Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
OPRE 6301
Introduction . . .
1
Two Means, Independent Samples . . .
H0 : µ1 = µ2 (or µ1 − µ2 = 0)
H1 : µ1 =
6 µ2 (or µ1 − µ2 6= 0)
2
Such a test, however, is hardly useful because in most
applications the population variances are unknown.
3
Equal Variances
The idea here is that if the two variances are the same,
we can pool data for both samples together to produce a
pooled variance estimate, defined as:
2 2
(n 1 − 1)s1 + (n 2 − 1)s2
s2p ≡ . (3)
n1 + n2 − 2
With s2p replacing both s21 and s22 in (2), we then have
4
It can be shown that if both populations are normally dis-
tributed, then the statistic in (4) follows the t distribution
with n1 + n2 − 2 degrees of freedom.
5
Unequal Variances
When the two population variances are not the same, the
statistic in (2) is neither normally nor t distributed. (In
fact, the exact sampling distribution in this case has not
yet been found!) However, it has been shown that the
sampling distribution can be well approximated by a t
distribution with the following degrees of freedom:
2
s21 s22
+
n1 n2
ν= 2 2 2 2 . (5)
(s1/n1) (s /n2)
+ 2
n1 − 1 n2 − 1
6
The Excel Data Analysis tool
A Recommendation
7
Two Means, Paired Samples . . .
8
Even if two populations are not “naturally” coupled, it
is often desirable to design experiments that are “con-
trolled” by some factors. The intent is to reduce “noise”
and hence variability in the sample differences. A good
example of this is that when testing the performance of
a new piece of equipment, the same operator should test
both the new machine and the old machine. This serves
to reduce potential noise coming from variability in oper-
ator skills, and is often referred to as “controlling for the
operator” (or for whatever).
Formally, . . .
9
Consider n pairs of observations (X1i, X2i) for i = 1, 2,
. . . , n. Since the data is paired, we can define for each
pair:
H0 : µD = 0
H1 : µD =
6 0
10
Now, define the test statistic:
D̄ − 0
t=p 2 , (7)
sD /n
where n
1 X
D̄ ≡ Di (8)
n i=1
and n
1 X
s2D ≡ (Di − D̄)2. (9)
n − 1 i=1
11
Two Proportions . . .
12
Consider the “generic framework”:
H0 : p1 = p2 (or p1 − p2 = 0)
H1 : p1 =
6 p2 (or p1 − p2 6= 0)
13
It can be shown that if the two samples are taken inde-
pendently, then the variance of our sample statistic is
p̂1 − p̂2
z=s , (11)
p̂1(1 − p̂1) p̂2(1 − p̂2)
+
n1 n2
which, when both n1p̂1 and n2p̂2 are greater than 5, can
be shown to approximately follow the standard normal
distribution.
14
The test procedure is therefore:
15