Prospects and Pitfalls of Statistical Testing: Insights from Replicating the Demographic Prisoner's Dilemma

This paper documents our efforts (and troubles) in replicating Epstein's (1998) demographic prisoner's dilemma model. Confronted with a number of ambiguous descriptions of model features we introduce a method for systematically generating a large number of model replications and testing for their equivalence to the original model. While, qualitatively speaking, a number of our replicated models resemble the results of the original model reasonably well, statistical testing reveals that in quantitative terms our endeavor was only partially successful. This fact hints towards some unstated assumptions regarding the original model. Finally we conduct a number of statistical tests with respect to the influence of certain design choices like the method of updating, the timing of events and the randomization of the activation order. The results of these tests highlight the importance of an explicit documentation of design choices and especially of the timing of events. A central lesson learned from this exercise is that the power of statistical replication analysis is to a large degree determined by the available data.