1 Intelligent Design The Relationship of Economic Theory to Experiments : Treatment driven Experiments

When I interact with colleagues and friends who are venturing into experimental economics, either as they prepare for their own experiment, or aim for a critical view of experiments run by others, I often hear the question: " What is a good experiment? " My first reaction is to answer my empirically minded colleagues " Well, let me ask you: What is a good regression? " Clearly a good experiment (or regression) is one that allows testing for the main effect while controlling for other plausible alternatives. This helps ensure that the original hypothesis is reached for the right reasons and the initial theory is not wrongly confirmed. However, there is an aspect of experimental design that is probably closer connected to theory than empirical work: As designers, we are responsible for the environment in which the data are generated. As such a good experimental design also needs to fulfill requirements one may impose on good theory papers: The environment should be such that it is easy to see what drives the result, and as simple as possible to make the point. The design has to provide a good environment for studying the questions at hand. Furthermore, ideally, the design (just like good theory) should be such that it seems plausible that the results could serve as prediction for behavior in other environments. The design of experiments has some interplay with empirical methods as a good experimental design foreshadows how the data to be generated can be analyzed. As such, good design can often reduce the need for fancy econometrics, or, at times, allow econometrics to be more powerful. It also often implies that the experimental design has to take a stand on what it means to accept or reject an initial theory or hypothesis. When deciding about a design, there are basically no fixed rules. The one exception probably being that economic experiments which use deception are really frowned upon, hard to