Engineering

data analysis
Han Baucas Bonoy
Instructor
 CHAPTER IV:
Experimental DesignS
“Statistical thinking will one day be
necessary for efficient citizenship as the
ability to read and write”

Samuel Wilks, 1951

Adapted from H.G. Well’s book
“Mankind in the Making”
EXPERIMENT
An operation or procedure carried out
under CONTROLLED
CONDITION in order to discover
unknown effect, to test or establish
a hypothesis, or to illustrate a
known fact
CONTROLLED CONDITION
Known or regulated by the
experimenter
What we need…

1. Accuracy: closeness to the

absolute or true value of
the quantity measured

2. Precision: the closeness to

one another of a set of
separate measurements of a
quantity
In an EXPERIMENT, there are …
EXPERIMENTAL ERROR
Variation produced by unknown factors beyond the control or not considered
by the experimenter (e.g. extraneous variables, confounding variables)

Systematic error: not calibrated

measuring tool; observer’s error
due to fatigue; wrong implementation of LESS ACCURATE
protocols; inappropriate design;
inappropriate statistical analysis, etc.

Random error: unpredictable incidence

of pests and diseases; meteorological LESS PRECISE
events; etc…
Principles of Experimentation
“DEVICES” to estimate the experimental
errors (systematic and random)
A. Replication
B. Randomization

“DEVICES” to control the error

A. Blocking
B. Proper Plot Technique
C. Data Analysis
 REPLICATION
It refers to number of distinct experimental units
under the same treatment.
Why replication is necessary?
1. More precise estimates of the experimental error
A
2. Decreases the experimental error, and thus
increase precision of the variability.
3. Improves the precision of effects estimation B
How many replication required?
• At least two replications are C
needed to measure experimental
error
1. ABSOLUTE MINIMUM: 3 replications
• Minimum numbers of replication
varies with discipline and researches. 2. FOLIAR FUNGICIDE: at least 4
replications
3. SOILBORNE DISEASE: 5 and even 6
replications
-uneven distribution of organisms in the
soil.
Factors that determine the
number of replication.
1. Uniformity of experimental materials/
pattern or magnitude of variability in some
aspects of the experiments i.e. soli gradient.
2. Number of treatments.
3. Degrees of precision required/ level of
significance required.
4. Time allotment, cost and availability of
fund

NOTE: The higher the number of replication,

the lower variance become. Hence,
precision in the experiment increases

©Casler,
2015
RANDOMIZATION
Random process of assigning
treatments to the experimental
units

Why randomization is necessary?

1. To remove bias/ random error
2. To lessen extraneous sources of
variability (confounding variables)
3. Forms a basis for valid statistical tests
or estimates of random error

©Tad-awan, 2017
(presentation)
CONTROL OF ERROR
• Process to reduce or control the variation due to extraneous
factors and increase the precision of the experiment.

BALANCING: arrangement of the treatments are

balanced; making experimental units uniform/
homogeneous.

BLOCKING AND GROUPING: forming a

relative uniform groups. A block is also a
replicate.
CONTROL OF ERROR
Field Plot Techniques
If the fertility gradient is going in
one direction,
blocks with long and narrow plots
should be made perpendicular to
the gradient.

If the slope is going on both directions

(north to south and east to west), or when
fertility gradient is not uniform, make the
block as square as possible.

©Tad-awan, 2017
(presentation)
CONTROL OF ERROR
BORDER EFFECTS
A • Plants along the edges of plots often perform
differently than those in the center of the plot
• Border rows on the edge of a field or end of a plot–
less competition for resources
• Plants on the perimeter of the plot can be influenced
by adjacent plots
•
B •
Fertilizer or irrigation can move from one plot to the
next
Impact of border effect is greater with very small plots

MINIMIZING BORDER EFFECTS

• Leave alleys between plots to minimize strain

• Remove plot edges and measure characters only on
C •
center portion
Plant border plots surrounding the experiment

©Tad-awan, 2017
(presentation)
DATA ANALYSIS
LEVEL OF SIGNIFICANCE

• The probability of rejecting the

null hypothesis when it is true.
Type I Error: False positive; when
concluding a significant
difference when there is none.
Type II Error: False negative;
when concluding a no
significant difference when
there is.
DATA ANALYSIS
NON-SIGNIFICANT RESULTS

The NONSIGNIFICANCE result in the analysis indicates the failure

of the experiment, or the gathered data, to detect any differences
among treatments. It does not in any way prove that all the
treatment are the same because the failure to detect treatment
difference could be the result of either very small or nil treatment
or a very large experimental error, or both.

The researcher should examine the size of the

experimental error and the numerical difference among
treatment means. If the mean values are considered large,
evaluate the possible sources of experimental errors and
the experiment may be repeated employing all efforts to
reduce the experimental errors.
Gomez and Gomez (1974)
DATA ANALYSIS
SIGNIFICANT RESULTS

The SIGNIFICANCE result in the analysis at p-value = 0.01,

means that chances are less than 1 in 100 that the observed
differences among the treatments could be due to chance. Note
that the test verifies the existence of some differences among
the treatments but does not indicate which treatments are
‘different’ and not.

To obtain which treatments are significantly different and

not, there are procedures, post hoc tests, that are commonly
utilized, namely; Duncan Mean Range test (DMRT), Least
Significant Difference (LSD), and Tukey Honest Significant
Difference (Tukey HSD).
Gomez and Gomez (1974)
DATA ANALYSIS
POST HOC TESTS
DMRT is applicable for unplanned pairwise comparison. Unplanned
comparison means that there is no specific pairs of treatments that
are chosen to be compared prior to the conduct of the experiment.

LSD is the simplest, most common, and most appropriate for planned
pairwise comparison. Planned comparison means that there are
specific pairs of treatments that are identified to be compared prior to
the conduct of the experiment. The very common is the pair
comparison between control treatments against all other treatments.

Tukey HSD most advisable when assumptions of normality and

homogeneity is formally tested. Also, this test is commonly used
for comparison of treatments for pests and diseases where
researchers tend to consider the possibility of committing Type I
Error (False Positive).
Gomez and Gomez (1974)
Coefficient of Variation (CV)
• Measure of relative variability/ dispersion of data points in a data
series around the mean
• Ratio of the standard deviation to the mean expressed in
percentage.
• It indicates precision and accuracy of your data.

ACCEPTABLE CV?
• Not more than 15% for field crops(rice, corn, soybean, wheat)
• Not more than 25% for horticultural crops (fruit and leafy
vegetables, fruit trees)
• Not more than 40% for crop grown under the soil

NOTE: No established so far. Some statistician argue, that it must

be below 50%.
Coefficient of Variation (CV)
VERY HIGH CV VERY LOW CV

Indicates something wrong in It indicates that very

the conduct of the experiment: small differences among
improper lay-out, too much
variability in the area probably treatment means will be
due to plot size, topography, detected to be significant
uneven or severe insect or which may lead to a
disease population which are not conclusion or
part of the experiment, severe recommendation that is
weather conditions, loss of not appropriate or
samples due to animals, etc. practical.

Experiment is invalid and

results are unacceptable.
 Coefficient of Determination (r2)
• It indicates the percentage variation in the character of interest (e.g.
yield) which is explained by the sources of variation (e.g.
treatments).

• Simply the ratio of the sum of square of a particular sources of

variation (e.g. Treatment) with the Total Sum of Square.

= between Grp. There is 24.86% variation in the yield

SS/TotalSS explained by the treatments (between
= 91.467/367.867 groups).
= 0.2486
COMMON EXPERIMENTAL
DESIGNS
 Completely Randomized Design (CRD)
Single-Factor Randomized Complete Block Design (RCBD)
Latin Square Design (LSD)
Factorial Design
Two Factors
Split Plot Design
Factorial Design
Three Factors
Split-Split Design
Completely Randomized Design
HOMOGENOUS ENVIRONMENT
Treatments are assigned completely at
random so that each experimental units
has the same chance of receiving any one A
treatment.

Only appropriate for experiments with

B
homogenous experimental units, such as
laboratory experiments, where C
environmentally effects are relatively
easy to control.
Randomized Complete Block Design
The design is especially suited for field
experiments where number of
treatments is not large and the A B
experimental area has a predictable
productivity gradient.

Unique feature against CRD: presence of

blocks of equal size where each of C
which contains all the treatments.
Latin Square Design
COLUMNS

The design is applicable only for

experiments in which the number
of treatments is not less than four A
D B C D
or more than eight.

ROWS
All experimental units (treatment
and replication combinations) are
completely randomized in each
row and column. Thus, each
treatment must ONLY occur
once in every row and in every
column.
Factorial Design
(2-factors)
• All experimental units (treatment and
replication combinations) are
completely randomized where the
effects of two or more factors were
taken into consideration. Also, each of
the given factors are of EQUAL
importance.
• Factorial design investigates all
treatment combinations of two or
more variables
• Factorial design allows us to test for
interactions between treatment
variables
Lay-out example:
3 x 3 x 4 Factorial in RCBD
Split-Plot Design

• All experimental units

(treatment and replication
combinations) are
completely randomized
wherein there are two
factors of which one is of
more importance than the
other.

Lay-out example:
Split Plot in RCBD
Thank you…

Ronald A. Fisher (1890 – 1962)

A British Statistician and Geneticist who pioneered experimental designs.