Experiments, Laboratory

Laboratory experiments are used in political science to empirically test the results and assumptions of formal models. These types of experiments are conducted using human subjects in a specific location, usually in a computer laboratory, where the experimenter has complete control over the experimental environment. These types of experiments often use a computer interface to deliver the experimental stimuli to subjects. For example, laboratory experiments are used to empirically test formal models of elections, committees, bargaining, and other political topics. This entry presents some of the special features and advantages of this method.

Formal models are mathematically defined and derived by assumptions where the assumptions (sometimes) result in an equilibrium prediction. In formal models, the assumptions are of two types: assumptions about institutional and other exogenous factors such as voting rules, bargaining procedures, preference distributions, and so forth, and assumptions about the behavioral choices of the political actors in the context of these exogenous factors. The model makes assumptions about both electoral institutions and voters' rationality, and the predictions are behavioral.

Laboratory experiments are ideal to test formal models since the assumptions of the model can be operationalized in the laboratory, and human subjects can be used to play the role of actors within the model. Hence, researchers can empirically determine how the assumptions and predictions of the model hold up when human subjects play the game. In laboratory experiments, researchers are able to establish controls over the experimental [Page 873]environment that allow them to establish and measure causality, which is how one variable(s) affects another variable(s). In experiments, causality is established by specifying different control variables and then having subjects randomly assigned to various treatments. For instance, if a researcher is interested in how the cost of voting affects voting behavior, the researcher can specify two treatments—one with a voting cost and another without a voting cost. Then, holding all other parameters of the experiment constant, it is possible to compare the two treatments and isolate the effects of voting cost, given the parameters introduced in the experiment.

Most laboratory experiments use a within-subject design as opposed to a between-subject design. In a within-subject design, subjects in the experiment experience all treatments; in a between-subject design, subjects experience only one treatment. In a within-subject design, subjects are randomly assigned to all the treatments over multiple periods. One advantage of a within-subject design is that it greatly increases the data that are collected. Another advantage as opposed to a between-subject design is that it reduces the error variance associated with differences in subject behavior (i.e., in a between-subject design, subjects may not be randomly mixed and might have shared factors that are unknown to the experimenter). Third, a within-subject design reduces repeated game effects, which means that most models are a one-shot game, but experiments usually test the model over multiple periods. To keep the one-shot nature of the model intact, subjects are assigned to different treatments over multiple periods, so that they will experience a different experimental environment in each period. Hence, each period will be a new experience, and subjects should not feel like they are playing the same experiment in each period.

...