Advanced Machine Learning for Business Professionals
Research Design Optimization
1.
2. Research Design
The Research design is the detailed plan of investigation. In fact, it is the
blueprint of the detailed procedures of testing the hypotheses and analyzing
the obtained data.
The research design thus maybe defined as the sequence of those steps
taken ahead of time to ensure that the relevant data will be collected in a
way that permits objective analysis of the different hypotheses formulated
with respect to the research problem.
The purpose of any research design is to provide a maximum amount of
information relevant to the problem under investigation at minimum cost.
Basically research designs serve two purposes:
First, It answers the research questions as objectively , validly, economically
as it is possible.
Second, a research design acts as a control mechanism.
3. Basic Principles of Research Design
Professor Fisher has enumerated three principles of experimental designs:
(1) the Principle of Replication; (2) the Principle of Randomization; and the
(3) Principle of Local Control.
(1) Replication: the term replication is a fusion of two words, namely
duplication and repetition. It refers to the deliberate repetition of an
experiment, using a nearly identical procedure with a different set of
subjects ,In a different setting and at a different time.
(2) Randomization: It is the second basic principle of research design. It
makes the test valid . Randomization ensures the independence of the
observation which in turn makes the statistical test valid. Not only this when
subjects are randomly assigned to the experimental treatments, this
automatically controls the extraneous variables, which otherwise are left
uncontrolled. Sometimes it has been found that the complete randomization
becomes difficult. This is especially true when one is dealing with
organismic variables. n such a case investigator is suggested to take the
middle path between complete randomization and complete non-
randomization.
4. (3) Local Control: By local control we mean balancing, blocking and
grouping of the subjects or the experimental units employed in the
experimental design.
The term grouping refers to the assignment of homogenous subjects or
experimental units into a group so that different groups of homogenous
subjects or experimental units into a group so that different groups of
homogenous subjects maybe available for differential treatment.
Blocking refers to the assignment of experimental units to different blocs in
such a manner that the assigned experimental units within a block maybe
homogenous.
Balancing refers to the fact that blocking, grouping and assignment of
experimental units have been done in such a fashion that the research design
appears to be a balanced.
5. Important experiment designs are as follows:
(a) Informal experimental designs:
(i) Before-and-after without control design.
(ii) After-only with control design.
(iii) Before-and-after with control design.
(b) Formal experimental designs:
(i) Completely randomized design (C.R. Design).
(ii) Randomized block design (R.B. Design).
(iii) Latin square design (L.S. Design).
(iv) Factorial designs.
7. Between Groups Design
We will discuss only two types of Between Groups
design:
Randomized Groups Design
Matched Groups design or Randomized Block design.
Randomized Groups Design: A Randomized groups design is one in
which subjects are randomly assigned to the different groups meant for the
different conditions or values of the independent variable. The randomized
groups design is based upon the assumption that the random assignment of
subjects into two or more groups will make these groups statistically
equivalent on the subject relevant variables(attitude, ability, motivation
emotions etc) which may produce variations in the dependent variable.
8. When the subjects are randomly assigned to only two groups, the resulting
design is known as two-randomized groups design and when the subjects are
randomly assigned into more than two groups, the resulting design is called
multi-group design or more than two randomized groups design.
So in two-randomized groups design the experimenter selects two values of an
independent variable. These two values may also be called “conditions” or
“treatments” of the experiment. His main interest is to examine whether or not
these two conditions affect the dependent variable in a differential way.
o Usually t-test or a non-parametric substitute namely Mann whitney –U test is
used as a statistical analysis in two-randomized groups design.
o Underwood(1966) has suggested two primary ways through which unbiased
groups or random groups of subjects can be formed in order to achieve the
successful random assignment. These are:
o (A) Captive Assignment (B) Sequential Assignment
9. Matched Groups Design or randomized-block design:This may be a two-
matched or More-than two-matched groups design.
In a matched- groups design all subjects are first tested on a common task or
a pretest measure(also called a matching variable) and then, they are formed
into groups(as many as needed for the experiment) on the basis of
performance on the common task or the matching variable.
The groups thus formed are said to be equivalent groups.
Subsequently, the different conditions or values of the independent variable
are introduced to each group.
If these groups have equivalent means on the dependent variable before the
experimental treatment is given and if a significant difference occurs after
administering the experimental treatment and controlling the relevant
variables, the resulting differences in the dependent variable wil be
attributed to the experimental treatment.
10. Methods of Matching:
Having selected a matching variable a set of scores earned by all subjects
on the matching variable , the next step is to match them . There are two
ways of matching :
(1) Matching By Pairs: On the basis of the obtained scores by the subjects ,
the experimenter matches subjects in a way that each subject has a
corresponding partner in a matched group or groups.
(2) Matching on the basis of Mean and Standard deviation: The
experimenter forms as many groups as needed for the experiment in a way
that their subsequent means and standard deviations do not differ
significantly.