Sign method (Two Factor with level two: without replication)

Factorial Treatment Combination Divisor

Effect (1) a b ab
M + + + + 4
A - + - + 2
B - - + + 2
AB + - - + 2

Main effect of A = {ab – b + a – (1)}/2

Main effect of B = {ab + b - a – (1)}/2
Interaction effect AB = {ab – b - a + (1)}/2

Sign Table for 3-Factor Factorial design

Factorial Treatment Combination Divisor
Effect (1) a b ab c ac bc abc
M + + + + + + + + 8
A - + - + - + - + 4
B - - + + - - + + 4
AB + - - + + - - + 4
C - - - - + + + + 4
AC + - + - - + - + 4
BC + + - - - - + + 4
ABC - + + - + - - + 4

Main effect of A = {abc – bc + ac – c + ab – b + a – (1)}/4

Interaction effect of AB = {abc – bc – ac + c + ab – b – a + (1)}/4
Interaction effect of ABC = {abc – bc – ac + c - ab + b + a - (1)}/4

Yates Method for Two Factor (A, B) with level two including replication

Treatment Total Col-1 Col-2 Main & SS

Combinatio Yield interaction effect
n
(1) [1] [a] + [1] [ab] + [b] + [a] + [1]
a [a] [ab] + [b] [ab] - [b] + [a] - [1] = [A] [A]/2n = A [A]2 /4n
b [b] [a] - [1] [ab] + [b] - [a] - [1] = [B] [B]/2n = B [B]2 /4n
ab [ab] [ab] - [b] [ab] - [b] - [a] + [1] = [AB] [AB]/2n = AB [AB]2 /4n

n = number of replication
[ ] Indicate yields total
 Yates Method for Three Factors(A, B, C) with level two

TC Total Col-1 Col-2 Col-3 Main & SS

Yiel interaction effect
d
[1] [1] [a] + [1] [ab] + [b] + [a] + [1] [abc] + [bc] + [ac] + [c]+
[ab] + [b] + [a] + [1]
a [a] [ab] + [b] [abc] + [bc] + [ac] + [c] [abc] - [bc] + [ac] - [c] + [A]/4n = A [A]2 /8n
[ab] - [b] + [a] - [1]
b [b] [ac] + [c] [ab] - [b] + [a] - [1] [abc] + [bc] – {[ac] + [c]} [B]/4n = B [B]2 /8n
+ [ab] + [b] – {[a] + [1]}
ab [ab] [abc] + [abc] - [bc] + [ac] - [c] [abc] - [bc] – {[ac] - [c]} + [AB]/4n = AB [AB]2 /8n
[bc] [ab] - [b] – {[a] - [1]}
c [c] [a] - [1] [ab] + [b] – {[a] + [1]} [abc] + [bc] + [ac] + [c] - [C]/4n = C [C]2 /8n
([ab] + [b] + [a] + [1])
ac [ac] [ab] - [b] [abc] + [bc] – {[ac] + [abc] - [bc] + [ac] - [c] – [AC]/4n = AC [AC]2 /8n
[c]} ([ab] - [b] + [a] - [1])
bc [bc] [ac] - [c] [ab] - [b] – {[a] - [1]} [abc] + [bc] – {[ac] + [c]} [BC]/4n = BC [BC]2 /8n
- ([ab] + [b] – {[a] + [1]})
abc [abc] [abc] - [bc] [abc] - [bc] – {[ac] - [abc] - [bc] – {[ac] - [c]} - [ABC]/4n = ABC [ABC]2 /8n
[c]} ([ab] - [b] – {[a] - [1]})

Example: A soft drink bottler is interested in obtaining more uniform fill heights in the bottles produced
by his manufacturing process. The filling machine theoretically fills each bottle to the correct target height,
but in practice, there is variation around this target, and the bottler would like to understand better the
sources of this variability and eventually reduce it.
The process engineer can control three variables during the filling process: the percentage carbonation (A),
the operation pressure in the filter (B), and the bottles produced per minute or line speed (C). The pressure
and speed are easy to control but the percentage carbonation is more difficult to control during actual
manufacturing because it varies with product temperature. However, for purposes of an example, the
engineer can control carbonation at two levels: 10 and 12. She chooses two levels for pressure: 25 and 30
psi and two levels for line speed: 200 and 250 bpm. She decided to run two replicates of a factorial design
in these three factors with all 16 runs taken at random order. The data that resulted from are shown in the
table below. Analyze the data
From the above table we can summarize the results in the tabular form
Treatment combination Replication 1 Replication 2 Total
(1) -3 -1 -4
a 0 1 1
b -1 0 -1
ab 2 3 5
c -1 0 -1
ac 2 1 3
bc 1 1 2
abc 6 5 11

Main Factor effect

1
A=
4n
[ a− (1 )+ ab−b+ac−c+ abc−bc ]= 4 1x 2 [ 1−(−4 )+5−(−1 ) +3−(−1 ) +11−2 ]= 24
8
=3

Similarly, B = 2.25 and C = 1.75

Again Interaction effect

 1
AB=
4n
[ ab−a−b+ ( 1 ) + abc−bc−ac +c ]= 18 [ 5−1− (−1 )+ (−4 )+ 11−2−3+ (−1 ) ]= 68 =0.75
AC = 0.25 and BC = 0.50

ABC = [(abc-bc-ac+c-ab+b+a-(1)]/4n = [11-2-3+(-1)-5+(-1)+1-(-4)]/8 = 0.50

The largest effects are for carbonation (A = 3.00), pressure (B = 2.25), speed (C = 1.75) and the
carbonation-pressure interaction (AB = 0.75), although the interaction effect does not appear to have as
large an impact on fill height deviation as the main effects.

The sum of squares

( 24 )2 (18 )2 ( 14 )2
SS A = =36 , SS B = =20.25 , SSC = =12.25
8∗2 8∗2 8∗2
( 6 )2 ( 2 )2 ( 4 )2 ( 4 )2
SS AB= =2.25 , SS AC = =0.25 , SS BC = =1.00∧SS ABC = =1.00
8∗2 8∗2 8∗2 8∗2

H0: β 1=β 2=β 3=β 12=β 13=β23= β123

H1 At least one β i ' s is not true

Analysis of Variance

SV Effect df SS MS CalF P-value

estimate
A 3.00 1 36.00 36.00 57.60 <0.0001
B 2.25 1 20.25 20.25 32.40 0.005
C 1.75 1 12.25 12.25 19.60 0.0022
AB 0.75 1 2.25 2.25 3.60 0.094
AC 0.25 1 0.25 0.25 0.40 0.545
BC 0.50 1 1.00 1.00 1.60 0.24
ABC 0.50 1 1.00 1.00 1.60 0.24
Error 8 5.00 0.625
Total 15 78.00

Fitted regression model

( ) ( ) ( ) ( )
^y = β^ 0 + β^ 1 x 1 + ^β 2 x 2 + ^β 3 x 3 + ^β 12 x 12=1.00+
3.00
2
x 1+
2.25
2
x 2+
1.75
2
x 3+
0.75
2
x 12

Analysis of 2k Factorial design

Two Factors (A, B) with level two means 22 Factorial Design {treatment combinations: 4 = (1), a, b, ab}.

Two Main Effect and One () 2

2
Interaction Effect (AB)

Three Factors (A, B, C) with level two each means 23 Factorial Design {treatment combinations: 8 = (1), a,

b, ab, c, ac, bc, abc}. Three main effects, three ()3

2
two-factor interaction (AB, AC, BC) and one
3
3()three-

factor interaction (ABC)

Four Factors (A, B, C, D) with level two means 24 Factorial Design {treatment combinations: 16 = (1), a,

b, ab, c, ac, bc, abc, d, ad, bd, abd, cd, acd, bcd,abcd }. Four main effects, six ()
4
2
two-factor interaction

(AB, AC, BC, AD, BD, CD), four ( 43) three-factor interaction (ABC, ABD, ACD, BCD) and one( 44) four-
factor interaction (ABCD)

So for 2k factors, we have, k main effects, ( k2) two-factor interaction, (3k) three-factor interaction,…, one –
k factor interaction. So for 2k there are 2k -1 effects

For three factors (A, B, C) the AB and ABC contrast can be written excluding the divisor as
abc – bc – ac + c + ab – b – a + (1). If we use 1 in the place of (1) we have
Contrast AB = abc – bc – ac + c + ab – b – a + 1 = (a-1)(b-1)(c+1)
Contrast ABC = abc – bc – ac + c - ab + b + a - 1 = (a-1)(b-1)(c-1)
In general for K factors
Contrast ABC…K = (a±1)(b±1)(c±1)…,¿)

For Four factors (A, B, C, D) analyze the model?

Analysis of Variance for a 2k Design
SV SS df
K main effects
A SSA 1
B SSB 1
… … …
K SSK 1

()
k
2
two-factor interaction
AB SSAB 1
AC SSAC 1
… … …
JK SSJK 1

()
k
3
three-factor interaction
ABC SSABC 1
ABD SSABD 1
…
IJK SSIJK 1
------------- ------------- -------------

()
k
k
1, k - factor interaction
ABC,…,K SSABC,…,K 1
Error SSE 2k(n-1)
Total SST n2k-1
 Confounding
Chemical process experiment in three blocks
Block 1 Block 2 Block 3
(1) = 28 (1) = 25 (1) = 27
a = 36 a = 32 a = 32
b = 18 b = 19 b = 23
ab =31 ab =30 ab =29
Total 113 106 111

3 2 2
Bi y … 1132 +1062 +1112 330 2
SS Block=∑ − = − =6.50
i −1 4 12 4 12

ANOVA
SV df SS MS CalF P-value
Block 2 6.50 3.25
A 1 208.33 208.33 50.32 0.0004
B 1 75.00 75.00 18.12 0.005
AB 1 8.33 8.33 2.00 0.206
Error 6 24.84 4.14
Total 11 323.00

Experiment

High + ab
b

Factor B
2-Factors (A, B)
Circle goes to Block 1 and square on Block 2

Block 1 Block 2
(1) a
ab b

Sign method (Two Factor with level two: without replication)

Factorial Treatment Combination Divisor
Effect (1) a b ab
M + + + + 4
A - + - + 2
B - - + + 2
AB + - - + 2

Main effect of A = {ab – b + a – (1)}/2

Main effect of B = {ab + b - a – (1)}/2
Interaction effect AB = {ab – b - a + (1)}/2

Look at Block 1 and Block 2, and see the sign:

For effect A there are (-, +) for Block 1 and (+, - ) for Block 2
For effect B there are (-, +) for Block 1 and (-, + ) for Block 2
For effect AB there are (+, +) for Block 1 and (-, - ) for Block 2
Two + sign for Block 1 and Two – signs for Block 2 implies block effect and interaction effects
are identical. That is, AB is confounded with block.

Three Factor Factorial Design

Example 1:

Replication
Block 1 Block 2
abc (1)
a bc
b ac
c ab

(1) a b ab c ac bc abc
AB + - - + + - - +

For AB interaction look at Block 1: 2 positive Sign (abc, c) and 2 negative Sign (a,b) so Block effect
cancel out. Again for Block 2: 2 positive signs [(1), ab] and 2 negative sign (bc, ac) so Block effect
cancel out. AB is not confounded with blocks.
 (1) a b ab c ac bc abc
AC + - + - - + - +

For AC interaction look at Block 1: 2 positive Sign (abc, b) and 2 negative Sign (a, c) so Block effect
cancel out. Again for Block 2: 2 positive signs [(1), ac] and 2 negative sign (ab, bc) so Block effect
cancel out. AC is not confounded with blocks.

(1) a b ab c ac bc abc
BC + + - - - - + +

For BC interaction look at Block 1: 2 positive Sign (abc, a) and 2 negative Sign (b, c) so Block effect
cancel out. Again for Block 2: 2 positive signs [(1), bc] and 2 negative sign (ab, ac) so Block effect
cancel out. BC is not confounded with blocks.

(1) a b ab c ac bc abc
ABC - + + - + - - +
For ABC interaction look at Block 1: 4 positive Signs (abc, a, b, c) and for Block 2: 4 negative signs
[(1), ab, ac, bc] so Block effect can’t be canceled out that mean there exist block effect. Since block
effect and interaction effect cannot be identified separately that mean confounding exists. That is ABC
is confounded with blocks.

Example 2:

Block 1 Block 2
abc ac
c bc
ab b
(1) a

(1) a b ab c ac bc abc
AB + - - + + - - +

For AB interaction look at Block 1: 4 positive Sign (abc, c, ab, (1)) and for Block 2: 4 negative signs
[a, b, ac, bc] so AB is confounded with block effects and can’t be estimated independently whereas all
other effects (interaction) AC, BC and ABC can be estimated independently.

Partial confounding of 23 factorial design with four replications

 Replication 1 Replication 2 Replication 3 Replication 4
Block 1 Block 2 Block 3 Block 4 Block 5 Block 6 Block 7 Block 8
(1) a (1) a (1) b (1) a
ab b c b a c b c
ac c ab ac bc ab ac ab
bc abc abc bc abc ac abc bc
ABC Confounding AB Confounding BC Confounding AC Confounding

23 factorial design with ABC (complete confounding) confounding with four replications

Replication 1 Replication 2 Replication 3 Replication 4

Block 1 Block 2 Block 3 Block 4 Block 5 Block 6 Block 7 Block 8
(1) a (1) abc (1) abc (1) abc
ac b ac a ac a ac a
ab c ab b ab b ab b
bc abc bc c bc c bc c
ABC Confounding ABC Confounding ABC Confounding ABC Confounding

Problem
Replication 1 Replication 2
Block 1 Block 2 Block 3 Block 4
(1) = 550 a = 669 (1) = 604 a = 650
ab = 642 b = 633 c = 1052 b = 601
ac = 749 c = 1037 ab = 635 ac = 868
bc = 1075 abc = 729 abc = 860 bc = 1063
ABC Confounding AB Confounding

2
( a+b+ c+ abc−ab−ac−bc−(1) ) ( 650+ 601+ 1052+ 860−635−868−1063−604 )2
SS ABC ( Rep 2)= = =6.125
n 2k 1∗23
( ( 1 )+ abc−ac +c−a−b+ab−bc )2 (550+ 729−749+1037−669−633+642−1075 )2
SS AB (Rep 1)= = =3528.0
n 2k 1∗23

n 2 2 2 2 2
Rh y … ( 6084 ) + ( 6333 ) ( 12417 )
SS Rep=∑ k
− = − =3875.0625
h=1 2 N 8 16
SS Blocks =SS ABC ( Rep 1 ) +SS AB ( Rep 2 ) =338+120.125=458.125

2
( a+b+ c+ abc−ab−ac−bc−(1) ) ( 669+633+ 1037+729−550−642−749−1075 )2
SS ABC ( Rep 1)= = =338
2k 23
( ( 1 ) +abc−ac + c−a−b+ab−bc )2 ( 604+860−868+1052−650−601+ 635−1063 )2
SS AB (Rep 2)= = =120.125
n 2k 1∗23

ANOVA

SV DF SS MS CalF P
Replicates 1 3875.0625 3875.0625
Blocks within Replicates 2 458.125
A 1 41310.5625 41310.5625 16.20 0.01
B 1 217.5625 217.5625 0.08 0.78
C 1 374850.5625 374850.5625 146.97 <0.001
AB(Rep1 only) 1 3528.0 3528.0 1.38 0.29
AC 1 94404.5625 94404.5625 37.01 <0.001
BC 1 18.0625 18.0625 0.007 0.94
ABC (Rep2 only) 1 6.125 6.125 0.002 0.96
Error 5 12752.3125 2550.4625
Total 15 531420.9375