Hypothesis Testing Introduction by ubb16013

VIEWS: 15 PAGES: 15

• pg 1
```									Hypothesis Testing:
Introduction
Part 1 of 5
Lecture 27
Section 11.1

1
ECO220Y: Overview
First Steps:     Theory:         Inference:

Form a
Research       Probability
Question                        Estimation
Distribution of
Data         a Random
Collection       Variable
Hypothesis
Describe                         Testing
the          Sampling
Sample        Distributions

2
Hypothesis Testing Overview
• Make hypotheses about the unknown
population parameters
– Ex: Hypothesize that population mean is 20
• Collect data and compute sample statistics
• Conduct statistical tests of hypotheses
– Use sample statistics, sample size, sampling
distributions
– What does evidence say about hypotheses?

3
Analogy: Parameters and Trial
• Population parameter     • Guilt of the defendant
is unknown                 is unknown
• Make hypotheses          • Make hypotheses on
• Collect a sample from    • Collect evidence
the population and         relevant to the
calculate statistics       determination of guilt
– Observe entire pop.?     – Complete evidence?
• Test hypotheses          • Test hypotheses

4
Two Hypotheses
• Null hypothesis (H0): Initial presumption
about the unknown parameter, which is
not based on evidence (data)
• Alternative or Research Hypothesis (H1):
Can be proved based on evidence (data)
• Example:
– H0: Defendant is innocent
– H1: Defendant is guilty

5
Jury Makes an Inference
• H0: Person is innocent; H1: Person is guilty
• Acquit: insufficient evidence to infer guilt
– Fail to reject presumption of innocence (H0)
• Does this mean there is no evidence of guilt?
• If you are acquitted does that prove innocence?
• Does failing to reject H0 mean it is true?
• Convict: enough evidence to infer guilt (H1)
– Reject presumption of innocence (H0)
• Jury convinced of guilt beyond a reasonable doubt
6
Setting Up Hypotheses
• Usual legal standard:       • If legal standard:
“Innocent until proven        “Guilty until proven
guilty beyond a               innocent beyond a
reasonable doubt”             reasonable doubt”
– H0:                         – H0:
– H1:                         – H1:
– If no evidence, what         – If no evidence, what
must the jury do:              must the jury do:
acquit or convict?             acquit or convict?
H0 is initial presumption about the population:
it is not based on any evidence               7
Take Notice of Language
• In statistics, like in a trial by jury, we either:
– Reject the null hypothesis in favor of the
research hypothesis
– Or fail to find enough evidence to reject the
null hypothesis (find insufficient evidence to
support the research hypothesis)
• What we cannot do with a sample:

X
– “Prove” the null hypothesis
– “Accept” the null hypothesis
8
R. A. Fisher
“In relation to any experiment we may speak of this
hypothesis as the ‘null hypothesis,’ and it should be
noted that the null hypothesis is never proved or
established, but is possibly disproved, in the course
of experimentation. Every experiment may be said
to exist only in order to give the facts a chance of
disproving the null hypothesis.”
H1 is important hypothesis: the research
hypothesis is what we hope to show is true
Asymmetry: cannot infer H0 is true but
can reject H0 and infer H1 is true         9
Type I and Type II Errors
• Type I Error: Reject a
true null hypothesis                 Guilty    Innocent
– Innocent defendant
wrongly convicted
– DNA tests exonerate                No        Type I
Convict
• Type II Error: Fail to               Error      Error
reject a false null hyp.
– Guilty defendant
wrongly set free                  Type II     No
Acquit
Error     Error
– OJ Simpson?
10
Type I and Type II Errors
• P(Type I Error) = α
• α is the significance                Guilty    Innocent
level of the statistical
test
• P(Type II Error) = β       Convict
No        Type I
Error      Error
• What is probability
jury makes no errors
(reaches the right                   Type II     No
Acquit
conclusion)?                          Error     Error

11
Statistician Chooses α
• Researcher chooses significance level (α)
– Researcher selects the probability of Type I
error that is tolerable for the question at hand
• For example, if a researcher chooses a
conventional significance level α = 0.05
– 5% of the time the researcher will reject a true
null hypothesis
• What kind of error is this: Type I or Type II?
• What can be done to reduce this error?
12
α and β
• Decreasing α (probability of Type I error)
increases β (probability of Type II error)
– If raise burden of proof (decrease α) so as not
to convict an innocent person we increase the
chance we let a guilty one go free (increase β)
– If we lower the burden of proof (increase α) so
that we can “put criminals behind bars”
(decrease β) we increase the chance we send
innocent people to jail

13
Setting Up Hypotheses
• Inference about unknown parameter θ :
– H0: θ   = some number (Ex. H0: θ   = 25)
– H1: θ   ≠ some number (Ex. H1: θ   ≠ 25)
– H1: θ   > some number (Ex. H1: θ   > 25)
– H1: θ   < some number (Ex. H1: θ   < 25)
• Show that the demand for apples is elastic
– H0:
– H1:

14
Ex: Hotel Profit Margins
• Economist wants to                          Histogram of Hotel Data
show hotel industry                                 n = 25

0 5 101520
profit margins are on

Density
average negative in
April 2005
– H0: μ = 0                               -.12       -.06         0
– H1: μ < 0                               April 2005 Profit Margin

• Random sample: 25
What should we infer?