students
are given an English test and are awarded an “S” grade if they pass the test. The
table below provides some of the information.
Class A Class B
Number S-grade Rate (%) Number S-grade Rate (%)
Males 200 40 20 100 10 10
Females w x y z
You are given the following rates (for males) as shown in the table above:
40
rate(S-grade | Class A) = 200 = 20%.
10
rate(S-grade | Class B) = 100 = 10%.
By considering similar rates for females, which of the following are possible values
of w, x, y, z such that Simpson’s Paradox will be observed in the above table when
combining males and females together within Class A and Class B?
(A) w = 100, x = 20, y = 80, z = 8.
(B) w = 200, x = 140, y = 100, z = 20.
(C) w = 80, x = 30, y = 200, z = 50.
(D) w = 50, x = 40, y = 200, z = 150.
Answer is (D). We simply have to test each option and see which of the options results
in Simpson’s Paradox being observed. The calculations are shown in the four tables
below, one table for each given option.
Testing w = 100, x = 20, y = 80, z = 8:
Class A Class B
Number S-grade Rate (%) Number S-grade Rate (%)
Males 200 40 20 100 10 10
Females 100 20 20 80 8 10
Overall 300 60 20 180 18 10
No Simpson’s Paradox observed as the direction of association is the same across the
two subgroups and overall.
Testing w = 200, x = 140, y = 100, z = 20:
13
Class A Class B
Number S-grade Rate (%) Number S-grade Rate (%)
Males 200 40 20 100 10 10
Females 200 140 70 100 20 20
Overall 400 180 45 200 30 15
No Simpson’s Paradox observed as the direction of association is the same across the
two subgroups and overall.
Testing w = 80, x = 30, y = 200, z = 50:
Class A Class B
Number S-grade Rate (%) Number S-grade Rate (%)
Males 200 40 20 100 10 10
Females 80 30 37.5 200 50 25
Overall 280 70 25 300 60 20
No Simpson’s Paradox observed as the direction of association is the same across the
two subgroups and overall.
Testing w = 50, x = 40, y = 200, z = 150:
Class A Class B
Number S-grade Rate (%) Number S-grade Rate (%)
Males 200 40 20 100 10 10
Females 50 40 80 200 150 75
Overall 250 80 32 300 160 53.3
Simpson’s Paradox is observed since the overall trend reverses compared to the direc-
tion observed in both subgroups.
25. A study was conducted to understand the relationship between a patient’s age and hav-
ing cardiovascular disease (CVD). The information on the variables “Age” (Young/Old)
and “CVD” (Has CVD/No CVD) was collected in the table below.
Young Old Total
Has CVD 100 50 150
No CVD 100 200 300
Total 200 250 450
Furthermore, it is known that a third variable, “Smoking”, is associated with “CVD”.
Using only the information given, which of the following statements must be true?
Select all that apply.
(A) Young patients are positively associated with having CVD.
(B) “Smoking” is a confounder when examining the association between “Age” and
“CVD”.
(C) “CVD” is a confounder when examining the association between “Age” and
“Smoking”.
(A) and (C) are correct. Young patients are positively associated with having CVD
because
100 50
rate(Has CVD | Young) = > = rate(Has CVD | Old).
200 250
A variable is shown to be a confounder when it is associated with both the other two
variables of interest. We are given that “Smoking” is associated with “CVD”, and
we know that “Age” is associated with “CVD” from the table, so we can conclude
that “CVD” is a confounder when examining the association between “Age” and
14
“Smoking”. However, we do not know if “Age” and “Smoking” are associated, so
we cannot conclude that “Smoking” is a confounder when examining the association
between “Age” and “CVD”.
26. A study was conducted to determine if treatment types (A and B) were associated
with how successful they were in curing disease X. The age of the subjects was also
recorded as it is a possible confounder. Each subject, depending on his/her age was
classified as either “Old” or “Young”. The table below shows the result of the study.
Treatment A Treatment B
Age Number Success Number Success
Old 120 25 280 75
Young 190 45 250 65
Total 310 70 530 140
Which of the statements below is correct?
(A) Age is a confounder between treatment types and how successful they are. Simp-
son’s Paradox is observed in this study.
(B) Age is a confounder between treatment types and how successful they are. Simp-
son’s Paradox is not observed in this study.
(C) Age is not a confounder between treatment types and how successful they are.
(D) More information is needed before determining if age is a confounder between
treatment types and how successful they are.
Answer is (C). Notice that age is not associated with success, since
100
rate(Success | Old) = 400 = 0.25;
110
rate(Success | Young) = 440 = 0.25.
For age to be a confounder, it needs to be associated with both treatment types and
success.
27. Su is investigating the association between blood pressure and “workaholism” in a
certain population. Someone who works more than 75 hours per week is considered a
workaholic.
The income level and blood pressure (high or normal) for each subject and whether
or not they are classified as “workaholic” are recorded and summarised in the table
below. Here “HBP” denotes “high blood pressure” while “NBP” denotes “normal
blood pressure”.
Income Group
Low Middle High
HBP NBP HBP NBP HBP NBP
Workaholic 25 75 23 87 26 134
Non-workaholic 25 80 18 72 9 51
Which of the following statements is true?
(A) We have an instance of Simpson’s Paradox for this data set, when considering
the association between being a “workaholic” and having “high blood pressure”,
first for individual income levels (“Low”, “Middle”, “High”) and then overall.
(B) We do not have an instance of Simpson’s Paradox for this data set, when con-
sidering the association between being a “workaholic” and having “high blood
pressure”, first for individual income levels (“Low”, “Middle”, “High”) and then
overall.
15
(C) We are not able to determine if we have an instance of Simpson’s Paradox for
this data set (or not), when considering the association between being a “worka-
holic” and having “high blood pressure”, first for individual income levels (“Low”,
“Middle”, “High”) and then overall. There is insufficient information given.
Answer is (A). For “Low”, “Middle” and “High” income levels, we have
rate(HBP | Workaholic) > rate(HBP | Non-workaholic).
For example, for the “Low” income level group,
25 25
= rate(HBP | Workaholic) > rate(HBP | Non-workaholic) = .
25 + 75 25 + 80
But overall, we have
(25+23+26)
rate(HBP | Workaholic) = (100+110+160) = 0.2,
which is smaller than
(25+18+9)
rate(HBP | Non-workaholic) = (105+90+60) = 0.204.
28. In NUS, the rate of coffee drinking among female students is 60% and the rate of
coffee drinking among male students is also 60%. It was found that the rate of coffee
drinking among scholarship students is 90%. Which of the following statements must
be true?
(I) Coffee drinking is positively associated with scholarship students.
(II) When considering the association between coffee drinking/non-coffee drinking
and scholarship students/non-scholarship students, Simpson’s Paradox is not ob-
served when the male and female subgroups are combined.
(A) Only (I).
(B) Only (II).
(C) Both (I) and (II).
(D) Neither (I) nor (II).
Answer is (C). It is given that rate(coffee drinking | female) = 60% = rate(coffee
drinking | male). By basic rule on rates, rate(coffee drinking) = 60%. It is also given
that rate(coffee drinking) = 60% < rate(coffee drinking | scholarship) = 90%. By
basic rule on rates, rate(coffee drinking | non-scholarship) < rate(coffee drinking |
scholarship), thus coffee drinkers are positively associated with scholarship students.
Since rate(coffee drinking | female) = 60% = rate(coffee drinking | male), gender is not
associated with coffee drinking. Hence gender is not a confounder in the association
between coffee drinking/non-coffee drinking and scholarship/non-scholarship students.
Therefore, there will be no Simpson’s Paradox observed when gender subgroups are
combined.
29. In a study, there are 3 variables being recorded: Sex, Cancer Status, and Smoking
Status. The rate of cancer among females is 40%, while the rate of cancer among
males is also 40%. Researchers discovered that the rate of cancer among smokers is
greater than the rate of cancer among non-smokers. Finally, it was found that males
were more likely to be smokers as compared to females.
Which of the following statements is/are true? Select all that apply.
16
(A) The overall rate of cancer is 50%.
(B) There is a positive association between smoking and cancer.
(C) Sex is a confounder when looking at the relationship between Smoking Status
and Cancer Status.
(D) Smoking Status is a confounder when looking at the relationship between Sex
and Cancer Status.
(B) and (D) are correct. Since rate(cancer | F) = rate(cancer | M) = 40%, the overall
rate(cancer) = 40%. As stated in the question, rate(cancer | smokers) > rate(cancer |
non-smokers), thus there is a positive association between smoking and cancer. Finally,
we know that there is a positive association between being males and smoking. Thus,
smoking status is a confounder. Sex is not a confounder in the relationship between
Smoking Status and Cancer Status because sex is not associated with cancer.
30. Billy collected data from his friends and found out that height is associated with
self-esteem. In particular, being tall is positively associated with having high self-
esteem. Megan, one of his friends, suggested that gender may be a confounder. Billy
agreed with her and decided to investigate this. After calculating rate(Male | Tall) and
rate(Male | Short), he found that gender is associated with height. He then shared with
Megan that she was right, and that gender is indeed a confounder. However, Megan
informed Billy that he missed a step in his conclusion. Which one of the following is
a possible pair of rates Billy missed before concluding that gender is a confounder?
(A) rate(Male | High self-esteem) and rate(Female | Low self-esteem)
(B) rate(Male | High self-esteem) and rate(Female | High self-esteem)
(C) rate(High self-esteem | Male) and rate(High self-esteem | Female)
(D) rate(High self-esteem | Male) and rate(Low self-esteem | Female)
Answer is (C). Billy has already found out that gender is associated with height. He
needs to show that gender is associated with self-esteem before he can conclude that
gender is a confounder. From the four options above, the only possible pair of rates
Billy can use to check for association between gender and self-esteem is (C). The pair
of rates in (C) compare the proportion of those with high self-esteem among males
and the proportion of those with high self-esteem among females, which is the correct
pair of proportions to compare, out of the four options, when checking for association
between gender and self-esteem.
17