The classic forms of experimental design in scientific research include the completely randomized design, the quasi experiment, and the non experiment. These classic forms are based on the principles of (1) being able to insure that only random selection, and no other factor, bias or variable led to the selection of the test population. (2) being able to insure internal validity in establishing cause and effect relationships, or that the cause and effect relationship that is being tested has no other explanation or other possible cause other than the one that is hypothesized. (3) The level of control over the variables, the actions of the test and control populations, as well as the elimination of bias in making the observations.
Randomization is considered to be the gold standard of experimental design in all of the sciences. But a completely randomized design is nearly impossible to accomplish in the social sciences, especially with large populations, diverse populations, or even populations which cannot be controlled for specific variables because they are human beings. In addition, when complex programs are introduced to complex test populations in order to confirm a cause and effect hypothesis, no amount of randomization will matter or be possible, since the program is supposed to be applicable to large and diverse populations which operate in random ways.
A quasi experiment is one which, first, does not have complete randomization in the selection of the population. However, various levels and forms of screening and selection may go on, and these can be quite exacting and complex. If multiple groups are involved, they may be selected to make sure they are equivalent, or they may not be equivalent at all. If testing poor children for reading levels, for example, then the selection of only wealthy children would not make sense.
Second, there are multiple populations. The most basic form of multiple population quasi experiment is the test group and the control group. The test group is exposed to the causative variable and the control group is not. There can be program tests, where there are multiple causative variables that each have dependent variables or multiple causative variables that are expected to have an overall effect on a largely defined dependent variable. A social program to tutor poor children with multiple sub programs might be tested to see if the students overall performance improves. This would be a weak design, however, as too many other variables can intervene.
Third, there is a pre test of both groups to see where they stand on the hypothesized issue. For a reading tutoring program, both groups would be tested for reading level before the experiment. Then the reading program is completed. Both groups are tested to see if there is support for the hypothesis that tutoring in reading improves reading scores in the test group. The control group would be expected to have the original reading level, with some factoring for the fact that they have continued in school while the rest were in school and tutoring.
A non experiment involves situations where there is no randomization whatsoever, since only certain people are available. There may no way to conduct pre and post testing of a hypothesis by introducing variables to determine cause and effect. The only way to support a hypothesis is from interviews or surveys where people can identify or describe changes that they encountered after an event.
For example, after a natural disaster, where no one could have foreseen to test or observe people’s original status, interviews and surveys provide information about changes in the lives of the people who survived. There is also no control or comparison group, because only that particular town or region functioned in the way that it did before the disaster.
Well designed non experiments provide valuable data and information, even when they do not come close to meeting the gold standard for experimentation.