Health Care Experimental Research Design
What Is Experimental Research,Manipulation in Research,Research Control,Study Designs,Qualities of Research Study Designs.
What Is Experimental Research
True
experiments have the potential to provide strong evidence about the
hypothesized causal relationship between independent and dependent variables.
Experiments are characterized by manipulation, control, and randomization. The
quality of experiments depends on the validity of their design.
Manipulation in Research
Manipulation
means the researcher actively initiates, implements, and terminates procedures.
In most instances, manipulation is linked to the independent variable(s) under
consideration. Essential to manipulation is that the researcher has complete
control over the process.
The researcher decides what is to be manipulated (eg,
selected nursing intervention protocols), to whom the manipulation applies (eg,
samples and subsamples of subjects), when the manipulation is to occur
according to the specification of the research design, and how the manipulation
is to be implemented.
Manipulation
implies and is impossible without researcher control over extraneous sources
that might affect and lead to incorrect scientific conclusions. Control aims
“to rule out threats to valid inference.”
It also adds precision, the
“ability to detect true effects of smaller magnitude” (Cook, T.,
& Campbell, 1979, p. 8). Unlike laboratory studies where total control is
often possible, in clinical research control is a relative matter. The
researcher has the responsibility for ensuring as much control over extraneous
forces as possible.
Research Control
Control
also includes “the ability to determine mine which units receive a
particular treatment at a particular time” (Cook, T., & Campbell,
1979, p. 8). This refers to control over two processes that determine who gets
what at what time. The first process is the researcher’s use of random methods
to assign subjects to treatments.
This is the preferred method of exerting
control over subjects and their treatment as, theoretically, it ensures that
known and unknown extraneous forces inherent to subjects are dispersed equally
across the different treatment options.
This may not always be possible, in
which case the second process comes into play-that of structuring the
assignment process in such a way that major, known extraneous forces are
controlled.
Study Designs
Commonly
used design strategies include blocking, matching, and counterbalancing. In
blocking the potentially confounding variable is incorporated into the study
design as: an independent variable. Subjects are then randomly assigned within
each block.
In matching, a weaker but very common method of control, the
researcher identifies one or more extraneous (usually up to three) variables to
be controlled.
As soon as a subject is recruited for one of the treatment
groups, the researcher then tries to find subjects for the other group(s)
identical to the first subject on the specified matching variables.
Counterbalancing occurs when the researcher is concerned that the order in
which treatments are administered influences the results.
When counterbalancing
is used, all subjects receive all treatments; however, the order of
administration of treatments is varied. Randomization entails two separate
processes:
(a) random selection of subjects from the population.
(b) random
assignment of subjects to treatment and control conditions.
Random selection is
the process of randomly drawing research subjects from the population about which
the researcher wants to gain knowledge and to which the researcher hopes to
generalize the findings of a study.
Random assignment entails allocating
sampling units (eg, patients) to treatment and control conditions by using a
decision method that is known to be random (eg, coin toss, random drawing, use
of random tables, computer-generated random sequences of options).
Random
selection is virtually nonexistent in intervention studies in nursing;
moreover, a large proportion (55.3%) of nursing intervention studies do not
even use random assignment methods (Abraham, Chalifoux, & Evers, 1992).
Qualities of Research Study Designs
T
Cook and Campbell (1979) reviewed four types of validity of research designs,
potential threats to each, and strategies to remedy these threats. Statistical
conclusion validity addresses the extent to which, at the
mathematical/statistical level, covariation is present between the independent
and dependent variables (ie, the extent to which a relationship exists
between the independent and dependent variables).
Internal validity refers to
whether an observed relationship between variables is indeed causal or, in the
absence of a relationship, that indeed there is no causal link.
Construct
validity of putative causes and effects refers to whether the causal relationship
between two variables is indeed “the one” and tries to refute the
possibility that a confounding variable may explain the presumed causal
relationship.
External validity refers to the generalizability of an observed
causal relationship “across alternate measures of the cause and effect and
across different types of persons, settings, and times” (Cook, T., &
Campbell, 1979, p. 37).
Validity of any type is not a yes/no issue of whether
or not it is present. Rather it is a matter of degree, determined by the extent
to which the researcher has tried to cope with the various potential threats to
each type of validity.