Factorial Design
Download
Report
Transcript Factorial Design
Factorial Design
Main Effects and Interaction Effects
Pre-test
Determine the appropriateness of
manipulations (operationalizations of
independent variables)
Test the validity and reliability of the
measurement scales
Pilot Study
Use a small group of participants to test
whether the questionnaire is clear and
understandable
Use the pilot study data to provide some
intuitive observations about whether the
experiment could work
Factorial Design
Much experimental psychology asks the
question:
What
effect does a single independent variable
have on a single dependent variable?
It is quite reasonable to ask the following
question as well.
What
effects do multiple independent variables
have on a single dependent variable?
Designs which include multiple independent
variables are known as factorial designs.
Factorial Design
All combinations of two or more values of two or
more IVs are created
Can be tested within, between or mixed
Reasons for factorial designs
– additional Ivs are included
Helps rule out several hypotheses at the same time
Allows the researcher to discover interactions
amongst the variables
Allows the researcher to include one or more
nuisance variables in the design
Efficiency
Picture of Toy Factorial Design
Bright colour
Dull Colour
Not Noisy
Noisy
Soft
Not Soft
Describing Toy Factorial Design
All possible combinations of soft, noisy and colourful
toys in a 2 x 2 x 2 design – i.e., three IVs
2 x 2 x 2 = 8 cells
Another factor (IV) with two levels
One factor (IV) with 2 levels
An Example Factorial Design
If we were looking at GENDER and TIME
OF EXAM, these would be two
independent factors
GENDER
would only have two levels: male or
female
TIME OF EXAM might have multiple levels,
e.g. morning, noon or night
This is a factorial design
Examples
If there are 2 levels of the first IV and 3 levels of
the second IV
It
is a 2x3 design
E.G.: coffee drinking x time of day
Factor coffee has two levels: cup of coffee or cup of water
Factor time of day has three levels: morning, noon and night
If there are 3 levels of the first IV, 2 levels of the
second IV and 4 levels of the third IV
It
is a 3x2x4 design
E.G.: coffee drinking x time of day x exam duration
Factor coffee has three levels: 1 cup, 2 cup 3 cups
Factor time of day has two levels: morning or night
Factor exam duration has 4 levels: 30min, 60min, 90min,
120min
Difficulties with Factorial Designs
More complex, more time
Difficult to analyze if some data is missing
Need more Ss for each factor, for each
extra level
Interactions, especially higher order
interactions (3, 4 or higher), can be difficult
to understand
Main Effects and Interaction Effects
The effect of a single variable is known as a
main effect
The effect of two variables considered
together is known as an interaction
For the two-way between groups design, an
F-ratio is calculated for each of the
following:
The
main effect of the first variable
The main effect of the second variable
The interaction between the first and second
variables
Analysis of a 2-way BetweenSubjects Design Using ANOVA
To analyse the two-way between groups
design we have to follow the same steps
as the one-way between groups design
State
the Null Hypotheses
Partition the Variability
Calculate the Mean Squares
Calculate the F-Ratios
Null Hypotheses
There are 3 null hypotheses for the twoway (between groups design.
The
means of the different levels of the first IV
will be the same, e.g. A A A
1
2
The
means of thedifferent levels of the
second IV will be the same, e.g. B B B
1
2
means of the
differences between the
different levels of the interaction are not the
same, e.g. AB AB AB AB
The
11
21
21
22
An Example Null Hypothesis for an
Interaction
140
120
100
Lectures
80
No Lectures
60
40
20
0
Worksheets
No Practice
AB11 AB21 AB21 AB22
The differences betweens the levels of
factor A are not the same.
Partitioning the Variability
If we consider the different levels of a oneway ANOVA then we can look at the
deviations due to the between groups
variability and the within groups variability.
ASij T (Ai T )(ASij Ai )
If wesubstitute AB into the above equation
we get ABS T (AB T ) (ABS AB )
ijk
ij
ij
ij
This provides the deviations associated
with between and within groups variability
for the two-way between groups design.
Partitioning the Variability (Cont.)
The between groups deviation can be
thought of as a deviation that is comprised
of three effects.
(AB ij T ) Ai effect B j effect Ai xB j effect
In other words the between groups
variability
is due to the effect of the first
independent variable A, the effect of the
second variable B, and the interaction
between the two variables AxB.
Partitioning the Variability
The effect of A is given by
Similarly the effect of Bis given by
The effect of the interaction AxBequals
which is known as a residual
Ai T
Bj T
( ABij T ) ( Ai T ) ( Bk T )
The Sum of Squares
The sums of squares associated with the twoway between groups design follows the same
form as the one-way
SST SSA SSB SSAB SSS / AB
We need to calculate a sum of squares
associatedwith the main effect of A, a sum of
squares associated with the main effect of B, a
sum of squares associated with the effect of the
interaction.
From these we can estimate the variability due
to the two variables and the interaction and an
independent estimate of the variability due to the
error.
The Mean Squares
In order to calculate F-Ratios we must
calculate an Mean Square associated with
The
Main Effect of the first IV
The Main Effect of the second IV
The Interaction.
The Error Term
The mean squares
The main effect mean squares are given
SS
by: MS SS where df a 1
MS
where df b 1
A
A
df A
MSAB
A
B
dfB
B
SSAB
where df AB (a 1)(b 1)
df AB
The interaction mean squares is given by:
MSS / AB
B
SSS / AB
where dfS / AB ab(s 1)
dfS / AB
The error mean square is given by:
The F-ratios
The F-ratio for the first main effect is:
MSA
MSS / AB
The F-ratio forthe second main effect is:
MSB
MSS / AB
The F-ratio forthe interaction is:
MSAB
MSS / AB
An Example 2x2 Between-Groups
ANOVA
Factor A - Lectures (2 levels: yes, no)
Factor B - Worksheets (2 levels: yes, no)
Dependent Variable - Exam performance
(0…30)
Mean
Std Error
LECTURES
WORKSHEETS
yes
yes
19.200
2.04
no
25.000
1.23
yes
16.000
1.70
no
9.600
0.81
no
Results of ANOVA
• When an analysis of variance is conducted on
the data, the following results are obtained
Source
Sum of
Squares
df
Mean
F
Squares
p
A (Lectures)
432.450
1
432.450
37.604
0.000
B (Worksheets)
0.450
1
0.450
0.039
0.846
AB
186.050
1
186.050
16.178
0.001
Error
184.000
16
11.500
What Does It Mean? - Main effects
A significant main effect of Factor A
(lectures)
“There
was a significant main effect of lectures
(F1,16=37.604, MSe=11.500, p<0.001). The
students who attended lectures on average
scored higher (mean=22.100) than those who
did not (mean=12.800).
No significant main effect of Factor B
(worksheets)
“The
main effect of worksheets was not
significant (F1,16=0.039, MSe=11.500, p=0.846)”
What Does It Mean? - Interaction
A significant interaction effect
“There
was a significant interaction between the
lecture and worksheet factors (F1,16=16.178,
MSe=11.500, p=0.001)”
However, we cannot at this point say anything
specific about the differences between the
means unless we
look
at the null
hypothesis
AB11
LECTURES
yes
no
AB21
AB21
AB22
Mean
Std Error
yes
19.200
2.04
no
25.000
1.23
yes
16.000
1.70
no
9.600
0.81
WORKSHEETS
Many researches prefer to continue to make
more specific observations.
Analytic Comparisons in General
If there are more than two levels of a
Factor
And, if there is a significant effect (either
main effect or simple main effect)
Analytical
comparisons are required.
Post hoc comparisons include Tukey tests,
Scheffé test or t-tests (Bonferroni corrected).
2X2
Independent Variable A
3X3X2
B1
Independent Variable
C
A1
Independent
Variable A
A2
A3
A1
A2
C1
Independent Variable B
B1
B2
A1B1
A1 B2
A2B1
A2 B2
Independent Variable B
B2
C2
C1
C2
C1
C2
A1B1C
A1B1C
A1B2C
A1B2C
A1B3C
A1B3C
1
2
1
2
1
2
A2B1C
A2B1C
A2B2C
A2B2C
A2B3C
A2B3C
1
2
1
2
1
2
A3B1C
A3B1C
A3B2C
A3B2C
A3B3C
A3B3C
1
2
1
2
1
2
Assignment to conditions
B3
Randomized group—completely randomized factorial
Matched factorial design—form groups based on
matching and randomly assign to conditions
The Purpose of Factorial Models
Two or more variables at once
Combinations of variables
Main effects
The
effect of one independent variable,
averaged over all levels of another
independent variable
The effect of one independent variable,
ignoring the other independent variables
Interaction
effect
of one independent variable
depends upon the levels of another
independent variable
the effects of one variable varies over
the levels of another independent
variable
Main effects
Model
Stimulus
R
Group A
T frightened model
R
Group B
T not frightened
model
R
Group A
R
Group B
T plastic
snake
T plastic
flower
M emotional
response
M emotional
response
M emotional
response
M emotional
response
Interaction
Stimulus
Plastic flower
Plastic snake
Model Emotion
Frightened
Not frightened
Group A
Group C
Group B
Group D
Three hypotheses
Main
effects
Stimulus
Model emotion
Interaction—Stimulus
X model
emotion
Actual interaction found
Stimulus
Plastic flower
Plastic snake
Model Emotion
Frightened
Not frightened
No Fright
No Fright
Fright
No Fright
Results from similar study
Stimulus
Plastic flower
Plastic snake
Model Emotion
Frightened
Not frightened
.1
0
1.2
.1
Interactions
Graphic representation
Model
Frightened
Not Frightened
Response
1.00
0.75
0.50
0.25
0.00
Flower
Snake
Stimulus
Main effects conditioned by interaction
Mood and Memory
Recall Mood
Sad
Happy
Recall Mood Sad
Learning Mood
Sad
Happy
.8
.46
.45
.78
Recall Mood Happy
0.85
0.8
0.75
0.7
0.65
0.6
0.55
0.5
0.45
0.4
Sad
Happy
Learning Mood
Factorial Design Analysis
2X2
Three questions
Two main effects
Interaction
Source table
Source
SS
df
MS
F
Significance
f
B
3
1
t
w
e
e
D
T
y
1
1
v
e
t
N
0 s
3 3 . t.
2 2.
2
T
4 6.
1 8.
3 0.
1 1.
.
o
1
2 5
2
7
2
e
0
p
r
V
0 u
. 35 h
0M
. 89 S
N
. 60
. 77
. 07
. 68
0M
S
N
tM
S
N
e
0
4 33
0 85
1
3 60
8 73
1
8 06
9 53
3
e
O
e
e
0
t
0
t
a
t
1
9
3S
n
n
3 33
9 79
51
3 70
0 22
51
8 06
9 62
03
re
d
e
d
le
d
.
9 6
2
No effects
1
i
.
5
5
4
1
1
D
I
is
o
1
1
1
6
T
d
g
L
W (
3 3 a
5 1 .
3 5
3 0 a
6 1 .
3
3
2
6
5
7
3
0
a
.
.
1
2
.
.
.
8
9
4
t
.
.
.
a
t
oS
3
7
0
3
8
0
3
3
0
e
V
o
s
H
n
tt
Ln
u
q
5
6 4
7
8
2
1
8
8
2
3
e
S
1C .
5
I .
3I .
S.
I .
E.
T
4
0
1
t
n
h
E
e w
c
ig
D
n
p
S
S
1
4
1
e
m
uS
t
p
5
1
1
a n
h
h
e
0
0
8
.0
7
.
.7
0
0
3
o5
s
a R
lo
aq
.3
.7
.T 8
. N8
r 9
5
o6
f
k
e
du
3r 6
3
3
t 11
1V 15
o
n
w
e
v
D
e
v
2
0
0
6
3
t r 9
n
g
a
n
c
l)
fr
4r 1
3
e 7
1O 0
8
4
9
r
c
r
v
h
I
f
n
sr ig e
8
c
0
c
L
3
5
9N
2
r
l
9
9
e
q
A
d
I
o
a
ea S c F
6
9e
0
8
r
0E 7
2O7
o0
a
i
S
u
c
s
i
a
o
o
r
G
L
u
g
b
lv
o
a
Estimated Marginal Means of PERSUAS2
v 20.0
o
D
V
I
s
o
ru
a
2R 1
7V 1
r 0
6
t 0
.
o
ar
S
I
e
o
.o 2
.n 8
.N 1
8
1
Ci
-
e
M
18.0
ia
t
io
n
ia
t
io
n
ia
t
io
n
16.0
n
Estimated Marginal Means
.
3
Possible Patterns
14.0
12.0
10.0
Strength of Argument
8.0
Weak
6.0
Strong
High
Low
Involvement in issue (own school versus other school)
Main Effect No Interaction
e
n
-
D
p
e
M
u
m
e
a
q
S
u
d
u
F
S
a
i
f
g
a
a
C
0
6
9
0
0
3
7
4
0
I
n
6
6
1
0
7
1
7
8
0
I
N
6
6
3
0
7
1
7
5
0
S
0
0
2
6
2
1
2
9
4
I
N
6
6
1
3
7
1
7
8
3
E
3
5
6
3
6
7
e
T
0
6 0
0
C
3
5 3
9
D
a
R
I
N
e
n
g
Estimated Marginal Means of PERSUAS2
i
s
o
S
W
t
a
20.0
o
t
1
M
.
0
0
0
Estimated Marginal Means
S
2
1
5
18.0
N
5
5
0
2
M
.
0
7
3
16.0
S
4
9
9
N
5
5
0
14.0
T
M
o
0
3
7
S
5
9
0
12.0
N
0
0
0
Strength of Argument
10.0
Weak
8.0
Strong
High
Low
Involvement in issue (own school versus other school)
Two Main Effects No Interaction
e
n
-
D
p
e
M
u
m
e
a
q
S
u
d
u
F
S
a
i
f
g
a
a
C
8
6
4
0
3
3
1
7
0
I
n
1
1
2
0
7
1
7
0
0
I
N
5
5
8
0
0
1
0
7
0
S
1
1
9
0
7
1
7
7
0
I
N
1
1
5
8
7
1
7
8
8
E
0
5
1
0
6
4
T
0
6 0
0
C
8
5 3
9
a
R
e
Estimated Marginal Means
D
I
N
e
n
g
22.0
i
s
o
S
W
t
a
o
t
1
M
.
3
3
3
20.0
S
8
4
9
18.0
N
5
5
0
2
M
.
7
0
3
16.0
S
2
1
3
N
5
5
0
14.0
T
M
o
0
7
3
12.0
S
5
3
5
N
0
0
0
10.0
Strength of Argument
8.0
Weak
6.0
Strong
High
Low
Involvement in issue (own school versus other school)
Interaction Only
e
n
-
D
p
e
M
u
m
e
a
q
S
u
d
u
F
S
a
i
f
g
a
C
3
1
4
0
3
3
1
1
0
I
n
6
6
3
0
7
1
7
2
0
I
N
0
0
8
8
0
1
0
7
9
S
6
6
2
5
7
1
7
7
4
I
N
6
6
1
0
7
1
7
0
0
E
0
5
1
0
6
4
T
0
6
0
0
e
C
3
5 3
9
D
a
R
I
N
e
n
g
i
s
Estimated
Marginal Means of PERSUAS2
o
S
W
t
o
t
1
M
.
7
7
7
20.0
Estimated Marginal Means
S
6
3
5
N
5
5
0
18.0
2
M
.
0
3
7
S
0
9
8
16.0
N
5
5
0
T
M
o
3
0
7
14.0
S
3
8
7
N
0
0
0
12.0
Strength of Argument
10.0
Weak
8.0
Strong
High
Low
Involvement in issue (own school versus other school)
Main Effect and Interaction
e
n
-
D
p
e
M
u
m
e
a
q
q
S
u
d
u
F
S
a
i
f
g
a
C
6
5
6
0
7
3
6
6
0
I
n
6
6
2
0
7
1
7
8
0
I
N
0
0
5
8
0
1
0
0
8
S
6
6
5
0
7
1
7
6
0
I
N
0
0
9
0
0
1
0
2
0
E
6
5
9
7
6
8
T
0
6
0
0
e
C
3
5 3
9
D
a
R
I
N
e
n
g
i
s
o
S
W
t
a
Estimated Marginal Means
of PERSUAS2
o
t
1
M
.
7
0
3
18.0
Estimated Marginal Means
S
6
5
6
N
5
5
0
16.0
2
M
.
7
0
3
S
1
8
4
14.0
N
5
5
0
T
M
o
7
0
3
12.0
S
6
0
2
N
0
0
0
10.0
Strength of Argument
8.0
Weak
6.0
Strong
High
Low
Involvement in issue (own school versus other school)
Two Main effects and Interaction
e
n
-
D
p
e
M
u
m
e
a
q
S
u
d
u
F
S
a
i
f
g
a
a
C
5
1
5
0
0
3
7
8
0
I
n
1
1
5
0
7
1
7
8
0
I
N
1
1
6
0
7
1
7
1
0
S
1
1
8
0
7
1
7
2
0
I
N
1
1
3
0
7
1
7
0
0
E
3
5
1
3
6
0
T
0
6 0
0
e
C
8
5 3
9
D
e
a
R
N
eI
n
g
i
s
s
o
S
W
t
a
t
Estimated Marginal t
Means of PERSUAS2
o
1
M
.
0
3
7
Estimated Marginal Means
20.0
S
2
2
1
N
5
5
0
18.0
2
M
.
0
0
0
S
8
7
1
16.0
N
5
5
0
T
M
o
0
7
3
14.0
S
0
3
7
N
0
0
0
12.0
10.0
Strength of Argument
8.0
Weak
6.0
Strong
High
Low
Involvement in issue (own school versus other school)
Interactions
Interactions
Antagonistic
Synergistic
Ceiling
effect
Higher Order Designs
Extensions within a two way design
More
3
levels 2 X 3
X3
Higher order designs
More
factors
2X2X2
Limitations
Interpretation of higher order interactions
Number of conditions and participants
3 X 3 X 3 X 3 = 81 cells
Higher order interactions
Involvement in
issue
High
Low
Strength of arguments
Weak
Strong
Number of
Number of
arguments
arguments
3
9
3
9
4.10
1.05
8.32
11.3
4.52
7.71
4.95
8.66
Low Involvement
10
9
8
7
6
Weak Arguments
5
Strong Arguments
4
3
2
1
0
3
9
Number of Arguments
High Involvement
12
10
8
Weak
Strong
6
4
2
0
3
9
Number of Arguments
What Is Multivariate Analysis of
Variance (MANOVA)?
Univariate Procedures For Assessing
Group Differences
The
t Test
Analysis of Variance (ANOVA)
Multivariate Analysis of Variance
(MANOVA)
The
Two-Group Case: Hotelling's T2
The k-Group Case: MANOVA
Group Comparison Analyses
ANOVA
ANCOVA
MANOVA
MANCOVA