Scribd
Upload
19 views

STAT 514 HW10 1. (1) : r σ abσ α σ rb α a−1 bσ rα σ bσ β σ ra β b−1 aσ rβ σ aσ

1. The document summarizes the results of an analysis of variance (ANOVA) model for a dataset. The main effects were found to be significant but the interaction was not. 2. A full factorial design with factors A, B, C is analyzed. Factors B, C, and the interaction AC were found to be significant based on the ANOVA table. Regression and residual plots showed adequate model fit. 3. A 2^2 factorial design with factors A and D is examined. The main effects of A and D and their interaction AD were all found to be significant based on the ANOVA table and model. Residual plots confirmed adequate model assumptions. Settings of A=- and D=+ were identified to optimize

Uploaded by

Mon Luffy
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as DOCX, PDF, TXT or read online on Scribd
19 views

STAT 514 HW10 1. (1) : r σ abσ α σ rb α a−1 bσ rα σ bσ β σ ra β b−1 aσ rβ σ aσ

1. The document summarizes the results of an analysis of variance (ANOVA) model for a dataset. The main effects were found to be significant but the interaction was not. 2. A full factorial design with factors A, B, C is analyzed. Factors B, C, and the interaction AC were found to be significant based on the ANOVA table. Regression and residual plots showed adequate model fit. 3. A 2^2 factorial design with factors A and D is examined. The main effects of A and D and their interaction AD were all found to be significant based on the ANOVA table and model. Residual plots confirmed adequate model assumptions. Settings of A=- and D=+ were identified to optimize

Uploaded by

Mon Luffy
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as DOCX, PDF, TXT or read online on Scribd
You are on page 1/ 11

STAT 514 HW10

1. (1)

(b)

Term R F F EMS
r a b
i j k
Whole ri 1 a b σ 2+ ab σ 2r
plot
αj r 0 b 2 rb ∑ α 2j 2
σ + + b σ rα
a−1
rα ij 1 0 b σ 2+b σ 2rα
subplot βk r a 0 2 ra ∑ β k
2
2
σ + + a σ rβ
b−1
rβ ik 1 a 0 σ 2+ a σ 2rβ
αβ jk r 0 0 2 2 r ∑ ∑ (αβ ) jk 2
σ +σ rαβ +
(a−1)( b−1)
rαβ ijk 1 0 0 σ 2+ σ 2rαβ
ε (ijk ) 1 1 1 σ2

The main effect method is significant (p<0.0001) and the main effect mix is also significant
(p<0.0001), but the interaction between those two main effects is not significant (p=0.1052).

(c)

Term R F R EMS
r a b
i j k
Whole ri 1 a b σ 2+ ab σ 2r +a σ 2αβ
plot
αj r 0 b 2 rb ∑ α 2j 2 2 2
σ + + b σ rα + r σ αβ +σ rαβ
a−1
rα ij 1 0 b σ 2+b σ 2rα + σ 2rαβ
subplot βk r a 1 σ 2+ra σ 2β +a σ 2rβ
rβ ik 1 a 1 σ 2+ a σ 2rβ
αβ jk r 0 1 σ 2+ σ 2rαβ +r σ 2αβ
rαβijk 1 0 1 σ 2+ σ 2rαβ
ε (ijk ) 1 1 1 σ2

Test α j , F=(MSA+MSrAB)/(MSrA+MSAB)=(111.0475+0.7321)/(0.4908+1.6727)=51.666
( 111.0475+ 0.7321 )2
p= =2.026
111.04752 0.73212
+
2 12
( 0.4908+ 1.6727 )2
q= =8.89
0.4908 2 1.67272
+
4 6
P<0.0001
So we reject Ho and conclude the main effect method is significant.

Test β k , F=MSB/MSrB
Test (αβ ) jk, F=MSAB/MSrAB

For main effect mix, it’s significant since p<0.0001.


For the main effect interaction, it is not significant since p=0.1052.

2. (a)
A B C Mean
- - - 26
+ - - 34.67
- + - 39.67
+ + - 49.33
- - + 42.33
+ - + 37.67
- + + 54.67
+ + + 42.33

1
A= ∗¿
4
1
AB= ∗¿
4

1
B= ∗¿
4
C=3.41667*2=6.83
AB=-0.8333*2=-1.67
AC=-4.41667*2=-8.83
BC=-1.41667*2=-2.83
ABC=-1.08333*2=-2.17
Effects B, C, AC appear to be large.

(b) Based on the ANOVA table provided under (a), B, C, and AC are significant since their p values are
below 0.05.
(c) We only consider A, B, C, AC in the model.
A
^ B
^ ^
C ^
AC
y= ý …+ x 1+ x 2+ x 3+ x1x 3
2 2 2 2
X1=1 if A=+, x1=-1 if A= - . Same for B and C.
Plug in the parameter estimates, the fitted regression model is
y=40.83+0.167 x 1+5.67 x 2+ 3.42 x 3−4.42 x 1 x 3

(d) QQ plot below shows the residual is normally distributed.

Residual plots below include residual vs predicted values, residual vs A, residual vs B, residual vs C.
We can’t see patterns from those plots. So the constant variance assumption was not violated.
(e) Interaction plot of A and C:

Main effect plot for B and C:


I would recommend “+” level for main effect B and C, but “ –“ level for effect A to make the
response largest.
(f) y=40.83+0.167 x 1+5.67 x 2+ 3.42 x 3−4.42 x 1 x 3
To maximize the response, let x2 at high level be 1.
y=46.5+0.167 x 1+3.42 x 3−4.42 x 1 x 3

From the contour plot, we can see when A=-1 and C=1, the response can get its highest value. The
conclusion is consistent with my answer in (e).
(g)
σ2 30.17
SE=
√ n∗2 k −2
√=
3∗2
=2.24

3. (a) Effect estimates are in the table below.


From the QQ plot above, we can see that effects of AD, A, and D are potentially significant.
(b) In the model which only includes effect A, D, and AD, we can see the main effects A and D, and
the interaction effect AD are all significant. This confirms my findings in (a).
(c) We only consider A, D, AD in the model.
^
A D
^ ^
AD
y= ý …+ x 1+ x 4 + x 1x 4
2 2 2
X1=1 if A=+, x1=-1 if A= - . Same for effect D.
Plug in the parameter estimates, the fitted regression model is
y=776.06−50.81 x 1+153.06 x 4−76.81 x 1 x 4

(d) The residual is normally distributed based on the QQ plot below. The residual plots of residual vs
predicted values, residual vs A, residual vs B are displayed below as well. We don’t see any violation
of the constant variance assumption.
(e) The ANOVA table of the 22 design is as below. We can see that factor A, D and interaction effect
AD are significant in this model.

(f) The difference of level means of effect D depends on levels of effect A. So effect A and D have
interaction.
(g) When A= - , D= +, the response has the largest value.

(h) If it’s necessary to let the response be 800, I would recommend the A= +, D= + based on the
contour plot in (g). ¿ 776.06−50.81+153.06−76.81=801.5

You might also like