Agricultural Experimentation

Introduction Objective: This lesson is to help you to understand a few principles in

statistical procedures and the fundamentals of experimental designs.

Note: This lesson is very basic and is meant for technicians in agriculture to learn
statistical concepts and methods for use field-level research. For a comprehensive study,
you should refer to textbooks on statistical methods and experimental designs. You can
also visit available on-line statistical tutorials such as:

http://statpages.org/javasta4.html

Procedure for Agricultural Experimentation

In agriculture research a scientist identifies solutions to problems through

experimentation. Research can be broadly defined as a systematic inquiry into a subject
to discover new facts or principles.

Research involves the following steps:

Formation of a hypothesis - a tentative explanation or solution.

Planning an experiment to objectively test the hypothesis.
Careful observation and collection of data from the experiment.
Interpretation of the experimental results.

The procedure for experimentation involves:

Defining the problem

Stating the objectives
Selection of treatments, experimental material, and experimental design
Deciding number of replications
Consideration of data to be collected
Outlining statistical analysis needed
Conducting the experiment
Analyzing data and interpreting the results, and reporting research findings

The experiment is an important tool of research. Some important characteristics of a well-

planned experiment are:

Simplicity,
High precision,
Absence of systematic error,
Range of validity of conclusions, and
Calculation of degree of uncertainty.
Some Basic Terms

Variability is a characteristic of biological material. Hence we need to decide whether

differences between experimental units result from unaccounted variability or real
treatment effects.

Statistical science helps overcome this difficulty by requiring the collection of data in a
way so as to provide unbiased estimates of treatment effects and the valuation of
treatment differences by tests of significance based on measuring variability.

An experimental unit refers to the unit of material to which a treatment is applied. It can
be a single leaf, a whole plant, an area of ground containing many plants, a pot, or in the
green house. The term plot is synonymous with experimental unit. A treatment may be an
amount of material or a method that is to be tested and compared with other treatments in
the experiment; e.g., cultivar fertilizer doses, etc.

Variable: A measurable characteristic of an experimental unit is a variable; plant height,

days to flowering, panicle length, or grain yield etc.

Individual measurements of a variable are data; 150 cm plant height, 45 days to 50%
flowering, 12 cm panicle length, or 3240 kg/ha of grain etc.

Data: A set of observations or measurements of a particular variable in an experiment;

105 cm, 95 cm, 120 cm, 75 cm, 100 cm, ...., represent data about the variable (plant
height).
Population: In a statistical sense, a population is a set of measurements or counts of a
single variable on all the units in the specified population.
Sample: A sample is a set of measurements (observations) obtained from part of the
specified population.

For example: All 150 plants in a plot form the population. Ten plants used for recording
the plant height from this population form the sample. We obtain information from the
sample.

Mean: Mean is the simple arithmetic average (of the sample or population).
Standard Deviation (SD): The measure of dispersion of the data around their mean is the
Standard Deviation. SD tells us how scattered the sample observations are around the
mean.
Variance: Square of the Standard Deviation.

Standard Error (SE): The SD of the population of means is its standard error. SE tells us
how scattered the treatment means (for example) will be.
Degrees of freedom (df): Represents the freedom with which the variability in a data set
could be accounted for. Usually, but not necessarily, df is one less than the number of
observations (n-1).
For example, with 24 observations on plant height, the degree of freedom to account for
the variability in plant height is 23 (24 observations - 1).

Probability level: We use a probability level to tell us whether an observed result is likely
to have happened by mere chance.

For example, a probability of 5% (0.05) tells us that a particular result observed may not
happen in 5 trials out of 100 trials purely by chance. In other words, the result would
normally be true in 95 cases out of 100 trials.

In agricultural experimentation, it is customary to describe the results that would be

expected by chance 5% or less as SIGNIFICANT and those expected by 1% or less as
HIGHLY SIGNIFICANT. However it is up to the scientist to select the odds at which it
is believed there are real effects.

Distribution of data

Data on a variable may commonly follow one of the following distributions.

Binomial distribution

This type of distribution is expected in data, which describes the proportion of

occurrences in which each occurrence can only be one of two possible outcomes. For
example in data representing percent survival of insects, one can expect either dead or
live insects.

Poisson Distribution

This is a distribution, which represents occurrences of rare events. For example, count
data such as the number of infested plants, the number of lesions per leaf, or number of
weeds per unit area.

Normal Distribution

These data represent a continuous distribution. Most biological data, when plotted in a
frequency curve, represent a bell shaped and symmetrical curve. Such data are said to be
normally distributed. For example, grain yield, plant height, etc.

Test of Significance:

Variability is a characteristic feature of nature. Two plants growing side by side are not
alike even under similar conditions. We also know that data on plant growth characters
reveal this variability. For example, from data on five samples plant height such as 115
cm, 95 cm, 82 cm, 108 cm, and 72 cm, ....., can we say the growth was good (115 cm
plant height) or the growth was poor (72 cm plant height)? What is the truth?

A statistical procedure to estimate the probability of a result happening by chance is

called a test of significance.

't' test, 'F' test, and Chi-square test are commonly used statistical procedures for testing
significance of observed results.

't' test of significance:

The single-sample 't' is a statistics computed from a sample, which expresses the
difference between the sample mean and a hypothetical population mean in standard error
units.

Thus:

't' = (sample mean - population mean) / standard error

The theoretical 't' values for the degrees of freedom for different sample sizes at different
probability levels are provided in a 't' Table.

Comparing the calculated 't' value with that of the theoretical 't' value from the 't' table,
one could test the significance of the sample mean being different from the hypothetical
population mean.

If the calculated 't' is > 't' table value, then the two mean values that are being tested are
significantly different.

If the calculated 't' is < 't' table value, then the two mean values that are being tested are
not significantly different from each other.

'F' test of significance:

The 'F' test is a ratio between two variances and is used to determine whether two
independent estimates of variance can be assumed to be estimates of the same variance.

In the analysis of variance, the 'F' test is used to test equality of means of two or more
treatments; that is, to answer the question, can it reasonably be assumed that the treatment
means resulted from populations with equal means?

Like the theoretical 't' values, theoretical 'F' ratios are also computed for different sample
sizes at a given probability level in the form of a 'F' Table.

Comparing a calculated 'F' ratio with that of the theoretical 'F' ratio from the 'F' Table, the
significance among a group of treatments could be tested.
If the 'F' calculated is > 'F' Table values, then the treatments are significantly different at
that probability level.

If the 'F' calculated is < 'F' Table value, then the treatments are not significantly different
at that probability level.

Analysis of Variance:

A partitioning of the total variability in the data arising from various sources is presented
in the form of a Table. This Table is called the Analysis of Variance Table and consists
of:

Source of Degrees of Sum of squares Mean Sum of 'F' ratio

variation freedom squares

The usage of an analysis of variance could be better understood if you review the lessons
on experimental designs.

We hope that this tutorial helped you to understand some of the basic principles in the
statistical procedures used in agricultural experimentation.

You may also refer to the following books:

Statistical Procedures For Agricultural Research By Gomez K.A . and A.A.

Gomez.
Statistical Methods For Agricultural Workers By Panse V.G and P.V. Sukhatme.