Documente Academic
Documente Profesional
Documente Cultură
2 2 5 5
15.2.3 8, 1 5, 2 1, 3 4, 1 , 2 , 11 , 12 ,
3 3 3 3
1 1 4 4
21 , 22 , 31 , 32 ; A and B are not additive for any
3 3 3 3
values of A and B.
r r
L r n
L
15.2.5 Since 0 ij , 2 Yijk j 1 . Setting 0 shows that
j j
i
i 1 i 1 i 1 k 1
r n
Y
i 1 k 1
ijk
Y . j . Y
T. j. rn rn j , so j
15.2.7 (a) A = 5, B = 2
2
(b) 2 Y221k is larger
k 1
15.2.9 Yes, Equation 15.2.3 assumes that the underlying within-cell error variances are all equal. For
these data, though, the sample within-cell error variances are considerably different.
1.875,
(b) ˆ1 0.25, ˆ 2 0.625, ˆ 3 0.375, 0.792,
0.958,
1 2 3
0.125
4
ij
B1 B2 B3 B4
A1 –1.25 0.42 0.58 0.25
A2 0.38 –0.46 –0.30 0.38
A3 0.88 0.05 –0.30 –0.63
(c) No
15.3.5 Yes, because the main events and interactions are all independent, so knowing the numerical
value of one implies nothing about the numerical value of another.
15.3.7
Source df SS MS
A 5 60 12
B 4 44 11
AB 20 18 0.9
Error 60 42
Total 89 164
k n
F R
j i EMS
2
j 0 n 2 n *
j (i ) 0 1 2
so the “error” term is the proper denomination for an F test of the A effect.
15.4.3 Below is the data’s ANOVA constructed using the computing formulas on p. 588. The EMS
column is derived from the rules given on p. 15-28.
Source df SS MS EMS
District (Ai) 4 64.45 16.11 2 4 A2
Error 15 15.66 1.04
Total 19 80.11 2
Setting the mean squares equal to their expected values gives ˆ 2 1.04 and ˆ A2 3.77.
15.4.5 (a)
r c µ n
F F F R
i j k l
i 0 c µ n *2
j r 0 µ n *2
ij 0 0 µ n
*2
rk r c 0 n r*2
r ik 0 c 0 n *2r
r jk r 0 0 n
*2r
r ijk 0 0 0 n
*2
r
l (ijk ) 1 1 1 1 2
EMS(A) 2 c n *2
EMS(B) 2 r n *2
EMS A B 2 n
*2
EMS(error) 2
(b) The appropriate denominator for testing each of the main effects or the interactions is
EMS (error).
r c u n 2
15.6.1 L Yijkl i j k ij ik jk ijk
i 1 j 1 k 1 l 1
L
r c u n
2 Yijkl i j k ij ik jk ijk ( 1)
i 1 j 1 k 1 l 1
L
Setting 0 reduces to
r c u n
Y
i 1 j 1 k 1 l 1
ijkl rcun 0
so
1 r c u l
ˆ Yijlk Y ....
rcun i 1 j 1 k 1 l 1
The MLE’s for the parameters are derived similarly.
15.6.3 (a) No, except that all three response pattern are parallel.
(b)
Y 111. Y .... Y 1... Y .1.. Y ..1. Y 11.. Y 1.1. Y .11.
111
66 30 39 12 15 8 10
6 0
12 6 6 4 3 2 2
112
Y 112. Y .... Y 1... Y .1.. Y ..2. Y 11.. Y 1.2. Y .12.
66 30 33 26 15 10 9
3 0
12 6 6 4 3 2 2
If 111
0 and 112
0, then
0. But if those three are 0, the remaining
113
interactions must also be 0 because of summation constraints. Yes, these data do illustrate the
comment on p. 15-43.
15.6.5 (a) SSA = 8.334; SSB = 16.334; SSC = 45.5; SSD = 8.334; SSAB = 0; SSAC = 3.166;
SSBC = 6.166
(b) SSE = 78; 24
2 2 3 2 T2
(c) SSABCD ijkl . C SSA SSB SSC SSD SSAB SSAC SSAD
i 1 j 1 k 1 l 1 2
15.7.3
obs. B D BD
(1) 11 – – +
a 5 – – +
b 9 + – –
ab 19 + – –
c 4 – – +
ac 2 – – +
bc 0 + – –
abc 6 + – –
d 3 – + –
ad 16 – + –
bd 14 + + +
abd 20 + + +
cd 8 – + –
acd 7 – + –
bcd 1 + + +
abcd 12 + + +
C BD 1 11 5 9 19 4 2 0 6 3 16 14 20 8 7 1 12
8
1 1
1 ,
8
8
so factors B and D are not additive.
15.7.5
Effect df SS
A 1 4.5
B 1 12.5
AB 1 18
C 1 12.5
AC 1 2
BC 1 18
ABC 1 0.5
Total 7 68
A reasonable error term would be formed by pooling the AC and ABC interactions. It would
have 2 df and a mean square equal to
2 0.5 1.25.
2
15.8.3 A linear combination of treatments that does not represent a main effect or interaction should
never be used as a defining contrast. Doing so would eliminate the orthogonality that exists
among the columns in the effect matrix and would bias the estimates of those effects.