Module 2:

Experimental Designs and Layout

 Learning Outcomes
At the end of this module participant should:
• At the end of this module, participants should be competent to
undertake the following:
• Apply the fundamental principles of replication, randomization
and blocking in design of experiments
• Statistical experimental design using completely randomized
designs (CRD), randomized complete block design (RCBD) and
split plots
• Develop a detailed experimental protocol for use for data
collection
• Layout a laboratory of field experiment based on an
experimental protocol
• Collect statistically sound samples from laboratory experiments
and field experimental trials
 Facilitation and evaluation methods
This module will entail:

-Lectures on fundamental principles

-Group discussions

-Group practicals on experimental designs and layout

-Questions and answer sessions to identify and address

key skills/knowledge gaps

Material covered in this module will be evaluated through

the mid-and post-course evaluation test
 Experimental Designs
• Statistical experimental design forms the basis for the collection of
reliable and unbiased data can be obtained from the subject of
study

• Experimental design determines the layout or set up of an

experiment in the physical space, and subsequent statistical analysis
of the data

• Experimental designs should address fundamental issues;

-(1) unbiased collection of data;

-(2) enable estimation of descriptive and inferential statistics;

-(3) minimize the effects of confounding factors, and in some cases

(4) Provide a basis for comparison with background or baseline

condition
 Experimental designs
• An experiment is designed to test a hypothesis and entails collecting data on effect of
experimental factor(s) on the response variables

• Experimental factor: the phenomena whose effect is being investigated e.g. type of
irrigation system, type of emitter design, crop variety

• Experiments can have single or multiple factors being investigated simultaneously

• Multiple factor experiments are called factorial experiments

• Knowing the number of experimental factors is key to selection of statistical tests for
hypothesis testing

• In any experiment, each factor will have at least two levels

• A level is a quantity or variant of a factor

• Assume an experiment testing the effects of three emitter types 1, 2 and 3 on

discharge

• In this case, Emitter is the Factor

• The factor has three levels: 1, 2 and 3.
 Experimental designs
• List other experimental factors and levels
common in irrigation based on your experience
(1)

(2)

(3)

(4)
 Experimental Designs
• Factorial experiments: entail more than one experimental factors
investigated simultaneously

• For example, if you intend to test the effects of irrigation type and wheat
variety you need to set up a factorial experiment with the two factors:

• (1) Factor 1: irrigation type – how many types of irrigation (levels)?

• (2) factor 2: wheat variety: how many varieties (levels) of the wheat)?

• In summary, factorial experiment have factors (n)>1. i.e. n=2, 3, 4 etc.,

but higher order factorial experiments are difficult to manage and analyze
statistically.

• For a start consider going up to a maximum of 3 for on-station

experiments and at most two for on-farm or farmer-managed experiments
 Experimental Designs
• An experimental treatment is what is allocated to an
experimental unit or a plot for field-based experiments

• Single factor experiment: level is the treatment

• E.g. Emitter type 1 is a treatment that will be allocated to a
plot/experimental unit,

• Factorial experiment: A combination of levels of factors an

experimental Treatment. E.g. for an irrigation type x wheat
variety experiment the following are treatments:
(1) Drip irrigation and wheat variety Induna
(2) Sprinkler irrigation and Wheat variety SC Sanai

For other factorial experiments of your choice identify, the

factors, levels and treatments
Experimental designs: Fundamental principles
• Regardless of the number of factors and levels, all
experiments should adhere to the following
fundamental experimental design principles:
• (1) Replication:

• (2) Randomization

• (3) Blocking or minimization of confounding non-

experimental factors, and in most cases in irrigation
and agriculture

• (4) Control
 Experimental Designs: Fundamentals
(1) Replication:

• Means having more than one experimental unit for each

treatment. Consider a single experiment to test effect of
three irrigation method on wheat yields

• Factor: irrigation method

• Levels (3): (1) drip, (2) Sprinkler; (3) Furrow

• Assume number of replicates = 3
• Then we need 3 similar plots for each of the irrigation
methods i.e. each treatment has replicates, 1, 2 and 3.

• How many plots or experimental units are needed for the

whole experiment?
 Experimental Design Fundamentals
• (1) Replication

• Statistical data analysis entails two processes:

• (1) calculation of descriptive statistics (average, standard deviation, mode,

median)

Can we calculate the mean yield of wheat for drip irrigation without replicated
plots allocated the same treatment – Try it and see

(2) Inferential statistics i.e. Comparison of average behaviour of treatments

requires statistical data analysis
Statistical data analysis requires calculation of a term called degrees of freedom
(DF)

DF = number of replicates (n) minus one i.e. DF = n-1

If DF = 0 data cannot be analyzed statistically i.e. the data are just values but we
cannot attached statistical inference to it.

In summary, why then do we replicate??

 Experimental Design Fundamentals
(2) Randomization

• Randomization is meant to avoid bias in allocation of treatments to

experimental units or plots, and hence ensure collection of reliable
experimental data.

• Bias: an non-experimental conditions that may favour or disadvantage

other treatments

• If we don’t control bias, our research results won’t be reliable i.e. we are
likely to draw wrong conclusions that cannot be reproduced in reality

• What examples of experimental bias can you think of in typical irrigation

research. One is bias introduced by different soil types. What other
examples can be considered as bias?

• In summary to avoid bias, we allocate treatments to experiment units

randomly
 Experimental Design fundamentals
• Randomization:
• Assume we have 3 treatments replicated 3 times, giving 9 experimental units
or plots. How then do we do randomization?

• First determine number of experimental units (plots) required

• Number plots sequentially from 1, 2…N
• Have an array of your treatments corresponding to N total number of
experimental units

• Randomize either by:

• (1) Put cards with treatments in hat and draw without replacement. i.e., first
one goes to plot # 1, 2nd to plot # 2 etc. until all plots/units are allocated a
treatment
• (2) use a computer or scientific calculator to generate random numbers,
arrange numbers in descending and ascending orders and use that
sequence to allocate treatments to units/plots

• We will do a practical on this!

 Experimental design fundamentals
(3) Blocking:
• In some instances, we have well-developed environmental
gradients that may interfere or confound our experimental
results e.g. prevailing wind in sprinkler irrigation

• Consider an experiment to investigate effects of wheat variety

on yield under sprinkler irrigation

• If not properly designed, some varieties will end up received

more water than others due to wind drift. Consequently we
may erroneously conclude that certain wheat varieties
perform better than others, but such results can’t be
reproduced when you have no wind drift.
• Therefore to overcome gradients effects we use blocking
 Experimental Design Fundamentals
• (3) Blocking

• Blocking means that each set of replicates is aligned perpendicular to the

environmental gradient
• In such an experimental design, the blocks represent the replicates

• Blocking ensures that each treatment experiences the whole gradation of

the gradient (i.e. high, moderate And low)

• In this case randomization is done per plot rather than for the whole
experimental area/physical space

• In summary: blocking is used to avoid the confounding effects of known

environmental gradient. In principle, blocks should be aligned
perpendicular to the gradient.

• If you have two know gradients, block using the one which you think has
the strongest potential confounding effects on your experimental results
 Experimental design Fundamentals
• (3) Blocking:

• Consider a wheat variety experiment under sprinkler irrigation

• Factor: wheat variety

• Levels 3: (1) SC Nduna, (2) SC Sanai and (3 SC Stallion

• Assuming 3 replicates, then number of plots = 9

• Field site has a strong prevailing wind from East to West

• To address problem, we need to design an experiment entailing

blocking

• Experimental designs with blocking have the term ‘’block’’ in theer

name e.g. randomized complete block design (RCBD)

• We will cover details when discussing specific experimental designs

and during one of our practicals
 Experimental Design Fundamentals
(4) Control:

• In most instances, we conduct research to improve on an existing system

• Such existing system or current practice, provides background or baseline

condition/behaviour.

• Therefore it is important to compared an improved system or new technology to that

background/baseline condition.

• In cases where the improved systems or new intervention is compared to the baseline
condition, such condition will serve as a Control treatment.

• The purpose of a control is to measure the level of improvement (or lack of it)
attributable to the new intervention

• In irrigation/agricultural research its is prudent to include a control treatment to an

experiment design. In some cases we may have two controls: negative and positive
control. For example if you need to investigate the effects of introducing drip irrigation
on maize yields, you may need data from rainfed maize plots (negative control), and
another from furrow irrigated maize. Both are controls representing different baseline
conditions: one is no irrigation, the other is with a low-technology irrigation system.

• Like other treatments, the control will need to be replicated.

 Classical Experimental designs
• Based on the fundamental principles discussed, several classical
experiment designs have been developed. These include:

• Most common and widely used:

• (1) Completely Randomized Design (CRD)

• (2) Randomized complete Block design (RCBD)

• (3) Split-plots or Split-Units design

• These are robust designs that are adequate for most research in
irrigation and agriculture, and form the core of this training course.

• Advanced but less used designs include:

• (1) lattice design
• (2) Latin square design, among others
 Classical Experimental Designs
(1) Completely Randomized Design (CRD):
• Most simplest experimental design

• Experimental space is divided into units/plots to fit

replicated treatments

• Treatments are randomly allocated to units/plots over the

whole area hence ‘’completely randomized’’

• Application: This design is ideal where there are no know

systematic gradients in the physical space. I.e.
experimental space is considered relatively uniform or an
variation in environmental conditions is considered to be
random
 Classical Experimental Designs
• Completely randomized design (CRD) Example:

• Consider a maize variety evaluation experiment

under furrow irrigation. Field site has no known
gradient and soils vary randomly from fine sandy to
medium sand

• Factor: maize variety

• Levels: 3 (PAN413, SC403, farmers variety (FV)

• You are required to design an experiment to test the

effects of maize variety on grain yield and yield
components under smallholder furrow irrigation
 Classical Experimental Designs
• Completely randomized Design Example

• Treatments: PAN413, SC403, farmers variety (FV)

• Replicates: 4
• Total number of plots = 3 treatments x 4 replicates per treatment
= 12 plots

• So, the experimental area will need to be divided into 12 equal

plots

• Maize spacing: 90 cm inter-row x 25 cm intra-row

• Gross plot: 5.4 m (6 rows) x 5 m (20 planting stations)
• Three seeds per station thinned to two(2) two weeks after
emergence
• What is the plant population per ha??
• Sketch the design or field plan for the experiment
 Classical Experimental Designs
• Completely randomized design (CRD) Example
• Fieldplan: Evaluation of maize verities under smallholder furrow irrigation
• Researchers: Technician responsible
• Year: start………………..End………………..: Site:…….
N
1 2 3 4

PAN413 SC403 FV PAN413

8 7 5 5

SC403 SC403 PAN413 FV

9 10 11 12

5m PAN413 FV SC403 FV

5.4 m
Note: Fieldplan not drawn to scale
 Classical experimental Designs
(2) Randomized Complete Block Design (RCBD)

As the name implies this design entail blocking and is

used where there are known gradients

Blocks are synonymous with replicates in this design

Blocks are perpendicular to the gradients

Unlike in CRD, randomization for RCBD is done per

plot not over the whole area
 Classical experimental designs
• Randomized complete Block design
• Again let’s consider a maize variety evaluation
experiment under furrow irrigation. Field site
has a distinct gradient of soil texture running
from East to West
• Factor: maize variety
• Levels: 3 (PAN413, SC403, farmers variety (FV)

• You are required to develop the experimental

design and fieldplan for the experiment
 Classical Experimental designs
• Randomized complete block design
• Treatments: PAN413, SC403, farmers variety (FV)
• Replicates: 4
• Total number of plots = 3 treatments x 4 replicates per
treatment = 12 plots

• So, here we will have four blocks/replicates with each set of

replicates perpendicular to gradient. Contrast with CRD

• Maize spacing: 90 cm inter-row x 25 cm intra-row

• Gross plot: 5.4 m (6 rows) x 5 m (20 planting stations)
• Three seeds per station thinned to two(2) two weeks after
emergence

• Sketch the design fieldplan for the experiment

 Generic Experimental Designs
(2) Randomized complete block design (RCBD):

• In most cases there are known gradients among experimental

units

• Use of CRD may favour certain treatments, leading to biased

results

• RCBD overcomes the gradient problem:

(1) Align blocks perpendicular to known gradient

(2) Randomize within a block – each block becomes a replicate

(3) After blocking and randomization, there is no bias

If ever you use this design, you should state the blocking factor
 Classical Experimental designs
(3) Split-plots design
• This is used for factorial experiments
• In some cases, the treatments cannot be applied
over a small area typical of experimental plots
without treatments interfering with each other
• What are typical examples?
• To overcome this we can use a split-plots design:
• Plots split into e.g., Two:
• (1) main plot
• (2) Sub-plot (we can have sub-subplots too!)
 Classical Experimental designs
• Split-plots design

• Factor than cannot be applied over small area

is allocated to main plot (i.e. main plot factor)

• The other factor is then allocated to sub-plot

(subplot factor)

• Randomization is done at two levels: main

plot
 Classical Experimental designs
• Consider a factorial experiment with two factors
(1) irrigation method, and (2) maize variety

• How feasible is it to use CRD or RCBD and allocate

an irrigation treatment method to a plot of size
5.4 x 5 m without interfering with neighbouring
plots

• Split-plots design is meant to solve such problems

 Classical Experimental designs
• Split-plots example:

• Develop an experimental design and fieldplan for the experiment

with the following factors

• Irrigation method: Furrow and sprinkler

• Maize varieties: PAN413, SC403, farmers variety (FV)

• Note: irrigation cannot be applied over small areas i.e. use split-
plots design
• Main plot: irrigation method
• Sub-plot: variety
• Replicates =4
• Randomize irrigation in main plots, then maize variety in sub-plot.
Process of randomization is as described before.
Classical Experimental Designs
• Experimental design is part of materials and methods
in research proposal

• It also appears in an experimental protocol

• An experimental protocol provides the ‘’how to ’

details of how to set up experiments, materials to be
used and response variables/indicators to be measured

• This is the document used by the technician during

implementation of research trial and data collection
and should be detailed and written in sequential order.
• Consider previous experimental design examples, and
develop an experimental protocol
 Summary
• Fundamental principles of experimental designs

• Experimental designs meant to ensure collection of

reliable and unbiased data

• Key classical designs (CRD, RCBD and Split-plots

• Given an experimental design problem, consider

experimental physical space in terms of homogeneity or
variation, decide on appropriate design.

• In all cases ensure you replicate and randomize, minimize

confounding factors and in some cases include a control
 Practical 2: Experimental designs and
layout
• Learning Outcomes
•
• At the end of this practical, participants should be competent to
undertake the following:
•
• Statistical experimental design using completely randomized
designs (CRD), randomized complete block design (RCBD) and split
plots
• Apply the fundamental principles of replication, randomization and
blocking in design of experiments
• Develop a detailed experimental protocol for use for data collection
• Layout a laboratory of field experiment based on an experimental
protocol
• Collect statistically sound samples from laboratory experiments and
field experimental trials
 Activities
• Design at least one experiment using each of the statistical
designs: CRD, RCBD and split plots
• Undertake replication, randomization and blocking during
experimental design

• Develop experimental protocols and datasheets for experiments

• Layout/set up of a laboratory or field experiment including pegging

• Demonstration of sampling (net vs gross plots) in experimental

plots

• Visit research laboratory and field experimental plots to critique

current experimental designs in view of skills and knowledge
gained
 Materials and facilities
• Calculators or PCs/laptops with Excel for generation of
random numbers
• Markers and manila for making draw cards for use in
randomization

• Notebooks/rough papers for making design sketches

• 20-50 m long tape measures for measuring plot sizes x 2

• 20-30 wooden/metal pegs to mark experimental plots

• Range rods for marking straight lines

• 50-100 m string/rope with end pegs for making straight lines

• A field site/clear ground of about 50 m x 50 m for setting up
experimental plots
•