Haruvy’s Experimental Philosophy

My experimental philosophy is a hybrid of the different views held by my mentors and co-authors in the field regarding how experiments should be run and what experiments should accomplish. My resulting view of experiments can be summarized as follows.

1. Pay attention to detail. The old guard of experimental economics used to spend a great deal of time and attention on the mechanics of experiments. That means agonizing over every word in the instructions, making sure subjects understand what is expected of them (through carful pre-experiment quiz and a post-experiment questionnaire), trying different monetary incentives, including binary lotteries and other incentive schemes to minimize wealth effect and portfolio building (see various works by Stahl, including working papers), trying different matching protocols to minimize contagion effects (again, Stahl), etc.  When I first became aware of experimental economics and started attending conferences (mid 1990s), such economists were the majority of the field. Today this group is shrinking. Perhaps because the field has become so big, new entrants feel comfortable following established rules and protocols and maybe this is the way it should be for a large field. I think there is some danger in that.  I am also sometimes a little suspicious of (often famous) results that I think could go away with a different design. Dale Stahl and I have shown that some ultimatum and dictator game results can radically change with different (possible better) designs. We have since extended these results to other games. 

2. Replicability.  Replicability is necessary in experiments. Otherwise, each study is a stand alone and there is no progression from one study to the next. Replicability means writing instructions in the simplest language possible and giving subjects many repetitions to learn. It also means focusing on a set of problems that can be replicated. In general, the more complex an experiment is, the less likely it is to be easily replicated. Lastly, sharing datasets and programs with other researchers is part of replicability as well.

3. Generality. Generality means that studying one game and fitting it econometrically (something that became popular in the late 1990’s) may be counterproductive. It may be a lot more scientifically relevant to have a model that does not fit well for any one game but does well predicting new games in advance. After all, we are in the business of making predictions.

4. Usefulness. To make ourselves useful we must study economically relevant problems. This means identifying real world issues and designing experiments that might provide answers, rather than running experiment and then finding real-world examples that loosely correspond.