Math3806 Lecture Note 4 Appendix

Heng Peng

March 9, 2020
P2. The selection of significant level α is relatively subjective.
Accepting the null hypothesis cannot say that the null
hypothesis is right, and only can say there is no strong
evidence to reject it.
P3. Two degree of freedoms of Hotelling’s T 2 or F distribution: p
is from the sample mean, n − p is the degree of freedom of
the sample covariance matrix.
P5. Example 4.1.
6+10+8
     
x̄1 3 8
x̄ = = 9+6+3 =
x¯2 3 6

(6 − 8)2 + (10 − 8)2 + (8 − 8)2

s11 = = 4,
3−1
(9 − 6)2 + (6 − 6)2 + (3 − 6)2
s22 = = 9,
3−1
(6 − 8)(9 − 6) + (10 − 8)(6 − 6) + (8 − 8)(3 − 6
s11 = = −3
3−1
I Example 4.1. (continuous) Hence
1 1
     
4 −3 1 9 3
S= , and S−1 = = 3
1
9
4
−3 9 4 · 9 − (−3)(−3) 3 4 9 27

Hence
1 1
− 92
    
8−9 7
T 2 = 3[8 − 9, 6 − 5] 3
1
9
4 = 3[−1, 1] 1 =
9 27
6−5 27 9

and Since n = 3 and p = 2, T 2 has the distribution of a

(3 − 1)2
F2,3−2 = 4F2,1
3−2
random variable.
P5. Example 4.2. n = 20, p = 3,
   
4.640 2.879 10.010 −1.810
x̄ =  45.400  , S =  10.010 199.788 −5.640 
9.965 −1.810 −5.640 3.628
and  
.586 −.022 .258
S−1 =  −.022 .006 −.002  .
.258 −.002 .402

T2 = 20[4.640 − 4, 45.400 − 50, 9.965 − 10]

  
.586 −.022 .258 4.640 − 4
×  −.022 .006 −.002   45.400 − 50  = 9.74.
.258 −.002 .402 9.965 − 10
The critical value is
(n − 1)p 19 · 3
Fp,n−p (.10) = F3,17 (.10) = 3.353 · 2.44 = 8.18
n−p 17
But T 2 = 9.74 > 8.18, and hence consequently, we reject H0 at the
10% level of significance.
P9. Example 4.3. n = 42, p = 2.
     
.563 .0144 .0117 −1 203.018 −163.391
x̄ = ,S = ,S =
.603 .0117 .0146 −163.391 200.228

The eigenvalue and eigenvector pairs for S are

λ1 = .026, eT T
1 = [.704, .710], and λ2 = .002, e2 = [−.710, .704].

The 95% confidence ellipse for µ consists of all values (µ1 , µ2 )

satisfying
  
203.018 −163.391 .563 − µ1
42[.564 − µ1 , .603 − µ2 ]
−163.391 200.228 .603 − µ2
2 · 41
≤ F2,40 (.05)
40
Since F2,40 (.05) = 3.23,

42(203.018)(.564 − µ1 )2 + 42(200.228)(.603 − µ2 )2
− 84(163.391)(.564 − µ1 )(.603 − µ2 ) ≤ 6.62
I Example 4.3. continuous. If µ0 = [.562, .589]T , then

42(203.018)(.564 − .562)2 + 42(200.228)(.603 − .589)2

− 84(163.391)(.564 − .562)(.603 − .589)
= 1.30 ≤ 6.62

Hence µ0 = [.562, .589]T is in the confidence region.

Equivalently, for a test of H0 : µ = µ0 vs H1 : µ 6= µ0 at
α = 0.05 level of significance, we will not reject the null
hypothesis H0 .
I The center of the joint confidence ellipsoid is at
x̄T = [.564, .603], the axes lie along e1 and e2 with the
half-lengths of the major and minor axes are given by
s s
p p(n − 1) √ 2(41)
λ1 Fp,n−p (α) = 0.026 (3.23) = 0.64
n(n − p) 42(40)
s s
p p(n − 1) √ 2(41)
λ2 Fp,n−p (α) = 0.002 (3.23) = 0.018
n(n − p) 42(40)
P14. Example 4.4. The 95% simultaneous T 2 intervals for the two
component means are
s r s r !
p(n − 1) s11 p(n − 1) s11
x̄1 − Fp,n−p (.05) , x̄1 + Fp,n−p (.05)
(n − p) n (n − p) n
r r r r !
2(41) 0.0144 2(41) 0.0144
= .564 − 3.23 , .564 + 3.23
40 42 40 42
= (.516, .612)

s r s r !
p(n − 1) s22 p(n − 1) s22
x̄2 − Fp,n−p (.05) , x̄2 + Fp,n−p (.05)
(n − p) n (n − p) n
r r r r !
2(41) 0.0146 2(41) 0.0146
= .603 − 3.23 , .603 + 3.23
40 42 40 42
= (.555, .651)
P14. Example 4.5. n = 87, p = 3,
   
526.29 5808.06 597.84 222.03
x̄ =  54.69  , S =  597.84 126.05 23.39 
25.13 222.03 23.39 23.11

p(n − 1) 3(87 − 1) 3(86)

Fp,n−p (α) = F3,84 (.05) = (2.7) = 8.29.
n−p 87 − 3 84
Then we obtain the simultaneous confidence statements
√ √
r r
5808.06 5808.06
526.29 − 8.29 ≤ µ1 ≤ 526.29 + 8.29 ,
87 87
or 503.06 ≤ µ1 ≤ 550.12
√ √
r r
126.05 126.05
54.69 − 8.29 ≤ µ2 ≤ 54.69 + 8.29 ,
87 87
or 51.22 ≤ µ2 ≤ 58.16
√ √
r r
23.11 23.11
25.13 − 8.29 ≤ µ3 ≤ 25.13 + 8.29 ,
87 87
or 23.65 ≤ µ3 ≤ 26.61
I The simultaneous T 2 -intervals above are wider than univariate
intervals because all three must hold with 95% confidence.
P18. Example 4.6, n = 96, p = 7. From the result 4.5,
simultaneous
q 90% confidence limits are given by
x̄i ± χ7 (.10) snii , i = 1, . . . , 7 where χ27 (.10) = 12.02.
2
p

Thus, with approximately 90% confidence,

√ 5.76
28.1 ± 12.02 √ , contains µ1 , or 26.06 ≤ µ1 ≤ 30.14
96
√ 5.85
26.6 ± 12.02 √ , contains µ2 , or 24.53 ≤ µ2 ≤ 28.67
96
√ 3.82
35.4 ± 12.02 √ , contains µ3 , or 34.05 ≤ µ3 ≤ 36.75
96
√ 5.12
34.2 ± 12.02 √ , contains µ4 , or 32.39 ≤ µ4 ≤ 36.01
96
√ 3.76
23.6 ± 12.02 √ , contains µ5 , or 22.27 ≤ µ5 ≤ 24.93
96
√ 3.93
22.0 ± 12.02 √ , contains µ6 , or 20.61 ≤ µ6 ≤ 23.39
96
√ 4.03
22.7 ± 12.02 √ , contains µ7 , or 21.27 ≤ µ7 ≤ 24.13
96
P23. Example 4.7. The T 2 -statistic for testing
H0 : δ T = [δ1 , δ2 ] = [0, 0] is constructed from the difference of
paired observation
dj1 = X1j1 − X2j1 : −19, −22, −18, −27, −4, −10, −14, 17, 9, 4, −19
dj2 = X1j2 − X2j2 : 12, 10, 42, 15, −1, 11, −4, 60, −2, 10, −7

Then
d¯1
     
−9.36 199.26 88.38
d̄ = = , Sd =
d¯2 13.27 88.38 418.61

and
  
2 .0055 −.0012 −9.36
T = 11[−9.36, 13.27] = 13.6
−.0012 .0026 13.27

Taking α = .05, we find that [p(n − 1)/(n − p)]Fp,n−p (.05) =

[2(10)/9]F2,9 (0.05) = 9.47 < T 2 . Hence we reject H0 and
conclude that there is a nonzero mean difference between the
measurements of the two laboratories.
I Example 4.7 continuous. The 95% simultaneous confidence
intervals for the mean differences δ1 and δ2 are
s s
sd21 √
r
¯ (n − 1)p 199.26
δ1 : d1 ± Fp,n−p (α) = −9.36± 9.47
n−p n 11

or (−22.46, 3.74)
√
r
418.61
δ2 : 13.27 ± 9.47 , or (−5.71, 32.25).
11
The 95% simultaneous confidence interval include zero, yet
the hypothesis H0 : δ = 0 was rejected at the 5% level. What
are we to conclude ?
P27. Example 4.8. First, note that S1 and S2 are approximately
equal, so that it is reasonable to pool them. Hence
 
49 49 2 1
Spooled = S1 + S2 =
98 98 1 5

and
x̄1 − x̄2 = [−1.9, .2]T .
So the confidence ellipse is centred at [−1.9, .2]T . The
eigenvalues and eigenvectors of Spooled are obtained from the
equation
0 = |Spooled − λI| = λ2 − 7λ + 9
consequently λ1 = 5.303 and λ2 = 1.697. The corresponding
eigenvectors e1 and e2 determined from
Spooled ei = λi ei , i = 1, 2.

eT T
1 = [.290, .957], ande1 = [.957, −.290]
 I Example 4.8.continuous. By
   
1 1 2 1 1 (98)(2)
+ c = + F2,97 (.05) = .25
n1 n2 50 50 (97)

since F2,97 (.05) = 3.1. The confidence ellipse extends

s
p 1 1
 p √
λi + c 2 = λi .25
n1 n2

unit along the eigenvector ei , or 1.15 units in the e1 direction

and .65 units in the e2 .
P29. Result 4.9, Type Error: [(1/n1 + 1/n2 ) Spooled ] should be

1 1
S1 + S2 .
n1 n2
P33. Example 4.9. n1 = 271, n2 = 138, n3 = 107 and
|S1 | = 2.783 × 10−8 , |S2 | = 89.539 × 10−8 ,
|S3 | = 14.579 × 10−8 , and |Spooled | = 17.398 × 10−8 . Taking
the nature logarithms of the determinants gives
ln |S1 | = −17.397, ln |S2 | = −13.926, ln |S3 | = −15.741, and
ln |Spooled | = −15.564. Then we calculate

2(42 ) + 3(4) − 1
  
1 1 1 1
u = + + −
270 137 106 270 + 137 + 106 6(4 + 1)(3 − 1)
= 0.0133
I Example 4.9. Continuous.

M = [270 + 137 + 106](−15.564) − [270(−17.397)

+137(−13.926) + 106(−15.741)]
= 289.3

and C = (1 − .0133)289.3 = 285.5. Referring C to a χ2 with

the degree of freedom ν = 4(4 + 1)(3 − 1)/2 = 20, it is clear
that H0 is rejected at any reasonable level of significance.