Statistical Inference for FE
Professor S. Kou, Department of IEOR, Columbia University
Lecture 3. Hypothesis Testing
1
Introduction to Hypothesis Testing
1.1
Con
f
dence Intervals and Hypothesis Testing
How to test the hypothesis
H
0
:
μ
=3
versus the alternative hypothesis
H
a
:
μ
6
=3
. Assume we have large sample size. One possible solution is simply
by using the idea of con
f
dence interval. More precisely,
f
rst we construct a
(
1
−
α
) c.i. for
μ
:
¯
X
±
z
α
/
2
S
√
n
.
Then we reject
H
0
if 3 is not within the (
1
−
α
) c.i; otherwise, we cannot
reject it as a plausible value for
μ
. In other words, we reject
H
0
if
3
<
¯
X
−
z
α
/
2
S
√
n
or
3
>
¯
X
+
z
α
/
2
S
√
n
.
The above formula can be rewritten as
¯
X
−
3
S/
√
n
<
−
z
α
/
2
or
¯
X
−
3
S/
√
n
>z
α
/
2
.
By using such a test, we may make a wrong decision with probability
α
,
if we reject the null hypothesis when it is indeed true. This is called type
I error. There is another possible error, which is the error we fail to reject
the null hypothesis when it is false. This is called type II error.
1.2 Connection between Hypothesis Testing and Jury Trials
The jury trail of an accused murderer is very similar to the statistical hy
pothesis testing.
1.
H
0
In a jury trial, the null hypothesis is that the accused is not guilty.
The statusquo hypothesis in the American system of justice is innocence,
which is assumed to be true until proven otherwise.
2.
H
a
. The alternative hypothesis is guilt, which is accepted only when
su
ﬃ
cient evidence exists to establish its truth.
3. Test Statistics. The test statistics in a trial is the
f
na
lvo
teo
fthe
jury, i.e., the number of the jury members who vote “guilty.”
4. Rejection Region. In a murder trial the jury vote must be unanimous
in favor of guilt before the null hypothesis of innocence is rejected in favor
of the alternative hypothesis of guilt. Thus, for a 12member jury trial, the
rejection region is
x
This preview has intentionally blurred sections. Sign up to view the full version.
View Full Document5. Assumption. The primary assumption made in jury trail concerns the
method of selecting the jury. The jury is assumed to represent a random
sample of citizens who have no prejudice concerning the case.
6. Sample and calculate the test statistics. Sample data are generated
through the trial jury’s deliberations. The
f
nal vote of the jury is analogous
to the calculation of the test statistics.
7.Conclusion. If the vote of the jury is unanimous in favor of guilt, the
null hypothesis of innocence is rejected and the court concludes that the
accused murderer is guilty.
8. Although the court does not, in general, know the probability, the
type one error, that the conclusion is in error, the system relies on the belief
that the value is made very small by requiring a unanimous vote before guilt
is concluded.
9. Any vote other than a unanimous one for guilt results in the court
This is the end of the preview.
Sign up
to
access the rest of the document.
 Spring '10
 kou
 Null hypothesis, Hypothesis testing, Statistical hypothesis testing, Ri, mle

Click to edit the document details