Handout #9: - Experimental
Research
Wiersma (2000) defines an experiment as a research situation in which
at least one independent variable, called the experimental variable, is
deliberately manipulated or varied by the researcher to test its possible
effect on another. In other words experiments test cause and effect
relationships. A well-designed experiment should have good experimental
validity. For it to have good internal
validity, an experiment should establish without doubt that the experimental
variable is the cause of the effect by controlling the effects of all other
factors. For it to have good external
validity it should be possible to generalise the findings to a range of
conditions. Internal validity can be
improved by:
a) Keeping factors
constant. In an experiment to see
the effect of sales pitch on the sale value of houses, the background of the
subjects (persons taking part in the experiment) as well as the design of the
houses is important. An experimenter
could select two groups of persons, expose one set to the sales pitch (the
treatment) and not expose the other set and compare the price both sets are
willing to pay. The background of both
groups should be similar and the design of houses should be similar or both background
and design would have an effect on sale value and would mask or confound the
effect of sales pitch. Keeping factors
constant leads to problems with external validity, as the findings would only
be applicable to the conditions used.
To improve external validity the experiment should be repeated using
other conditions.
b) Including a control
group. In the experiment
described above the group that is not exposed to the sales pitch is known as
the control. The control group provides
a basis for comparison. If an effect is
observed in the test subjects and not in the control it can be assumed that the
effect is due to the treatment if both groups are similar.
c) Random assignment.
Random assignment of subjects to the groups is used to ensure that the
groups are similar. In random
assignment each individual has an equal chance of being placed in either group
so the characteristics should be spread evenly between both groups.
Experimental studies fall under three categories: pre-experimental designs,
experimental designs and quasi-experimental designs.
PRE-EXPERIMENTAL DESIGNS
1.
The One-Shot Experimental Case Study
Treatment Observation
This is not to be confused with case studies
in qualitative research. In this design
a group is exposed to a treatment and an observation made. It is a very poor design as factors are not
kept constant, there is no control and no random assignment. It is not certain
whether a change occurred as no observation was made before the treatment.
2.
The One-Group Pretest Posttest Design
Observation Treatment Observation
In this design an observation (pretest) is
made before the treatment is applied and after the treatment is applied
(posttest). The effect of other factors
cannot be ruled out. However, it can
show that a change occurred.
3.
The Static-Group Comparison
Group 1 Observation
Treatment Observation
Group 2 Observation
In this design a control group, to which a
final observation is made, is used. As
no attempt is made to ensure that both groups are similar, the effects of other
factors is not ruled out.
1.
The Pretest Posttest Control Group Design
Random Group
1 Observation Treatment Observation
Assignment Group
2 Observation Observation
In this design subjects are randomly assigned
to both groups and observations are made before and after the treatment is
administered to the test group. Cause
and effect is established.
2.
The Posttest only Control Group Design
Random Group
1 Treatment Observation
Assignment Group
2 Observation
In this design the pretest observation is
removed. In some experiments a pretest
may influence the results and so the pretest is removed.
3. The
Solomon Four-Group Design
Random Group
1 Observation Treatment Observation
Assignment Group
2 Observation Observation
Group
3 -
Treatment Observation
Group
4 Observation
Instead of removing the pretest the effect of the pretest may be
assessed by adding two other groups.
This design is expensive.
The Solomon Four Group Design is an example of a factorial design in
which the effect of more than one independent variables (in this case two;
treatment and pretest) are measured.
Factorial designs allow for interactions between independent variables
to be assessed.
QUASI-EXPERIMENTAL DESIGNS
1. Non-randomised
Control Group Pretest-Posttest Design
Group
1 Observation Treatment Observation
Group
2 Observation Observation
This is similar to the pretest posttest
control group design except that subjects are not randomly assigned.
2. Time-Series Experiments
Obs Obs Obs Treatment Obs Obs Obs
In time series designs several observations
are made before and after the treatment.
A control group may or may not be used.
Leedy, P.D. & Ormrod, J.E. (2001) Practical Research Planning and
Design (7th ed.). New Jersey:
Merrill Prentice Hall.
Wiersma, W. (2000). Research Methods in Education An Introduction. (7th ed.) Boston: Allyn and Bacon.