X

MS922 Quantitative Business Analysis

T
SHTR
E APTLHACCLEY DOEF BUUSSEIFNUELS S

Hypothesis testing
L ESACRHNOI N
OGL

Dr Itamar Megiddo
Department of Management Science
[email protected]

My office Hours:
X

Objectives
S T R AT H C LY D E B U S I N E S S S C H O O L

• Develop an understanding of Hypothesis Testing:

• One-tailed normal-distribution based test

• One-tailed t-distribution based test
• Two-tailed normal distribution based test
• Two-tailed t-distribution based test

• Contingency Tables – Chi-squared test

• Type 1 and type 2 errors

 Inferential statistics
X
S T R AT H C LY D E B U S I N E S S S C H O O L

Probability distributions as models of

populations

POPULATION

Estimating Collecting
population data
parameters
using sample
statistics
SAMPLE
Analysing sample
data
X

Hypothesis testing framework

S T R AT H C LY D E B U S I N E S S S C H O O L

1. Make an initial assumption about the data generating

mechanism – a null hypothesis and an alternative hypotheis

2. Collect a sample of data as evidence

3. Calculate the probability of observing such evidence, given

the initial assumption

4. Depending on how likely or unlikely the evidence is, we reject

or fail to reject the initial assumption
X

Tests covered
S T R AT H C LY D E B U S I N E S S S C H O O L

• One-tailed normal-distribution based test

• One-tailed t-distribution based test

• Two-tailed normal distribution based test

• Two-tailed t-distribution based test

• Chi-square test of independence

X

Example: time spend reading

S T R AT H C LY D E B U S I N E S S S C H O O L

A national student survey indicated that the average time spent

reading per week was 5.42 hours .

At a local university, a professor suspects that the students at his

institution spend longer than 5.42 hours per week reading.

He conducts a survey of 45 students, and observes a mean

reading time of 6.02 hours per week, with standard deviation of
1.45 hours.
X

Example: time spend reading

S T R AT H C LY D E B U S I N E S S S C H O O L

Is this evidence of a true increase in the mean reading time per

week at the professor’s institution, when compared to the
national average?

Conduct a hypothesis test at a significance level of 5% to

examine this idea.
 Hypothesis testing framework
X

Example: time spend reading

S T R AT H C LY D E B U S I N E S S S C H O O L

1. Make an initial assumption about the data generating mechanism – a null

hypothesis and an alternative hypothesis

Null hypothesis (𝐻! ): average time spent reading is 5.42 hours per
week.
That is, 𝜇 = 5.42 ℎ𝑜𝑢𝑟𝑠 This relates to the
national average

Alternative hypothesis (𝐻" ): average time spent reading is greater

than 5.42 hours.
This relates to the
That is, 𝜇 > 5.42 ℎ𝑜𝑢𝑟𝑠 professor’s views

The ”greater than” inequality identifies

this is a one-tailed test
 Hypothesis testing framework
X

Example: time spend reading

S T R AT H C LY D E B U S I N E S S S C H O O L

1. Make an initial assumption about the data generating mechanism – a null

hypothesis and an alternative hypotheis
2. Collect a sample of data as evidence

Revisiting our sample evidence:

He conducts a survey of 45 students, and observes a mean
reading time of 6.02 hours per week, with standard deviation of
1.45 hours.
The sample size of 45 means that
the central limit theorem applies

𝜎
𝑿~𝑁 𝜇,
𝑛
 Hypothesis testing framework
X

Example: time spend reading

S T R AT H C LY D E B U S I N E S S S C H O O L

1. Make an initial assumption about the data generating mechanism – a null

hypothesis and an alternative hypotheis
2. Collect a sample of data as evidence
3. Calculate the probability of observing such evidence, given the initial
assumption

Use the CLT to calculate the Z-value and use the standard normal
table to obtain the probability of observing a sample mean of 6.02
hours or higher.

6.02 − 5.42
𝑧= = 2.776 From z-tables:
1.45 𝑷(𝒁 > 𝟐. 𝟕𝟕𝟔) = 𝟎. 𝟎𝟎𝟐𝟕
45
 Hypothesis testing framework
X

Example: time spend reading

S T R AT H C LY D E B U S I N E S S S C H O O L

1. Make an initial assumption about the data generating mechanism – a null hypothesis
and an alternative hypotheis
2. Collect a sample of data as evidence
3. Calculate the probability of observing such evidence, given the initial assumption
4. Depending on how likely or unlikely the evidence is, we reject or fail to reject the
initial assumption

The probability we calculated is called the p-value

The p-value is the probability of observing evidence (e.g., our
sample mean) which is as extreme or more extreme than our
present case.
 Hypothesis testing framework
X

Example: time spend reading

S T R AT H C LY D E B U S I N E S S S C H O O L

The probability we calculated is called the p-value

The p-value is the probability of observing evidence (e.g., our sample
mean) which is as extreme or more extreme than our present case.

Our p-value is 0.0027.

We are asked to conduct a test at significance level 5% or 0.05.

As our p-value is less than our significance level, we reject the null
hypothesis (𝐻! ).
X

Tests covered
S T R AT H C LY D E B U S I N E S S S C H O O L

• One-tailed normal-distribution based test

• One-tailed t-distribution based test

• Two-tailed normal distribution based test

• Two-tailed t-distribution based test

• Chi-squared test of independence

X

Smaller samples
S T R AT H C LY D E B U S I N E S S S C H O O L

• In the previous case, we had a sample of 45. As with

confidence intervals, this allowed us to:

• Use the central limit theorem

• Calculate our z-value
• Use the standard normal distribution table to obtain our
• p-value
• Accept/reject our null hypothesis on this basis

What if we have a sample size of 7?

X

What to use when?

S T R AT H C LY D E B U S I N E S S S C H O O L

• If we know the true population standard deviation, we calculate

a z-value.

• If we do not know the true population standard deviation (and

so estimate it from the sample standard deviation), we look at
the sample size:

• If we have a large sample size (n>30), we calculate a z-value

• If we have a small sample size (n<30), we calculate a t-value
 Revisiting the time spent
X

example (small sample)

S T R AT H C LY D E B U S I N E S S S C H O O L

• Suppose we have the same values, the same sample mean,

and the same sample standard deviation, but these are now
based on a sample size of 7 instead of 45.

6.02 − 5.42
𝑡= = 1.0948
1.45
7

Now, we will use the critical value approach to

deciding whether to accept or reject.
X

Critical value approach

S T R AT H C LY D E B U S I N E S S S C H O O L

• We are also working with a small sample size, and thus need to
assume that our sample mean follows a t-distribution.

• Here, we must work out the appropriate “critical value” that

sections off 0.05 probability in the upper tail of the appropriate
t-distribution.

That is, one with a degrees of

freedom of 𝒏 – 𝟏 = 𝟔
X

Critical value approach

S T R AT H C LY D E B U S I N E S S S C H O O L

• We are also working with a small sample size, and thus need to
assume that our sample mean follows a t-distribution.

• Here, we must work out the appropriate “critical value” that

sections off 0.05 probability in the upper tail of the appropriate
t-distribution.

• Once we have this critical value, if the value of our test statistic,
our t-value, exceeds it, we reject the null hypothesis.
X

Critical value approach

S T R AT H C LY D E B U S I N E S S S C H O O L

• Here, we must work out the appropriate “critical value” that

sections off 0.05 probability in the upper tail of the appropriate
t-distribution.
X

Critical value approach

S T R AT H C LY D E B U S I N E S S S C H O O L

From before, t-value = 1.0948

Critical value (the value that portions of an area of 0.05 in the upper
area of the t-distribution) = 1.943

Do we accept or reject the null hypothesis?

Given our observed t-value is less than our critical t-value, we do not
reject the null hypothesis in this case.

Why the change in decision?

X

Tests covered
S T R AT H C LY D E B U S I N E S S S C H O O L

• One-tailed normal-distribution based test

• One-tailed t-distribution based test

• Two-tailed normal distribution based test

• Two-tailed t-distribution based test

• Chi-squared test of independence

X

One-tailed vs Two-tailed tests

S T R AT H C LY D E B U S I N E S S S C H O O L

A quick recap of our first testing problem…

The professor suspected that the true mean reading time was above the national
average.

This led to form the following assumptions:

Null hypothesis (𝐻! ): averageThis

time is what
spent is termed
reading is 5.42ahours
one-tailed test,
per week.
That is, 𝜇 = 5.42 ℎ𝑜𝑢𝑟𝑠because of the structure of the
alternative hypothesis – the great than
Alternative hypothesis (𝐻" ): average time spent reading is greater than 5.42 hours.
element of the alternative hypothesis
That is, 𝜇 > 5.42 ℎ𝑜𝑢𝑟𝑠
 Example: time spend reading
X

(version 2)
S T R AT H C LY D E B U S I N E S S S C H O O L

A national student survey indicated that the average time spent

reading per week was 5.42 hours .

At a local university, a professor would like to investigate whether

this holds true for his students. He does not know whether he
will find his students read more or less than the national
average, but just wants to test this idea that students spend 5.42
hours per week reading.
 Example: time spend reading
X

(version 2)
S T R AT H C LY D E B U S I N E S S S C H O O L

He conducts a study of 45 students, and observes a mean

reading time per week of 6.02 hours, and a standard deviation of
1.45 hours.

Conduct a hypothesis test (at a significance level of 5%) to

examine how the reading time at the Professor’s university
compares to the national average.
X
Hypothesis testing framework
Example: time spend reading
S T R AT H C LY D E B U S I N E S S S C H O O L

(version 2)
1. Make an initial assumption about the data generating mechanism – a null hypothesis and
an alternative hypothesis

Null hypothesis (𝐻! ): average time spent reading is 5.42 hours per
week.
That is, 𝜇 = 5.42 ℎ𝑜𝑢𝑟𝑠

Alternative hypothesis (𝐻" ): average time spent reading is greater

than 5.42 hours.

That is, 𝜇 ≠ 5.42 ℎ𝑜𝑢𝑟𝑠

X
Hypothesis testing framework
Example: time spend reading
S T R AT H C LY D E B U S I N E S S S C H O O L

(version 2)
1. Make an initial assumption about the data generating mechanism – a null hypothesis and
an alternative hypothesis

Rereading:
At a local university, a professor would like to investigate whether
this holds true for his students. He does not know whether he
will find his students read more or less than the national
average, but just wants to test this idea that students spend 5.42
hours per week reading.
Nothing here suggests that he thinks the reading time at his institution will
be higher, and nothing suggests he thinks it will be lower à a one-tailed
test is inappropriate.
X
Hypothesis testing framework
Example: time spend reading
S T R AT H C LY D E B U S I N E S S S C H O O L

(version 2)
1. Make an initial assumption about the data generating mechanism – a null hypothesis and
an alternative hypothesis

Important notes:
• Hypotheses should be set up before we observe the data, rather than after
• If there is nothing in the situation to suggest a one-tailed test, we adopt a two-
tailed test as default

If we are interested in checking for an increase or decrease, but not both, from the
null, use a one-tailed test

If you are interested in any difference from the null value, then the test should be
two-tailed
X
Hypothesis testing framework
Example: time spend reading
S T R AT H C LY D E B U S I N E S S S C H O O L

(version 2)
1. Make an initial assumption about the data generating mechanism – a null hypothesis and an
alternative hypotheis
2. Collect a sample of data as evidence

Revisiting our sample evidence:

He conducts a survey of 45 students, and observes a mean
reading time of 6.02 hours per week, with standard deviation of
1.45 hours.
The sample size of 45 means that
the central limit theorem applies

𝜎
𝑿~𝑁 𝜇,
𝑛
X
Hypothesis testing framework
Example: time spend reading
S T R AT H C LY D E B U S I N E S S S C H O O L

(version 2)
1. Make an initial assumption about the data generating mechanism – a null hypothesis and an
alternative hypotheis
2. Collect a sample of data as evidence
3. Calculate the probability of observing such evidence, given the initial assumption

Proceed as did for one-tailed test.

Use the CLT to calculate the Z-value and use the standard normal
table to obtain the probability of observing a sample mean of 6.02
hours or higher.

6.02 − 5.42 From z-tables:

𝑧= = 2.776 𝑷(𝒁 > 𝟐. 𝟕𝟕𝟔) = 𝟎. 𝟎𝟎𝟐𝟕
1.45
45
X
Hypothesis testing framework
Example: time spend reading
S T R AT H C LY D E B U S I N E S S S C H O O L

(version 2)
1. Make an initial assumption about the data generating mechanism – a null hypothesis and an
alternative hypotheis
2. Collect a sample of data as evidence
3. Calculate the probability of observing such evidence, given the initial assumption

What’s different for the two-tailed test?

For a two-tailed test you must adjust the p-value calculation or

your critical value calculation accordingly.
X
Hypothesis testing framework
Example: time spend reading
S T R AT H C LY D E B U S I N E S S S C H O O L

(version 2)
1. Make an initial assumption about the data generating mechanism – a null hypothesis and an
alternative hypotheis
2. Collect a sample of data as evidence
3. Calculate the probability of observing such evidence, given the initial assumption

p-value is the probability of observing a result as extreme or

more extreme
Acceptance region

For the one-tailed p-value = 𝑃 𝑍 > 2.776

test, as prior = 0.0027
example
Rejection region
X
Hypothesis testing framework
Example: time spend reading
S T R AT H C LY D E B U S I N E S S S C H O O L

(version 2)
1. Make an initial assumption about the data generating mechanism – a null hypothesis and an
alternative hypotheis
2. Collect a sample of data as evidence
3. Calculate the probability of observing such evidence, given the initial assumption

p-value is the probability of observing a result as extreme or

Acceptance region
more extreme
For the one-tailed test, if our
alternative hypothesis is
smaller than (not our example)
Rejection region
For the one-tailed p-value = 𝑃 𝑍 > 2.776
test, as prior = 0.0027
example
X
Hypothesis testing framework
Example: time spend reading
S T R AT H C LY D E B U S I N E S S S C H O O L

(version 2)
1. Make an initial assumption about the data generating mechanism – a null hypothesis and an
alternative hypotheis
2. Collect a sample of data as evidence
3. Calculate the probability of observing such evidence, given the initial assumption

p-value is the probability of observing a result as extreme or

more extreme
For the one-tailed test, if our
alternative hypothesis is
smaller than (not our example)

For the one-tailed p-value = 𝑃 𝑍 > 2.776

test, as prior = 0.0027
example
p-value= 𝑃 𝑍 > 2.776 + Acceptance region
For the two-tailed test, 𝑃 𝑍 < −2.776
adjusting the p-value, = 2×𝑃 𝑍 > 2.776
Rejection
we need to double it = 2×0.0027 = 0.0054 region
X
Hypothesis testing framework
Example: time spend reading
S T R AT H C LY D E B U S I N E S S S C H O O L

(version 2)
1. Make an initial assumption about the data generating mechanism – a null hypothesis and
an alternative hypotheis
2. Collect a sample of data as evidence
3. Calculate the probability of observing such evidence, given the initial assumption
4. Depending on how likely or unlikely the evidence is, we reject or fail to reject the
initial assumption

Our p-value = 2×𝑃 𝑍 > 2.776 + 𝑃 𝑍 < −2.776

= 2×𝑃 𝑍 > 2.776
= 2×0.0027
= 0.0054
Our p-value is less than our significance level of 5%, thus we reject the null hypothesis
X

Tests covered
S T R AT H C LY D E B U S I N E S S S C H O O L

• One-tailed normal-distribution based test

• One-tailed t-distribution based test

• Two-tailed normal distribution based test

• Two-tailed t-distribution based test

• Chi-squared test of independence

X
Smaller samples
Example: time spend reading
S T R AT H C LY D E B U S I N E S S S C H O O L

(version 2; small sample)

Lets look at the problem but suppose we have a sample size of 7 instead of 45.

A national student survey indicated that the average time spent reading per week was 5.42
hours .

At a local university, a professor would like to investigate whether this holds true for his
students. He does not know whether he will find his students read more or less than the
national average, but just wants to test this idea that students spend 5.42 hours per week
reading.

He conducts a study of 7 students and observes a mean reading time per week of 6.02
hours, and a standard deviation of 1.45 hours.

Conduct a hypothesis test (at a significance level of 5%) to examine how the reading time
at the Professor’s university compares to the national average.
X

What to use when?

S T R AT H C LY D E B U S I N E S S S C H O O L

• If we know the true population standard deviation, we calculate

a z-value.

• If we do not know the true population standard deviation (and

so estimate it from the sample standard deviation), we look at
the sample size:

• If we have a large sample size (n>30), we calculate a z-value

• If we have a small sample size (n<30), we calculate a t-value
X
Hypothesis testing framework
Example: time spend reading
S T R AT H C LY D E B U S I N E S S S C H O O L

(version 2; small sample)

1. Make an initial assumption about the data generating mechanism – a null hypothesis and
an alternative hypothesis

Null hypothesis (𝐻! ): average time spent reading is 5.42 hours per
week.
That is, 𝜇 = 5.42 ℎ𝑜𝑢𝑟𝑠

Alternative hypothesis (𝐻" ): average time spent reading is greater

than 5.42 hours.

That is, 𝜇 ≠ 5.42 ℎ𝑜𝑢𝑟𝑠

X
Hypothesis testing framework
Example: time spend reading
S T R AT H C LY D E B U S I N E S S S C H O O L

(version 2; small sample)

1. Make an initial assumption about the data generating mechanism – a null hypothesis and an alternative
hypotheis
2. Collect a sample of data as evidence
3. Calculate the probability of observing such evidence, given the initial assumption

6.02
t= = 1.0948
1.45
7
Adjust the p-value calculation or critical value calculation accordingly.

Half 0.05 to give 0.025, and calculate the value of t that gives a region of
0.025 in the upper tail of the distribution, given a sample size of 7.

Degrees of freedom: 𝑣 = 7 − 1 = 6
X
Hypothesis testing framework
Example: time spend reading
S T R AT H C LY D E B U S I N E S S S C H O O L

(version 2; small sample)

1. Make an initial assumption about the data generating mechanism – a null hypothesis and an alternative hypotheis
2. Collect a sample of data as evidence
3. Calculate the probability of observing such evidence, given the initial assumption

6.02
t= = 1.0948
1.45
7

Adjust the p-value calculation or critical value calculation accordingly.

Adjusting critical value: Half 0.05 to give 0.025, and calculate the value of t that gives a
region of 0.025 in the upper tail of the distribution, given a sample size of 7.

Degrees of freedom: 𝑣 = 7 − 1 = 6
X
Hypothesis testing framework
Example: time spend reading
S T R AT H C LY D E B U S I N E S S S C H O O L

(version 2; small sample)

… Calcluate the value of t that gives a region of 0.025 in the upper tail of the t-distribution
given a sample size of 7.
X
Hypothesis testing framework
Example: time spend reading
S T R AT H C LY D E B U S I N E S S S C H O O L

(version 2; small sample)

1. Make an initial assumption about the data generating mechanism – a null hypothesis and an alternative hypotheis
2. Collect a sample of data as evidence
3. Calculate the probability of observing such evidence, given the initial assumption
4. Depending on how likely or unlikely the evidence is, we reject or fail to reject the initial assumption

Our t-value is 1.0948

Our critical value 2.447 (the value that portions of an area of 0.025 in the upper
area of the t-distribution)

Given our observed t-value is less than our critical value we do not reject the Null
hypothesis

Note: the area portioned off by 2.447 in the upper tail of the t-distribution is called
the critical region.
X

Tests covered
S T R AT H C LY D E B U S I N E S S S C H O O L

• One-tailed normal-distribution based test

• One-tailed t-distribution based test

• Two-tailed normal distribution based test

• Two-tailed t-distribution based test

• Chi-squared test of independence

X

Chi-squared test
S T R AT H C LY D E B U S I N E S S S C H O O L

An experiment was conducted to determine whether gender was

related to political party preference. The following data was
collected on which party people preferred. There are three
political parties, namely Labour, Lib-Dem and Conservative.
Labour Lib-Dem Conservative
Male 90 70 20
Female 40 80 100

Conduct a Pearson’s chi-squared test to test the independence

of gender and political party preference at the 5% level
X

Chi-squared test - steps

S T R AT H C LY D E B U S I N E S S S C H O O L

1. Calculate totals along margins of the contingency table

2. Calculate expected frequencies
3. Calculate the Chi-squared statistic
4. Calculate the number of degrees of freedom
5. Identify the critical value
6. Compare the test statistic to the critical value to make
accept/reject decision
X

Chi-squared test – but first

S T R AT H C LY D E B U S I N E S S S C H O O L

Null hypothesis (𝐻/ ): Political party preference and gender are

independent

Alternative hypothesis (𝐻0 ): Political party preference and gender

are not independent
X

Chi-squared test – step 1

S T R AT H C LY D E B U S I N E S S S C H O O L

Calculate totals along margins of the contingency table

Labour Lib-Dem Conservative Sum
Male 90 70 20 180
Female 40 80 100 220
Sum 130 150 120 400
X

Chi-squared test – step 2

S T R AT H C LY D E B U S I N E S S S C H O O L

Calculate expected frequencies

Labour Lib-Dem Conservative Sum
Male 90 70 20 180
Female 40 80 100 220
Sum 130 150 120 400

Male = 180/400 Female = 220/400

Expected number of female labour supporters:

220
130× = 71.5
400
X

Chi-squared test – step 2

S T R AT H C LY D E B U S I N E S S S C H O O L

Calculate expected frequencies

Labour Lib-Dem Conservative Sum
Male 90 70 20 180
Female 40 80 100 220
Sum 130 150 120 400

Or getting the same results:

Labour = 130/400 Lib-dem = 150/400 Con = 120/400

Expected number of female labour supporters:

130
220× = 71.5
400
X

Chi-squared test – step 2

S T R AT H C LY D E B U S I N E S S S C H O O L

Calculate expected frequencies

Labour Lib-Dem Conservative
Male 58.5 (90) 67.5 (70) 54 (20)
Female 71.5 (40) 82.5 (80) 66 (100)
X

Chi-squared test – step 3

S T R AT H C LY D E B U S I N E S S S C H O O L

Calculate the Chi-squared statistic

Labour Lib-Dem Conservative
Male 58.5 (90) 67.5 (70) 54 (20)
Female 71.5 (40) 82.5 (80) 66 (100)

1 1 1 1
1
𝑜2 − 𝑒2 90 − 58.5 70 − 67.5 100 − 66
𝜒 =6 = + + ⋯+
𝑒2 58.5 67.5 66
2

= 69.63
X

Chi-squared test – steps 4

S T R AT H C LY D E B U S I N E S S S C H O O L

Calculate the number of degrees of freedom

Labour Lib-Dem Conservative
Male 58.5 (90) 67.5 (70) 54 (20)
Female 71.5 (40) 82.5 (80) 66 (100)

Degrees of freedom
= 𝑟−1 𝑐−1
= 2−1 3−1
= 1×2
=2
X

Chi-squared test – step 5

S T R AT H C LY D E B U S I N E S S S C H O O L

Accept area

Identify the critical value

Degrees of freedom = 2, and upper tail area = 0.05 (5% Reject area
significance level)
X

Chi-squared test – step 6

S T R AT H C LY D E B U S I N E S S S C H O O L

Accept area

Compare the test statistic to the critical value to make

accept/reject decision
Reject area
Test statistic = 69.93, which is greater than our critical value of
5.99

As such, we reject the null hypothesis and conclude that there is

an association between party political preference and gender.
X

Type I and Type II errors

S T R AT H C LY D E B U S I N E S S S C H O O L

• Type I – mistakenly reject a null hypothesis 𝐻/

• Also called a false positive.
• E.g., mistakenly convicting an innocent defendant after a trial

• Type II – mistakenly fail to reject null hypothesis 𝐻/

• Also called false negative
• E.g., mistakenly acquitting a guilty defendant after trial
 Questions?
X S T R AT H C LY D E B U S I N E S S S C H O O L
 What to cover in week
6 review session?
X S T R AT H C LY D E B U S I N E S S S C H O O L
X T
SH E APTLHACCLEY DOEF BUUSSEIFNUELS S
TR L ESACRHNOI N
OGL
X

Objectives
S T R AT H C LY D E B U S I N E S S S C H O O L

•H0= Average time spent reading is 5.42 hours per week µ = 5.42
hours H1= Average time spent reading is greater than 5.42 hours
per week µ > 5.42 hours