School of Engineering

Numerical Methods an d Data Analysis

Lecture 3 - Full Factorial

 1 The Full Factorial Design

2 23 Factorial Design

Table of
Contents 3 2 4 Factorial Design

3 O t h er 2k Design
 Full Factorial Design

The 2 k Fact orial Design

• 2 k is p artic ularly useful in the early stages o f experimental
work

• Too many fac tors a n d interactions t o investigate

• Special (short-cut) me t ho d s fo r analysis

• a.k.a Factor Screening Experiments.

 Full Factorial Design

The 2 k Fact orial Design

• Special case o f t h e gene ral fa c t o ria l design; k factors, all
a t two levels

• The two levels are usually called low a n d h i g h

• Because 2 factors, all response is approximat ely linear over

the ran ge o f the f a c t o r level chosen.

• When we are s t a rtin g t o stu dy a process/system, usually

this is o u r first assumption: Linearism o f the two factors

• They c o u l d b e either:
– Q u a n t i t a t i v e (2 values o f time, temp, pressure etc)
– Qu a lit at iv e (2 machines, two operators, t wo batches)

• Very widely u sed in in du s t ria l ex p e rime n ta tio n

 Full Factorial Design

The 2 2 Fact orial Design

• Yate’s Order, b y Dr. Frank Yates
• Arrange as O r t h o g o n a l Constant
• The ‘+’ a n d ‘-’ signs show ‘high’ a n d ‘low’
• The h ig h a n d low shows the ‘level’ o f th e experiment
• The word ‘orthogonal’ presumes t h a t we c alc ulated th e m by
m a t c h i n g the hig h a n d lows o f all possible f ac to r
co mb in a tio n
 Full Factorial Design

S t a n d a r d o r d e r (Yate’s Order)

• (1): Treatment mod el (the whole experiment model)

• a : tre a t me nt f o r f a c t o r A
• b : tre a t me nt f o r f a c t o r B
• ab : t re at me n t fo r intera ctio n ab
• I : Mo de l effect
• A : F actor A effect
• B : F ac tor B effect
• AB : in tera ction AB effect
 Full Factorial Design

O r t h o g o n a l Co ntr ast
• Is basically tak in g the Factorial effect combinatio ns

(+)

(0) (0) (0)

 2 4 Factorial Design

O r t h o g o n a l Co ntr ast
 2 k Factorial Design

2 2 - O r t h o g o n a l C o n t ra s t
 2 k Factorial Design

2 3 - O r t h o g o n a l C o n t ra s t
 Concept

Concept
• In a f ac t orial experiment, all
possible c o mb in at i o ns of
f a c t o r levels are tested
• Varied t og et he r 2 n factorial
design
• 22 = 2 levels 2 f a ct o rs = 4 p o in t s = 4
runs
• Two factors: O -R, two level: B-T
• Randomisation - Replication
 Concept

Concept
B (-,+) Ab (+,+)

(1) (-,-) A (+,-)

 Full Factorial Design

Analysis Procedure f o r Full Factorial

1. Estimate f a c t o r design
2. Form initial model
3. Perform statistical testing
4. Analyse results
5. Interpret results
 Full Factorial Design

The 2 2 Fact orial Design

A = r e ac t a n t concentration, B = catalys t amount, y = (observation) recovery

Investigating the effect of the c o nc en tra tio n of re a c t a n t a n d a m o u n t o f catalyst.
To fi nd the best c o m b i n a t i o n of fa ct ors t o maximize yield.

A is f a c t o r A, B is f a c t o r B a n d AB is f a c t o r AB
+ = Hi gh levels of A a n d B Level of A = 15% a n d 25%
- = Low levels o f A a n d B Level o f B = 1 p o u n d a n d 2 p o u n d s
How d oe s this ‘high’
a n d ‘low’ look like?
 Full Factorial Design

1) The simplest FF case : 22 - t o f in d

m a in f a c t o r
• “-” a n d “+” d e no t e the low
a n d h ig h levels of a factor,
respectively
• Low a n d h ig h are a rb it rary
terms – doesn’t m e a n ‘low’ o r
‘high’ values b u t ‘positions’
• a is A a t h ig h a n d B a t low
• b is A a t low a n d B a t High
• a b is A a n d B b o t h high:
interaction
• (1) is A a n d B b o t h low
• Factors c a n b e qu antitative
o r qualitative, a l t ho u g h
their t r e a t me n t in the final
mo d el will b e different
 Full Factorial Design

1) The simplest FF case : 22 - t o f in d

m a in f a c t o r
 Full Factorial Design

2) Initial Model: Effect Estimates for

FF 22

• Effect Estimates shows the positive and

negative modes of the factor and interaction.

• Positive means increasing outcome while

negative means decreasing outcomes.
 Full Factorial Design

2) Initial Model: Sum of Squares for

FF 22

• Sum o f squares c a n be calculated t o study the

residual c o n t r ib u t i o n towards the experiment.

• Usually f or simple 22, the O r t h o g o n a l Cont ras t is

e n o u g h t o justify the M a i n Effects of the
experiment
 Full Factorial Design

3) Analyse Results

(1) (1)
a a
b b
ab ab
 Full Factorial Design

4) a n d 5) Analyse a n d Interpret
Results
• The effect of A is +ve: concentration will
increase the yield (MAIN FACTOR)
• The effect of B is –ve: 15% a n d 20% will
decrease the yield
• Interaction a p p e a rs t o be small in this case
relative t o two effects
FACTORS AND EFFECT SUM OF %
INTERACTIONS ESTIMATES SQUARES CONTRIBUTION
A 8.33 208 71.4%
B -5.00 75 25.7%
AB 1.67 8.33 0.03%
291.33
 Full Factorial Design

2 k FF
• In ma ny experiments involving 2 k design, we
are interested in t h e m a g n i t u d e a n d
direc tion o f t h e f ac tors effect

• OR in o t h e r words, which variables likely t o

b e imp ortant.
 1 The Full Factorial Design

2 23 Factorial Design

Table of
Contents 3 2 4 Factorial Design

3 O t h er 2k Design
 23 Factorial Design

23
• Suppose there are three f acto rs A, B a n d C o f interest.
• The eight tr eatment s c a n b e a r r a n g e d as below.
• The Right is of ten call Yate’s o rd e r th e Left is called Geometric
Notation.
• They are all just m e t h o d t o achieve O r t h o g o n a l Cont ras t t o be
Zero.
• Like previously if all balanc e o u t as Zero, t h a t means Hypothesis
Null are assumed t o b e acc ep ted
 23 Factorial Design

23
 23 Factorial Design

23 - Estimate F a ct o r Design a n d Initial

M o de l

• The development of estimates in 23 f ac torial

• 1st consider A:
– Effect of A when B a n d C is low level is [ a – (1)]/n.
– Effect of A when B is high a n d C is low is [ ab – b ]/ n .
– Effect o f A when B is low a n d C is high is [ ac – c ]/ n
– Effect of A when B is high a n d C is high is [ abc – bc ]/ n

• O r using contrast, same results

 23 Factorial Design

23 - Estimate F a ct o r Design a n d Initial

M o de l

Analysis d o n e via
c om p u t er
…technically
 23 Factorial Design

23 - Estimate F a ct o r Design a n d Initial

M o de l

Calculate these O r t h o g o n a l Contrast

Run these c onf igurat ions
 23 Factorial Design

2 3 - Statistical Testing
23
● F o r e s t i m a t e s : [….]/4n
● F o r s u m s q : [….]2/8n
 23 Factorial Design

Example 1
A 23 fa cto rial design was used t o develop a nitrid e e tch process o n a single-wafer p la sm a etching
tool. Plasma e tching is a f o r m o f p la sm a processing used t o f a b r i c a te inte gra te d circuits. It involves
a h igh -speed stream o f glow d ischarge (plasma) o f a n a p p r o p r i a t e gas mixture be ing sh ot (in pulses)
a t a sample. The design f a c t o r are th e g a p between th e electrodes, th e Hexafluoroethane gas flow
a n d t he Radio Frequency (RF) power a p p l i ed t o th e cathode. Each r u n a t two levels a n d replicated
twice. The response variable is th e e tch ra te f o r silicone nitride. The d a t a a re shown below

Initial model A = gap, B = Flow of Hexafluoroethane, C = Power, y = Etch Rate

Further info: http://www.youtube.com/watch?v=BfZfqMDERQM
 23 Factorial Design

23 - Initial M o d e l (DON’T DRAW THIS IN

FINAL EXAM)

Cube plots are often

useful visual displays
o f experimental
results.

N o t t o b e repr oduc e
in exam!
 23 Factorial Design

23 - Calc ulation

23
● F o r e s t i m a t e s : [….]/4n
● F o r s u m s q : [….]2/8n
 23 Factorial Design

23 - Effect Estimates, Statistical Analysis

a n d I nt e rpret ati on

• ‘Percent contribution’ shows the percentage c o n t r i b u t i o n o f eac h model

t o the t o t a l sum o f squares
• Rough g ui de TO STUDY RESIDUAL a n d effects c on trib ut ion
• Effective t o see the relative imp orta nc e of each model term
• Here we know f a c t o r C is the mai n effect of the whole experiment a n d
any in te rac ti on of the least a n d most.
 23 Factorial Design

23 - Effect Estimates, Statistical Analysis

a n d I n t er pr e t at i o n

• The full s ummary proves o u r estimates

• Power is the ma in effect o f this experiments
A = gap,
• The significants are A, C a n d AC B = Flow o f Hexafluoroethane,
• The a c c e p t an c e are B, AB, BC a n d ABC
C = Power
• Int eraction ABC doesn’t n o t exist
 1 The Full Factorial Design

2 23 Factorial Design

Table of
Contents 3 2 4 Factorial Design

3 O t h er 2k Design
 2 4 Factorial Design

2 4
 2 4 Factorial Design

23 – O r t h o g o n a l Cont rast
 2 4 Factorial Design

2 4 – O r t h o g o n a l Cont ras t
 2 4 Factorial Design

2 4 – Effect Estimates a n d Sum o f

Squares

● For estimates: [….]/8n

● For sumsq: [….]2/16n
 2 4 Factorial Design

2 4 – Example
• Example 6.2 (A single replicate of the 2 4 design)
– A 2 4 fac torial was used t o investigate the effects of
f o u r fac tors o n the filtration ra te o f a resin
– The facto rs are A = temperature, B = pressure, C =
co n c e nt ra tio n of formaldehyde, D= stirring rate
 2 4 Factorial Design

2 4 – Example (After calc ulations &

general answers)

• B is n o t significant a n d
all interac tions involving
B are negligible
• A is the highest c o n t r i b u t o r
• Small possibility o f
interaction
• Interaction between AC is
the highest a m o n g them
– A = temperature,

– B = pressure,

– C = concentrati on of
form aldehyde,

– D= s t i r r i n g r a t e
 2 4 Factorial Design

2 4 – Example (ANOVA Table -

for comparison )

Response: Filtration Rate

ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of Mean F
Source Squares DF Square Value Prob >F
Model 5535.81 5 1107.16 56.74 < 0.0001
A 1870.56 1 1870.56 95.86 < 0.0001
C 390.06 1 390.06 19.99 0.0012
D 855.56 1 855.56 43.85 < 0.0001
AC 1314.06 1 1314.06 67.34 < 0.0001
AD 1105.56 1 1105.56 56.66 < 0.0001
Residual 195.12 10 19.51
Cor Total 5730.94 15

Std. Dev. 4.42 R-Squared 0.9660

Mean 70.06 Adj R-Squared 0.9489
C.V. 6.30 Pred R-Squared 0.9128
PRESS 499.52 Adeq Precision 20.841
 2 4 Factorial Design

Summary of 2 k
Ef f ec t E s tim at es a n d Sum o f Sq u a r es

Summary of 2 2
● For estimates: [….]/2n
● For sumsq: [….]2/4n
Summary of 2 3
● For estimates: [….]/4n
● For sumsq: [….]2/8n
Summary of 2 4
● For estimates: [….]/8n
● For sumsq: [….]2/16n
 1 The Full Factorial Design

2 23 Factorial Design

Table of
Contents 3 2 4 Factorial Design

3 O t h er 2k Design