MAIN POINTS

Introduction

The controlled experiment allows the most unequivocal evaluation of causal relationships between two or more variables. However, many phenomena that are of interest to social scientists are not amenable to the straightforward application of experimental designs. Quasi experiments, cross-sectional designs, and pre experiments are designs that are generally weaker on internal validity and have limited causal inferential powers.

Types of Relations and Designs

Stimulus response relationships are well suited for experimental investigation, but property disposition relationships are not. There are four reasons for this; in stimulus response situations: 1) the interval between cause and effect is generally short; 2) the independent variable is often specific and easy to identify; 3) it is quite simple to create groups that differ according to exposure to the independent variable but that are similar otherwise; and 4) it is easy to determine whether the independent variable actually occurred first.

Cross-Sectional Designs

The cross-sectional design is perhaps the most predominant design employed in the social sciences. This design is often identified with survey research. The most common alternatives to experimental methods of control and the drawing of causal inference in cross-sectional designs are multivariate analytic techniques such as cross tabulation and path analysis. In cross-sectional design, researchers cannot establish time order of the variables by performing statistical analyses, but rather this must be done on the basis of theoretical and logical considerations.

Quasi Experimental Designs

Quasi experimental designs are weaker on internal validity than experimental designs, and like cross-sectional designs, they depend on data analysis techniques as a method of control and do not require randomization. They are superior to cross-sectional designs because they usually involve the study of more than one sample, often over an extended period of time.

In contrasted-groups designs, researchers observe a dependent variable in various groups that differ on independent variables. This design is weak on internal validity, but often researchers succeed in compensating for this fault by the use of matching, by expanding the number and variety of contrasted groups, or by making repeated observations of the dependent variable.

Planned variation designs are similar to contrasted groups designs, except that in the former, a stimulus, such as a program, is systematically presented to varying groups with different characteristics. Planned variation techniques are most effective when important variables are equally distributed across the groups involved.

Time series designs involve a number of observations of the dependent variable both before and after the introduction of the independent variable. This type of design generally allows researchers to rule out testing and maturation as possible explanations for changes after the introduction of the independent variable. History and regression effects are harder to rule out, particularly if the researcher has not made many observations over a long period of time.

In control series designs, time series are developed both for the "experimental" group and for a number of "control" groups that appear to be essentially equivalent to the "experimental"group. The groups are not really equivalent, because they were not formed by randomization. Still, they provide some protection against the threats to internal validity of history, maturation, and testing.

Combined Designs

All quasi experimental designs have certain weaknesses. At times, researchers can effectively combine two or more of these designs in the same investigation in order to allow the strengths of one to compensate for the weaknesses of another. This may entail the development of smaller scale experiments within a large quasi experimental framework.

Preexperimental Designs

Preexperimental designs are not suitable for experimental manipulations and do not allow for the random allocation of cases to an experimental and a control group. In fact, most often these designs do not include a comparison group. Preexperiments are the weakest kinds of research designs since most of the sources of internal and external validity are not controlled for. The risk of drawing causal inferences from preexperimental designs is extremely high, and they are primarily useful as a basis for pretesting some research hypotheses and for exploratory research.

An example of a preexperimental design is the one shot case study, which involves an observation of a single group or event at a single point in time, usually subsequent to some phenomenon that allegedly produced change. This technique is useful in exploratory research and may lead to insights that, in turn, could be studied as research hypotheses.

A Comparison of Designs

Two extremely basic problems in scientific research are inferring causation and generalizing the findings, and these problems pose an equally basic dilemma: in order to secure unambiguous evidence about causation, researchers frequently sacrifice generalizability. This is the problematic relationship between internal and external validity. Designs that are strong on internal validity tend to be weak on external validity, and vice versa.

Experimental designs and their counterparts tend to be internally valid but hard to generalize from, whereas surveys and some of the "weaker" quasi experimental designs tend to be less internally valid but easy to generalize from. The contrast hinges largely on whether randomization or representative sampling is used and on the artificiality of the research setting. The dilemma of internal versus external validity can be partly resolved by using representative samples of well defined populations in experiments and by seeking additional information to rule out certain rival hypotheses in survey investigations.