Data Analysis for

Business
Week 3 Session 6

Hypothesis Tests – Classical t-tests

 In this section, we will examine:

• The use of hypothesis testing for a mean or

proportion.
• Classical t-tests: Independent sample test, and
paired sample test.
• Using p-value for the hypothesis test.
• Procedures in hypothesis testing.
• Type I and II Error in a hypothesis test.

BEO6000 - Session 6 2
 Types of Hypothesis Test for the Mean

Methods Null hypothesis (Ho) Alternative hypothesis (H1)

One-sample test µ0 = test value µ0 ¹ test value (two-sided test)

µ0 ≤ test value (one-sided test) or
or µ0 > test value (one-sided test)
µ0 ≥ test value (one-sided test) or
µ0 < test value (one-sided test)
Independent µ1 = µ2 µ1 ¹ µ2 (two-sided test) or
sample test µ1 ≤ µ2 (one-sided test) or µ1 > µ2 (one-sided test) or
(2 sub- µ1 ≥ µ2 (one-sided test) µ1 < µ2 (one-sided test)
populations)
Paired sample µD = 0 µD ¹ 0 (two-sided test) or
tests µD ≤ 0 (one-sided test) or µD > 0 (one-sided test) or
(often before and µD ≥ 0 (one-sided test) µD < 0 (one-sided test)
after test)

BEO6000 - Session 6 3
Types of Hypothesis Test for the Proportion

Methods Null hypothesis (Ho) Alternative hypothesis (H1)

One-sample test r0 = test value r0 ¹ test value (two-sided test)

r0 ≤ test value (one-sided test) or
or r0 > test value (one-sided test)
r0 ≥ test value (one-sided test) or
r0 < test value (one-sided test)
Independent r1 = r2 r1 ¹ r2 (two-sided test) or
sample test r1 ≤ r2 (one-sided test) or r1 > r2 (one-sided test) or
(2 sub- r1 ≥ r2 (one-sided test) r1 < r2 (one-sided test)
populations)
Paired sample rD = 0 rD ¹ 0 (two-sided test) or
tests rD ≤ 0 (one-sided test) or rD > 0 (one-sided test) or
(often before and rD ≥ 0 (one-sided test) rD < 0 (one-sided test)
after test)

BEO6000 - Session 6 4
To perform a hypothesis test in BEO6000, we use:

The modern approach

† p-value (Sig)

Prob[HO is true] = p.
From the data in an SPSS output, you could perform the hypothesis
test very easily by referring to the SPSS’s “Sig.” for the p-value.

BEO6000 - Session 6 5
 Testing Hypotheses Using SPSS

Step 1: Formulate the null hypothesis (H0) and alternative hypothesis (H1 or HA).
Step 2: Key data into SPSS spreadsheet or import data from Excel spreadsheet.
Step 3: Ensure all data are in proper format. For example, group codes must be
created if the hypothesis test involves means of 2 independent populations or 2
related populations.
Step 4: Enter the confidence interval percentage (example, 95% for significance value
of 5% or 99% for significance value of 1%) for your hypothesis test in SPSS.
Step 5: Make a statistical decision by evaluating the SPSS output. Decide (with
explanation) whether to reject (or not) the null hypothesis determined by
whether the p-value indicates, when compared to the significance value (α),
whether or not the p-value > α or p-value < α .
Step 6: State the conclusion referring to the sample evidence supporting (or not) the
null/alternative hypothesis (the test scenario in the question) with specific
reference to the level of significance used to arrive at the conclusion.
Step 7: A Post-hoc analysis is required if the null hypothesis was not supported by
sample evidence. Interpret the Confidence Interval of Difference in the SPSS
output.

BEO6000 - Session 6 6
 Using computers
Statistics in professional life:
Absolutely no one would perform a task like this manually using formulas and
tables in “real life”.
The only justification for solving a problem manually is so that you might
understand what SPSS (or any other statistical software) is doing.

BEO6000 - Session 6 7
 One-tail or two-tail test
Null Hypothesis Alternative Level of
(Ho) Hypothesis (H1) significance Regions of rejection

µ = a µ ¹ a 𝛼 𝛼
2 2
or or
P = a P ¹ a
𝛼
! x or r
a

left
µ≥ a µ < a
tail
or or rejection 𝛼 𝛼
P≥ a P < a region x or r
!
a

µ≤ a µ > a right
or or tail
P≤ a P >
rejection
a region 𝛼 𝛼
! x or r
a

Note:
BEO6000 “direction”
- Session 6 of “arrow” gives tail in which rejection region occurs. 8
 Hypothesis testing: One-tail test or two-tail test?
• When you conduct a test of statistical significance in SPSS,
the p-value is for a two-tailed test.
• The easiest way to convert a two-tailed test into a one-
tailed test is to divide in half the p-value provided in the
output.

• If the question requires you to conduct the test at 0.05

significance level (𝛼):

Do not reject H0 Reject H0

(or Accept H1)

Two-tailed test p-value > 0.05 p-value < 0.05

One-tailed test (p-value/2) > 0.05 (p-value/2) < 0.05

BEO6000 - Session 6 9
 Independent sample test (2 sub-populations)

The independent-samples t-test (or independent t-test, for

short) compares the means between two unrelated groups
on the same continuous, dependent variable.
For example, you could use an independent t-test to
understand whether there is a significant difference between
male and female first year graduate salaries.

BEO6000 - Session 6 10
Independent sample test (2 independent
populations)

Formula for tests concerning Post test (when Ho is rejected)

means Interval estimates.
Equal variance: Equal variance:
( 1 ) 2 ( 1) 2
2 n1 S1 n 2 S 2
Sp 2 1 1
n1 n 2 2 X X t S p( )
1 2
n
1 n 2

( X1 X2 ) ( 1 2
)
t
2 1 1
Sp ( )
n1 n 2
Unequal variance: Unequal variance:
( X1 X2 ) ( 1 2)
t S12 S2
2
S12 S2
2 X1 X2 t
n1 n2
n1 n2

Formula for6 degree of freedom à df = n1 + n2 - 2

BEO6000 - Session 11
 Important to note:

For Independent Sample Tests, you have to conduct two steps:

Firstly, Preliminary F test – Test for Equality of Variances

Secondly, test for Equality of the Means (t-test)

Let’s do the Preliminary F test – Test for Equality of Variances first!

BEO6000 - Session 6 12
 Independent sample test (2 independent populations)
Example 2: We are to test the null hypothesis at 5% level of significance that that the average
weekly expenditure on lunches of males and females in a target population are equal. We use a
random sample of 20 males and 20 females.

Note: The following slides, the SPSS output has been edited.

Levene’s Test
For Equality of t-test for Equality of Means
Variances

95% Confidence
Std. Interval of the
Sig.
Mean Error Difference
(2-tailed)
F Sig. t df Difference Difference
Lower Upper

Equal variances 3.522 0.068 -3.329 38 0.002 -4.55 1.05 -7.32 -1.78
Weekly
assumed
lunch
expenditure
Equal variances
($)
not assumed
-3.329 37.4 0.002 -4.55 1.05 -7.33 -1.77

BEO6000 - Session 6 13
 Independent sample test (2 independent populations)
Example 2: We are to test the null hypothesis at 5% level of significance that that the average
weekly expenditure on lunches of males and females in a target population are equal. We use a
random sample of 20 males and 20 females.

Note: The following slides, the SPSS output has been edited.

Levene’s Test
For Equality of t-test for Equality of Means
Variances

95% Confidence
Std. Interval of the
Sig.
Mean Error Difference
(2-tailed)
F Sig. t df Difference Difference
Lower Upper

Equal variances 3.522 0.068 -3.329 38 0.002 -4.55 1.05 -7.32 -1.78
Weekly
assumed
lunch
expenditure
Equal variances
($)
not assumed
-3.329 37.4 0.002 -4.55 1.05 -7.33 -1.77

BEO6000 - Session 6 14
First test à Preliminary F test – Test for Equality of Variances
H0: Equal variances assumed
H1: Equal variances not assumed

Levene’s Test
For Equality of t-test for Equality of Means
Variances

95% Confidence
Std. Interval of the
Sig.
Mean Error Difference
(2-tailed)
F Sig. t df Difference Difference
Lower Upper

Equal variances 3.522 0.068 -3.329 38 0.002 -4.55 1.05 -7.32 -1.78
Weekly
assumed
lunch
expenditure
Equal variances
($)
not assumed
-3.329 37.4 0.002 -4.55 1.05 -7.33 -1.77

Statistical decision: We do not reject the null hypothesis that the variances are equal because p-value (0.068) >
0.05.
Conclusion: At 5% level of significance, sample evidence supports the null hypothesis that equal variances
assumed.

BEO6000 - Session 6 15
 Independent sample test (2 independent populations)
H0: Average weekly expenditure on lunches are equal for males and females.
H1: Average weekly expenditure on lunches are not equal for males and females.

Levene’s Test
For Equality of t-test for Equality of Means
Variances

95% Confidence
Std. Interval of the
Sig.
Mean Error Difference
(2-tailed)
F Sig. t df Difference Difference
Lower Upper

Equal variances 3.522 0.068 -3.329 38 0.002 -4.55 1.05 -7.32 -1.78
Weekly
assumed
lunch
expenditure
Equal variances
($)
not assumed
-3.329 37.4 0.002 -4.55 1.05 -7.33 -1.77

Chance that the null hypothesis is true.

BEO6000 - Session 6 16
 Independent sample test (2 independent populations)
H0: Average weekly expenditure on lunches are equal for males and females.
H1: Average weekly expenditure on lunches are not equal for males and females.

Levene’s Test
For Equality of t-test for Equality of Means
Variances

95% Confidence
Std. Interval of the
Sig.
Mean Error Difference
(2-tailed)
F Sig. t df Difference Difference
Lower Upper

Equal variances 3.522 0.068 -3.329 38 0.002 -4.55 1.05 -7.32 -1.78
Weekly
assumed
lunch
expenditure
Equal variances
($)
not assumed
-3.329 37.4 0.002 -4.55 1.05 -7.33 -1.77

Statistical decision: We reject the null hypothesis because the p-value (0.002) < 0.05

BEO6000 - Session 6 17
 Independent sample test (2 independent populations)
H0: Average weekly expenditure on lunches are equal for males and females.
H1: Average weekly expenditure on lunches are not equal for males and females.

Levene’s Test
For Equality of t-test for Equality of Means
Variances

95% Confidence
Std. Interval of the
Sig.
Mean Error Difference
(2-tailed)
F Sig. t df Difference Difference
Lower Upper

Equal variances 3.522 0.068 -3.329 38 0.002 -4.55 1.05 -7.32 -1.78
Weekly
assumed
lunch
expenditure
Equal variances
($)
not assumed
-3.329 37.4 0.002 -4.55 1.05 -7.33 -1.77

Conclusion: At 5% level of significance, the sample evidence does not support the null
hypothesis that the average weekly expenditure on lunches are equal for males and
female.
BEO6000 - Session 6 18
 Independent sample test (2 independent populations)

Size of the difference

Post hoc.
We have just concluded that the average weekly expenditure of
males and females is different.
How different?
Confidence interval.

BEO6000 - Session 6 19
Independent sample test (2 independent
populations)
95% confidence
Independent Samples Test interval.

Levene’s Test
For Equality of t-test for Equality of Means
Variances

95% Confidence
Interval of the
Sig. Difference
(2-tailed) Mean
F Sig. t df Difference
Lower Upper

Equal variances 3.522 0.068 -3.329 38 0.002 -4.55 -7.32 -1.78

Weekly
assumed
lunch
expenditure
Equal variances
($)
not assumed -3.329 37.4 0.002 -4.55 -7.33 -1.77

BEO6000 - Session 6 20
What is a confidence interval?

A confidence interval, in statistics, refers to the probability that a population parameter (population
mean or proportion) will fall between a set of values for a variable. It allows you to estimate the
population means and proportions based on sample evidence.
They are most often constructed using confidence levels of 95% or 99%.

Reason for estimating population means and proportions:

The reason why estimating population means and proportions is of such importance is because
more often than not we do not have access to population data and so these parameters cannot
usually be obtained directly from the data.

Confidence interval At 95% confidence level, population mean or proportion is estimated to

constructed with 95% lie within this confidence interval.
confidence:

Lower limit Upper limit

Confidence interval At 99% confidence level, population mean or proportion is estimated to

constructed with 99% lie within this confidence interval.
confidence:

Lower limit Upper limit

BEO6000 - Session 6 21
Independent sample test (2 independent populations)
95% confidence
interval.
Independent Samples Test

Levene’s Test
For Equality of t-test for Equality of Means
Variances

95% Confidence
Interval of the
Sig. Difference
(2-tailed) Mean
F Sig. t df Difference
Lower Upper

Equal variances 3.522 0.068 -3.329 38 0.002 -4.55 -7.32 -1.78

Weekly
assumed
lunch
expenditure
Equal variances
($)
not assumed -3.329 37.4 0.002 -4.55 -7.33 -1.77

At 95% level of confidence, the difference in the average weekly lunch expenditure
of males and females in the target population is between $1.78 and $7.32.

BEO6000 - Session 6 22
 Independent sample test (2 independent populations)

Firstly, Preliminary F test – Test for Equality of Variances

Requirements for each hypothesis test:

1. Formulate the null and alternative hypotheses.
H0: Equal variances assumed
H1: Equal variances not assumed
2. State your statistical decision using significant value (𝛼) of 5% for each test.
Do not reject the null hypothesis that the variances are equal because p-value (0.068) > 0.05.

3. State your conclusion in context.

At significance level of 5%, the sample evidence supports the null hypothesis that equal
variances assumed.

BEO6000 - Session 6 23
Independent sample test (2 independent populations)
Secondly, test for Equality of the Means (t-test)
Requirements for each hypothesis test:
i) Formulate the null and alternative hypotheses.
H0: Average weekly expenditure on lunches are equal for males and
females.
H1: Average weekly expenditure on lunches are not equal for males and
females.

ii) State your statistical decision using significant value (𝛼) of 5% for each
test.
Reject the null hypothesis because p-value (0.002) < 0.05.

iii) State your conclusion in context.

At significance level of 5%, the sample evidence does not supports
the null hypothesis that average weekly expenditure on lunches are
equal for males and females. The samples lead us to conclude that in the
target population, the mean weekly expenditures of males and females
are different.
BEO6000 - Session 6 24
 Independent sample test (2 independent populations)

Post hoc

Confidence interval:
At 95% level of confidence, the difference in the average
weekly lunch expenditure of males and females in the target
population is between $1.78 and $7.32.

Note: The interval is fairly wide but this is a consequence of

only having a small sample.

BEO6000 - Session 6 25
 Note
If we want to perform a 1-tail test of the alternative hypothesis that
females spend more on average than male, divide the p-value by 2.

p-value
Ho: µfemale ≤ µ male
2
H1: µfemale > µ male

We would still reject the null hypothesis in this exercise.

If there is a possibility of performing a 1-sided or a 2-sided
test, SPSS always do a 2-tail test.

BEO6000 - Session 6 26
 Proportions using SPSS

The trick:
Code the “yes” answer as 1 and the “no” answer as 0.
The mean of the variable is the proportion who responded “yes”.
Proceed using SPSS to test the means in the usual way.

BEO6000 - Session 6 27
 Paired sample tests (2 related populations)
Formula for tests concerning mean of Post test (when Ho is rejected)
difference (µD) Interval estimates.

X
t SD
D D
t
S XD
D n
n
Rationale:
If HO is true, i.e. µD = 0.

XD » µD Þ XD - µD » 0.

XD - µD » 0 Þ t = small value.

t = small value Þ do not reject HO.

Formula for degree of freedom à df = n- 1

BEO6000 - Session 6 28
 Paired sample tests (2 related populations)
Example 3:
We test whether average expenditure on a product increases after an
advertising campaign amongst a target population.

A random sample of 40 households is selected from the target population.

Weekly expenditure on the product is recorded before and after the advertising
campaign for each household in the sample.

One-tail or two-tail test? H0: µbefore - µafter ≥ 0

H1: µbefore - µafter < 0

Ho: There is no difference in the weekly expenditure on a product before and after
the advertising campaign or there is a reduction in the weekly expenditure on a
product after the advertising campaign.
H1: There is an increase in the weekly expenditure on a product after the advertising
campaign.

BEO6000 - Session 6 29
 Paired sample tests (2 related populations)

Paired Samples Test

Paired Differences

95% Confidence
Interval of the
Difference

Std. Std. Error

Sig.
Mean Deviation Mean Lower Upper t df (2-tailed)

Pair Before –
1 After -3.53 1.20 0.78 -5.10 -1.95 -4.53 39 .000

Mean of paired differences (before - after).

BEO6000 - Session 6 30
 Paired sample tests (2 related populations)
Ho: There is no difference in the weekly expenditure on a product before and after
the advertising campaign or there is a reduction in the weekly expenditure on a
product after the advertising campaign.
H1: There is an increase in the weekly expenditure on a product after the advertising
campaign.
Paired Samples Test

Paired Differences

95% Confidence
Interval of the
Difference

Std. Std. Error

Sig.
Mean Deviation Mean Lower Upper t df (2-tailed)

Pair Before –
1 After -3.53 1.20 0.78 -5.10 -1.95 -4.53 39 .000

Check to test whether the null

hypothesis is true.

BEO6000 - Session 6 31
 Paired sample tests (2 related populations)
Ho: There is no difference in the weekly expenditure on a product before and after
the advertising campaign or there is a reduction in the weekly expenditure on a
product after the advertising campaign.
H1: There is an increase in the weekly expenditure on a product after the advertising
campaign.
Paired Samples Test

Paired Differences

95% Confidence
Interval of the
Difference

Std. Std. Error

Sig.
Mean Deviation Mean Lower Upper t df (2-tailed)

Pair Before –
1 After -3.53 1.20 0.78 -5.10 -1.95 -4.53 39 .000

Probability that the null

hypothesis is true.

BEO6000 - Session 6 32
 Paired sample tests (2 related populations)
Ho: There is no difference in the weekly expenditure on a product before and after the
advertising campaign or there is a reduction in the weekly expenditure on a product after the
advertising campaign.
H1: There is an increase in the weekly expenditure on a product after the advertising campaign.

Paired Samples Test

Paired Differences

95% Confidence
Interval of the
Difference

Std. Std. Error

Sig.
Mean Deviation Mean Lower Upper t df (2-tailed)

Pair Before –
1 After -3.53 1.20 0.78 --5.10 -1.95 -4.53 39 .000

Statistical decision: Reject the null hypothesis that the average difference (before - after) is 0
because p-value (0.000/2 =0.000) <0.05.
Conclusion: At 5% level of significance, the sample evidence does not support the null hypothesis
that the average difference.
BEO6000 - Session 6 33
Paired sample tests (2 related populations)
One-sided test

Recall:
This is a one-side test so we have to divide p-value by 2.
Do not reject H0 Reject H0
(or Accept H1)

Two-tail test p-value > 0.05 p-value < 0.05

One-tail test (p-value/2) > 0.05 (p-value/2) < 0.05

p-value/2 =0.000 Þ reject the null hypothesis.

Conclusion: At significance level of 5%, the sample evidence does not supports the null
hypothesis that there is no difference in the weekly expenditure on a product before and after the
advertising campaign or there is a reduction in the weekly expenditure on a product after the
advertising campaign. The sample leads us to conclude that in the target population, the mean
expenditure is greater after the advertising campaign.

BEO6000 - Session 6 34
 Paired sample tests (2 related populations)
Size of the difference
Post hoc.
We have just concluded that the average expenditure is greater after the
advertising campaign.
How different?
Confidence interval.

BEO6000 - Session 6 35
Paired sample tests (2 related populations)
Paired Samples Test

Paired Differences

95% Confidence
Interval of the
Difference

Std. Std. Error

Sig.
Mean Deviation Mean Lower Upper t df (2-tailed)

Pair Before –
1 After -3.53 1.20 0.78 -5.10 -1.95 -4.53 39 .000

95% confidence interval.

At 95% level of confidence, the difference in average weekly expenditure on the

product increased by between $1.95 and $5.10 after the advertising campaign.

BEO6000 - Session 6 36
Paired sample tests (2 related populations)

Requirements for each hypothesis test:

i) Formulate the null and alternative hypotheses.

Ho: There is no difference in the weekly expenditure on a product before and after the
advertising campaign or there is a reduction in the weekly expenditure on a product after the
advertising campaign.
H1: There is an increase in the weekly expenditure on a product after the advertising
campaign.

ii) State your statistical decision using significant value (𝛼) of 5% for each test.
For this one-tail test à (p-value/2) = (0.000/2)
=0
Therefore, reject H0 because 0 is less than significant value of 0.05.

BEO6000 - Session 6 37
 Paired sample tests (2 related populations)

Requirements for each hypothesis test (cont’d):

iii) State your conclusion in context.

At significance level of 5%, the sample evidence does not supports the null
hypothesis that there is no difference in the weekly expenditure on a product
before and after the advertising campaign or there is a reduction in the weekly
expenditure on a product after the advertising campaign. The sample leads us to
conclude that in the target population, the mean expenditure is greater after the
advertising campaign.

BEO6000 - Session 6 38
 Paired sample tests (2 related populations)
Post hoc
Confidence interval:
At 95% level of confidence, the difference in average weekly
expenditure on the product increased by between $1.95 and
$5.10 after the advertising campaign.

BEO6000 - Session 6 39
 Paired sample tests (2 related populations)

Paired tests
Confused about which to subtract from which?
I.e. first - second or second - first?
And which was first anyway?
Answer:
It doesn’t really matter.
If one way gives positive values, the other way will give negative values of the
same absolute size?

BEO6000 - Session 6 40
 Confidence intervals

If you perform a t test using SPSS:

You automatically get confidence intervals too.
Independent and paired samples.
Means and proportions.

BEO6000 - Session 6 41
 Type I and Type II Errors
H0 is either true or false, and at the end of the test procedure we either reject it or do not reject
it. However, our decision which is based on sample evidence might not be correct.

Altogether there are four possibilities:

H0 is true H0 is false

Fail to Reject H0 Correct decision Incorrect decision

Reject H0 Incorrect decision Correct decision

Type I error: Type II error:

rejecting a true H0. non-rejecting a false H0.

P(Type I error) = α P(Type II error) = β

The probability of the type I error, α, is the significance level.

In Example 1, the type I error is 0.05

This is the risk we are prepared to take of making a Type 1 error.

BEO6000 - Session 6 42
 All hypothesis tests are the same!
Remember:
Hypotheses and the test statistics may differ.
The test statistic is calculated assuming the null is true.
We always reject the null if the test statistic seems larger than could reasonably be explaine
by chance.
If the chance that the null is true is, say, 0.027, it probably isn’t true.
We might be wrong (the risk of Type I Error).
Remember also that if the sample is not random, all of this is a meaningless exercise.

The independent sample test, and pair-sample test used here are known as classical
tests or parametric tests. They require the data be normally distributed. In other words,
data must not be skewed. If data is skewed, apply non-parametric test (next week
session).

BEO6000 - Session 6 43
 Sig or p-value
In all tests:
sig = Pr[HO is true] given the sample results.
In all tests, reject the null hypothesis if the sig value (p-
value) is less than 0.05 (or whatever you have chosen as
the level of significance).

BEO6000 - Session 6 44
 Non-parametric tests
Tests of hypotheses concerning the distributions of a
population (non-normal distribution).

Topic in next week lecture.

BEO6000 - Session 6 45
 References:
Cronk, B. C. (2020). How to Use SPSS: A Step-By-Step Guide to Analysis and Interpretation, 11th
edition. New York: Routledge. Retrieved from
http://ebookcentral.proquest.com/lib/vu/detail.action?docID=5904712.
Black, K., Asafu-Adjaye, J., Burke, P., Khan, N., King, G., Perera, N., Papadimos, A., Sherwood, C. and
Wasimi, S. (2019). Business Analytics and Statistics. Australia: Wiley.
Hair, J. F., Page, M. and Brunsveld, N. (2019). Essentials of Business Research Methods, 4th edition.
New York: Routledge.

BEO6000 - Session 6 46