RESEARCH DESIGNS
Thomas Bevins
Summer 1999
Research Designs are platforms to use to
explore new knowledge in order to better understand phenomena, clarify
explanations and identify causative factors. Although there are no real
rules for choosing a design; one must realize the consequences for choosing
one design over the other. One should choose the design that best attempts
to address the conceptual issues presented.
Some questions that would help you decide which quantitative design
is most appropriate for your study include:
-
How much do you know about the variables of interest?
-
Are you manipulating the levels of the independent variable?
-
How many independent variables are being tested?
-
How many levels do each independent variable have, and are these levels
experimental or control?
-
How many groups of subjects are being tested?
-
How will subjects be selected?
-
Can subjects be randomly assigned to groups?
-
Is pretest data being collected?
-
Are you interested in examining differences or similarities between groups?
-
How often will observations of responses be made?
-
What is the temporal (time) sequence of interventions and measurements?
The research design gives backbone structure to a study. There are several
concepts to understand when choosing a research design. These are summarized
below:
Concepts to Consider when designing a Research
Study
-
Manipulation. What is actively directed or managed by the researcher.
Usually refers to giving a treatment or not giving a treatment (or giving
one of many levels of treatment) to a group of subjects. Assigning
groups to various levels of the independent variable(s).
-
Control. The extent to which the researcher can manage extraneous
sources that might affect a study and lead to incorrect scientific conclusions.
-
Random selection. Randomly drawing research subjects from the target
population.
-
Random assignment. Allocating subjects to treatment and control
conditions in a nonsystematic way, using a method that is known to be random.
-
Probability. (as related to research findings) The likelihood that
research findings are low in uncertainty and error; that the findings are
trustworthy and believable.
-
Bias. The difference between the true and the observed. Undesirable
influences on outcomes in research.
-
Causality. Determining the cause-and-effect relationship(s) that
exist between variables.
QUANTITATIVE RESEARCH DESIGNS
Experimenatl v. nonexperimental
EXPERIMENTAL DESIGN |
NON EXPERIMENTAL DESIGN |
A. Manipulate variables to bring about effect |
A. Observe variables and effects |
B. All relevant variables have been defined so that they can be manipulated,
controlled and studied |
B. Used to identify/measure/describe variables and/or determine relationships
for further (experimental) study |
C. Random selection and random assignment occur to improve control |
C. Manipulation, control and randomization are lacking. |
EXPERIMENTAL DESIGN
The purpose of an experimental design is
to provide a structure for evaluating the cause-and-effect relationship
between a set of independent and dependent variables.
Some of the elements of an experimental design:
-
Manipulation - the researcher manipulates the levels of the independent
variable. This usually means that you are looking at the effect of some
treatment on one group of subjects (the treatment group), and comparing
that to another group of subjects who do not receive the treatment (the
control group). The type of treatment here is the independent variable
that gets "manipulated". The idea of manipulation is that the researcher
"manipulates" by assigning some subjects to the treatment group, and the
other subjects to the control group. The researcher does not have to use
a control group, the design may incorporate two or more "treatments", or
various levels of the same treatment that are compared. It is the effect
of this "manipulation" that is measured to determine the result of the
experimental treatment. Another variation on the control group is the "attention
control group". This group would get some "neutral" experimental
attention/treatment, but not the treatment variable being studied.
This allows the researcher to look at three groups (experimental, attention
control, and "silent" control) and better control for the Hawthorne effect.
-
Control - the researcher incorporates elements of control so that
the evidence supporting a causal relationship can be interpreted with confidence.
Using a control group is only one aspect of control. Control is acquired
through manipulation, randomization, the use of control groups, and methods
to handle extraneous variables. (more on control below).
-
Randomization - subjects are randomly assigned to at least
two comparison groups.
-
With randomization, extraneous variables are evenly distributed between
or among groups; even variables you have not thought about yet may be controled
through randomization.
-
Randomization (or random assignment) is not the same thing as random sampling
or random selection.
An extraneous variable is a variable that you
did not initially intend to include in your design, but this variable might
have an influence on your study in a way that would invalidate the results
of the study. The researcher attempts to exert control over these
extraneous variables in one of three ways:
-
Selective control - the use of randomization (see above). For instance:
if gender is an extraneous variable, then by randomly assigning subjects
to groups, the number of males and females in each group should be evenly
distributed, and the variable gender should not effect the outcome for
one group versus the other.
-
Physical control - control of an extraneous variable by making it
a constant. For instance: if gender is an extraneous variable, then
only study females. This way the variable gender will not effect the outcome
of one group versus the other. The disadvantage of this approach is that
it limits your ability to generalize your results. If you only study females,
you do not know how this treatment will effect males.
-
Statistical control - include the extraneous variables in the design.
For instance: add gender as another independent variable in your study.
This can be a very powerful way to control the effect of an extraneous
variable like gender. You will actually analyze the effect of the variable,
and know how gender effects the outcome of your study.
Cause and Effect
-
In order to postulate a cause and effect relationship, the following
conditions must be met:
-
the causal variable and the effect variable must be associated with each
other
-
the cause must precede the effect
-
the relationship/association must not be explainable by another (extraneous)
variable
-
The validity of the conclusion of cause and effect depends on the
control of extraneous variables.
Advantages of an experimental design:
-
can test cause and effect relationships
-
provides better control of the experiment and minimizes threats to validity.
Disadvantages of an experimental design:
-
for many studies, all extraneous variables cannot be identified or controlled
-
not all variables can be experimentally manipulated
-
some experimental procedures may not be practical or ethical in a clinical
setting
-
the Hawthorne effect may be strong in an experimental situation, and not
so in an ex post facto study.
Be able to compare and contrast the following types of experimental designs:
True experimental design:
including random selection (random sampling), pretest/posttest, random
assignment, manipulation of the levels of the independent variable(s),
including a control group.
Posttest-only design:
(after-only design): you must assume that randomization assures pre-experimental
group equivalence.
Solomon four-group design:
combines the true experimental and the posttest only. Allows you
to evaluate the effect of the pretest on the posttest scores, and any interaction
betwen the test and experimental condition.
Factorial designs:
Allows the researcher to examine the effects of one or more intervention
on different factors or levels of variables in the study. Used for
statistical control. Tends to increase sample size because you want
to have enough subjects in each "cell" of the design.
Counterbalanced (crossover) designs:
when more than one intervention (treatment) is used, and you want to know
the effect of manipulating the order in which the treatments are given.
DePoy gives three criteria that need to be met to determine that a true
experimental design is appropriate:
-
there is sufficient development of theory to warrant
the proposition of causality
-
you have asked a causal quantitative question
-
conditions exist, and legal and ethical issues allow
for the random selection of subjects (random sampling), random
assignment to groups, use of a pretest/posttest, and manipulation
of the levels of the independent variable(s), including a control group
The basic characteristics distinguishing a true experimental design
from a quasi-experimental or nonexperimental design are randomization
and comparison of groups. If you
are trying to distinguish between experimental and quasi-experimental look
to see if there is random assignment of subjects to groups, and
there is a comparison of these groups. Comparison: is there some
experimental group getting an experimental treatment, and another group
getting a different experimental treatment, or a group getting no experimental
treatment (control group)?
QUASI EXPERIMENTAL DESIGNS
It is not always possible to implement a design that meets the three
criteria of a true experimental study (manipulation, control and randomization).
Quasi-experimental designs differ from experimental designs because either
there is no control group or randomization cannot occur.
Types of quasi experimental designs:
1. Nonequivalent control group design:
-
One of the most frequently used designs.
-
It is the same as experimental except control group subjects are not randomly
assigned.
-
Advantages -
-
in spite of the lack of randomization, the use of a control group and posttest
increases the strength of the design
-
good pretest data allows for improvement of analysis results
-
Disadvantages -
-
Threats to internal and external validity, i.e., selection - threat as
subjects are not randomly selected.
-
There are limits to extent one can infer causality.
2. After only non equivalent control group design:
-
Similar to after only experimental group design but randomization is not
used to assign subjects to groups.
-
This design assumes groups are equivalent after the independent variable
is introduced.
-
This type of design can be used to study effects of natural or man made
catastrophes such as hurricanes, floods or plane crashes.
-
Use of good demographic data to further describe the subjects in each group
would strengthen this type of study design.
-
Advantage - Design is simple and quick.
-
Disadvantage - Cannot be used to assess causality. Reduces internal
validity.
3. Time series design:
-
This is useful for determining trends where it is not possible to have
a control group.
-
No randomization occurs when only one group is available.
-
Time series design significantly increases the strength of the study.
-
Advantages -
-
Repeated pretreatment observations help to control for maturation.
-
Repeated post treatment measures allow one to determine if change is maintained
over time.
-
Disadvantage - History may threaten internal validity.
PRE-EXPERIMENTAL DESIGNS
One-shot case study: there is manipulation
in that a "treatment" (independent variable) is given, and the dependent
variable is then measured, but there is no randomization, and essentially
no control.
Pretest-posttest design: no randomization,
not much control. Like a one-shot case study with a pretest.
Static group comparison: still no randomization,
but maybe a little more control. You have a "control group", but
those subjects are non-equivalent, and there is no pretest to see how equivalent
they might have been.
NONEXPERIMENTAL DESIGNS
These designs are used in situations where manipulation of an independent
variable, control or randomization are not involved. These designs are
focused to describe and measure independent and dependent variables.
They are sometimes called descriptive research designs. Nonexperimental
research does not prove causality. The goal is to describe phenomena and
explore and explain relationships between variables.
Types of non experimental designs:
1. Descriptive/Exploratory Survey Studies-
-
This is used to examine differences or similarities between variables.
-
This is a comparatively weak quantitative design, often used when little
is known about a topic or to initially explore a research question.
-
There is no random assignment or control.
-
The design can be used with both quantitative and qualitative methods.
Rember that because of the differences between qualitative and quantitative
research in philosophy, and the order in which steps are taken, the two
studies would be very different.
-
Survey designs gather information from a segment of the population.
-
Random selection of subjects for the survey can increase the ability to
generalize information. (Random selection is not the same as random
assignment - assignment occurs before the experimental condition is experienced.)
-
Instruments commonly used for this type design include interviews (telephone
and in person), questionnaires (mailed or administered in person).
2. Correlational Studies-
-
A correlation is the examination the strength of the relationship(s) between
2 or more variables.
-
If subjects were assigned to treatment and control groups, that group assignment
and manipulation of the independent variable would turn this into an experimental
or quasi-experimental study (depending on the level of control of extraneous
variables).
-
If there is no manipulation of the independent variable and there is no
assignment of groups this is a nonnexperimental design.
-
Data can be collected through the use of a questionnaire, interviews, or
it can be measured with any of a variety of instruments.
-
Advantage of nonexperimental correlational designs is they are straightforward,
usually inexpensive and quick. They may be used as preliminary research
for future studies.
-
Disadvantage is only the relationship between variables can be determined;
not causation
3. Ex Post Facto Studies -
-
A form of correlational study where the event of interest has already occurred.
-
There is no assignment of subjects to groups - the researcher is studying
the groups after they have already been formed.
-
This design is used to describe existing relationships between variables
and to determine if there is a correlation between variables. Remember,
correlation does not indicate causation.
4. Prediction Studies -
-
an exploration of relationships among independent variables to make predictions
about the dependent variable
-
advantage - helps make intelligent decisions
-
disadvantage - a relationship does not ensure cause-and-effect (see
section on cause-and -effect)
5. Cross-sectional Studies -
-
examine data across different groups at one point in time
-
advantage - less time consuming or expensive
-
disadvantage - less control of extraneous variables
6. Longitudinal Studies -
-
collect data from one group over different points in time
-
advantage - a subject serves as his/her own control, which greatly
controls extraneous variables
-
disadvantage - costly and time consuming - may take decades to finish
study, depending on the study
Cause-and-Effect
-
In order to postulate a cause-and-effect relationship, the following
conditions must be met:
-
the causal variable and the effect variable must be associated with each
other
-
the cause must precede the effect
-
the relationship/association must not be explainable by another (extraneous)
variable
-
The validity of the conclusion of cause and effect depends on the
control of extraneous variables. For the most part, nonexperimental studies
do not control extraneous variables. Therefore, even though a statistically
significant relationship may be shown, there are not grounds to postulate
a cause and effect relationship unless conditions 1 through 3 are met.
This is one of the most common mistakes made in research.
Case Studies -
-
A research method that provides a detailed analysis of a single individual,
group, institution or problem of interest. In the last 2 decades, the case
study method has evolved into a well respected set of design strategies.
They are used in health and social science inquiry.
-
Case studies may yield either quantitative or qualitative data. The key
lies in how the research question or query is framed.
-
Criteria for determining when to chose this design -
-
If interested in examining phenomenon in it's current context
-
If want to contribute to the support and development of theory
-
If want to explore in-depth an atypical, different case or when it is impossible
to conduct a design group.
-
Cannot generalize from case study findings to a population!
-
Case study designs includes single and multiple case designs.
-
They may be descriptive, exploratory, explanatory or causal.
-
Data is content analyzed and presented in narrative form.
References:
DePoy, E. & Gitlin, L. (1998).
Introduction to research: Understanding and applying multiple
strategies. St. Louis: Mosby.
and
LoBiondo-Wood, G., & Haber, J. (1998).
Nursing research: Methods, critical appraisal, and utilization (4th
ed.). St. Louis: Mosby.