Basic Statistics

Introdu
tion
Estimation - Conden e Interval
Hypothesis testing - p value
Determination of sample size
Summary
Basi statisti s: p value and onden e interval

Nguyen Quang Vinh
February, 2012
Nguyen Quang Vinh
Introdu tion
Summary
Outline
1
Introdu tion
Statisti s
Obje tives
Statisti s

Estimation
A point estimate
An interval estimate
Interpretation a onden e interval

Hypothesis testing
Hypotheses
p value
The hypothesis testing pro edure
Power of Test
Nguyen Quang Vinh
Introdu tion
Summary
Statisti s
Obje tives Statisti s
Outline
1
Introdu tion
Statisti s
Obje tives
Statisti s

Estimation
A point estimate

Hypothesis testing
Hypotheses
p value
Power of Test
Nguyen Quang Vinh
Introdu tion
Summary
Statisti s
Statisti s
Statisti s:
s ien e of data
study of un ertainty
Biostatisti s: data from: Medi ine, Biologi al s ien es
(business, edu ation, psy hology, agri ulture, e onomi s...)
Modern so iety:
Reading
Writing
Statisti al thinking: to make the strongest possible on lusions
from limited amounts of data.
Nguyen Quang Vinh
Introdu tion
Summary
Statisti s
Outline
1
Introdu tion
Statisti s
Obje tives
Statisti s

Estimation
A point estimate

Hypothesis testing
Hypotheses
p value
Power of Test
Nguyen Quang Vinh
Introdu tion
Summary
Statisti s

Obje tives:
(1) Organize & summarize data
(2) Rea h inferen es: sample
Statisti s:
population
Des riptive statisti s(1)
Inferential statisti s: drawing of inferen es(2)
Estimation (95% C.I.)

Hypothesis testing rea hing a de ision (p value)
Parametri statisti s
Non-parametri statisti s << Distribution-free statisti s
Modeling, Predi ting
Nguyen Quang Vinh
Introdu tion
Summary
Estimation
A point estimate
Outline
1
Introdu tion
Statisti s
Obje tives
Statisti s

Estimation
A point estimate

Hypothesis testing
Hypotheses
p value
Power of Test
Nguyen Quang Vinh
Introdu tion
Summary
Estimation
A point estimate
Why estimation?
Two reasons:
Innite populations: in apable of omplete examination

Finite populations: ost, time
In addition, estimation an help not to defer a on lusion, until
the entire population has been observed
Nguyen Quang Vinh
Introdu tion
Summary
Estimation
A point estimate
Estimation of
mean(s):
a single population mean

the dieren e between two population means: unpaired, paired
proportion(s):
a single population proportion

the dieren e of two population proportions
varian e(s):
a single population varian e

the ratio of two population varian es
Nguyen Quang Vinh
Introdu tion
Summary
Estimation
A point estimate
An estimation of these parameters
An estimation of these parameters:

Point estimate
Interval estimate
Nguyen Quang Vinh
Introdu tion
Summary
Estimation
A point estimate
Outline
1
Introdu tion
Statisti s
Obje tives
Statisti s

Estimation
A point estimate

Hypothesis testing
Hypotheses
p value
Power of Test
Nguyen Quang Vinh
Introdu tion
Summary
Estimation
A point estimate
A point estimate
Estimator Parameter
In many ases, a parameter may be estimated by more than one

estimator.
Example:
Sample mean estimate population mean

Sample median estimate population mean
Nguyen Quang Vinh
Introdu tion
Summary
Estimation
A point estimate
The riteria of good estimator (opt.)
(1) E (x )
= without
systemati error
E (x ) is alled systemati error
(2) Mean square error

E (x )2
E (x
)2
must be small in omparison to
must be small
Nguyen Quang Vinh
Introdu tion
Summary
Estimation
A point estimate
Outline
1
Introdu tion
Statisti s
Obje tives
Statisti s

Estimation
A point estimate

Hypothesis testing
Hypotheses
p value
Power of Test
Nguyen Quang Vinh
Introdu tion
Summary
Estimation
A point estimate
In general, an interval estimate is obtained by the formula:

estimator
(reliability oe ient) x (standard error)
What is dierent is the sour e of the reliability oe ient:

In parti ular, when sampling is from a normal distribution with
known varian e, an interval estimate for
as: x
z /2 x
Nguyen Quang Vinh
may be expressed
Introdu tion
Summary
Estimation
A point estimate
Outline
1
Introdu tion
Statisti s
Obje tives
Statisti s

Estimation
A point estimate

Hypothesis testing
Hypotheses
p value
Power of Test
Nguyen Quang Vinh
Introdu tion
Summary
Estimation
A point estimate
How to interpret the interval given by this expression
In repeated sampling 100(1 )% of all intervals of the form

will in the long run in lude the population mean,
The quantity
(1 ), is alled the onden e oe ient &

z /2 x , is alled the onden e interval for
The interval x
The most frequently used values are: .90, .95, .99, whi h have
asso iated reliability fa tors, respe tively, of 1.645, 1.96, 2.58
Nguyen Quang Vinh
Introdu tion
Summary
Estimation
A point estimate
The pra ti al interpretation
We are 100(1 )% ondent that the single omputed
interval x
z /2 x
ontains the population mean,
Example: ...
E = margin error = maximum error = pra ti al / lini al

a eptable error:
E
= z /2 x = z /2 n
Nguyen Quang Vinh
Introdu tion
Summary
Hypothesis testing
Hypotheses
p value
Power of Test
Outline
1
Introdu tion
Statisti s
Obje tives
Statisti s

Estimation
A point estimate

Hypothesis testing
Hypotheses
p value
Power of Test
Nguyen Quang Vinh
Introdu tion
Summary
Hypothesis testing
Hypotheses
p value
Power of Test
Why hypothesis testing
Hypothesis (H.): a statement on erns about some one or

more populations
Testing hypothesis: to aid resear her in rea hing a de ision
on erning a population by examining a sample from that
population
Nguyen Quang Vinh
Introdu tion
Summary
Hypothesis testing
Hypotheses
p value
Power of Test
Hypothesis testing for

mean(s)
a single population mean

the dieren e between two population means: unpaired, paired
proportion(s)
a single population proportion
unpaired: a small sample, a su iently large sample

paired
the dieren e of two population proportions

varian e(s)
a single population varian e

the dieren e of two population varian es
Nguyen Quang Vinh
Introdu tion
Summary
Hypothesis testing
Hypotheses
p value
Power of Test
Outline
1
Introdu tion
Statisti s
Obje tives
Statisti s

Estimation
A point estimate

Hypothesis testing
Hypotheses
p value
Power of Test
Nguyen Quang Vinh
Introdu tion
Summary
Hypothesis testing
Hypotheses
p value
Power of Test
Two types of hypotheses
(1) Resear h Hypotheses:

The onje ture or supposition
It may be the results of years of observation
Resear h H. leads dire tly to Statisti al H.
(2) Statisti al Hypotheses: Hypotheses are stated in su h a way
that they may be evaluated by appropriate statisti al te hniques.
HO
HA
Nguyen Quang Vinh
Introdu tion
Summary
Hypothesis testing
Hypotheses
p value
Power of Test
Statisti al Hypotheses
HO
The H O is the hypothesis that is tested

The H O should ontain either =, ,
(The statement on erns about some one or more population's

parameters in term of equality or inequality)
HA
What we hope or expe t to be able to on lude as a result of

the test usually should be pla ed in the H A
The H O
&
HA are omplementary
One-sided vs. Two-sided Hypothesis Tests (opt.)
Nguyen Quang Vinh
Introdu tion
Summary
Hypothesis testing
Hypotheses
p value
Power of Test
Notes
Neither hypothesis testing nor statisti al inferen e, in general,

leads to proof a hypothesis
It merely indi ates whether the hypothesis is supported or not
supported by the available data
When we fail to reje t the H O , we do not say that it is true,
but that it may be true
When we speak of a epting a H O , we have this limitation
in mind & do not wish to onvey the idea that a epting
implies proof
Nguyen Quang Vinh
Introdu tion
Summary
Hypothesis testing
Hypotheses
p value
Power of Test
Outline
1
Introdu tion
Statisti s
Obje tives
Statisti s

Estimation
A point estimate

Hypothesis testing
Hypotheses
p value
Power of Test
Nguyen Quang Vinh
Introdu tion
Summary
Hypothesis testing
Hypotheses
p value
Power of Test
p value
Test statisti =p value
General formula:
hypothesizedparameter
= relevantstatisti
S .E .oftherelevantstatisti
= x0
Teststatisti
Example: z
Test statisti
p value
De ision maker, sin e the de ision to reje t or not to reje t the

H O depends on the magnitude of the test statisti
Nguyen Quang Vinh
Introdu tion
Summary
Hypothesis testing
Hypotheses
p value
Power of Test
De ision rule for a reje tion or not the
HO
= type I error = level of signi an e (say, .01, .05, .10)

= type II error (say, .05, .10, .20)
When we reje t a H O p
< ,
risk of ommitting a type I
error, reje ting a true H O
When we fail to reje t a H O : risk of ommitting a type II

error, a epting a false H O
Nguyen Quang Vinh
Introdu tion
Summary
Hypothesis testing
Hypotheses
p value
Power of Test
Type I & Type II error

Conditions under
which type I & type
II errors may be
committed (the four
possibilities)
The results
in the study
sample
Conclusion:
Actual Situation
(Truth in the population)
Ho false
Ho true
Reject
Ho
Correct
decision
Type I
error
Fail to
reject Ho
Type II
error
Correct
decision
Nguyen Quang Vinh
Introdu tion
Summary
Hypothesis testing
Hypotheses
p value
Power of Test
Testing Hypothesis Reje ted or not reje ted
HO
In the testing pro ess the H O either is reje ted or is not

reje ted
If H O is not reje ted, we will say that the data on whi h the
test is based do not provide su ient eviden e to ause
reje tion
If the testing pro ess leads to reje tion, we will say that the
data at hand are not ompatible with the H O , but are
supportive of some other hypothesis & may be designated by
H A (H A a ontradi tion statement of H O )
Nguyen Quang Vinh
Introdu tion
Summary
Hypothesis testing
Hypotheses
p value
Power of Test
Outline
1
Introdu tion
Statisti s
Obje tives
Statisti s

Estimation
A point estimate

Hypothesis testing
Hypotheses
p value
Power of Test
Nguyen Quang Vinh
Introdu tion
Summary
Hypothesis testing
Hypotheses
p value
Power of Test
The Five-Step pra ti al pro edure for Hypothesis Testing

(opt.)
Step 1: Set up H O , H A
1. Data: The nature of the data ( lassi ation)

2. Assumptions: The normality of the population distribution,
equality of varian es, independen e of samples. . .
3. Hypotheses: H O , H A
Step 2: Dene the test statisti
4. Test statisti
5. Distribution of the Test Statisti
Nguyen Quang Vinh
Introdu tion
Summary
Hypothesis testing
Hypotheses
p value
Power of Test
The Five-Step pra ti al pro edure for Hypothesis Testing,

ont. (opt.)
Step 3: Dene a reje tion region: having determined a value
for
6. De ision rule
Step 4:
7. Cal ulate the value of the test statisti , and ompare it with
the a eptan e & reje tion regions that have already been
spe ied.
8. State our de ision: to reje t H O or to fail to reje t H O
Step 5:
9. Give a on lusion: this statement should be free of

statisti al jargon & should merely summarize the results of the
analysis.
Nguyen Quang Vinh
Introdu tion
Summary
Hypothesis testing
Hypotheses
p value
Power of Test
Outline
1
Introdu tion
Statisti s
Obje tives
Statisti s

Estimation
A point estimate

Hypothesis testing
Hypotheses
p value
Power of Test
Nguyen Quang Vinh
Introdu tion
Summary
Hypothesis testing
Hypotheses
p value
Power of Test
The Power of a Statisti al Test
(opt.)
The probability of a type II error,
b,
has remained a phantom:
we know it is there,
but we don't know what it is
One thing we an say is that: a wide C.I. for
means that the
orresponding 2-tailed test of Ho versus HA has a large han e

of failing to reje t a false Ho; that is
Nguyen Quang Vinh
is large.
Introdu tion
Summary
Determining b
(opt.)
b = P(fail to reje t H O
1 - b = P(reje ting
1 -
Hypothesis testing
Hypotheses
p value
Power of Test
when H O is false)
H O when H O is false)
represents the probability of making a orre t de ision in
the event that H O is false

Sin e we like
to be small, that is we prefer 1 -
The value of 1 -
to be large
is referred to as the power of test
Nguyen Quang Vinh
Introdu tion
Summary
Power of test
Hypothesis testing
Hypotheses
p value
Power of Test
(opt.)
when = B
z2
z1
P(Z<z2)
P(Z>z1)
Power of test = P(Z>z1) + P(Z<z2)
Nguyen Quang Vinh
Introdu tion
Summary
(opt.)
Estimating a onden e interval

Testing a hypothesis
Nguyen Quang Vinh
Introdu tion
Summary
Summary
1. Introdu tion to Statisti s

2. Estimation - onden e interval
3. Hypothesis testing - p value
Nguyen Quang Vinh

Basic Statistics

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Basic Statistics

Uploaded by

Copyright:

Available Formats

Introdu

Basi statisti s: p value and onden e interval

Nguyen Quang Vinh

Basi statisti s: p value and onden e interval

Estimation - Conden e Interval

Hypothesis testing - p value

Nguyen Quang Vinh

Basi statisti s: p value and onden e interval

Estimation - Conden e Interval

Hypothesis testing - p value

Nguyen Quang Vinh

Basi statisti s: p value and onden e interval

Nguyen Quang Vinh

Basi statisti s: p value and onden e interval

Estimation - Conden e Interval

Hypothesis testing - p value

Nguyen Quang Vinh

Basi statisti s: p value and onden e interval

Obje tives Statisti s

Des riptive statisti s(1)

Inferential statisti s: drawing of inferen es(2)

Estimation (95% C.I.)

Modeling, Predi ting

Nguyen Quang Vinh

Basi statisti s: p value and onden e interval

Estimation - Conden e Interval

Hypothesis testing - p value

Nguyen Quang Vinh

Basi statisti s: p value and onden e interval

Innite populations: in apable of omplete examination

Nguyen Quang Vinh

Basi statisti s: p value and onden e interval

a single population mean

a single population proportion

a single population varian e

Nguyen Quang Vinh

Basi statisti s: p value and onden e interval

An estimation of these parameters

An estimation of these parameters:

Nguyen Quang Vinh

Basi statisti s: p value and onden e interval

Estimation - Conden e Interval

Hypothesis testing - p value

Nguyen Quang Vinh

Basi statisti s: p value and onden e interval

In many ases, a parameter may be estimated by more than one

Sample mean estimate population mean

Nguyen Quang Vinh

Basi statisti s: p value and onden e interval

The riteria of good estimator (opt.)

E (x ) is alled systemati error

(2) Mean square error

must be small in omparison to

Nguyen Quang Vinh

Basi statisti s: p value and onden e interval

Estimation - Conden e Interval

Hypothesis testing - p value

Nguyen Quang Vinh

Basi statisti s: p value and onden e interval

In general, an interval estimate is obtained by the formula:

(reliability oe ient) x (standard error)

What is dierent is the sour e of the reliability oe ient:

Nguyen Quang Vinh

Basi statisti s: p value and onden e interval

Basi statisti s: p value and onden e interval

Basi statisti s: p value and onden e interval

Estimation - Conden e Interval

Basi statisti s: p value and onden e interval

Estimation - Conden e Interval

Basi statisti s: p value and onden e interval

Basi statisti s: p value and onden e interval

Estimation - Conden e Interval

Basi statisti s: p value and onden e interval

Basi statisti s: p value and onden e interval

Estimation - Conden e Interval

Basi statisti s: p value and onden e interval

Innite populations: in apable of omplete examination

Basi statisti s: p value and onden e interval

Basi statisti s: p value and onden e interval

Basi statisti s: p value and onden e interval

Estimation - Conden e Interval

Basi statisti s: p value and onden e interval

Basi statisti s: p value and onden e interval

Basi statisti s: p value and onden e interval

Estimation - Conden e Interval

Basi statisti s: p value and onden e interval

(reliability oe ient) x (standard error)

What is dierent is the sour e of the reliability oe ient:

Basi statisti s: p value and onden e interval

Estimation - Conden e Interval

Basi statisti s: p value and onden e interval

(1 ), is alled the onden e oe ient &

Basi statisti s: p value and onden e interval

We are 100(1 )% ondent that the single omputed

Basi statisti s: p value and onden e interval

Estimation - Conden e Interval

Basi statisti s: p value and onden e interval

Basi statisti s: p value and onden e interval

unpaired: a small sample, a su iently large sample

the dieren e of two population proportions

Basi statisti s: p value and onden e interval

Estimation - Conden e Interval

Basi statisti s: p value and onden e interval

Basi statisti s: p value and onden e interval

Basi statisti s: p value and onden e interval

Basi statisti s: p value and onden e interval

Estimation - Conden e Interval

Basi statisti s: p value and onden e interval

Basi statisti s: p value and onden e interval

= type I error = level of signi an e (say, .01, .05, .10)

Basi statisti s: p value and onden e interval

Basi statisti s: p value and onden e interval

Basi statisti s: p value and onden e interval

Estimation - Conden e Interval