Quasi-Experimental Design

Conceptual Overview and Discussion

The proper use of a quasi-experiment depends on establishing the equivalence between units. This is often achieved through the use of control variables in a multivariate regression framework, or through the use of extensive case histories. Analogous to the between- and within-subject design, some quasi-experiments will compare differences across regions or subjects. For example, one could examine the effects of smoking on community health in one state that has a smoking ban and another that does not have such a ban, provided the states are otherwise similar on key variables. By comparing community health measures, some inference could be made about the effects of smoking bans. Alternately, an analyst could compare changes over time in the same region, for example, gauging the effects of a smoking ban in an isolated community before and after a ban is introduced, as was performed by Richard Sargent, Robert Shepard, and Stanton Glantz in their study of a smoking ban in an isolated Montana town.

The difficulty of inference from quasi-experiments is clear. Unlike in the laboratory, the analyst here has to contend with endless sources of interference, lack of control, and multiple plausible explanations of observed changes. In their 1968 study of a crack-down on speeding in Connecticut, Donald T. Campbell and H. Laurence Ross provide a definitive analysis of a quasi-experiment. The fundamental credo in drawing causal inference from a quasi-experiment, they contend, is that because we lack control and randomization it becomes much more difficult to establish cause and effect, especially in the face of competing explanations. Causal inference is possible, then, only because analysts can use observations before and after the event as well as auxiliary data to establish trends, minimize noise, pick up regressions to the mean, and rule out other explanations.

...