You are on page 1of 3

By inverting these expression, it is possible to derive 100 (1 )% confidence intervals for

x y , x2 / y2
p py
and x
. Theorems 8.5,8.6, and 8.7 state the results. The proofs are very
straightforward and will be left as exercises.
THEOREM 8.5 let X1, X2 , Xn ~ N

and let the Xs and Ys be independent. Let s


2
x

be the pooled standard deviation. Then a 100 1 % confidence interval for


is

1 1
1 1
x y t / 2 ,n m 2 s p

,
x

/
2
,
n

2
s

n
m
n
m
given by.
.
Case study
Occasionally in forensic medicine, or in the aftermath of a bad accident, identifying the sex
of a very difficult task. In some of these cases, dental structure provides a useful criterion,
since individual teeth will remain in good condition long after other tissues have deteriorated.
Furthermore, studies have shown that female teeth and male teeth have different physical and
chemical characteristics.
The extent to which X-rays can penetrate tooth enamel, for instance, is different for men that
it is for women. Listed in Table 8.6 are spectropenetration gradients for eight female teeth
and eight male teeth (53). These number are measures of the rate of change in the amount of
X-ray penetration through a 500-micron section of tooth enamel at a wavelength of 600 m
as opposed to 400 m .
TABLE 8.6 Enamel spectropenetration gradients
4.9
4.8
5.4
5.3
5.0
3.7
5.5
4.1
5.4
5.6
6.6
4.0
6.3
3.6
4.3
5.0

x y

Let x and y be the population means of the spectropenatration gradient associated with
male teeth and with female teeth, respectively. Note that
8

x 43.4. x
i 1

i 1

2
1

239.32,

From which
x

43.4
5 .4
8

And
s x2

8(239.32) ( 43.4) 2
0.55.
8(7)

Similary,
8

y i 36.1.
i 1

So that

y
i 1

2
1

166.95.

36.1
4.5
8

And
s y2

8(166.95) (36.1) 2
0.58.
8(7)

Therefore, the pooled standar deviation is e qual to 0.75 :


7(0.55) 7(0.58)
s p
0.565 0.75.
882
We know that the ratio
Will be approximated by a Student t curva with 14 degrees of freedom. 0..0.0.0.0.
y
Since t0.025.14 = 2.14, the 95% confidance interval for x
is given by

1 1
1 1
x y 2.14 s p

,
x

1
.
14
s

p

8
8
8
8

5.4 4.5 2.14 0.75 0,25, 5.4 4.5 2.14 0,75 0.25
0.1,1.7

Comment. Here the 95% confidance interval does not include the value 0. This means that
head we tested.
H0 : x y
Versus
H1 : x y

At the = 0.05 level of significance, H0 would have been rejected.


QUESTION 8.5.1 Refer to the data on hermoluminescance described in Case Study 8.2
y
Construct a 90% confidance interval for x
X , X ..., X n N x , x2 and Y1 , Y2 ..., Yn N x , y2 and let the Xs and Ys
THOREM 8.6 Let 1 2
2 / 2
be independent. A 100 1 % confidance interval for the variance ratio, x y , is given by
sx2

sx2

F
F1 a / 2, m 1, n 1,
a
/
2
,
m

1
,
n

1
,
2
2
s

sy
y

CASE STUDY
The easiest way to measure the movement, or flow, of a glacier is with a camera. Firs a set of
reference points is marked off at vaious sites near the glaciers edge. Then these points, along
with the glacier, are photographed for an airplane. The problem is this : How long should the
time interval be between photographs? If too short a period has elapsed, the glacier will not
have moved very far and the errors associated with the photographic technique will be
reletively large. If too long a period has elapsed, parts of the glacier might be deformed by the
surrounding terrain, an eventuality that could introduce substantial variability into the point
to point velocity estimates.

In this example, two sets of flow retes for the Antarctics Hoseason Glacier are
examined (113), one besed on photographs take thre years apart, the other, five years apart.
Both sets of date were taken under identical conditions. Also, on the basis of other
considerations, it can be assumed that the true flow rate for the glacier was constant for the
eight years in question. The data are listed in Table 8.7
TABLE 8.7 Enamel retes estimated
For the Hoseason Glacier (meters per day)
Thee Year Span (X1)
Five-Year Span (Y1)
0.73
0.72
0.76
0.74
0.75
0.74
0.77
0.72
0.73
0.72
0.75
0.74
The objective here is to assess the relative variabilities associated with the three and
five year time periods. One way to do this assuming the data to be normal is to
counstruct, say, a 95% confidence interval for the variance ratio. If that interval does not
contain the value 1, we infer that the two time periods lead to flow rate estimates of
significantly different precision.
Form Table 8.7
7

y i 5.23
i 1

and

x
i 1

2
i

3.9089,

So that

s y2

7(3.9089) (5.23) 2
0.000224.
7 ( 6)

Similary,
5

y i 3.64
i 1

and

x
i 1

2
i

2.6504,

Miking

5(2.6504) (3.64) 2
0.000120
5(4)
The two critical values come from Table A.4 in the Appendix :
F0,025.b.6 0.109 and F0.975.4.6 6.23
s 2y

When all of these quantities are substituted into the statement of Theorem 8.6 we get a 95%
2 /2 :
confidence interval for x y
0.000224
0.000224

0.109.
6.23 0.203,11.629.

0.0000120
0.000120

Thus, although three year data had a larger sample variance than the five year data, no
conclusions can be drawn about the true