Before you begin

• These slides are used in presentations at workshops.

• They are best viewed with a pdf reader like Acrobat Reader (free download).

• Make sure that “Single Page View” or “Fit to Window” is selected.

• Navigation buttons are provided at the bottom of each screen if needed (see below).

• Viewing in web browsers is not recommended.

Do not try to print the slides

There are many more pages than the number of slides listed at the bottom right of each
screen!

Apologies for any inconvenience.

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 1 / 33
 Inferential Statistics (testing hypotheses)
(mα+hs)Smart Workshop
Semester 2, 2016

Geoff Coates

(This workshop is a follow-up to the earlier “Introduction to Statistical Inference”

session.) These slides go through the steps used to conduct the one sample t−test and
demonstrates how to extract the necessary information from a description of an
experiment.

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 2 / 33
What can (mα+hs)Smart do for you?

Online Stuff Drop-in Study Sessions

presentation slides from Monday, Wednesday, Friday,
workshops on many topics 10am-12pm, Ground Floor
practice exercises Barry J Marshall Library,
teaching weeks and study
short videos
breaks.
and more!

Email: [email protected]
Workshops
Can’t find what you want?
See our current
Got a question?
Workshop Calendar for this
Semester’s topics. Drop us a line!

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 3 / 33
Contents

What is inference? Populations and samples Go

General structure of hypothesis testing Go

A typical test/exam-style question Go

Go Setting up the hypotheses

Go Data and test statistic
Go p−value
Go Decision
Go Using a t−table to find the p−value [some stats units only]
Go Checking the assumptions for the t−test [some stats units only]
Another typical test/exam-style question
Go Setting up the hypotheses
Go Data and test statistic
Go p−value
Go Decision
Go Using a t−table to find the p−value [some stats units only]
Go Checking the assumptions for the t−test [some stats units only]
Appendix: A tip for understanding t−tests Go

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 4 / 33
What is inference? Populations and samples
Statistical inference is a technique for inferring something about an entire population.

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 5 / 33
What is inference? Populations and samples
Statistical inference is a technique for inferring something about an entire population.

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 5 / 33
What is inference? Populations and samples
Statistical inference is a technique for inferring something about an entire population.
The “something” is a numerical characteristic called a population parameter.

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 5 / 33
What is inference? Populations and samples
Statistical inference is a technique for inferring something about an entire population.
The “something” is a numerical characteristic called a population parameter.

population parameters:

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 5 / 33
What is inference? Populations and samples
Statistical inference is a technique for inferring something about an entire population.
The “something” is a numerical characteristic called a population parameter.

population parameters:
population mean: µ

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 5 / 33
What is inference? Populations and samples
Statistical inference is a technique for inferring something about an entire population.
The “something” is a numerical characteristic called a population parameter.

population parameters:
population mean: µ
population standard deviation: σ

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 5 / 33
What is inference? Populations and samples
Statistical inference is a technique for inferring something about an entire population.
The “something” is a numerical characteristic called a population parameter.

population parameters:
population mean: µ
population standard deviation: σ
population proportion: p

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 5 / 33
What is inference? Populations and samples
Statistical inference is a technique for inferring something about an entire population.
The “something” is a numerical characteristic called a population parameter. We do this
by collecting a random sample of population members

population parameters:
population mean: µ
population standard deviation: σ
population proportion: p

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 5 / 33
What is inference? Populations and samples
Statistical inference is a technique for inferring something about an entire population.
The “something” is a numerical characteristic called a population parameter. We do this
by collecting a random sample of population members

population parameters:
population mean: µ
population standard deviation: σ
population proportion: p

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 5 / 33
What is inference? Populations and samples
Statistical inference is a technique for inferring something about an entire population.
The “something” is a numerical characteristic called a population parameter. We do this
by collecting a random sample of population members and examining an estimate of the
parameter called a sample statistic.

population parameters:
population mean: µ
population standard deviation: σ
population proportion: p

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 5 / 33
What is inference? Populations and samples
Statistical inference is a technique for inferring something about an entire population.
The “something” is a numerical characteristic called a population parameter. We do this
by collecting a random sample of population members and examining an estimate of the
parameter called a sample statistic.

sample statistics:

population parameters:
population mean: µ
population standard deviation: σ
population proportion: p

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 5 / 33
What is inference? Populations and samples
Statistical inference is a technique for inferring something about an entire population.
The “something” is a numerical characteristic called a population parameter. We do this
by collecting a random sample of population members and examining an estimate of the
parameter called a sample statistic.

sample statistics:
sample mean: x

population parameters:
population mean: µ
population standard deviation: σ
population proportion: p

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 5 / 33
What is inference? Populations and samples
Statistical inference is a technique for inferring something about an entire population.
The “something” is a numerical characteristic called a population parameter. We do this
by collecting a random sample of population members and examining an estimate of the
parameter called a sample statistic.

sample statistics:
sample mean: x
sample standard deviation: s

population parameters:
population mean: µ
population standard deviation: σ
population proportion: p

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 5 / 33
What is inference? Populations and samples
Statistical inference is a technique for inferring something about an entire population.
The “something” is a numerical characteristic called a population parameter. We do this
by collecting a random sample of population members and examining an estimate of the
parameter called a sample statistic.

sample statistics:
sample mean: x
sample standard deviation: s
sample proportion: p̂
population parameters:
population mean: µ
population standard deviation: σ
population proportion: p

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 5 / 33
What is inference? Populations and samples
Statistical inference is a technique for inferring something about an entire population.
The “something” is a numerical characteristic called a population parameter. We do this
by collecting a random sample of population members and examining an estimate of the
parameter called a sample statistic.

sample statistics:
sample mean: x
sample standard deviation: s
sample proportion: p̂
population parameters:
population mean: µ
population standard deviation: σ
population proportion: p

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 5 / 33
What is inference? Populations and samples
Statistical inference is a technique for inferring something about an entire population.
The “something” is a numerical characteristic called a population parameter. We do this
by collecting a random sample of population members and examining an estimate of the
parameter called a sample statistic.

sample statistics:
sample mean: x
sample standard deviation: s
sample proportion: p̂
population parameters:
population mean: µ
population standard deviation: σ
population proportion: p

Note: separating population parameters and sample statistics helps organise all the
notation used in statistics.

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 5 / 33
General structure of hypothesis testing

All hypothesis testing procedures follow the same general structure:

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 6 / 33
General structure of hypothesis testing

All hypothesis testing procedures follow the same general structure:

Null Hyp. (H0 ): A specific statement about a population parameter (or param-
eters).

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 6 / 33
General structure of hypothesis testing

All hypothesis testing procedures follow the same general structure:

Null Hyp. (H0 ): A specific statement about a population parameter (or param-
eters). We would like to prove this wrong if possible.

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 6 / 33
General structure of hypothesis testing

All hypothesis testing procedures follow the same general structure:

Null Hyp. (H0 ): A specific statement about a population parameter (or param-
eters). We would like to prove this wrong if possible.
Alt. Hyp. (H1 ): A general statement about a population parameter (or param-
eters) opposing H0 .

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 6 / 33
General structure of hypothesis testing

All hypothesis testing procedures follow the same general structure:

Null Hyp. (H0 ): A specific statement about a population parameter (or param-
eters). We would like to prove this wrong if possible.
Alt. Hyp. (H1 ): A general statement about a population parameter (or param-
eters) opposing H0 .
Data: Random sample(s) from the population(s).

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 6 / 33
General structure of hypothesis testing

All hypothesis testing procedures follow the same general structure:

Null Hyp. (H0 ): A specific statement about a population parameter (or param-
eters). We would like to prove this wrong if possible.
Alt. Hyp. (H1 ): A general statement about a population parameter (or param-
eters) opposing H0 .
Data: Random sample(s) from the population(s).
Test Statistic: Suitable estimate of the population parameter (or combination
of parameters) derived from these data.

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 6 / 33
General structure of hypothesis testing

All hypothesis testing procedures follow the same general structure:

Null Hyp. (H0 ): A specific statement about a population parameter (or param-
eters). We would like to prove this wrong if possible.
Alt. Hyp. (H1 ): A general statement about a population parameter (or param-
eters) opposing H0 .
Data: Random sample(s) from the population(s).
Test Statistic: Suitable estimate of the population parameter (or combination
of parameters) derived from these data.
Sampling Distn : Describes the probability structure for the test statistic when
H0 is true.

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 6 / 33
General structure of hypothesis testing

All hypothesis testing procedures follow the same general structure:

Null Hyp. (H0 ): A specific statement about a population parameter (or param-
eters). We would like to prove this wrong if possible.
Alt. Hyp. (H1 ): A general statement about a population parameter (or param-
eters) opposing H0 .
Data: Random sample(s) from the population(s).
Test Statistic: Suitable estimate of the population parameter (or combination
of parameters) derived from these data.
Sampling Distn : Describes the probability structure for the test statistic when
H0 is true.
p-value: probability of observed test statistic value or one more
favourable to H1 from the sampling distribution.

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 6 / 33
General structure of hypothesis testing

All hypothesis testing procedures follow the same general structure:

Null Hyp. (H0 ): A specific statement about a population parameter (or param-
eters). We would like to prove this wrong if possible.
Alt. Hyp. (H1 ): A general statement about a population parameter (or param-
eters) opposing H0 .
Data: Random sample(s) from the population(s).
Test Statistic: Suitable estimate of the population parameter (or combination
of parameters) derived from these data.
Sampling Distn : Describes the probability structure for the test statistic when
H0 is true.
p-value: probability of observed test statistic value or one more
favourable to H1 from the sampling distribution.
Decision: if p−value “too small” (ie. < significance level α), we reject
H0 in favour of H1 at the 100α% significance level.

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 6 / 33
A typical test/exam-style question

We will now apply this general framework to a commonly used test, the one-sample
t−test, using data from a published paper:

Franklin, D et al 2000, ‘Oral Health Status of Children in a Paediatric Intensive Care

Unit’, Intensive Care Medicine, vol. 26, pp. 319-324.

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 7 / 33
A typical test/exam-style question

We will now apply this general framework to a commonly used test, the one-sample
t−test, using data from a published paper:

Franklin, D et al 2000, ‘Oral Health Status of Children in a Paediatric Intensive Care

Unit’, Intensive Care Medicine, vol. 26, pp. 319-324.

Some of the analysis in this paper has been re-cast as a typical test/exam-style question.

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 7 / 33
A typical test/exam-style question: setting up the hypotheses

A study of the dental status of critically ill children in a Paediatric Intensive Care Unit
examined 16 children with permanent teeth and found that the mean number of missing
or filled teeth was 1.2 with a standard deviation of 1.9. Extensive analysis has established
that the mean number of such teeth in the wider population of children is 1.4. Test
whether the mean for critically ill children differs from this.

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 8 / 33
A typical test/exam-style question: setting up the hypotheses

A study of the dental status of critically ill children in a Paediatric Intensive Care Unit
examined 16 children with permanent teeth and found that the mean number of missing
or filled teeth was 1.2 with a standard deviation of 1.9. Extensive analysis has established
that the mean number of such teeth in the wider population of children is 1.4. Test
whether the mean for critically ill children differs from this.

Null Hyp. A specific statement about a population parameter (or param-

eters). We would like to prove this wrong if possible.
H0 :

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 8 / 33
A typical test/exam-style question: setting up the hypotheses

A study of the dental status of critically ill children in a Paediatric Intensive Care Unit
examined 16 children with permanent teeth and found that the mean number of missing
or filled teeth was 1.2 with a standard deviation of 1.9. Extensive analysis has established
that the mean number of such teeth in the wider population of children is 1.4. Test
whether the mean for critically ill children differs from this.

Null Hyp. A specific statement about a population parameter (or param-

eters). We would like to prove this wrong if possible.
H0 :

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 8 / 33
A typical test/exam-style question: setting up the hypotheses

A study of the dental status of critically ill children in a Paediatric Intensive Care Unit
examined 16 children with permanent teeth and found that the mean number of missing
or filled teeth was 1.2 with a standard deviation of 1.9. Extensive analysis has established
that the mean number of such teeth in the wider population of children is 1.4. Test
whether the mean for critically ill children differs from this.

Null Hyp. A specific statement about a population parameter (or param-

eters). We would like to prove this wrong if possible.
H0 :

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 8 / 33
A typical test/exam-style question: setting up the hypotheses

A study of the dental status of critically ill children in a Paediatric Intensive Care Unit
examined 16 children with permanent teeth and found that the mean number of missing
or filled teeth was 1.2 with a standard deviation of 1.9. Extensive analysis has established
that the mean number of such teeth in the wider population of children is 1.4. Test
whether the mean for critically ill children differs from this.

Null Hyp. A specific statement about a population parameter (or param-

eters). We would like to prove this wrong if possible.
H0 : µ

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 8 / 33
A typical test/exam-style question: setting up the hypotheses

A study of the dental status of critically ill children in a Paediatric Intensive Care Unit
examined 16 children with permanent teeth and found that the mean number of missing
or filled teeth was 1.2 with a standard deviation of 1.9. Extensive analysis has established
that the mean number of such teeth in the wider population of children is 1.4. Test
whether the mean for critically ill children differs from this.

Null Hyp. A specific statement about a population parameter (or param-

eters). We would like to prove this wrong if possible.
H0 : µ

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 8 / 33
A typical test/exam-style question: setting up the hypotheses

A study of the dental status of critically ill children in a Paediatric Intensive Care Unit
examined 16 children with permanent teeth and found that the mean number of missing
or filled teeth was 1.2 with a standard deviation of 1.9. Extensive analysis has established
that the mean number of such teeth in the wider population of children is 1.4. Test
whether the mean for critically ill children differs from this.

Null Hyp. A specific statement about a population parameter (or param-

eters). We would like to prove this wrong if possible.
H0 : µ =

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 8 / 33
A typical test/exam-style question: setting up the hypotheses

A study of the dental status of critically ill children in a Paediatric Intensive Care Unit
examined 16 children with permanent teeth and found that the mean number of missing
or filled teeth was 1.2 with a standard deviation of 1.9. Extensive analysis has established
that the mean number of such teeth in the wider population of children is 1.4. Test
whether the mean for critically ill children differs from this.

Null Hyp. A specific statement about a population parameter (or param-

eters). We would like to prove this wrong if possible.
H0 : µ =

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 8 / 33
A typical test/exam-style question: setting up the hypotheses

A study of the dental status of critically ill children in a Paediatric Intensive Care Unit
examined 16 children with permanent teeth and found that the mean number of missing
or filled teeth was 1.2 with a standard deviation of 1.9. Extensive analysis has established
that the mean number of such teeth in the wider population of children is 1.4. Test
whether the mean for critically ill children differs from this.

Null Hyp. A specific statement about a population parameter (or param-

eters). We would like to prove this wrong if possible.
H0 : µ = 1.4 missing/filled teeth

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 8 / 33
A typical test/exam-style question: setting up the hypotheses

A study of the dental status of critically ill children in a Paediatric Intensive Care Unit
examined 16 children with permanent teeth and found that the mean number of missing
or filled teeth was 1.2 with a standard deviation of 1.9. Extensive analysis has established
that the mean number of such teeth in the wider population of children is 1.4. Test
whether the mean for critically ill children differs from this.

Null Hyp. A specific statement about a population parameter (or param-

eters). We would like to prove this wrong if possible.
H0 : µ = 1.4 missing/filled teeth

Alt Hyp. A general statement about a population parameter (or param-

eters) opposing H0 .
H1 : µ

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 8 / 33
A typical test/exam-style question: setting up the hypotheses

A study of the dental status of critically ill children in a Paediatric Intensive Care Unit
examined 16 children with permanent teeth and found that the mean number of missing
or filled teeth was 1.2 with a standard deviation of 1.9. Extensive analysis has established
that the mean number of such teeth in the wider population of children is 1.4. Test
whether the mean for critically ill children differs from this.

Null Hyp. A specific statement about a population parameter (or param-

eters). We would like to prove this wrong if possible.
H0 : µ = 1.4 missing/filled teeth

Alt Hyp. A general statement about a population parameter (or param-

eters) opposing H0 .
H1 : µ

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 8 / 33
A typical test/exam-style question: setting up the hypotheses

A study of the dental status of critically ill children in a Paediatric Intensive Care Unit
examined 16 children with permanent teeth and found that the mean number of missing
or filled teeth was 1.2 with a standard deviation of 1.9. Extensive analysis has established
that the mean number of such teeth in the wider population of children is 1.4. Test
whether the mean for critically ill children differs from this.

Null Hyp. A specific statement about a population parameter (or param-

eters). We would like to prove this wrong if possible.
H0 : µ = 1.4 missing/filled teeth

Alt Hyp. A general statement about a population parameter (or param-

eters) opposing H0 .
H1 : µ

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 8 / 33
A typical test/exam-style question: setting up the hypotheses

A study of the dental status of critically ill children in a Paediatric Intensive Care Unit
examined 16 children with permanent teeth and found that the mean number of missing
or filled teeth was 1.2 with a standard deviation of 1.9. Extensive analysis has established
that the mean number of such teeth in the wider population of children is 1.4. Test
whether the mean for critically ill children differs from this.

Null Hyp. A specific statement about a population parameter (or param-

eters). We would like to prove this wrong if possible.
H0 : µ = 1.4 missing/filled teeth

Alt Hyp. A general statement about a population parameter (or param-

eters) opposing H0 .
H1 : µ 6= 1.4 missing/filled teeth

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 8 / 33
A typical test/exam-style question: data and test statistic

A study of the dental status of critically ill children in a Paediatric Intensive Care Unit
examined 16 children with permanent teeth and found that the mean number of missing
or filled teeth was 1.2 with a standard deviation of 1.9. Extensive analysis has established
that the mean number of such teeth in the wider population of children is 1.4. Test
whether the mean for critically ill children differs from this.

H0 : µ = 1.4 missing/filled teeth H1 : µ 6= 1.4 missing/filled teeth

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 8 / 33
A typical test/exam-style question: data and test statistic

A study of the dental status of critically ill children in a Paediatric Intensive Care Unit
examined 16 children with permanent teeth and found that the mean number of missing
or filled teeth was 1.2 with a standard deviation of 1.9. Extensive analysis has established
that the mean number of such teeth in the wider population of children is 1.4. Test
whether the mean for critically ill children differs from this.

H0 : µ = 1.4 missing/filled teeth H1 : µ 6= 1.4 missing/filled teeth

Data:

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 8 / 33
A typical test/exam-style question: data and test statistic

A study of the dental status of critically ill children in a Paediatric Intensive Care Unit
examined 16 children with permanent teeth and found that the mean number of missing
or filled teeth was 1.2 with a standard deviation of 1.9. Extensive analysis has established
that the mean number of such teeth in the wider population of children is 1.4. Test
whether the mean for critically ill children differs from this.

H0 : µ = 1.4 missing/filled teeth H1 : µ 6= 1.4 missing/filled teeth

Data:

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 8 / 33
A typical test/exam-style question: data and test statistic

A study of the dental status of critically ill children in a Paediatric Intensive Care Unit
examined 16 children with permanent teeth and found that the mean number of missing
or filled teeth was 1.2 with a standard deviation of 1.9. Extensive analysis has established
that the mean number of such teeth in the wider population of children is 1.4. Test
whether the mean for critically ill children differs from this.

H0 : µ = 1.4 missing/filled teeth H1 : µ 6= 1.4 missing/filled teeth

Data: n = 16 critically ill children with permanent teeth.

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 8 / 33
A typical test/exam-style question: data and test statistic

A study of the dental status of critically ill children in a Paediatric Intensive Care Unit
examined 16 children with permanent teeth and found that the mean number of missing
or filled teeth was 1.2 with a standard deviation of 1.9. Extensive analysis has established
that the mean number of such teeth in the wider population of children is 1.4. Test
whether the mean for critically ill children differs from this.

H0 : µ = 1.4 missing/filled teeth H1 : µ 6= 1.4 missing/filled teeth

Data: n = 16 critically ill children with permanent teeth.

Test Statistic: Suitable estimate of the population parameter (or combination
of parameters) derived from these data.

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 8 / 33
A typical test/exam-style question: data and test statistic

A study of the dental status of critically ill children in a Paediatric Intensive Care Unit
examined 16 children with permanent teeth and found that the mean number of missing
or filled teeth was 1.2 with a standard deviation of 1.9. Extensive analysis has established
that the mean number of such teeth in the wider population of children is 1.4. Test
whether the mean for critically ill children differs from this.

H0 : µ = 1.4 missing/filled teeth H1 : µ 6= 1.4 missing/filled teeth

Data: n = 16 critically ill children with permanent teeth.

Test Statistic: Suitable estimate of the population parameter (or combination
of parameters) derived from these data.

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 8 / 33
A typical test/exam-style question: data and test statistic

A study of the dental status of critically ill children in a Paediatric Intensive Care Unit
examined 16 children with permanent teeth and found that the mean number of missing
or filled teeth was 1.2 with a standard deviation of 1.9. Extensive analysis has established
that the mean number of such teeth in the wider population of children is 1.4. Test
whether the mean for critically ill children differs from this.

H0 : µ = 1.4 missing/filled teeth H1 : µ 6= 1.4 missing/filled teeth

Data: n = 16 critically ill children with permanent teeth.

Test Statistic: Suitable estimate of the population parameter (or combination
of parameters) derived from these data.
x = 1.2 teeth.

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 8 / 33
A typical test/exam-style question: data and test statistic

A study of the dental status of critically ill children in a Paediatric Intensive Care Unit
examined 16 children with permanent teeth and found that the mean number of missing
or filled teeth was 1.2 with a standard deviation of 1.9. Extensive analysis has established
that the mean number of such teeth in the wider population of children is 1.4. Test
whether the mean for critically ill children differs from this.

H0 : µ = 1.4 missing/filled teeth H1 : µ 6= 1.4 missing/filled teeth

Data: n = 16 critically ill children with permanent teeth.

Test Statistic: Suitable estimate of the population parameter (or combination
of parameters) derived from these data.
x = 1.2 teeth. In most Intro Stats units the only available
testing procedure is the one-sample t−test. This uses a “stan-
dardized” version of x :
x − µ0
t= =
√s
n

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 8 / 33
A typical test/exam-style question: data and test statistic

A study of the dental status of critically ill children in a Paediatric Intensive Care Unit
examined 16 children with permanent teeth and found that the mean number of missing
or filled teeth was 1.2 with a standard deviation of 1.9. Extensive analysis has established
that the mean number of such teeth in the wider population of children is 1.4. Test
whether the mean for critically ill children differs from this.

H0 : µ = 1.4 missing/filled teeth H1 : µ 6= 1.4 missing/filled teeth

Data: n = 16 critically ill children with permanent teeth.

Test Statistic: Suitable estimate of the population parameter (or combination
of parameters) derived from these data.
x = 1.2 teeth. In most Intro Stats units the only available
testing procedure is the one-sample t−test. This uses a “stan-
dardized” version of x :
x − µ0 1.2 − 1.4
t= =
√s 1.9
√
n 16

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 8 / 33
A typical test/exam-style question: data and test statistic

A study of the dental status of critically ill children in a Paediatric Intensive Care Unit
examined 16 children with permanent teeth and found that the mean number of missing
or filled teeth was 1.2 with a standard deviation of 1.9. Extensive analysis has established
that the mean number of such teeth in the wider population of children is 1.4. Test
whether the mean for critically ill children differs from this.

H0 : µ = 1.4 missing/filled teeth H1 : µ 6= 1.4 missing/filled teeth

Data: n = 16 critically ill children with permanent teeth.

Test Statistic: Suitable estimate of the population parameter (or combination
of parameters) derived from these data.
x = 1.2 teeth. In most Intro Stats units the only available
testing procedure is the one-sample t−test. This uses a “stan-
dardized” version of x :
x − µ0 1.2 − 1.4
t= = = −0.421
√s 1.9
√
n 16

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 8 / 33
A typical test/exam-style question: p−value

p-value: probability of the observed test statistic value or one more

favourable to H1 : µ 6= 1.4 when H0 is true.

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 9 / 33
A typical test/exam-style question: p−value

p-value: probability of the observed test statistic value or one more

favourable to H1 : µ 6= 1.4 when H0 is true.

x −1.4
We observed x = 1.2 (or equivalently t = 0.475
= −0.421):

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 9 / 33
A typical test/exam-style question: p−value

p-value: probability of the observed test statistic value or one more

favourable to H1 : µ 6= 1.4 when H0 is true.

x −1.4
We observed x = 1.2 (or equivalently t = 0.475
= −0.421):

What sort of values of x would have been more favourable to H1 ?

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 9 / 33
A typical test/exam-style question: p−value

p-value: probability of the observed test statistic value or one more

favourable to H1 : µ 6= 1.4 when H0 is true.

x −1.4
We observed x = 1.2 (or equivalently t = 0.475
= −0.421):

What sort of values of x would have been more favourable to H1 ? Well, anything further
away from 1.4 than the observed 1.2, such as 1.1 or 0.9, etc (ie. any x < 1.2).

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 9 / 33
A typical test/exam-style question: p−value

p-value: probability of the observed test statistic value or one more

favourable to H1 : µ 6= 1.4 when H0 is true.

x −1.4
We observed x = 1.2 (or equivalently t = 0.475
= −0.421):

What sort of values of x would have been more favourable to H1 ? Well, anything further
away from 1.4 than the observed 1.2, such as 1.1 or 0.9, etc (ie. any x < 1.2).
These values correspond to t < −0.421.

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 9 / 33
A typical test/exam-style question: p−value

p-value: probability of the observed test statistic value or one more

favourable to H1 : µ 6= 1.4 when H0 is true.

x −1.4
We observed x = 1.2 (or equivalently t = 0.475
= −0.421):

What sort of values of x would have been more favourable to H1 ? Well, anything further
away from 1.4 than the observed 1.2, such as 1.1 or 0.9, etc (ie. any x < 1.2).
These values correspond to t < −0.421.
In fact, since H1 is two sided, any value further away from 1.4 on the other side (ie.
greater than 1.6) would have been more favourable to H1 as well.

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 9 / 33
A typical test/exam-style question: p−value

p-value: probability of the observed test statistic value or one more

favourable to H1 : µ 6= 1.4 when H0 is true.

x −1.4
We observed x = 1.2 (or equivalently t = 0.475
= −0.421):

What sort of values of x would have been more favourable to H1 ? Well, anything further
away from 1.4 than the observed 1.2, such as 1.1 or 0.9, etc (ie. any x < 1.2).
These values correspond to t < −0.421.
In fact, since H1 is two sided, any value further away from 1.4 on the other side (ie.
greater than 1.6) would have been more favourable to H1 as well.

These values correspond to t > 0.421 (a useful feature of t).

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 9 / 33
A typical test/exam-style question: p−value

p-value: probability of the observed test statistic value or one more

favourable to H1 : µ 6= 1.4 when H0 is true.

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 10 / 33
A typical test/exam-style question: p−value

p-value: probability of the observed test statistic value or one more

favourable to H1 : µ 6= 1.4 when H0 is true.

Now, the question is “what is the probability of getting a t value in the red zone when H0
is true?”

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 10 / 33
A typical test/exam-style question: p−value

p-value: probability of the observed test statistic value or one more

favourable to H1 : µ 6= 1.4 when H0 is true.

Now, the question is “what is the probability of getting a t value in the red zone when H0
is true?” To answer that, we need to know the Sampling Distribution of t. Statistical
theory says that the distribution looks like this . . .

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 10 / 33
A typical test/exam-style question: p−value

p-value: probability of the observed test statistic value or one more

favourable to H1 : µ 6= 1.4 when H0 is true.

Now, the question is “what is the probability of getting a t value in the red zone when H0
is true?” To answer that, we need to know the Sampling Distribution of t. Statistical
theory says that the distribution looks like this . . .

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 10 / 33
A typical test/exam-style question: p−value

p-value: probability of the observed test statistic value or one more

favourable to H1 : µ 6= 1.4 when H0 is true.

Now, the question is “what is the probability of getting a t value in the red zone when H0
is true?” To answer that, we need to know the Sampling Distribution of t. Statistical
theory says that the distribution looks like this . . . a “t−distribution with 16 − 1 = 15
degrees of freedom (df)”.

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 10 / 33
A typical test/exam-style question: p−value

p-value: probability of the observed test statistic value or one more

favourable to H1 : µ 6= 1.4 when H0 is true.

Now, the question is “what is the probability of getting a t value in the red zone when H0
is true?” To answer that, we need to know the Sampling Distribution of t. Statistical
theory says that the distribution looks like this . . . a “t−distribution with 16 − 1 = 15
degrees of freedom (df)”. So the p−value is this shaded area.

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 10 / 33
A typical test/exam-style question: p−value

Before we calculate the p−value, have a guess at what you think it is. (Hint: the total
area under the curve is 1).

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 11 / 33
A typical test/exam-style question: p−value

Before we calculate the p−value, have a guess at what you think it is. (Hint: the total
area under the curve is 1).
Technology such as Excel can give an exact answer: the cell function
“=T.DIST.2T(0.421,15)” returns p − value = 0.680.

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 11 / 33
A typical test/exam-style question: p−value

Before we calculate the p−value, have a guess at what you think it is. (Hint: the total
area under the curve is 1).
Technology such as Excel can give an exact answer: the cell function
“=T.DIST.2T(0.421,15)” returns p − value = 0.680. Was your guess close?

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 11 / 33
A typical test/exam-style question: decision

p − value = 0.680

Decision: if p−value “too small” (ie. < significance level α), we reject
H0 in favour of H1 at the 100α% significance level.

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 12 / 33
A typical test/exam-style question: decision

p − value = 0.680

Decision: if p−value “too small” (ie. < significance level α), we reject
H0 in favour of H1 at the 100α% significance level.

Clearly a p−value this big does not meet the criterion of “too small” and we would
“retain H0 at any sensible significance level”.

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 12 / 33
A typical test/exam-style question: decision

p − value = 0.680

Decision: if p−value “too small” (ie. < significance level α), we reject
H0 in favour of H1 at the 100α% significance level.

Clearly a p−value this big does not meet the criterion of “too small” and we would
“retain H0 at any sensible significance level”.

• Some Intro Stats units use a t−table to find a suitable approximation for the
p−value. If you don’t use this method, you can skip the next section by clicking
here .

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 12 / 33
Using a t−table to find the p−value
• A t−table provides enough information about a p−value to make decisions:

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 13 / 33
Using a t−table to find the p−value
• A t−table provides enough information about a p−value to make decisions:

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 13 / 33
Using a t−table to find the p−value
• A t−table provides enough information about a p−value to make decisions:

• Each row refers to a different t−distribution. In this case we need the row for 15 df.

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 13 / 33
Using a t−table to find the p−value
• A t−table provides enough information about a p−value to make decisions:

• Each row refers to a different t−distribution. In this case we need the row for 15 df.

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 13 / 33
Using a t−table to find the p−value
• A t−table provides enough information about a p−value to make decisions:

• Each row refers to a different t−distribution. In this case we need the row for 15 df.

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 13 / 33
Using a t−table to find the p−value
• A t−table provides enough information about a p−value to make decisions:

• Each row refers to a different t−distribution. In this case we need the row for 15 df.

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 13 / 33
Using a t−table to find the p−value

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 14 / 33
Using a t−table to find the p−value

In this case, the positive version of t is 0.421

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 14 / 33
Using a t−table to find the p−value

In this case, the positive version of t is 0.421 and the smallest value of t in this row is
0.691.

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 14 / 33
Using a t−table to find the p−value

In this case, the positive version of t is 0.421 and the smallest value of t in this row is
0.691. We can now say that the blue shaded area is 0.25.

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 14 / 33
Using a t−table to find the p−value

In this case, the positive version of t is 0.421 and the smallest value of t in this row is
0.691. We can now say that the blue shaded area is 0.25.

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 14 / 33
Using a t−table to find the p−value

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 15 / 33
Using a t−table to find the p−value

So, we can say that the p−value (grey area partially obscured by the blue area) is greater
than the blue area:

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 15 / 33
Using a t−table to find the p−value

So, we can say that the p−value (grey area partially obscured by the blue area) is greater
than the blue area:
p − value > 2 × 0.25
p − value > 0.5.

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 15 / 33
Using a t−table to find the p−value

So, we can say that the p−value (grey area partially obscured by the blue area) is greater
than the blue area:
p − value > 2 × 0.25
p − value > 0.5.
(Remember that the exact answer is p − value = 0.680).

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 15 / 33
Using a t−table to find the p−value

This is enough to make our decision:

Decision: if p−value “too small” (ie. < significance level α), we reject
H0 in favour of H1 at the 100α% significance level.

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 16 / 33
Using a t−table to find the p−value

This is enough to make our decision:

Decision: if p−value “too small” (ie. < significance level α), we reject
H0 in favour of H1 at the 100α% significance level.

Clearly, since p − value > 0.5 does not meet the criterion of “too small”, we would
“retain H0 at any sensible significance level” as before.

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 16 / 33
Assumptions for the t−test

• You may also be asked to consider whether the assumptions required for a
hypothesis test to work have been met. If you don’t discuss this topic in your unit,
you can skip the next section by clicking here .

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 17 / 33
Checking the assumptions for the t−test

Most testing procedures come with assumptions. These are conditions that the
population(s) and sample(s) have to meet for the p−values to be reliable.

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 18 / 33
Checking the assumptions for the t−test

Most testing procedures come with assumptions. These are conditions that the
population(s) and sample(s) have to meet for the p−values to be reliable.

For all tests, a key assumption is

the data is a random sample of the relevent population.

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 18 / 33
Checking the assumptions for the t−test

Most testing procedures come with assumptions. These are conditions that the
population(s) and sample(s) have to meet for the p−values to be reliable.

For all tests, a key assumption is

the data is a random sample of the relevent population.

The population here is “all critically ill children with permanent teeth in Paediatric
Intensive Care”.

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 18 / 33
Checking the assumptions for the t−test

Most testing procedures come with assumptions. These are conditions that the
population(s) and sample(s) have to meet for the p−values to be reliable.

For all tests, a key assumption is

the data is a random sample of the relevent population.

The population here is “all critically ill children with permanent teeth in Paediatric
Intensive Care”.

We have to assume that the chosen 16 children represent a random sample of such
children . . .

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 18 / 33
Checking the assumptions for the t−test
For the one sample t−test we just performed, a key assumption is

the population follows a Normal Distribution or the sample size is large.

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 19 / 33
Checking the assumptions for the t−test
For the one sample t−test we just performed, a key assumption is

the population follows a Normal Distribution or the sample size is large.

n = 16 is not a very large sample so a good statistician would examine the sample itself
for evidence of non-Normality and/or outliers.

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 19 / 33
Checking the assumptions for the t−test
For the one sample t−test we just performed, a key assumption is

the population follows a Normal Distribution or the sample size is large.

n = 16 is not a very large sample so a good statistician would examine the sample itself
for evidence of non-Normality and/or outliers.

Now, it’s common for your lecturers (and published articles) not to include the raw data
in questions. This is mainly to avoid distraction from the skills they want to test.

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 19 / 33
Checking the assumptions for the t−test
For the one sample t−test we just performed, a key assumption is

the population follows a Normal Distribution or the sample size is large.

n = 16 is not a very large sample so a good statistician would examine the sample itself
for evidence of non-Normality and/or outliers.

Now, it’s common for your lecturers (and published articles) not to include the raw data
in questions. This is mainly to avoid distraction from the skills they want to test.

However, a keen reader might be able to detect evidence of non-Normality and/or

outliers just from the summary information provided:

x = 1.2 teeth s = 1.9 teeth

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 19 / 33
Checking the assumptions for the t−test
For the one sample t−test we just performed, a key assumption is

the population follows a Normal Distribution or the sample size is large.

n = 16 is not a very large sample so a good statistician would examine the sample itself
for evidence of non-Normality and/or outliers.

Now, it’s common for your lecturers (and published articles) not to include the raw data
in questions. This is mainly to avoid distraction from the skills they want to test.

However, a keen reader might be able to detect evidence of non-Normality and/or

outliers just from the summary information provided:

x = 1.2 teeth s = 1.9 teeth

Still can’t see it?

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 19 / 33
Checking the assumptions for the t−test
For the one sample t−test we just performed, a key assumption is

the population follows a Normal Distribution or the sample size is large.

n = 16 is not a very large sample so a good statistician would examine the sample itself
for evidence of non-Normality and/or outliers.

Now, it’s common for your lecturers (and published articles) not to include the raw data
in questions. This is mainly to avoid distraction from the skills they want to test.

However, a keen reader might be able to detect evidence of non-Normality and/or

outliers just from the summary information provided:

x = 1.2 teeth s = 1.9 teeth

Still can’t see it? If not, here is another clue:

the smallest possible data value here is 0 missing/filled teeth.

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 19 / 33
Checking the assumptions for the t−test
For the one sample t−test we just performed, a key assumption is

the population follows a Normal Distribution or the sample size is large.

n = 16 is not a very large sample so a good statistician would examine the sample itself
for evidence of non-Normality and/or outliers.

Now, it’s common for your lecturers (and published articles) not to include the raw data
in questions. This is mainly to avoid distraction from the skills they want to test.

However, a keen reader might be able to detect evidence of non-Normality and/or

outliers just from the summary information provided:

x = 1.2 teeth s = 1.9 teeth

Still can’t see it? If not, here is another clue:

the smallest possible data value here is 0 missing/filled teeth.

You can’t go backwards from the mean by even one standard deviation! This means
that there must be data values much bigger than 1.2 in order to create such a large
standard deviation.

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 19 / 33
Checking the assumptions for the t−test

Here’s a possible data set that fits these summary figures:

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 20 / 33
Checking the assumptions for the t−test

Here’s a possible data set that fits these summary figures:

Technically, the t−test should not have been used here.

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 20 / 33
Checking the assumptions for the t−test

Here’s a possible data set that fits these summary figures:

Technically, the t−test should not have been used here. Alternative methods do exist
when these assumptions are not met. (If your unit covers “non-parametric tests” or “data
transformations” you may know some of them.)

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 20 / 33
Another t−test question: setting up the hypotheses

Plaque develops on teeth in response to the presence of bacteria and can lead to harmful
effects. The difference in plaque coverage (% of all teeth surfaces with plaque) between
admission and discharge was measured for each of the 16 critically ill children in the
previous example. The mean of these differences (discharge − admission) was 4.0% with
a standard deviation of 7.4%. Test whether there was a mean change in plaque coverage
between admission and discharge.

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 21 / 33
Another t−test question: setting up the hypotheses

Plaque develops on teeth in response to the presence of bacteria and can lead to harmful
effects. The difference in plaque coverage (% of all teeth surfaces with plaque) between
admission and discharge was measured for each of the 16 critically ill children in the
previous example. The mean of these differences (discharge − admission) was 4.0% with
a standard deviation of 7.4%. Test whether there was a mean change in plaque coverage
between admission and discharge.

Null Hyp. A specific statement about a population parameter (or param-

eters). We would like to prove this wrong if possible.
H0 :

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 21 / 33
Another t−test question: setting up the hypotheses

Plaque develops on teeth in response to the presence of bacteria and can lead to harmful
effects. The difference in plaque coverage (% of all teeth surfaces with plaque) between
admission and discharge was measured for each of the 16 critically ill children in the
previous example. The mean of these differences (discharge − admission) was 4.0% with
a standard deviation of 7.4%. Test whether there was a mean change in plaque coverage
between admission and discharge.

Null Hyp. A specific statement about a population parameter (or param-

eters). We would like to prove this wrong if possible.
H0 :

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 21 / 33
Another t−test question: setting up the hypotheses

Plaque develops on teeth in response to the presence of bacteria and can lead to harmful
effects. The difference in plaque coverage (% of all teeth surfaces with plaque) between
admission and discharge was measured for each of the 16 critically ill children in the
previous example. The mean of these differences (discharge − admission) was 4.0% with
a standard deviation of 7.4%. Test whether there was a mean change in plaque coverage
between admission and discharge.

Null Hyp. A specific statement about a population parameter (or param-

eters). We would like to prove this wrong if possible.
H0 :

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 21 / 33
Another t−test question: setting up the hypotheses

Plaque develops on teeth in response to the presence of bacteria and can lead to harmful
effects. The difference in plaque coverage (% of all teeth surfaces with plaque) between
admission and discharge was measured for each of the 16 critically ill children in the
previous example. The mean of these differences (discharge − admission) was 4.0% with
a standard deviation of 7.4%. Test whether there was a mean change in plaque coverage
between admission and discharge.

Null Hyp. A specific statement about a population parameter (or param-

eters). We would like to prove this wrong if possible.
H 0 : µd

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 21 / 33
Another t−test question: setting up the hypotheses

Plaque develops on teeth in response to the presence of bacteria and can lead to harmful
effects. The difference in plaque coverage (% of all teeth surfaces with plaque) between
admission and discharge was measured for each of the 16 critically ill children in the
previous example. The mean of these differences (discharge − admission) was 4.0% with
a standard deviation of 7.4%. Test whether there was a mean change in plaque coverage
between admission and discharge.

Null Hyp. A specific statement about a population parameter (or param-

eters). We would like to prove this wrong if possible.
H 0 : µd

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 21 / 33
Another t−test question: setting up the hypotheses

Plaque develops on teeth in response to the presence of bacteria and can lead to harmful
effects. The difference in plaque coverage (% of all teeth surfaces with plaque) between
admission and discharge was measured for each of the 16 critically ill children in the
previous example. The mean of these differences (discharge − admission) was 4.0% with
a standard deviation of 7.4%. Test whether there was a mean change in plaque coverage
between admission and discharge.

Null Hyp. A specific statement about a population parameter (or param-

eters). We would like to prove this wrong if possible.
H 0 : µd =

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 21 / 33
Another t−test question: setting up the hypotheses

Plaque develops on teeth in response to the presence of bacteria and can lead to harmful
effects. The difference in plaque coverage (% of all teeth surfaces with plaque) between
admission and discharge was measured for each of the 16 critically ill children in the
previous example. The mean of these differences (discharge − admission) was 4.0% with
a standard deviation of 7.4%. Test whether there was a mean change in plaque coverage
between admission and discharge.

Null Hyp. A specific statement about a population parameter (or param-

eters). We would like to prove this wrong if possible.
H 0 : µd =
The opposite of “change” is “no change” which translates to “µd = 0”.

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 21 / 33
Another t−test question: setting up the hypotheses

Plaque develops on teeth in response to the presence of bacteria and can lead to harmful
effects. The difference in plaque coverage (% of all teeth surfaces with plaque) between
admission and discharge was measured for each of the 16 critically ill children in the
previous example. The mean of these differences (discharge − admission) was 4.0% with
a standard deviation of 7.4%. Test whether there was a mean change in plaque coverage
between admission and discharge.

Null Hyp. A specific statement about a population parameter (or param-

eters). We would like to prove this wrong if possible.
H 0 : µd = 0
The opposite of “change” is “no change” which translates to “µd = 0”.

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 21 / 33
Another t−test question: setting up the hypotheses

Plaque develops on teeth in response to the presence of bacteria and can lead to harmful
effects. The difference in plaque coverage (% of all teeth surfaces with plaque) between
admission and discharge was measured for each of the 16 critically ill children in the
previous example. The mean of these differences (discharge − admission) was 4.0% with
a standard deviation of 7.4%. Test whether there was a mean change in plaque coverage
between admission and discharge.

Null Hyp. A specific statement about a population parameter (or param-

eters). We would like to prove this wrong if possible.
H 0 : µd = 0
The opposite of “change” is “no change” which translates to “µd = 0”.

Alt Hyp. A general statement about a population parameter (or param-

eters) opposing H0 .
H 1 : µd

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 21 / 33
Another t−test question: setting up the hypotheses

Plaque develops on teeth in response to the presence of bacteria and can lead to harmful
effects. The difference in plaque coverage (% of all teeth surfaces with plaque) between
admission and discharge was measured for each of the 16 critically ill children in the
previous example. The mean of these differences (discharge − admission) was 4.0% with
a standard deviation of 7.4%. Test whether there was a mean change in plaque coverage
between admission and discharge.

Null Hyp. A specific statement about a population parameter (or param-

eters). We would like to prove this wrong if possible.
H 0 : µd = 0
The opposite of “change” is “no change” which translates to “µd = 0”.

Alt Hyp. A general statement about a population parameter (or param-

eters) opposing H0 .
H 1 : µd

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 21 / 33
Another t−test question: setting up the hypotheses

Plaque develops on teeth in response to the presence of bacteria and can lead to harmful
effects. The difference in plaque coverage (% of all teeth surfaces with plaque) between
admission and discharge was measured for each of the 16 critically ill children in the
previous example. The mean of these differences (discharge − admission) was 4.0% with
a standard deviation of 7.4%. Test whether there was a mean change in plaque coverage
between admission and discharge.

Null Hyp. A specific statement about a population parameter (or param-

eters). We would like to prove this wrong if possible.
H 0 : µd = 0
The opposite of “change” is “no change” which translates to “µd = 0”.

Alt Hyp. A general statement about a population parameter (or param-

eters) opposing H0 .
H 1 : µd
The word “change” suggests that the children could have acquired more
plaque or less plaque whilst in hospital.

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 21 / 33
Another t−test question: setting up the hypotheses

Plaque develops on teeth in response to the presence of bacteria and can lead to harmful
effects. The difference in plaque coverage (% of all teeth surfaces with plaque) between
admission and discharge was measured for each of the 16 critically ill children in the
previous example. The mean of these differences (discharge − admission) was 4.0% with
a standard deviation of 7.4%. Test whether there was a mean change in plaque coverage
between admission and discharge.

Null Hyp. A specific statement about a population parameter (or param-

eters). We would like to prove this wrong if possible.
H 0 : µd = 0
The opposite of “change” is “no change” which translates to “µd = 0”.

Alt Hyp. A general statement about a population parameter (or param-

eters) opposing H0 .
H1 : µd 6= 0
The word “change” suggests that the children could have acquired more
plaque or less plaque whilst in hospital.

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 21 / 33
Another t−test question: data and test statistic

Plaque develops on teeth in response to the presence of bacteria and can lead to harmful
effects. The difference in plaque coverage (% of all teeth surfaces with plaque) between
admission and discharge was measured for each of the 16 critically ill children in the
previous example. The mean of these differences (discharge − admission) was 4.0% with
a standard deviation of 7.4%. Test whether there was a mean change in plaque coverage
between admission and discharge.

H 0 : µd = 0 H1 : µd 6= 0

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 21 / 33
Another t−test question: data and test statistic

Plaque develops on teeth in response to the presence of bacteria and can lead to harmful
effects. The difference in plaque coverage (% of all teeth surfaces with plaque) between
admission and discharge was measured for each of the 16 critically ill children in the
previous example. The mean of these differences (discharge − admission) was 4.0% with
a standard deviation of 7.4%. Test whether there was a mean change in plaque coverage
between admission and discharge.

H 0 : µd = 0 H1 : µd 6= 0

Data:

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 21 / 33
Another t−test question: data and test statistic

Plaque develops on teeth in response to the presence of bacteria and can lead to harmful
effects. The difference in plaque coverage (% of all teeth surfaces with plaque) between
admission and discharge was measured for each of the 16 critically ill children in the
previous example. The mean of these differences (discharge − admission) was 4.0% with
a standard deviation of 7.4%. Test whether there was a mean change in plaque coverage
between admission and discharge.

H 0 : µd = 0 H1 : µd 6= 0

Data:

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 21 / 33
Another t−test question: data and test statistic

Plaque develops on teeth in response to the presence of bacteria and can lead to harmful
effects. The difference in plaque coverage (% of all teeth surfaces with plaque) between
admission and discharge was measured for each of the 16 critically ill children in the
previous example. The mean of these differences (discharge − admission) was 4.0% with
a standard deviation of 7.4%. Test whether there was a mean change in plaque coverage
between admission and discharge.

H 0 : µd = 0 H1 : µd 6= 0

Data: n = 16 critically ill children with permanent teeth.

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 21 / 33
Another t−test question: data and test statistic

Plaque develops on teeth in response to the presence of bacteria and can lead to harmful
effects. The difference in plaque coverage (% of all teeth surfaces with plaque) between
admission and discharge was measured for each of the 16 critically ill children in the
previous example. The mean of these differences (discharge − admission) was 4.0% with
a standard deviation of 7.4%. Test whether there was a mean change in plaque coverage
between admission and discharge.

H 0 : µd = 0 H1 : µd 6= 0

Data: n = 16 critically ill children with permanent teeth.

Test Statistic: Suitable estimate of the population parameter (or combination

of parameters) derived from these data.

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 21 / 33
Another t−test question: data and test statistic

Plaque develops on teeth in response to the presence of bacteria and can lead to harmful
effects. The difference in plaque coverage (% of all teeth surfaces with plaque) between
admission and discharge was measured for each of the 16 critically ill children in the
previous example. The mean of these differences (discharge − admission) was 4.0% with
a standard deviation of 7.4%. Test whether there was a mean change in plaque coverage
between admission and discharge.

H 0 : µd = 0 H1 : µd 6= 0

Data: n = 16 critically ill children with permanent teeth.

Test Statistic: Suitable estimate of the population parameter (or combination

of parameters) derived from these data.

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 21 / 33
Another t−test question: data and test statistic

Plaque develops on teeth in response to the presence of bacteria and can lead to harmful
effects. The difference in plaque coverage (% of all teeth surfaces with plaque) between
admission and discharge was measured for each of the 16 critically ill children in the
previous example. The mean of these differences (discharge − admission) was 4.0% with
a standard deviation of 7.4%. Test whether there was a mean change in plaque coverage
between admission and discharge.

H 0 : µd = 0 H1 : µd 6= 0

Data: n = 16 critically ill children with permanent teeth.

Test Statistic: Suitable estimate of the population parameter (or combination

of parameters) derived from these data.
x d = 4.0.

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 21 / 33
Another t−test question: data and test statistic

Plaque develops on teeth in response to the presence of bacteria and can lead to harmful
effects. The difference in plaque coverage (% of all teeth surfaces with plaque) between
admission and discharge was measured for each of the 16 critically ill children in the
previous example. The mean of these differences (discharge − admission) was 4.0% with
a standard deviation of 7.4%. Test whether there was a mean change in plaque coverage
between admission and discharge.

H 0 : µd = 0 H1 : µd 6= 0

Data: n = 16 critically ill children with permanent teeth.

Test Statistic: Suitable estimate of the population parameter (or combination

of parameters) derived from these data.
x d = 4.0. Using the one-sample t−test again:

x d − µd
t= s =
√d
n

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 21 / 33
Another t−test question: data and test statistic

Plaque develops on teeth in response to the presence of bacteria and can lead to harmful
effects. The difference in plaque coverage (% of all teeth surfaces with plaque) between
admission and discharge was measured for each of the 16 critically ill children in the
previous example. The mean of these differences (discharge − admission) was 4.0% with
a standard deviation of 7.4%. Test whether there was a mean change in plaque coverage
between admission and discharge.

H 0 : µd = 0 H1 : µd 6= 0

Data: n = 16 critically ill children with permanent teeth.

Test Statistic: Suitable estimate of the population parameter (or combination

of parameters) derived from these data.
x d = 4.0. Using the one-sample t−test again:

x d − µd 4.0 − 0
t= s = 7.4
√d √
n 16

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 21 / 33
Another t−test question: data and test statistic

Plaque develops on teeth in response to the presence of bacteria and can lead to harmful
effects. The difference in plaque coverage (% of all teeth surfaces with plaque) between
admission and discharge was measured for each of the 16 critically ill children in the
previous example. The mean of these differences (discharge − admission) was 4.0% with
a standard deviation of 7.4%. Test whether there was a mean change in plaque coverage
between admission and discharge.

H 0 : µd = 0 H1 : µd 6= 0

Data: n = 16 critically ill children with permanent teeth.

Test Statistic: Suitable estimate of the population parameter (or combination

of parameters) derived from these data.
x d = 4.0. Using the one-sample t−test again:

x d − µd 4.0 − 0
t= s = 7.4
= 2.162
√d √
n 16

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 21 / 33
Another t−test question: p−value

p-value: probability of the observed test statistic value or one more

favourable to H1 : µd 6= 0 when H0 is true.

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 22 / 33
Another t−test question: p−value

p-value: probability of the observed test statistic value or one more

favourable to H1 : µd 6= 0 when H0 is true.

x d −0
We observed x = 4.0 (or equivalently t = 1.85
= 2.162):

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 22 / 33
Another t−test question: p−value

p-value: probability of the observed test statistic value or one more

favourable to H1 : µd 6= 0 when H0 is true.

x d −0
We observed x = 4.0 (or equivalently t = 1.85
= 2.162):

What sort of values of x would have been more favourable to H1 ?

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 22 / 33
Another t−test question: p−value

p-value: probability of the observed test statistic value or one more

favourable to H1 : µd 6= 0 when H0 is true.

x d −0
We observed x = 4.0 (or equivalently t = 1.85
= 2.162):

What sort of values of x would have been more favourable to H1 ? Well, anything further
away from 0 than the observed 4.0, such as 4.5 or 5.0, etc (ie. any x > 4.0).

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 22 / 33
Another t−test question: p−value

p-value: probability of the observed test statistic value or one more

favourable to H1 : µd 6= 0 when H0 is true.

x d −0
We observed x = 4.0 (or equivalently t = 1.85
= 2.162):

What sort of values of x would have been more favourable to H1 ? Well, anything further
away from 0 than the observed 4.0, such as 4.5 or 5.0, etc (ie. any x > 4.0).

These values correspond to t > 2.162.

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 22 / 33
Another t−test question: p−value

p-value: probability of the observed test statistic value or one more

favourable to H1 : µd 6= 0 when H0 is true.

x d −0
We observed x = 4.0 (or equivalently t = 1.85
= 2.162):

What sort of values of x would have been more favourable to H1 ? Well, anything further
away from 0 than the observed 4.0, such as 4.5 or 5.0, etc (ie. any x > 4.0).

These values correspond to t > 2.162.

In fact, since H1 is two sided, any value further away from 0 on the other side (ie. less
than -4.0) would have been more favourable to H1 as well.

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 22 / 33
Another t−test question: p−value

p-value: probability of the observed test statistic value or one more

favourable to H1 : µd 6= 0 when H0 is true.

x d −0
We observed x = 4.0 (or equivalently t = 1.85
= 2.162):

What sort of values of x would have been more favourable to H1 ? Well, anything further
away from 0 than the observed 4.0, such as 4.5 or 5.0, etc (ie. any x > 4.0).

These values correspond to t > 2.162.

In fact, since H1 is two sided, any value further away from 0 on the other side (ie. less
than -4.0) would have been more favourable to H1 as well.

These values correspond to t < −2.162 (a useful feature of t).

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 22 / 33
Another t−test question: p−value

p-value: probability of the observed test statistic value or one more

favourable to H1 : µd 6= 0 when H0 is true.

Since the sample size is still 16, the Sampling Distribution of t is still t(15)

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 23 / 33
Another t−test question: p−value

p-value: probability of the observed test statistic value or one more

favourable to H1 : µd 6= 0 when H0 is true.

Since the sample size is still 16, the Sampling Distribution of t is still t(15) and the
p−value is this shaded area.

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 23 / 33
Another t−test question: p−value

Before we calculate the p−value, have a guess at what you think it is. (Hint: the total
area under the curve is 1).

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 24 / 33
Another t−test question: p−value

Before we calculate the p−value, have a guess at what you think it is. (Hint: the total
area under the curve is 1).

• Using Excel again to get the exact answer: “=T.DIST.2T(2.162,15)” returns

p − value = 0.047.

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 24 / 33
Another t−test question: p−value

Before we calculate the p−value, have a guess at what you think it is. (Hint: the total
area under the curve is 1).

• Using Excel again to get the exact answer: “=T.DIST.2T(2.162,15)” returns

p − value = 0.047. Was your guess close?

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 24 / 33
Another t−test question: decision

p − value = 0.047

Decision: if p−value “too small” (ie. < significance level α), we reject
H0 in favour of H1 at the 100α% significance level.

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 25 / 33
Another t−test question: decision

p − value = 0.047

Decision: if p−value “too small” (ie. < significance level α), we reject
H0 in favour of H1 at the 100α% significance level.

If α is chosen to be 0.05 (5%), our p−value does meet the criterion of “too small” (just)
and we would “reject H0 at the 5% significance level”.

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 25 / 33
Another t−test question: decision

p − value = 0.047

Decision: if p−value “too small” (ie. < significance level α), we reject
H0 in favour of H1 at the 100α% significance level.

If α is chosen to be 0.05 (5%), our p−value does meet the criterion of “too small” (just)
and we would “reject H0 at the 5% significance level”.

We have convincing proof that the amount of plaque on the teeth of children with
permenant teeth tends to change during their stay in paediatric intensive care.

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 25 / 33
Another t−test question: decision

p − value = 0.047

Decision: if p−value “too small” (ie. < significance level α), we reject
H0 in favour of H1 at the 100α% significance level.

If α is chosen to be 0.05 (5%), our p−value does meet the criterion of “too small” (just)
and we would “reject H0 at the 5% significance level”.

We have convincing proof that the amount of plaque on the teeth of children with
permenant teeth tends to change during their stay in paediatric intensive care.

In fact, it seems that plaque tends to increase because the difference was calculated as
discharge % − admission%!

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 25 / 33
Another t−test question: decision

p − value = 0.047

Decision: if p−value “too small” (ie. < significance level α), we reject
H0 in favour of H1 at the 100α% significance level.

If α is chosen to be 0.05 (5%), our p−value does meet the criterion of “too small” (just)
and we would “reject H0 at the 5% significance level”.

We have convincing proof that the amount of plaque on the teeth of children with
permenant teeth tends to change during their stay in paediatric intensive care.

In fact, it seems that plaque tends to increase because the difference was calculated as
discharge % − admission%! Children in paediatric intensive care are too sick to clean
their own teeth and the paper in which these data appeared concluded that dental
management by staff needed to be improved.

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 25 / 33
Another t−test question: decision

p − value = 0.047

Decision: if p−value “too small” (ie. < significance level α), we reject
H0 in favour of H1 at the 100α% significance level.

If α is chosen to be 0.05 (5%), our p−value does meet the criterion of “too small” (just)
and we would “reject H0 at the 5% significance level”.

We have convincing proof that the amount of plaque on the teeth of children with
permenant teeth tends to change during their stay in paediatric intensive care.

In fact, it seems that plaque tends to increase because the difference was calculated as
discharge % − admission%! Children in paediatric intensive care are too sick to clean
their own teeth and the paper in which these data appeared concluded that dental
management by staff needed to be improved.
• If your unit doesn’t cover t−tables, you can skip the next section by clicking here .

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 25 / 33
Using a t−table to find the p−value

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 26 / 33
Using a t−table to find the p−value

Let’s take a close up look at the grey shaded area.

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 26 / 33
Using a t−table to find the p−value

Let’s take a close up look at the grey shaded area.

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 26 / 33
Using a t−table to find the p−value

Let’s take a close up look at the grey shaded area.

From the t−table, our t = 2.162 sits between 2.131

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 26 / 33
Using a t−table to find the p−value

Let’s take a close up look at the grey shaded area.

From the t−table, our t = 2.162 sits between 2.131 and 2.249.

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 26 / 33
Using a t−table to find the p−value

Let’s take a close up look at the grey shaded area.

From the t−table, our t = 2.162 sits between 2.131 and 2.249.

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 26 / 33
Using a t−table to find the p−value

Let’s take a close up look at the grey shaded area.

From the t−table, our t = 2.162 sits between 2.131 and 2.249.

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 26 / 33
Using a t−table to find the p−value

So, we don’t know the exact p−value but we can say that the p−value (grey area) is

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 27 / 33
Using a t−table to find the p−value

So, we don’t know the exact p−value but we can say that the p−value (grey area) is
larger than the blue area (0.02)

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 27 / 33
Using a t−table to find the p−value

So, we don’t know the exact p−value but we can say that the p−value (grey area) is
larger than the blue area (0.02) but smaller than the green area (0.025).

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 27 / 33
Using a t−table to find the p−value

So, we don’t know the exact p−value but we can say that the p−value (grey area) is
larger than the blue area (0.02) but smaller than the green area (0.025). Remembering to
double the above values to add in the left-hand tail, we can say:

2 × 0.02 < p − value < 2 × 0.025

0.04 < p − value < 0.05

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 27 / 33
Using a t−table to find the p−value

So, we don’t know the exact p−value but we can say that the p−value (grey area) is
larger than the blue area (0.02) but smaller than the green area (0.025). Remembering to
double the above values to add in the left-hand tail, we can say:

2 × 0.02 < p − value < 2 × 0.025

0.04 < p − value < 0.05

(Recall that the exact answer is p − value = 0.047.)

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 27 / 33
Using a t−table to find the p−value

0.04 < p − value < 0.05

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 28 / 33
Using a t−table to find the p−value

0.04 < p − value < 0.05

This is enough to make our decision:

Decision: if p−value “too small” (ie. < significance level α), we reject
H0 in favour of H1 at the 100α% significance level.

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 28 / 33
Using a t−table to find the p−value

0.04 < p − value < 0.05

This is enough to make our decision:

Decision: if p−value “too small” (ie. < significance level α), we reject
H0 in favour of H1 at the 100α% significance level.

If α is chosen to be 0.05 (5%), our p−value does meet the criterion of “too small” (just)
and we would “reject H0 at the 5% significance level” as before.

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 28 / 33
Assumptions for the t−test

• You may also be asked to consider whether the assumptions required for a
hypothesis test to work have been met. If you don’t discuss this topic in your unit,
you can skip the next section by clicking here .

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 29 / 33
Checking the assumptions for the t−test

For the one sample t−test we just performed, a key assumption is

the population follows a Normal Distribution or the sample size is large.

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 30 / 33
Checking the assumptions for the t−test

For the one sample t−test we just performed, a key assumption is

the population follows a Normal Distribution or the sample size is large.

n = 16 is not a very large sample so a good statistician would examine the sample itself
for evidence of non-Normality and/or outliers.

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 30 / 33
Checking the assumptions for the t−test

For the one sample t−test we just performed, a key assumption is

the population follows a Normal Distribution or the sample size is large.

n = 16 is not a very large sample so a good statistician would examine the sample itself
for evidence of non-Normality and/or outliers.

Luckily, differences in “before-and-after” studies like this are highly likely not to be
skewed (there’s no boundary at 0 for a start).

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 30 / 33
Checking the assumptions for the t−test

For the one sample t−test we just performed, a key assumption is

the population follows a Normal Distribution or the sample size is large.

n = 16 is not a very large sample so a good statistician would examine the sample itself
for evidence of non-Normality and/or outliers.

Luckily, differences in “before-and-after” studies like this are highly likely not to be
skewed (there’s no boundary at 0 for a start).

However, outliers are still a potential problem (in this case a child who acquires or loses
an unusually large amount of plaque while in hospital).

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 30 / 33
Checking the assumptions for the t−test

For the one sample t−test we just performed, a key assumption is

the population follows a Normal Distribution or the sample size is large.

n = 16 is not a very large sample so a good statistician would examine the sample itself
for evidence of non-Normality and/or outliers.

Luckily, differences in “before-and-after” studies like this are highly likely not to be
skewed (there’s no boundary at 0 for a start).

However, outliers are still a potential problem (in this case a child who acquires or loses
an unusually large amount of plaque while in hospital). You would need the raw data to
check.

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 30 / 33
Appendix: A tip for understanding t−tests

There should be something familiar about the t−distribution.

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 31 / 33
Appendix: A tip for understanding t−tests

There should be something familiar about the t−distribution. It looks a lot like a
Standard Normal distribution (shown above in grey).

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 31 / 33
Appendix: A tip for understanding t−tests

This means that a t−statistic is like a z−score for the sample mean (when µ = 1.4).
That is, it roughly follows the “68-95-97.5% Rule”:

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 32 / 33
Appendix: A tip for understanding t−tests

This means that a t−statistic is like a z−score for the sample mean (when µ = 1.4).
That is, it roughly follows the “68-95-97.5% Rule”:

There is a 68% chance that your data will produce a sample mean that is within
1 standard deviation of the population mean.

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 32 / 33
Appendix: A tip for understanding t−tests

This means that a t−statistic is like a z−score for the sample mean (when µ = 1.4).
That is, it roughly follows the “68-95-97.5% Rule”:

There is a 68% chance that your data will produce a sample mean that is within
1 standard deviation of the population mean.
There is a 95% chance that your data will produce a sample mean that is within
2 standard deviations of the population mean.

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 32 / 33
Appendix: A tip for understanding t−tests

This means that a t−statistic is like a z−score for the sample mean (when µ = 1.4).
That is, it roughly follows the “68-95-97.5% Rule”:

There is a 68% chance that your data will produce a sample mean that is within
1 standard deviation of the population mean.
There is a 95% chance that your data will produce a sample mean that is within
2 standard deviations of the population mean.
There is a 97.5% chance that your data will produce a sample mean that is within
3 standard deviations of the population mean.

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 32 / 33
Appendix: A tip for understanding t−tests

This means that a t−statistic is like a z−score for the sample mean (when µ = 1.4).
That is, it roughly follows the “68-95-97.5% Rule”:

There is a 68% chance that your data will produce a sample mean that is within
1 standard deviation of the population mean.
There is a 95% chance that your data will produce a sample mean that is within
2 standard deviations of the population mean.
There is a 97.5% chance that your data will produce a sample mean that is within
3 standard deviations of the population mean.

So, t = −0.421 (ie. half a standard deviation below the population mean) is a pretty
typical result when µ = 1.4.

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 32 / 33
Appendix: A tip for understanding t−tests

This means that a t−statistic is like a z−score for the sample mean (when µ = 1.4).
That is, it roughly follows the “68-95-97.5% Rule”:

There is a 68% chance that your data will produce a sample mean that is within
1 standard deviation of the population mean.
There is a 95% chance that your data will produce a sample mean that is within
2 standard deviations of the population mean.
There is a 97.5% chance that your data will produce a sample mean that is within
3 standard deviations of the population mean.

So, t = −0.421 (ie. half a standard deviation below the population mean) is a pretty
typical result when µ = 1.4. In other words, we have no compelling evidence against H0 .

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 32 / 33
Appendix: A tip for understanding t−tests

This means that a t−statistic is like a z−score for the sample mean (when µ = 1.4).
That is, it roughly follows the “68-95-97.5% Rule”:

There is a 68% chance that your data will produce a sample mean that is within
1 standard deviation of the population mean.
There is a 95% chance that your data will produce a sample mean that is within
2 standard deviations of the population mean.
There is a 97.5% chance that your data will produce a sample mean that is within
3 standard deviations of the population mean.

So, t = −0.421 (ie. half a standard deviation below the population mean) is a pretty
typical result when µ = 1.4. In other words, we have no compelling evidence against H0 .

On the other hand, t = 2.162 is (roughly) 2 standard deviations above the mean so it’s
an unusual result when µ = 1.4.

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 32 / 33
Appendix: A tip for understanding t−tests

This means that a t−statistic is like a z−score for the sample mean (when µ = 1.4).
That is, it roughly follows the “68-95-97.5% Rule”:

There is a 68% chance that your data will produce a sample mean that is within
1 standard deviation of the population mean.
There is a 95% chance that your data will produce a sample mean that is within
2 standard deviations of the population mean.
There is a 97.5% chance that your data will produce a sample mean that is within
3 standard deviations of the population mean.

So, t = −0.421 (ie. half a standard deviation below the population mean) is a pretty
typical result when µ = 1.4. In other words, we have no compelling evidence against H0 .

On the other hand, t = 2.162 is (roughly) 2 standard deviations above the mean so it’s
an unusual result when µ = 1.4. In other words, we probably do have compelling evidence
against H0 .

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 32 / 33
 Using STUDYSmarter Resources

This resource was developed for UWA students by the STUDYSmarter team for the
numeracy program. When using our resources, please retain them in their original form
with both the STUDYSmarter heading and the UWA crest.

Inferential Statistics
((mα+hs)Smart Workshop(testing
Semesterhypotheses)
2, 2016) Contents Prev Next 33 / 33