8.

FACTORIAL EXPERIMENT

The effect of two or more factors on a response can be investigated. Such multi factor
investigations occurs often within industrial and scientific experimentation when an outcome
may depend on many factors. The factors chosen for examination would be tested at a
number of level reflecting likely operating levels for each factor. Experiments of this type
come under the domain of Factorial Design where both main effects and interaction effects
can be investigated to determine the level of influenced, if any, such factors have and how
this may be occurring in the observed response measurements.

Experimental factors in factorial design are generally specified at a number of different levels
of example, temp: 40oC, 50oC, 600C and 70oC; cooling time; 10 minutes, 15 minutes and 20
minutes; raw material flow rate, slow and fast and operators; A, B and C. These quantitative
and qualitative Levels reflect the different factor settings to be controlled in the experiment.
It is the effect of these levels on a measured response, separately and in combination which
the factorial design structures are designed to assess and estimate.

Factorial designs at two levels

Factorial experimentation is a powerful technique for scientific and technological research.

However, when many factors are required to be assessed, full factorial design can be costly
to implement. Two-level factorial design is a form of factorial structure which can be useful
in multi-factor experimentation using only two levels of each factor reduces the number of
treatment combination within the experimental region. Factorial design at two levels is
usually refer to as 2k Factorial design. This implies that there are two levels each for k
number of factor.
Example.:- Let us consider a two-level factorial design to investigate the effect of
temperature, pressure and catalyst on the chemical yield of pilot plant. The temperature and
pressure are quantitative but catalyst is qualitative. The two levels of temperature considered
were 20oC and 40oc; pressure were; 100kpa and 200kpa and the two catalyst considered were
A and B.

The experimental region covered by the eight possible treatment combinations is shown in the
figure below

1
The levels of factorial design at two levels are normally represented by (-) for lower level and (+)
for higher level (for quantitative factors)

The two levels factorial Design

In this example, k = 3 ( 3 factors). Therefore the factor design is 23 factorial design. For this
design, we will have 2k experimental run and (2k-1) effects estimates.
=> No of experimental run =23 = 8
No of effect (main and interaction) determination = 23 - 1 = 7

Notation for a 23 factorial design

Run 2o Temp 21 Pressure 22 Catalyst Yield (y)

1 - - - 60
2 + - - 72
3 - + - 54
4 + + - 68
5 - - + 52
6 + - + 83
7 - + + 45
8 + + + 80

Calculation of main effects

Method 1: Averaging individual measures of Effects:

When we talk about the effect of a factor, we mean the change in the response as we move from
the A-A to the A+A version of that factor i.e from the low to the high level of the factor.

Temperature effect:-
Conditions at which
comparison is made
Individual measure of the effect of changing temp. from 20oc Pressure
Catalyst
to 40oC Kpa
Y1 -Y2 = 72-60 =12
Y4 - Y3 = 68-54 =14 100 A
Y6 -Y5 = 83-52 =31 200 A
Y8 -Y7 = 80-45 =35 100 B
--------- 200 B
Σ=92
main effect of temp T = 92/4 = 23

2
Pressure Effect:
Conditions at which
comparison is made
Individual measure of the effect of changing pressure from Temperature Catalyst
100 kpa to 200 kpa 0o
Y3 -Y1 = 54-60 = -6 20 A
Y 4 -Y 2 = 68 -72 = -4 40 A
Y7 -Y5 = 45-52 = -7 20 B
Y8 -Y6 = 80-83 =- 40 B
---------
Σ= -20

main effect of pressure = -20/4 = -5

Catalyst Effect:
Conditions at which
comparison is made

Individual measure of the effect of changing catalyst from A Temperature Pressure

to B o
C Kpa
Y5 -Y1 = 52-60 = -8 20 100
Y6 - Y2 = 83-72 =11 40 100
Y7 -Y3 = 45-54 = -9 20 200
Y8 -Y4 = 80-68 =12 40 200
---------
Σ= 6
main effect of catalyst = 6/4 = 1.5

Method 2: Difference between two averages

The main effect for each of the variables seen to be the difference between two averages:

:.Temperature effect, T = 72+68+83+80 - 60+54+52+45

4 4
= 75.75 - 52.75
= 23
Pressure effect ; P = 54 + 68 + 45 + 80 - 60 + 72 + 52 + 83
4 4
= 61.75 - 66.75 = -5

Catalyst effect ; K = 52+83+45+80 - 60 + 72+54+ 68

4 4
= 65.0 - 63.5 = 1.5

3
INTERACTION EFFECTS
The average effect of temperature is 23. It is obvious from the data however, that the temperature
effect is much greater catalyst B than with catalyst A. The variables, temperature and catalyst
do not behave additively and are therefore said to interact.

Calculation of interaction effects

T x K :- The interaction between temperature and catalyst is given by the difference the average
temperature effect with catalyst A and the average temperature effect with catalyst B i.e half the
difference.

Catalyst average Average

temperature effect Temp Catalyst effect
(+) B 33 i.e (31+35)/2 OR(+)40 11.5 i.e (11+12)/2
(-) A 13 i.e (12+14)2 (-) 20 -8.5 i.e (-8-9)2
diff 20 diff. = 20
T x C interaction = 20/2 = 10 C x T interaction = 20/2 = 10

:. T x P interaction:
Pressure Average
temperature effect
(+) 200 kPa (14+35)/2 = 24.5
(-) 100 kPa (12+31)/2 = 21.5
Difference = 30

T x P interaction = 3.0/2 = 1.5

Also
P x C interaction: = 0

Three - Factor interaction

Consider the T x P interaction. Two measures of the T x P interaction are available
from the experiment, one for each catalyst i.e
T xP interaction with catalyst B (+)
(Y8- Y7) - (Y6-Y5) = (80-45) - ( 83-52) = 35 - 31 = 2
2 2 2
T x P interaction with catalyst A(-)
(Y4-Y3) - (Y2 - Y1) = (68 - 54) - (72 - 60) = 40 - 12 = 1
2 2 2
:. T x P x C interaction = 2 - 1 = 0.5
2

4
QUICKER METHODS FOR CALCULATING EFFECTS
Method 3: Table of Signs
Run T P C mean TP TC PC TPC Yield
1 - - - + + + + - 60
2 + - - + - - + + 72
3 - + - + - + - + 54
4 + + - + + - - - 68
5 - - + + + - - + 52
6 + - + + - + - - 83
7 - + + + - - + - 45
8 + + + + + + + + 80
4 4 4 8 4 4 4 4
Divisor
e.g
Pressure effect, P = -60-72+54+68-52-83+45+80 = -5
4
T x P x C interaction = -60 + 72 + 54 - 68 + 52 - 83 - 45 + 80 = 0.5
4
etc.

Method 4: Yate’s Algorithm:

The steps in using this method are as follows:
(1) set up design in standard form
(2) associate observation with appropriate design
(3) consider observations in successive pairs
(4) in column 1 enter first the sum of pairs and then the differences (by taking next
number from the previous
(5) repeat (iv) in column 2
(6) carry on till column k is filled
(7) set up column of divisor. First element is 2k others are 2k-1
(8) Estimate is obtained by dividing elements on column k by element in divisor column
(9) identify estimate

Run T P C Yield C1 C2 C3 Divisor Estimate Effects

1 - - - 60 132 254 514 8 64.25 Mean
2 + - - 72 122 260 92 4 23 T
3 - + - 54 135 26 -20 4 -5 P
4 + + - 68 125 66 6 4 1.5 TP
5 - - + 42 12 -10 6 4 1.5 C
6 + - + 83 14 -10 40 4 10 TC
7 - + + 45 31 2 0 4 0 PC
8 + + + 80 35 4 2 4 0.5 TPC
5
Calculation of standard error
When run replicates are made under a given set of experimental conditions, the variation
between their associated observations may be used to estimate the standard deviation of as ingle
observation and hence the standard deviation of the effects.
Let us assume that the above experiment was replicated twice and the result is shown below.

diff.
Avera of
Run T P C 1 2 S2i df
ge repli.
1 - - - 59 61 60 2 2 1
2 + - - 74 70 72 4 8 1
3 - + - 50 58 54 8 32 1
4 + + - 69 67 68 2 2 1
5 - - + 50 54 52 4 9 1 ΣS2i = 64
6 + - + 81 85 83 4 9 1
7 - + + 46 44 45 2 2 1
8 + + + 79 81 80 2 2 1

Standard error is equivalent to estimated standard deviation of the estimate.

Suppose we have r = 22 experimental run with each treatment combination replicated ni times
where i = 1,2,3 .....
suppose the observations are independent with variance of σ2.
The estimate of σ2 from each point of the design must be calculated.

Since each run is replicated twice and we used the mean

Therefore
Since each main effect and interaction is a statistic of the form Y+ - Y_; where each average
contains eight observation (four positive and four negative) the variance of each effect (assuming
independent errors) is given by
V(Effect) = V(Y-+- Y-)
= V(Y+) + V(Y-)
V(Y+) from V(x) - σ2
8
V(y-1) = σ2
8
:. V (effect = σ2 + σ2 = σ2 /4
8 8
In general, if a total of N observations is made in conducting a two -level factorial or
replicated factorial design, then
V(effect) = 4 σ2
N where N = no of runs x no of replicates.
from the example, estimate of σ2 = 8

6
 :. variance of effect, V(effect) = S2p
4
= 8 =2
4
:. Stand deviation of effect = Standard error of the effect
= 21/2

INTERPRETATION OF RESULTS

Effect Estimate+ std. Error

Average 64.25 + 0.7
main effects
temperature, T 23.0 + 1.4
pressure, P -5.0 + 1.4
catalyst, C 1.5 + 1.4
two -factor interactions
TxP 1.5+ 1.4
TxC 10.0+ 1.4
PxC 0.0 + 1.4
three- factor interactions
TxPxC 0.5 + 1.4