1. Experimental Design
Submitted to:
Mr. Ajesh P. Joseph,
School of Social Work,
Marian College,
Kuttikkanam.
Submitted by:
Bimal Antony,
1st
MSW,
School of Social Work,
Marian College,
Kuttikkanam.
Date of Submission:
19th
April 2011.
2. 2Experimental Design
Introduction
An experiment is a process or study that results in the collection of data. The results of
experiments are not known in advance. Usually, statistical experiments are conducted in
situations in which researchers can manipulate the conditions of the experiment and can control
the factors that are irrelevant to the research objectives. For example, a rental car company
compares the tread wear of four brands of tires, while also controlling for the type of car, speed,
road surface, weather, and driver.
Experimental design is the process of planning a study to meet specified objectives.
Planning an experiment properly is very important in order to ensure that the right type of data
and a sufficient sample size and power are available to answer the research questions of interest
as clearly and efficiently as possible.
Purpose of Experimental Research Design
The aim of the experimental research is to investigate the possible cause-and-effect
relationship by manipulating one independent variable to influence the other variable(s) in the
experimental group, and by controlling the other relevant variables, and measuring the effects of
the manipulation by some statistical means. By manipulating the independent variable, the
researcher can see if the treatment makes a difference on the subjects.
If the average scores of two groups prove to be significantly different, and if there are not
any explanations for this difference, then it can be concluded that the effect of the treatment
caused this difference. This is where experimental research differs from correlational research,
For instance, correlational studies only describe or predict the strong relationship between
socioeconomic level and the academic achievement but cannot prove the direct cause-and-effect
relationship between these two variables. It is the experimental research which can demonstrate
that by changing the independent variable, a change is possible on the dependent variable.
In educational research the most frequently studied dependent variables are achievement,
motivation, attention, interest in learning, participation and attitudes. The common independent
variables that are manipulated are teaching methods, types of assignments, types of teaching
materials such as text books and visual aids, types of rewards, types of questions used by the
teacher, and evaluation techniques. There are however, some independent variables such as age
and gender that cannot be manipulated. When the independent variable that is chosen cannot be
manipulated, either a comparative research is conducted, or a second independent variable is
chosen for manipulation in order to conduct an experimental study.
3. 3Experimental Design
Categories of Experimental Design
Before-After or Pre-test - Post-test Experimental Design
This is normally called classical experimental design. It is more reliable and it represents
the so-called four-cell design (Fig. 13.1). Here, before experimentation all the groups are
selected, observed and measured. There is one independent variable - the treatment - and one
dependent variable. Subjects are assigned randomly to the control group and experimental group.
Then the dependent variable is measured for both the groups.
After pretest, the treatment is introduced to the subjects in the experimental group only.
The dependent variable is then measured again for both the groups and compared. This is the
Post- test. This design suffers from one limitation, that is, it does not ensure to be free from the
influence of external factors.
After-only or Pre-test only Experimental Design
After-only or Pre-test only Experimental Design In this type of design the study is carried
out under social conditions, which are not at all in the control of the physical or natural
conditions. Two groups of subjects, who are similar in all conditions, are chosen. One is called
experimental group and the other is called control group. Experiment is carried out on the
experimental group as per the pre-determined method.
After the prescribed period both groups are observed and the results are measured. The
results are compared and changes that are observed in the experimental group are recognised as a
result of manipulating the variable in the
experiment.
Quasi or Ex-Post Facto Experimental Design
The name of the quasi experiment has been given to those situations in which the
experimenter cannot randomly assign subjects to experimental groups, but can still manipulate
the independent variable. However when even such manipulation is impossible - that is to say,
when the stimulus is also beyond the control of the researcher – we can no longer speak of
experimentation, what we have is purely and simply, an analysis of co-variation. Never the less
there are research situations which, although lacking both features of experimentation (i.e.
randomisation and manipulation) involve a design that closely resembles that of experimentation.
Such designs are called ex post facto.
‘Ex Post Facto’ is a Latin phrase which means ‘done or made after a thing but retro
action upon it’. In this the experimenter does not achieve the change which studies, he invariably
chances upon the effect after it has already occurred. In Library Science, children reading habits
and behaviour of a fresh reader in the library can be studied with the application of this design.
4. 4Experimental Design
Completely Randomized Design
Frequently an investigator wishes to compare three or more treatments in a single
experiment. In a survey, too, he may wish to study several populations; for example, he may be
interested in IQ scores from a standard test for students at five schools, Such comparisons could
be accomplished by looking at the samples two at a time and comparing the means. Although
feasible, this is an inefficient method of comparison for more than two populations.
One reason for its inefficiency is that the standard deviation for the difference between
the two, sample means is not calculated from all the samples but instead uses samples only from
the two populations under immediate consideration. Second, we feel intuitively that we shall
almost find a significant difference between at least one pair of means (the extreme ones, e.g.) if
we consider enough identical populations. We can no longer trust our level of significance.
Therefore, instead of using two samples at a time, we wish to make a single test to find
out whether the students from the five schools are from five populations having the same
population mean.
Completely randomized design is primarily concerned with tests for population means.
To study the means, it is necessary to "analyze the variance".
Randomized Complete Block Design
In the completely randomized design, treatments are assigned at random. For example, if
the treatments are three drugs and there are 24 patients, eight patients are assigned at random to
each of the three treatments.
The 24 patients may vary widely in initial condition, and their initial condition may affect
their response to the drugs. In the completely randomized design, we try to take care of these
differences among the patients by assigning them at random into groups of eight patients.
Unfortunately, it is possible that all the patients receiving drug 1 may be comparatively healthy
and all those receiving drug 2 may be comparatively unhealthy, even though the assignment was
randomly made. By randomization, however, at least we have given each drug an equal chance
with respect to the initial condition of the groups. Further more, we can expect that if the
experiment is large enough, randomization will roughly equalize the initial condition of the three
groups. Besides initial condition, the experimenter may feel that other factors might influence the
response to the drugs (e.g., age or weight).
A block design is a much used method for dealing with factors that are known to be
important and which the investigator wishes to eliminate rather than to study.
In the randomized complete block design, still with three treatments and 24 patients, the
patients are divided into eight blocks, each consisting of three patients. These blocks are farmed
so that each block is as homogeneous as possible. Each block consists of as many experimental
units as there are treatments-three, in this case. The blocks might be easily formed on the basis of
age, for example, with blocks 1 and 8 consisting of the three youngest and the three oldest
patients, respectively. The individuals in a particular block are as alike as possible. On the other
hand, there may be wide differences between the individuals for different blocks.
5. 5Experimental Design
Latin Square Design
In the randomized complete block design, the effect of a single factor was removed. It is
occasionally possible to remove .the effects of two factors simultaneously in the same
experiment by using the Latin Square design. In order to use the Latin square design, however, it
is necessary to assume that no interaction exists between the treatment effect and either block
effect. In addition, the number of treatments must be equal to the number of categories for each
of the two factors. We might, for instance, wish to test four detergents, using four methods of
application, at four hospitals. A 4X4 Latin square design could then be employed, using each
detergent exactly once with each method and exactly once in each hospital. The assignment of
detergent could be made as shown in the following table; the roman numeral in the ith row and
jth column indicates the detergent that will be used by the ith application method in the jth
hospital. As assigned in the Table 2, the first detergent is used in hospital 1 by method 1, in
hospital 2 by method 4, in hospital 3 by method 3, and in hospital 4 by method 2. Only 16
observations are needed because of the balanced arrangement used and because of the
assumption of no interaction.
Factorial Design
Often a researcher can use a single experiment advantageously to study two or more
different kinds of treatments. For example, in investigating performance of two types of seeds,
he may wish to vary the level of fertilizer used during the experiment. If he chose three levels of
fertilizer-low, medium, and high-one factor would be "type of seed", the second factor "level of
fertilizer". A factorial design, with two factors, would consist of employing all six treatments
formed by using each type of seed with each level of fertilizer. Factorial designs can involve
more than two factors; however, we consider here the case of two factors only.
A factorial design can also be used in a survey. For example, we might wish to compare
three methods of teaching operations research, and at the same time compare the fast four grades.
We might have records on standardized tests for two classes in each grade taught by each
method. The class mean improvement from initial test to final test could be the measure of
success. Our data would then consist of two observations on each of 12 (3 x 4) different
treatment combinations.
The characteristic of the factorial design is that every level of one factor is used in
combination with every level of the other factor. The design is effective for studying the two
factors in combination. This implies that factorial designs are appropriate in finding out whether
interactions exist between factors.
Some factors can be measured quantitatively, and different levels for them are chosen on
an ordered scale; level of fertilizer, dosage level; and temperature are all factors of this type.
Other factors involve no obvious underlying continuum and can be said to be qualitative; drug
and type of seed are factors of the second type.
6. 6Experimental Design
Solomon Four Group Design
The Solomon four group design is a way of avoiding some of the difficulties associated
with the pretest-posttest design.
This design contains two extra control groups, which serve to reduce the influence of
confounding variables and allow the researcher to test whether the pretest itself has an effect on
the subjects.
The Solomon four group test is a standard pretest-posttest two-group design and the
posttest only control design. The various combinations of tested and untested groups with
treatment and control groups allows the researcher to ensure that confounding variables and
extraneous factors have not influenced the results
The first two groups of the Solomon four group design are designed and interpreted in
exactly the same way as in the pretest-post-test design, and provide the same checks upon
randomization.
The comparison between the posttest results of groups C and D, marked by line ‘D’,
allows the researcher to determine if the actual act of pretesting influenced the results. If the
difference between the posttest results of Groups C and D is different from the Groups A and B
difference, then the researcher can assume that the pretest has had some effect upon the results.
The comparison between the Group B pretest and the Group D posttest allows the
researcher to establish if any external factors have caused a temporal distortion. For example, it
shows if anything else could have caused the results shown and is a check upon causality.
The Comparison between Group A posttest and the Group C posttest allows the
researcher to determine the effect that the pretest has had upon the treatment. If the posttest
results for these two groups differ, then the pretest has had some effect upon the treatment and
the experiment is flawed.
The comparison between the Group B posttest and the Group D posttest shows whether
the pretest itself has affected behavior, independently of the treatment. If the results are
significantly different, then the act of pretesting has influenced the overall results and is in need
of refinement.
Analysis of Covariance
Analysis of covariance is a combination of the two techniques-analyses of variance and
regression. It is the simultaneous study of several regressions.
The purpose of analysis of covariance is to remove the effect of one or more unwanted
factors in an analysis of variance. For example, in studying the heights of three populations of
children (cyanotic heart disease children, sibs of heart-disease children, and "well children"), we
7. 7Experimental Design
may wish to eliminate the effect of age. A variable whose effect one wishes to eliminate by
means of a covariance analysis is called a covariate or a concomitant variable.
ADVANTAGES, DISADVANTAGES AND LIMITATIONS OF
EXPERIMENTAL METHOD
Experimentation has two basic advantages; firstly it is the research method that best
enables us to tackle the problem of the causal relationship; secondly it allows isolation
of specific phenomena, which could not be studied, systematically in their natural setting,
owing to the presence of other factors that hide, confuse and distort them.
Advantages
_ Its power to determine causal relationship is much better that that of all other
methods.
_ The influence of extraneous variables can be more effectively controlled.
_ The element of human error is more reduced.
_ More conditions may be created and tested in this method.
_ This method yields generally exact measurements and it can be repeated.
Disadvantages
_ It is very difficult to establish comparable control group and experimental group.
_ The scope of experimentation with human beings is extremely difficult.
_ Experiment is often difficult to design as it tends to be expensive and time consuming.
_ Experimentation can be used only in studies of the present but not in studies relating
to the past or future.
Limitations
_ Experimentation is applicable to certain phenomena and certain social situations.
_ Experimentation cannot be conducted if the independent variable cannot be
manipulated.
_ This approach is generally suitable to ‘micro’ issues (involving interpersonal
relationships) rather than to ‘macro’ situations (on account of the difficulty of
manipulating institutions or social groups).
An experiment conducted in a laboratory under artificial conditions may not truly
represent a situation. It is too simplistic to consider that there are only two variables. In natural
sciences it works where other variables can be kept under control. In behavioural or social
sciences the situation is too complex as best put by John W. Best (p. 92). Despite its appealing
simplicity and apparent logic, it did not provide an adequate method for studying complex
problem. It assumed a highly artificial and restricted relationship between single variables.
Rarely, if ever, are human events the result of single causes
8. 8Experimental Design
They are usually the result of the interaction of many variables, and an attempt to limit
variables so that one can be isolated and observed proves impossible. Hence, there are many
limitations and suppositions when applied to social situations including library and information
management.
Human beings cannot be put to experimentation on many psychological and ethical
grounds. Human beings when under a test or observation can easily manipulate their natural
behaviour. They cannot be put in test tubes and titrated like chemicals. Then a human being
experimenter has naturally inherent biases (likings and disliking) when other humans are
concerned. They tend to take sides in heart of hearts. Hence observations and conclusions may
not reflect the objective reality.