Automated Simulations

Fast, inexpensive computers have become the experimental laboratories of research statisticians. Exploratory simulations provide quick feedback on the validity of conjectures and often suggest modifications leading to new approaches. Simulations reject false conjectures and identify suitable candidates for the rigorous task of theorem proving. In spite of the growth in computing power, the tedium of simulation programming often results in studies that fail to identify important properties. Software that automates routine tasks dodges many of these problems and frees time for thinking about the problem rather than debugging code. This paper describes the design and use of a computing environment that automates the tasks of describing, running, and analyzing simulations. The paper begins with a short illustration intended to convey broad concepts needed in an efficient simulation program. Building on this discussion, I describe the design of the software and use a prototype implementation to run a small simulation. Interested readers may obtain this prototype as well as further documentation and examples via anonymous FTP from the system at hilbert.wharton.upenn.edu. The prototype is written in X-Lisp and requires the Lisp-Stat software described in Tierney (1990).