Professional Documents
Culture Documents
Nested-Factorial Experiments - Report v1
Nested-Factorial Experiments - Report v1
Design Of Experiments
A seminar report on:
Kaisar Katchi
(Roll No. 05)
Under The Guidance Of:
Professor S. B. Rane
Contents
Topic
The Nested Experiment
Page
2
6
10
12
13
17
M
H
a
e
c
a
h
d
i
n
1
2
e
A
B
In a nested experiment, factors are contained, or dependant, on other factors. That is, they are
unique enough to cause significant changes in the experimental reading on their own, but they
are also encapsulated inside other factors, which are under consideration.
Consider an experiment to study the effect of the readings of two machines designed to observe
the strain values on the supports used for fixing cathode ray tubes.
Further, each machine has two heads that individually measure the strain reading.
Four readings are observed per machine, that is, two per head (k).
The heads can be chosen from any number of heads. Hence, those heads are a random
sample. But, heads are not crossed (no interchanging; same head is not used for both
machines).
Machines can be considered as fixed. That is, not random.
With the experimental parameters set, the experiment is conducted, and the following
observations are noted:
Machine
Head
13
10
Head Total
16
19
Machine Total
24
22
Experimental Model
Yijk = + Mi + Hj(i) + k(ij)
Where:
Source
EMS
(Machine) Mi
2 + 4H2 + 16M
(Head) Hj(i)
2 + 4H2
(Error) k(ij)
Data Analysis:
The sum of squares for the factors is calculated:
ANOVA
The analysis of variance yields:
4
Source
df
SS
MS
EMS
(Machine) Mi
139.5
139.5
2 + 4H2 + 16M
4.72
(Head) Hj(i)
80
40
2 + 4H2
1.35
(Error) k(ij)
59
29.5
Totals
278.5
Experiment Conclusions:
From SS (SSHeads) calculation, we see significant difference between the heads of Machines A
and B. These results suggest a more careful adjustment between heads within the machines.
In a factorial experiment, multiple factors affect our variable; such as the effect of temperature
and altitude on current flow in an integrated circuit.
When both factors and nested factors appear in the same experiment, it is known as a nestedfactorial experiment.
Consider an experiment to test a new method of loading bullets in a gun, to improve the loading
speed.
Three groups of testers are selected, categorized by their build: slight, average and heavy build.
Each group is divided into three teams.
Each team uses two methods: old and new. Each method is used twice.
Experimental Model
Yijkm = + Mi + Gj + MGij + Tk(j) + MTik(j) + m(ijk)
Where:
Mi = methods, i = 1, 2.
Gj = groups, j = 1, 2, 3.
The following data represents the number of bullets loaded per minute (more is better):
GROUP
II
III
Team
Method I
20.2
26.2
23.8
22
22.6
22.9
23.1
22.9
21.8
24.1
26.9
24.9
23.5
24.6
25
22.9
23.7
23.5
14.2
18
12.5
14.1
14
13.7
14.1
12.2
12.7
16.2
19.1
15.4
16.1
18.1
16
16.1
13.8
15.1
Method II
EMS Table:
Source
EMS
Mi
2 + 2MT 2 + 18M
Gj
2 + 4T 2 + 12G
MGij
2 + 2MT 2 + 6MG
Tk(j)
2 + 4T 2
MTik(j)
2 + 2MT 2
m(ijk)
Group
II
III
SS (Cell)
256.18
199.96
246.96
SS (Method)
214.19
196.83
242.10
SS (Team)
35.74
1.62
1.90
SS (MxT Interaction)
6.25
1.51
10.72
ANOVA:
An Analysis of Variance leads to:
Source
df
SS
MS
EMS
Mi
651.95
651.95
2 + 2MT 2 + 18M
Gj
16.05
8.02
2 + 4T 2 + 12G
MGij
1.19
0.60
2 + 2MT 2 + 6MG
Tk(j)
39.26
6.54
2 + 4T 2
MTik(j)
10.72
1.79
2 + 2MT 2
m(ijk)
18
41.59
2.31
Totals
35
760.76
To calculate the effect of Methods and Teams, the group readings are retabulated:
Method
s
Teams
1
Method
3
Totals
Teams
4
M
6
Totals
Teams
7
M
9
Totals
II
Team
20.2
26.2
23.8
22
22.6
22.
9
23.1
22.9
21.8
24.1
26.9
24.9
23.5
24.6
25
22.9
23.7
23.5
44.3
53.1
48.7
45.5
47.2
47.
9
46
46.6
45.3
14.2
18
12.5
14.1
14
13.
7
14.1
12.2
12.7
16.2
19.1
15.4
16.1
18.1
16
16.1
13.8
15.1
30.4
37.1
27.9
95.4
30.2
32.1
29.
7
92
30.2
26
27.8
84
74.7
90.2
76.6
241.5
75.7
79.3
77.
6
232.6
76.2
72.6
73.1
221.9
146.1
Totals
140.6
137.9
Experiment Conclusion:
The mean number of bullets per minute for methods 1 (new) and 2 (old) are 23.58 and 15.08
respectively, showing significant improvement.
The significant difference between teams within the groups is concentrated in group 1 (SS
35.74).
Mean for teams in group I are: 18.68, 22.55, 19.15, suggesting that team 2 from group 1 is
exceptionally faster than the other two teams.
Repeated-Measures Design
Consider an experiment to determine the weight lifting capacity of strength trainers before and
after a new training method. Here, the same subjects repeat the experiment; hence, the same
experimental unit is repeated. There are two repeated measures on each subject. Such an
experiment is called a repeated-measures experiment.
Seven trainers undergo the experiment, and the following results are obtained:
Subjects
Pretest
100
110
90
110
125
130
105
Post-test
115
125
105
130
140
140
125
Experiment Model:
Yij = + Si + j(i)
Degress of Freedom for Si: 6 and j(i): 7.
Model for the within-subject experiment, that is, before and after test results:
Yij = + Si + Tj + STij
Degrees of Freedom for Si: 6, Tj: 1 and Stij: 6.
ANOVA
10
df
SS
MS
2084.71
347.45
Within Subjects
901.00
128.71
Tests (Tj)
864.29
864.29
Residual
35.71
5.96
Totals
13
145
2985.71
If treated as a two-factor factorial experiment (that is, treating the members as two factors: one
before the new training method, one after), with one observation per treatment, gives us the
following EMS table:
Source
df
EMS
Si
2 + 2S 2
Tj
2 + 2ST 2 + 7T
STij
2 + 2ST 2
k(ij)
Conclusion
As more factors are added to experiment, only the mathematical model needs to be expanded.
Nested and nested-factorial experiments may hence be designed in the following method:
Experiment
Design
Analysis
Completely randomized
Two or more
Factors:
A. Factorial
(crossed)
B. Nested
(hierarchical)
C. Nested
factorial
Completely randomized
Nested ANOVA
Nested-factorial ANOVA
12
13
16
References
Fundamental Concepts in the Design of Experiments Charles R. Hicks (Oxford University
Press, Fourth Edition).
17