interest. In another regression study degree classification may represent an explanatory or
independent variable such as in predicting the starting salary of a first job.
The essence of experimental design is a research situation in which at least one
independent variable is deliberately manipulated or varied by the researcher with the
purpose of observing the effect this has on at least one response variable. Other variables
which could influence the response variable are controlled. Variables can be controlled
by three main strategies, singly or in combination: allocation of subjects to conditions,
holding variables constant and statistical control (adjustment). Good experimental design
should allow a situation to be set up so that plausible alternative hypotheses can be ruled
out. The extent to which plausible alternatives have been explained or ruled out, a cause
and effect can be implied, i.e., ‘X’ was responsible for ‘Y’, and internal validity has
been established.
A variable which is not effectively controlled and which changes along with the
independent variable may account for an alternative explanation of the observed
experimental effect. This is called a confounding variable and it confounds or confuses
the effect of the independent variable on the response variable. This would threaten any
causal inference and hence any internal validity of an experiment. In many experimental
studies we may want to make statements about effects that apply beyond the particular
situation studied. This raises the question of external validity, the extent to which the
study findings can be generalized from the particular experimental setting to other similar
settings.
Example 2A researcher may be interested in the effectiveness of different ways of teaching
vocabulary to 6-year-old children. Three teaching methods could be compared: silent
reading of a story by children, story-telling by a teacher and storytelling by a teacher
which is also enhanced by pictures. Teaching method, an independent variable, could be
manipulated by the researcher, and pupils would be randomly assigned to one of three
independent teaching method groups. Sometimes groups in an experiment are referred to
as treatments. Randomization means that each pupil has an equal probability of being
assigned to any treatment.
If different teachers were involved in each teaching method group, then any effect of
teaching method may be confounded by teacher effects see Figure 1.1.
(^) Treatment (Group)
Silent reading Storytelling Storytelling enhanced by
pictures
Teacher 1 (35 pupils
randomly assigned to
treatment)
Teacher 2 (35 pupils
randomly assigned to
treatment)
Teacher 3 (35 pupils
randomly assigned to
treatment)
Figure 1.1: Effect of teacher
confounded with treatment (group)
Statistics and research design 11