Academic writers frequently run across a variety of sources involving experimental designs as they conduct research for their papers. Understanding what these experimenters are trying to do and why can prove useful as one assesses the validity and application of the material to the paper being researched. In addtion, understanding the basic principles of research design outlined here can help the writer design suitable experiments as well.

The classes of research design include:

Type O: The Simple Observation
Type I: The One Shot Study
Type II: The Single Group Pre Post Design
Type III: The Static Group Design
Type IV: Cross-Sectional (survey) Design
Type V: The Pre Post Control Group Design
Type VI: The Posttest Only Control Group Design
Type VII: The Solomon Four Group Design
Type VIII: Factorial Designs
Type IX: The Separate Sample Pre Post Design
Type X: The Nonequivalent Control Group Design
Type XI: The Interrupted Time Series Design
Type XII: The Interrupted Time Series Control Group Design

The cardinal principle of research design is summed up as  "doing only one thing at a time" or, in other words, allowing only one independent variable to vary while the experimenter controls all the other variables as he or she measures the effect on the dependent variable.

Threats to validity can be either internal, external, or both.  A threat to validity is, by definition, any factor that influences the results of the experiment. Internal threats are those created by the choice and design of the experiment, external threats are those influences beyond the experiment that are largely uncontrollable.

Study Types

The nature of experimental designs is controlled by the degree the experimenter takes to minimize the effect of validity threats. Ideally, a tightly controlled experiment should provide the most valid research data, however, depending on circumstances, other designs may provide the best attainable results.

NON RESEARCH

Non research designs do not address threats to validity.

PRE-EXPERIMENTAL

Pre-experimental designs are minimally designed to address threats to validity, thus one must approach the results with a healthy degree of skepticism. The advantage presented by these designs are the ease of execution, low cost, and speed. These designs rely exclusively on statistical controls only.

EXPERIMENTAL

Experimental designs are designed to minimize the threats to validity and present the best chance at accurate results. However, the painstaking effort to ensure validity guarantees the execution to be challenging, costly, and take time.  Experimental designs acheive experimental control by the combination of random selection and assignment, the use of both control and experimental groups, researcher manipulation of the stimulus, and pretest posttest measures.

QUASI-EXPERIMENTAL

Quasi-experimental designs may be indicated as an alternative to true experimental designs where ethical or practical considerations preclude administration of the experimental intervention to the experimental group or denial of the intervention from the control group. The quasi experimental design acheives a combination of experimental and statistical control. They combine at least one experimental control feature with sound random sampling techniques.

Internal Threats to Validity

Extraneous variables potentially threaten the validity of research studies and must be taken into account either when designing experimental research or when assessing the validity of an existing study. These threats fall into the following eight threat classes:

HISTORY

External events occurring before and during the experiment capable of skewing results.

MATURATION

The normal changes that occur over time regardless of the experiment.

TESTING EFFECTS

The effect of the experiment itself on the test subjects brought about by their participation.

INSTRUMENTATION

The errors resulting from inaccuracies in measurement instrument or changing procedures or criteria. Instrumentation effects are especially troublesome when the instrument is a human observer.

STATISTICAL REGRESSION

The tendency of test subjects with extreme scores to grow closer to the mean with each additional test. Also referred to as 'regression to the mean'

SELECTION

The group makeup (choice of test subjects) can potentially affect results, so, for example, a group made up of one demographic may respond very differently from another demographic. A related concern occurs if the makeup of the control and experimental groups is dissimilar. 

ATTRITION

The loss of test subjects during the experiment.

INTERACTION

Two or more threats working together may influence experimental outcomes even though the effect of individual threats remain negligible.

—By Douglas E. Greathouse, MA.(2007)