Chapter 8

Factorial and Fractional Factorial

Experiments for Process Design and
Improvement
8-1. What is Experimental Design?
Objectives of Experimental Design
– Determine which variables (x’s) are most
influential on the response, y
– Determine where to set the influential x’s so
that y is near the nominal requirement
– Determine where to set the influential x’s so
that variability is small
– Determine where to set the influential x’s so
that the effects of the uncontrollable variables
z are minimized
8-1. What is Experimental Design?

Results of Experimental Design (used early in

process development):
1. Improved yield
2. Reduced variability and closer conformance
to nominal
3. Reduced development time
4. Reduced overall costs
8-1. What is Experimental Design?

Example 1 Characterizing a Process

• SPC has been applied to a soldering process.
Through u-charts and Pareto analysis, statistical
control has been established and the number of
defective solder joints has been reduced to 1%.
The average board contains over 2000 solder
joints, 1% may still be too large.
• Desired to reduce the defects level more.
8-1. What is Experimental Design?
Example 1 Characterizing a Process
• Note: since the process is in statistical control,
not obvious what machine adjustments will be
necessary. There are several variables that may
affect the occurrence of defects:
– Solder Temp, Preheat temp, Conveyor speed, Flux
type, Flux specific gravity, Conveyor angle.
• A designed experiment involving these factors
could help determine which factors could help
significantly reduce defects. (Screening
experiment)
.
8-2. Guidelines for Designing
Experiments
Procedure for designing an experiment
1. Recognition of and statement of the problem.
2. Choice of factors and levels.
3. Selection of the response variable.
4. Performing the experiment
5. Data analysis
6. Conclusions and recommendations

• #1-#3 makeup pre-experimental planning

• #2 and #3 often done simultaneously, or in reverse order.
8-3. Factorial Experiments

• When there are several factors of interest

in an experiment, a factorial design
should be used.
• A complete trial or replicate of the
experiment for all possible combinations
of the levels of the factors are
investigated.
8-3. Factorial Experiments
• Main effect is the change in response produced
by a change in the level of a primary factor.
• An interaction is present among factors if a
change in the levels of one factor influences the
effect of another factor.
• Consider an experiment with two factors A and
B
– Interested in
• Main effect of A
• Main effect of B
• Interaction effect of AB
8-3. Factorial Experiments
8-3.2 Statistical Analysis
• Completely randomized design with two factors (A and
B) and n replicates.
• The model is
y ijk =  +  i +  j + (  ) ij +  ijk
where  = overall mean
i = effect of ith level of factor A
j = effect of jth level of factor B
()ij = effect of the interaction between A
and B.
 = random error component
 8-3. Factorial Experiments

8-3.2 Statistical Analysis

Factor B
1 2  b
1 y111, y112, …, y121, y122, …,  y1b1, y1b2, …,
y11n y12n y1bn
2 y211, y212, …, y221, y222, …, y2b1, y2b2, …,
Factor A y21n y22n  y2bn
    
a ya11, ya12, …, ya21, ya22, …, yab1, yab2, …,
ya1n ya2n  yabn
8-3. Factorial Experiments

8-3.2 Statistical Analysis

• Total corrected sum of squares
decomposition
a b n a b
   ijk
( y − y ... ) =
2
bn  i..
( y − y ... ) + an  . j.
(2
y − y ... ) 2

i =1 j=1 k =1 i =1 j=1
a b
+ n   ( y ij. − y i.. − y. j. + y... ) 2
i =1 j=1
a b n
+    ( y ijk − y ij. ) 2

i =1 j=1 k =1
8-3. Factorial Experiments

8-3.2 Statistical Analysis

• Total corrected sum of squares decomposition,
notation:
SST = SSA + SSB + SSAB + SSE
• The corresponding degree of freedom
decomposition is
abn – 1 = (a – 1) + (b – 1) + (a – 1)(b – 1) + ab(n – 1)
8-3. Factorial Experiments
8-3.2 Statistical Analysis
Source of Sum of Degrees of
Variation Squares Freedom Mean Square F0
A SSA a-1 SS MS A
MS A = A F0 =
a −1 MS E
B SSB b-1 SS MS B
MS B = B F0 =
b −1 MS E
Interaction SSAB (a – 1)(b – 1) SS AB MS AB
MSAB = F0 =
(a − 1)(b − 1) MS E
Error SSE ab(n-1) SS E
MSE =
ab(n − 1)
Total SST abn - 1
8-3. Factorial Experiments
8-3.2 Statistical Analysis
Sum of Squares Computing Formulas
2
a b n y
SS T =    yijk 2
− ...
i =1 j=1 k =1 abn
2
a y2 y 2 b y y 2
Main Effects SS A =  i.. − ... SS B =  − ...
. j.

i =1 bn abn j=1 an abn

2 2
a b y y
SS AB =   − ... − SS A − SS B
ij.
Interaction
i =1 j=1 n abn

Error SSE = SST – SSA - SSB - SSAB

8-3. Factorial Experiments
Example 8-5
Aircraft Primer Example
2
a b n y
SST =    y ijk 2
− ...
i =1 j=1 k =1 abn
2
(89 .8)
= (4.0) 2 + (4.5) 2 +  + (5.0) 2 − = 10 .72
18
a y i2.. y...2
SS primers =  −
i =1 bn abn
(28.7) 2 + (34.1) 2 + (27.0) 2 (89.8) 2
= − = 4.58
6 18
2
b y y...2
SS methods =  −
. j.

j=1 an abn
(40.2) 2 + (49.6) 2 (89.8) 2
= − = 4.91
9 18
8-3. Factorial Experiments
Example 8-5
Aircraft Primer Example

a b y ij2.
y...2
SS int eraction =  − − SS primers − SS methods
i =1 j=1 n abn
(12.8) 2 + (15.9) 2 + (11.5) 2 + (15.9) 2 + (18.2) 2 + (15.5) 2
=
3
(89.8) 2
− − 4.58 − 4.91 = 0.24
18

SSE = 10.72 – 4.58 – 4.91 – 0.24 = 0.99

 8-3. Factorial Experiments
Example 8-5
Aircraft Primer Example – Table 12-4. Analysis of Variance

Sum of Degrees of
Source of Variation Squares Freedom Mean Square F0 P-value
Primer types 4.58 2 2.29 28.63 2.71 x 10-5
Application methods 4.91 1 4.91 61.38 4.65 x 10-6
Interaction 0.24 2 0.12 1.5 0.269
Error 0.99 12 0.08
Total 10.72 17
8-3. Factorial Experiments
Example 8-5
Aircraft Primer Example – Figure 12-12. Graph of average adhesion
force versus primer types
8-3. Factorial Experiments
8-3.3 Residual Analysis
• Residuals are important in accessing model
adequacy
• The residuals from a two-factor factorial are

eijk = yijk − ŷ ijk

= y ijk − y ijk
8-3. Factorial Experiments

8-3.3 Residual Analysis

Residuals Versus Primer
(response is Force)

0.4

0.3

0.2

0.1
Residual

0.0

-0.1

-0.2

-0.3

-0.4

1 2 3

Primer
8-3. Factorial Experiments

8-3.3 Residual Analysis

Normal Probability Plot

.999
.99
.95
Probability

.80
.50
.20
.05
.01
.001

-0.4 -0.3 -0.2 -0.1 0.0 0.1 0.2 0.3

Residuals
8-4. 2k Factorial Design
8-4.1 The 22 Design
• 2k is the notation used to indicate that a certain
experimental design has k factors of interest,
each at two levels.
• 22 design: Two factors A and B, each at two
levels
A B
Low -1 -1
High +1 +1
There are a total of four possible combinations.
Introduction to Statistical Quality Control,
4th Edition
8-4. 2k Factorial Design
8-4.1 The 22 Design
• The simplest design involves two factors A and B and n
replicates.
• Interested in the main effect of A, the main effect of B,
and the interaction between A and B.
• Effects are calculated by:
Average Response at high level - Average
Response at the low level.
• A large effect would indicate a significant factor (or
interaction). (How large is large?)
• Contrasts can be calculated and used to estimate the
effects and then sums of squares.
8-4. 2k Factorial Design

8-4.1 The 22 Design

• Let the letters (1), a, b, and ab represent the
totals of all n observations taken at these design
points.
• Effect estimate of A:
A = yA+ − yA−
a + ab b + (1)
= −
2n 2n
=
1
a + ab − b − (1)
2n
8-4. 2k Factorial Design
8-4.1 The 22 Design
• Effect estimate of B: B = y B+ − y B−
b + ab a + (1)
= −
2n 2n
=
1
b + ab − a − (1)
2n
• Effect estimate of AB: ab + (1) a + b
AB = −
2n 2n
=
1
ab + (1) − a − b
2n
8-4. 2k Factorial Design
8-4.1 The 22 Design
• For the previous effects formulas, the quantities
in brackets are called contrasts.
• For example, ContrastA = a + ab – b – (1)
• The contrasts are used to calculate the sum of
squares for the factors and interaction.
(contrast) 2
SS =
n  (contrast coefficients) 2
8-4. 2k Factorial Design
8-4.1 The 22 Design
• The sum of squares for A, B, and AB are:

SS A =
a + ab − b − (1)
2

4n

SS B =
b + ab − a − (1)
2

4n

SS AB =
ab + (1) − a − b
2

4n
8-4. 2k Factorial Design
Example 8-6
• The effect estimates for A, B, and AB for the router
example are:
A=
a + ab − b − (1)
2n
=
1
96.1 + 161.1 − 59.7 − 64.4 = 16.64
2( 4)

B=
 b + ab − a − (1)
2n
=
1
59.7 + 161.1 − 96.1 − 64.4 = 7.54
2( 4)
8-4. 2k Factorial Design
Example 8-6
• The effect estimates for A, B, and AB for the
router example are:

AB =
ab + (1) − a − b
2n
=
1
161.1 + 64.4 − 96.1 − 59.7 = 8.71
2(4)
8-4. 2k Factorial Design
Example8
• The ANOVA table for the router example is
then
Source of Sum of Degrees of P-value
Variation Squares Freedom Mean Square F0
Bit Size (A) 1107.226 1 1107.226 185.25 1.17 x 10-8
Speed (B) 227.256 1 227.256 38.03 4.82 x 10-5
AB 303.631 1 303.631 50.80 1.20 x 10-5
Error 71.723 12 5.977
Total 1709.836 15
8-4. 2k Factorial Design
Regression Model
• A regression model could be fit to data from a factorial
design
y = 0 + 1x1 +  2 x 2 + 12 x1x 2 + 
• where 0 is the grand average of all observations and
each coefficient, j is  effect estimate.
• For Example 12-6, the fitted regression model is

 16.64   7.54   8.71 

ŷ = 23.83 +   x1 +  x 2 +   x1 x 2
 2   2   2 
8-4. 2k Factorial Design

Residual Analysis
• Residual plots are used to access the adequacy of
the model once again.
• Residuals are calculated using the fitted
regression model.
• The residual plots versus the factor levels,
interactions, predicted values, and a normal
probability plot are all useful in determining the
adequacy of the model and satisfaction of
assumptions.
8-4. 2k Factorial Design

Analysis Procedure for Factorial Designs

1. Estimate the factor effects
2. Form preliminary model
3. Test for significance of factor effects
4. Analyze residuals
5. Refine model, if necessary
6. Interpret results
8-4. 2k Factorial Design
8-4.2 The 2k Design for k  3 Factors
• When k  2, you could have a single
replicate, but some assumptions need to be
made. (Can’t estimate all interactions).
• For k = 3, the main effects and interactions
of interest are A, B, C, AB, AC, BC,
ABC.
• The main effects are again represented by
a, b, c, ab, ac, bc, abc, and (1)
8-4. 2k Factorial Design
8-4.2 The 2k Design for k  3 Factors
Effect Estimate for A:
A = y A+ − y A− =
1
a + ab + ac + abc − b − c − bc − (1)
4n
Effect Estimate for B:
B = y B+ − y B− =
1
b + ab + bc + abc − a − c − ac − (1)
4n
Effect Estimate for C:
C = y C+ − y C− =
1
c + ac + bc + abc − a − b − ab − (1)
4n
8-4. 2k Factorial Design
8-4.2 The 2k Design for k  3 Factors
Effect Estimate for AB:
AB = y AB + − y AB − =
1
ab + (1) + abc + c − b − a − bc − ac
4n
Effect Estimate for AC:
AC = y AC + − y AC − =
1
ac + (1) + abc + b − a − c − ab − bc
4n
Effect Estimate for BC:

BC = y BC + − y BC − =
1
bc + (1) + abc + a − b − c − ab − ac
4n
8-4. 2k Factorial Design
8-4.2 The 2k Design for k  3 Factors
Effect Estimate for ABC:
ABC = y ABC + − y ABC − =
1
abc − bc − ac + c − ab + b + a − (1)
4n
In general, the effects can be estimated using
Contrast
Effect =
n 2 k −1
The sum of squares for any effect is

SS =
(Contrast)
2

n 2k
8-4. 2k Factorial Design
Example 8-7

• An experiment was performed to

investigate the surface finish of a metal
part. The experiment is a 23 factorial
design in the factors feed rate (A), depth
of cut (B), and tool angle (C), with n = 2
replicates.
8-4. 2k Factorial Design
Example 8-7
Design Factors
Surface
Run A B C Finish Totals
1 (1) -1 -1 -1 9, 7 16
2 a 1 -1 -1 10, 12 22
3 b -1 1 -1 9, 11 20
4 ab 1 1 -1 12, 15 27
5 c -1 -1 1 11, 10 21
6 ac 1 -1 1 10, 13 23
7 bc -1 1 1 10, 8 18
8 abc 1 1 1 16, 14 30
8-4. 2k Factorial Design
Example 8-7 – Analysis of Variance Table

Source of Sum of Degrees of P-value

Variation Squares Freedom Mean Square F0
A 45.5625 1 45.5625 18.69 2.54 x 10-3
B 10.5625 1 10.5625 4.33 0.07
C 3.0625 1 3.0625 1.26 0.29
AB 7.5625 1 7.5625 3.10 0.12
AC 0.0625 1 0.0625 0.03 0.88
BC 1.5625 1 1.5625 0.64 0.45
ABC 5.5625 1 5.5625 2.08 0.19
Error 19.5000 8 2.4375
Total 92.9375 15
8-4. 2k Factorial Design
Regression Model
• For Example 12-7, the fitted regression model involving
only those factors found significant (A, B) and the next
significant interaction (AB) is
 3.375   1.625   1.375 
ŷ = 11.0625 +   x1 +  x 2 +   x1x 2
 2   2   2 
= 11.0625 + 1.6875x1 + 0.8125x 2 + 0.6875x1x 2
• Developing a model from a designed experiment can be
a valuable tool in determining optimal settings for the
factors.
8-4. 2k Factorial Design
Other Methods for Judging the Significance of
Effects
• The standard error of any effect estimate in a 2k
design is 2
ˆ
s.e.(Effect) =
n 2k −2

• Two standard deviation limits on any estimated

effect is
Effect estimate  2[s.e.(effect)]
8-4. 2k Factorial Design
Other Methods for Judging the Significance of
Effects
Effect estimate  2[s.e.(effect)]
• This interval is an approximate 95% confidence
interval on the estimated effect.
• Interpretation is simple: If zero is contained
within the 95% confidence interval, then that
effect is essentially zero and the corresponding
factor is not significant at the  = 0.05 level.
8-4. 2k Factorial Design
8-4.3 A Single Replicate of the 2k Design
• As the number of factors in a factorial
experiment increase, the number of effects that
can be estimated also increases.
• In most situations, the sparsity of effects
principle applies.
• For a large number of factors, say k > 5, it is
common practice to run only a single replicate
of the 2k design and pool or combine the higher-
order interactions as estimate of error.
8-4. 2k Factorial Design
Example 8-8
• Nitride etch process on a single-wafer plasma
etcher. There are four factors of interest. The
response is etch rate for silicon nitride. A single
replicate is used.

Gap Pressure C2F6 Flow Power

Level A (cm) B (m Torr) C (SCCM) D (W)
Low (-) 0.80 450 125 275
High (+) 1.20 550 200 325
8-4. 2k Factorial Design
Example 8-8
A B C D Etch Rate
Run (Gap) (Pressure) (C2F6 Flow) (Power) (? /min)
1 -1 -1 -1 -1 550
2 1 -1 -1 -1 669
3 -1 1 -1 -1 604
4 1 1 -1 -1 650
5 -1 -1 1 -1 633
6 1 -1 1 -1 642
7 -1 1 1 -1 601
8 1 1 1 -1 635
9 -1 -1 -1 1 1037
10 1 -1 -1 1 749
11 -1 1 -1 1 1052
12 1 1 -1 1 868
13 -1 -1 1 1 1075
14 1 -1 1 1 860
15 -1 1 1 1 1063
16 1 1 1 1 729
8-4. 2k Factorial Design
Example 8-8
• The estimated effects are found to be
A = -101.625 AD = -153.625
B = -1.625 BD = -0.625
AB = -7.875 ABD = 4.125
C= 7.375 CD = -2.125
AC = -24.875 ACD = 5.625
BC = -43.875 BCD = -25.375
ABC = -15.625 ABCD = -40.125
D = 306.125
8-4. 2k Factorial Design
Example 8-8
• Normal probability plot of effects
Normal Probability Plot of the Effects
(response is Etch, Alpha = .05)

A: A
D B: B
C: C
D: D
1
Normal Score

-1
A

AD

-100 0 100 200 300

Effect
8-4. 2k Factorial Design

Example 8-8
• Normal probability plot reveals that A, D, and AD
appear to be significant.
• To be sure that other main factors or two factor
interactions are not significant, pool the three- and four-
factor interactions to form the error mean square.
• (NOTE: if the normal probability plot had indicated that
any of these interactions were important, they would not
be included in the error term.)
8-4. 2k Factorial Design
Example 8-8- Analysis of Variance
Source of Sum of Degrees of
Variation Squares Freedom Mean Square F0
A 41,310.563 1 41,310.563 20.28
B 10.563 1 10.563 <1
C 217.563 1 217.563 <1
D 374,850.063 374,850.063 183.99
AB 248.063 1 248.063 <1
AC 2,475.063 1 2,475.063 1.21
AD 94,402.563 94,402.563 48.79
BC 7,700.063 1 7,700.063 3.78
BD 1.563 1.563 <1
CD 18.063 1 18.063 <1
Error 10,186.815 5 2,037.363
Total 531,420.936 15
 8-4. 2k Factorial Design

Example 8-8- Analysis of Variance

• Factors A, D, and the interaction AD are
significant.
• The fitted regression model for this experiment
is
 101.625   306.125   153.625 
ŷ = 776.0625 −   x1 +  x 2 −   x1x 2
 2   2   2 
where x1 represents A, x2 represents D.
8-4. 2k Factorial Design

8-4.4 Addition of Center Points to the 2k Design

• So far, the assumption of linearity in the factor
effects has been made.
• 2k works well when the linearity assumption
holds only approximately.
• 2k design will support the main effects and
interactions model providing some protection
against curvature.
8-4. 2k Factorial Design

8-4.4 Addition of Center Points to the 2k Design

• There may be situations where a second-order
model is appropriate.
• Consider the case of k = 2, a model including
second-order effects is
y = 0 + 1x1 +  2 x 2 + 12 x1x 2 + 11x12 +  22 x 22 + 
• The model cannot be fitted using a 22 design; in
order to fit a quadratic model, all factors must
be run at at least three levels.
8-4. 2k Factorial Design
8-4.4 Addition of Center Points to the 2k Design
• Center points can be added to the standard 2k design
• Center points can provide not only some protection
against curvature, if the center points are replicated,
then an independent estimate of experimental error can
be obtained.
• Center points consist of nc replicates run at the xi = 0 (i =
1, 2, …, k).
• Addition of center points does not have an impact on the
usual effects estimates in a 2k design.
• Assume the k factors are quantitative in order to have a
“center” or middle level of the factor.
8-4. 2k Factorial Design
8-4.4 Addition of Center Points to the 2k Design
• Sum of Squares for pure quadratic error

n f n c (y F − y c )
2
SS pure quadratic =
nF + nc

where nF = number of factorial design points

y F , y c = average of runs at factorial points,
and average of center points, respectively.
• SSpure quadratic has a single degree of freedom.
8-4. 2k Factorial Design
8-4.4 Addition of Center Points to the 2k Design
• When center points are added to the design, the
model that can be estimated is
k k
y = 0 +   j x j +  ij x i x j +   jj x 2j + 
j=1 i j j=1

where jj are pure quadratic effects

• The test for curvature then actually tests
k k
H 0 :   jj = 0 vs. H1 :   jj  0
j=1 j=1
8-4. 2k Factorial Design
Example 8-9
• Reconsider the plasma etch experiment from example
12-8. Four center points (nc = 4) have been added to the
design with the responses given in Table 12-18.
• Averages: y F = 776 .0625 y c = 752 .75
• Curvature Sum of Squares
n f n c (y F − y c )
2
SS pure quadratic =
nF + nc
16(4)(776.0625 − 752.75) 2
=
16 + 4
= 1739.1
8-4. 2k Factorial Design
Example 8-9
• An estimate of experimental error can be
obtained by calculating the sample variance of
the center points:
20
 i
( y − 752. 75) 2

ˆ 2 = i =17

3
= 3122.7
8-4. 2k Factorial Design
8-4.4 Addition of Center Points to the 2k Design
• The F-test for curvature is given by
MSCurvature
F0 =
MSRe sidual

where MSresidual = SSresidual/df, with SSresidual a

combination of sum of squares for pure error
and sum of squares for lack of fit. (See Minitab
output, next slide).
8-4. 2k Factorial Design
Minitab Output for Example 8-9
Estimated Effects and Coefficients for Etch (coded units)

Term Effect Coef StDev Coef T P

Constant 776.06 10.20 76.11 0.000
A -101.62 -50.81 10.20 -4.98 0.001
B -1.63 -0.81 10.20 -0.08 0.938
C 7.37 3.69 10.20 0.36 0.727
D 306.12 153.06 10.20 15.01 0.000
A*B -7.88 -3.94 10.20 -0.39 0.709
A*C -24.88 -12.44 10.20 -1.22 0.257
A*D -153.62 -76.81 10.20 -7.53 0.000
B*C -43.87 -21.94 10.20 -2.15 0.064
B*D -0.63 -0.31 10.20 -0.03 0.976
C*D -2.13 -1.06 10.20 -0.10 0.920
Ct Pt -23.31 22.80 -1.02 0.337

Analysis of Variance for Etch (coded units)

Source DF Seq SS Adj SS Adj MS F P

Main Effects 4 416389 416389 104097 62.57 0.000
2-Way Interactions 6 104845 104845 17474 10.50 0.002
Curvature 1 1739 1739 1739 1.05 0.337
Residual Error 8 13310 13310 1664
Lack of Fit 5 10187 10187 2037 1.96 0.308
Pure Error 3 3123 3123 1041
Total 19 536283
8-5. Fractional Replication of the
2k Design

• As the number of factors in a 2k increase,

the number of runs required increases
rapidly.
• If we can assume that some higher-order
interactions are negligible, then a
fractional factorial design can be used to
gain information on main effects and low-
order interactions.
8-5. Fractional Replication of the
2k Design

8-5.1 The One-Half Fraction of the 2k

• Contains 2k-1 runs
• Often called a 2k-1 fractional factorial
design
• Consider a three-factor design: 23-1; that
is one half of a full 23
8-5. Fractional Replication of the
2k Design
8-5.1 The One-Half Fraction of the 2k
• Table for 23 with all main effects and interactions:
Factorial Effect
Run I A B C AB AC BC ABC
a + + - - - - + +
b + - + - - + - +
c + - - + + - - +
abc + + + + + + + +
ab + + + - + - - -
ac + + - + - + - -
bc + - + + - - + -
(1) + - - - + + + -

• Suppose runs a, b, c, and abc are chosen as the one-half fraction

(shown in the top half of the table)
• Note that the runs selected yield a plus on the ABC effect.
• ABC is called a generator of this particular fraction.
8-5. Fractional Replication of the
2k Design
8-5.1 The One-Half Fraction of the 2k
• The identity element I is also plus for the four runs,
therefore I = ABC.
• I = ABC is the defining relation for the design.
• The defining relation can be used to find aliases of the
main effects and interactions.
• To find the aliases of a main effect or interaction,
multiply the entire defining relation by that term. (Note:
any column multiplied by itself always results in the
identity column, i.e. A•A = I)
8-5. Fractional Replication of the
2k Design
8-5.1 The One-Half Fraction of the 2k
• With defining relation, I = ABC, some of the aliases are
– The alias of main effect A is
A•I = A •ABC = BC
That is, A is aliased with (identical to) the BC interaction.
– The alias of the main effect B is
B •I = B •ABC = AC
That is, B is aliased with the AC interaction
– The alias of the AB interaction is
AB •I = AB •ABC = C
That is, the AB interaction is aliased with the main effect C.
8-5. Fractional Replication of the
2k Design
8-5.1 The One-Half Fraction of the 2k
• Normal Probability Plots and Residuals
– Normal probability plot can be useful in assessing the
significance of effects (especially when there are
many effects to be estimated)
– Residuals can be obtained by the regression model
shown previously.
– Residuals should be plotted against predicted values,
against levels of the factors, and on normal
probability paper for two reasons:
1. To assess the validity of the underlying model assumptions
2. Gain additional insight into the experimental situation.
8-5. Fractional Replication of the
2k Design

8-5.1 The One-Half Fraction of the 2k

• Projection of the 2k-1 Design
– If one or more factors from a one-half fraction
of a 2k can be dropped, the design will
project into a full factorial design.
– Projection is highly useful in screening
experiments.
8-5. Fractional Replication of the
2k Design
8-5.1 The One-Half Fraction of the 2k
• Design Resolution
– Design resolution is useful in categorizing designs.
– The resolution of a design usually denoted by Roman
numerals.
– Of particular importance are designs of Resolution
III, IV, and V.
– The resolution of a design indicates the alias
relationship among the factors.
8-5. Fractional Replication of the
2k Design
8-5.1 The One-Half Fraction of the 2k
• Design Resolution
– Resolution III designs. No main effects are aliased with
one another. Main effects are aliased with two factor
interactions.
– Resolution IV designs. No main effect is aliased with
any other main effect or two-factor interaction. Two-
factor interactions are aliased with each other.
– Resolution V designs. No main effects are aliased with
any other main effect or two-factor interactions. Two-
factor interactions are not aliased with one another. Two-
factor interactions are aliased with three-factor
interactions.
 8-5. Fractional Replication of the
2k Design
8-5.1 The One-Half Fraction of the 2k
• Design Resolution
3−1
– For example, the notation 2 III indicates that
the design is a one-half fraction of a 23 with
resolution III. From this, we know that main
effects are not aliased with one another, but
are aliased with two factor interactions.
8-5. Fractional Replication of the
2k Design
8-5.2 Smaller Fractions: The 2k-p Fractional Factorial
Design
• While the half fraction is useful, it may be more
economical to use even smaller fractions of the 2k.
• Consider a design with k = 11 factors.
– A full 2k would require 2,048 runs
– A half fraction, 2k-1 would require 1,024 runs: still an
unreasonable number.
• How about a design for all 11 factors that would only
require 32 runs? Some assumptions would have to be
made, but this is a possible design.
– 11 factors in 32 runs would be a 1/6 fraction of the 211 or a
211-6
8-5. Fractional Replication of the
2k Design

8-5.2 Smaller Fractions: The 2k-p Fractional

Factorial Design
• Setting up a 2k-p design requires:
1. Setting up a full factorial design for k-p
factors.
2. Generate the remaining p columns by
selecting appropriate design generators.
8-5. Fractional Replication of the
2k Design
To illustrate, consider a 26-2 fractional factorial design.
1. Set up a full factorial design for 6 – 2 = 4 factors:
Run A B C D
1 - - - -
2 + - - -
3 - + - -
4 + + - -
5 - - + -
6 + - + -
7 - + + -
8 + + + -
9 - - - +
10 + - - +
11 - + - +
12 + + - +
13 - - + +
14 + - + +
15 - + + +
16 + + + +
8-5. Fractional Replication of the
2k Design
2. Generate the remaining 2 columns using a design generator that
involves the 4 factors used in part 1. For example, the column for factor E
could be generated using ABC. The column for factor F could be
generated using BCD Run A B C D E F
1 - - - - - -
2 + - - - + -
3 - + - - + +
4 + + - - - +
5 - - + - + +
6 + - + - - +
7 - + + - - -
8 + + + - + -
9 - - - + - +
10 + - - + + +
11 - + - + + -
12 + + - + - -
13 - - + + + -
14 + - + + - -
15 - + + + - +
16 + + + + + +
8-5. Fractional Replication of the
2k Design
• In this example there are two design generators:
– E = ABC → I = ABCE
– F = BCD → I = BCDF
• To find the complete defining relation, multiply all pairs of
the design generators (in this case there are only two so we
would multiply (ABCE)(BCDF) = ADEF)
• Therefore, the complete defining relation from which all
aliases can be found for main effects and interactions is
• I = ABCE = BCDF = ADEF
• By definition, the length of the smallest “word” in the
defining relation is also the resolution of the design. In this
case, the resolution is IV.
8-5. Fractional Replication of the
2k Design

8-5.2 Smaller Fractions: The 2k-p Fractional

Factorial Design
• Selection of Design Generators
– Design generators should not be chosen arbitrarily.
– Choosing the incorrect design generator can result in
a design of smaller resolution than possible.
– Select the design generators that will result in the
highest possible design resolution.
 Homework
1. A process engineer is trying to improve
the life of a cutting tool. He has run a 23
experiment using cutting speed (A),
metal hardness (B), and cutting angle
(C) as the factors. The data from two
replicates are shown in Table.1.
(a) Do any of the three factors affect
tool life?
(b) What combination of factor levels
produces the longest tool life?
(c) Is there a combination of cutting
speed and cutting angle that always
gives good results regardless of
metal hardness?
2. An article in Industrial and Engineering Chemistry (“More on Planning
Experiments to Increase Research Efficiency,” 1970, pp. 60–65) uses a 2 5−2
design to investigate the effect of A = condensation temperature, B = amount
of material 1, C = solvent volume, D = condensation time, and E = amount
of material 2, on yield. The results obtained are as follows:

(a) Verify that the design generators used were I = ACE and I = BDE.
(b) Write down the complete defining relation and the aliases from this
design.
(c) Estimate the main effects.
(d) Prepare an analysis of variance table. Verify that the AB and AD
interactions are available to use as error.
3. A 24-1 design has been used to investigate the effect of four factors on
the resistivity of a silicon wafer. The data from this experiment are
shown in Table 2.
(a) Estimate the factor effects. Plot the effect estimates on a normal
probability scale.
(b) Identify a tentative model for this process. Fit the model and test for
curvature.
(c) Plot the residuals from the model in part (b) versus the predicted
resistivity. Is there any indication on this plot of model inadequacy?
(d) Construct a normal probability plot of the residuals. Is there any
reason to doubt the validity of the normality assumption?