Professional Documents
Culture Documents
DESIGNS
FRANCISCO S. ANTONIO
Associate Professor
CAGAYAN STATE UNIVERSITY
Experimental Designs
set of rules, plans and course of
action taken in the conduct of
experiment
Attributes of
ExperimentalDesigns
ensure cost effective collection of appropriate data
Linear Model:
Yij = μ + ti + eij
Testing Significance of
Treatments via ANOVA F test
1. State the statistical hypothesis
2. Group Comparisons
3. Trend Comparisons
Statistical Hypothesis for
Pairwise Comparision
Ho: u1-un=0 , the treatment means for u1 and un are
not different
Ha: u1- un =/= 0, the treatment means u1 and un
are different
Commonly Used Tests for
Pairwise Comparisons
1. Least Significant Difference (LSD) test
2. Orthogonal contrasts
3. Set of orthogonal contrasts
TREND COMPARISON
determine quantitative functional relationship
between the treatments and response variable by
trend comparisons which fit the orthogonal
polynomial trends such as linear, quadratic, cubic ,
quartic, and other higher degree of polynomials
Randomized
Complete Block
Design (RCBD)
Attributes of RCBD
a. precision is increased by proper grouping or
blocking of the experimental units as a form of local
control
b. eu’s are grouped into blocks such that differences
between units among different blocks are greater
than differences between the units within each block
Attributes of RCBD
c. blocking is done so that blocks cut across or ar
perpendicular to the direction of the eu’s gradient
d. with differences among the blocks, variability is
removed from the experimental error improving the
precision of the experiment
Randomization and
Layout
Conditions : no. of treatments = 4
no. of replication = 3
c. for each block, allocate the treatments into the eu’s at random and independently of
3. using the sequence of draws as the treatment numbers and the drawn numbers as
e. lab trials with replication over time such that differences among
experimental units conducted at the same time and among those
conducted over time constitute the two known sources of variability
Features of Blocking,
Randomization and Layout in
LSD
a. number of treatments = number of columns =
number of rows
b. treatments are then arranged in blocks in two ways;
by rows and columns such that each treatment occurs
only once in each column and row
c. each row is a complete block, each column is
likewise a complete block
LSD Randomization
and Layout
ILLUSTRATION of LSD
An experiment is to be conducted to
measure lead content of 4 seaweeds . It is
suspected that the source location (S1, S2, S3,
S4) and the chemists (C1, C2, C3, C4) that perform
the chemical analysis contribute to the variability
of the results. It is assumed that the source
location, the chemists and kinds of seaweeds have
no interaction.
Treatments= T1, T2, T3, T4
Row classification =S1, S2, S3, S4
Column Classification=C1, C2, C3, C4
Steps in Randomization
1. Randomize the assignment of the source location
to the rows on the plan
Column Column 2 Column 3 Column 4
1
Row 1 A B C D
Row 2 B C D A
Row 3 C D A B
Row 4 D A B C
Steps in Randomizaton
2. Randomize the assignment of the source
location to the rows on the plan
Draw 1 2 3 4= Course location
sequence:
Drawn 3 1 4 2= Plan rows
number :
Colum Column Column Column 4
n1 2 3
S3 A B C D
S1 B C D A
S4 C D A B
S2 D A B C
Steps in Randomization
of the chemist to
3. Randomize the assignment
the columns in the plan
Draw 1 2 3 4= chemists
sequence:
Drawn 3 1 4 2= Plan columns
number :
C3 C2 C4 C1
S3 A B C D
S1 B C D A
S4 C D A B
S2 D A B C
Steps in Randomization
4. Randomize the assignment of treatments to the
treatments on the plan
Draw 1 2 3 4 seaweeds
sequence:
Drawn B(4) 1(D) 4(C) 2(A) Plan columns
number :
C3 C2 C4 C1
S3 T4 T1 T3 T2
S1 T1 T3 T2 T4
S4 T3 T2 T4 T1
S2 T2 T4 T1 T3
Final Layout
of LSD
5. Rearrange the rows and columns in sequence
(LAYOUT)
C1 C2 C3 C4
S1 T2 T1 T4 T3
S2 T4 T3 T1 T2
S3 T1 T2 T3 T4
S4 T3 T4 T2 T1
LSD Data Presentation
Row
Class’n
Column Classification Row Total
1 2 3 4
1
2
3
4
Col Total
Treatment T1 T2 T3 T4 Total
Total
ANOVA TABLE
Source of Degree of Sum of Mean of Fc Fα=.05 Fα=.01
Variation freedom Squares Squares
Rows r-1 RSS MSR MSR/M
SE
Columns c-1 CSS MSC MSC/M
SE
Treatments t-1 TrSS MSTr MSTr/
MSE
Error (r-1)(c-2) ESS MSE
TOTAL t2-1 TSS
FACTORIAL
EXPERIMENTS
FACTORIAL
EXPERIMENTS
an experiment where the treatments consist of all
treatment combinations of two or more factors
Common Terms Used in Factorial
Experiments
*Factor - a kind of treatment whose effect is to
be measured in an experiment
*Levels – the different amounts or categories of a
factor
*Treatment Combinations – treatments formed by
combining the different levels of two or more
factors
KINDS OF TREATMENT EFFECTS
Simple effect –difference between two levels of
the factor at a given level of another factor
an experiment involving two factors where
the treatment combinations are randomized over
the eu’s using RCBD
ILLUSTRATION
An experiment to assess simultaneously effect of
Factor A three plantind densities (A1, A2, A3)
and Factor B the kind of fertilizer with 4 levels
(B1, B2, B3, B4) using RCBD with three (3)
replications
Treatment Combinations= 3 x 4 = 12 treatment
combinations
Illustration of
Factorial
Experiment
Treatment Combinations= 3 x 4 = 12 treatment
combinations
T1-A1B1 T1-A2B1 T1-A1B1
T1-A1B2 T1-A2B2 T1-A1B2
T1-A1B3 T1-A2B3 T1-A1B3
T1-A1B4 T1-A3B4 T1-A1B4
RANDOMIZATION AND LAYOUT
OF 2-FACTOR in RCBD
For block 1
Order of 1 2 3 4
5 6 7 8 9 10 11 12
Draws:
Trt 5 9 2 11 7 1 4 6 10 12 3 8
combinat’n:
For block 2
Order of 1 2 3 4 5 6 7 8 9 10 11 12
Draws:
Trt 5 9 2 11 7 1 4 6 10 12 3 8
combinat’n:
For block
Order of 1 32 3 4 5 6 7 8 9 10 11 12
Draws:
Trt 8 5 10 9 4 2 3 12 3 11 1 7
combinat’n:
Layout
1 2 3
1 2 3
1 2 3
A2B1 A3B A1B A1B3 A1B A2B3 A 2 B A 1B A3 B2
4
1 2 1 4
4 5 6 4 5 6 4 5 6
A1 B1
B2
B3
B4
A2 B1
B2
B3
B4
A3 B1
B2
B3
B4
ANOVA F-TEST FOR A 2-FACTORAL
RCBD
Construct the two- way table of total as follows
Factor Factor B A
A totals
1 2 3 4
1
2
3
4
B totals
Construct the ANOVA Table
Source of Degree of Sum of Mean of Fc Fα=.05 Fα=.01
Variation freedom Squares Squares
Block r-1 RSS MSR MSR/MSE P<.05 P<.01
Treatment ab-1 TrSS MSTr MSTr/MSE
A a-1 ASS MSA MSA/MSE
B b-1 BSS MSB MSB/MSE
A xB (a-1) (b-1) ABSS MSAB MSAB/MS
E
Error (r-1)(ab-1) ESS MSE
TOTAL Abr-1 TSS
NOTES ON INTERPRETATION
OF FACTORIAL EXPERIMENTS
In general, tests on main effects have
straightforward interpretation only when the
interactions are not significant
The presence or absence of main effect tellls
nothing about the presence or absence of
interactions
With a significant interaction, the factors do not
act indepenently of each other; th simple effect
depends upon the level of the other factor of the
interaction term
NOTES ON INTERPRETATION
OF FACTORIAL EXPERIMENTS
If the interaction is not significant, the factors
act independently of each other, the simple
effects of a factor are the same for all levels of
the other factor and so the simple effects are
equal to the corresponding main effect
Non-independent factors indicate the complexity
of the situation which have been missed had a
single-factor approach to experiment been used
for each factor
CRD WITH SUBSAMPLING
-
CRD WITH SUBSAMPLING
more than 1 observation on each experimental unit
is done in CRD with subsampling
ILLUSTRATION OF crd WITH
SUBSAMPLING
An experiment is conducted to determine the
sugar content of three varieties of sugarcane. At
maturity, four cnes of each variety are available
for refractometer brix test. From each cane,
three determinations are made: at the base,
middle and top portions.
DATA PRESENTATION OF CRD
WITH SUBSAMPLING
Treatment Sampling
Unit
Experimental Unit Treatment Treatment
Total Mean
1 2 3 4
1 1
2
3
Total
2 1
2
3
Total
3 1
2
3
Total
TOTAL
ANOVA TABLE
Source of
Degree Sum of Mean of Fc Fα=.0 Fα=.0
Variation of Squares Squares 5 1
freedom
Treatment t-1 TrSS MSTr
Experimenta t(r-1) ESS MSE
l Error
F2 F4 F1 F1 F4 F3 F3 F3 F2
F4 F3 F2 F2 F2 F1 F1 F4 F1
F3 F1 F4 F4 F3 F4 F2 F2 F3
T3 T1 T2 T1 T3 T2 T2 T1 T3
Data Presentation
Type of Kind of Farm Mean
Irrigation
Fertilizer
1 2 3
F1
T1 F2
F3
F4
F1
T2 F2
F3
F4
F1
T3 F2
F3
F4
Analysis of Results
Test the significance of treatments via
ANOVA F-test
SV df SS MS Fc
Mainplot
Block r-1 RSS MSR MSB/MSEa
A a-1 ASS MSA MSA/MSE
a
A1
A2
A3
Factor
B
means
Analysis of Variance
(ANOVA)
SV Df
SS MS F
Block r-1 RSS MSR MSR/MSEa
Horizontal a-1 ASS MSA MSA/MSEa
Factor (A)
Error(a) (r-1)(a-1) EaSS MSEa
Vertical b-1 BSS MSB MSB/MSEb
factor (B)
Error (b) (r-1)(b-1) EbSS MSEb
AXB (a-1)(b-1) ABSS MSAB MSAB/MSE
c
ErroR© (r-1) (a-1) EcSS MSEc
(b-1)
Total Rab-1 TSS
Strip-Split-Plot
Design
Randomization and Layout
B1 B2 B3
A2
A3
C2
C1
C1
C1
C2
C1
C1
C2
C2
C2 C2 C1
A1 C2 C2 C1
C1 C1 C2
B2 B1 B3
A3 C1 C2 C2
C2 C1 C1
A1 C2 C1 C1
C1 Block1
C2 C2
A2 C2 C1 C2
C1 C2 C1
Randomization and Layout
B3 B1 B2
A1 C1 C2 C1
C2 C1 C2
A3 C1 C2 C2
C2 C1 C1
A2 C2 C1 C1
C1 C2 C2
B1 B3 B2
A2 C1 C1 C2
C2 C2 C1
A1 C2 C2 C1
C1 C1 C2
Block3
A3 C1 C2 C1
C2 C1 C2
Block4
(ANOVA)
SV df SS MS F
Block
Horizontal
r-1
a-1 RSS
ASS
MSR
MSA
MSR/MSEa
MSA/MSEa
Factor (A)
Error(a) (r-1)(a-1) EaSS MSEa
Vertical factor b-1 BSS MSB MSB/MSEb
(B)
Error (b) (r-1)(b-1) EbSS MSEb
AXB (a-1)(b-1) ABSS MSAB MSAB/MSEc
Error (c) (r-1) (a-1)(b-1) EcSS MSEc
Split plot factor c-1 CSS MSC MSC/MSEd
(C)
AxC (a-1)(c-1) ACSS MSAC MSAC/MSEd
BxC (b-1)(c-1) BCSS MSBC MSBC/ MSEd
AxBxC (a-1)(b-1)(c-1) ABCSS MSABC MSABC/ MSEd
Error (C) ab(r-1)(c-1) EdSS MSEd
Total rabc-1 TSS
Data Presentation
Type of Depth of Fertilizer rate Block
Irrigation Plowing 1 2 3 4
Light
Shallow None
Average
Heavy
Medium None
Average
Heavy
Deep None
Average
Heavy
Heavy Shallow None
Average
Heavy
Medium None
Average
Heavy
Deep None
Average
Heavy
SUMMARY
STATISTICS ASIDE
FROM ANOVA
1. Coefficient of variation(%CV) – experimental error
expressed as percentage of the mean
-measures degree of precision of the experiment
-higher %CV, lower reliability of experimental result
2. Standard error of a treatment mean(se of a mean) –
measure the average error in estimating the true
treatment mean
-measures degree of a precision of the mean as a
estimate of the true treatment mean
3. Standard error of the difference between two
treatment means
-measure the average error in estimating the
difference between two treatment means
a. estimates of treatment means and effects
b. estimates of the effect of treatment