To facilitate the design of efficient simulation experiments, Schruben and Margolin (1978) recommend a correlation induction strategy for orthogonally blockable experimental designs. The objective of such experiments is to estimate a general linear regression model on the basis of a quantitative response variable generated by the simulation model. Nozari, Arnold, and Pegden (1987) develop optimal statistical procedures for analyzing simulation experiments performed under the Schruben-Margolin correlation induction strategy. Formulas are given for parameter estimation, hypothesis testing, and confidence interval estimation. The validity of this statistical analysis procedure is contingent upon the presence of a pure error component in the response. The goal of this thesis is to provide an appropriate statistical analysis technique for simulation experiments conducted under the Schruben-Margolin correlation induction strategy in the absence of pure error, and to identify conditions under which the pure error component is absent. Often, in order to construct valid inferences on the responses from a simulation experiment, the technique used to execute the simulation experiment must be properly identified. For purposes of this research, the identification problem takes the form of ensuring that the hypothesized metamodel is appropriate for the number of random number streams used to induce correlations between responses across design points. / Master of Science

The aim of this thesis is to research the issues involved in creating distributed problem solving environments for scientific computing. As part of our evaluation, we have developed a distributed problem solving environment called DPSolve which combines a very high level language, an interactive X Windows interface and a set of powerful problem solving methods into a single environment. The interface is designed to work on any system running X Windows, whilst the computations are done on a more powerful parallel computer. We implemented the interface on a DEC3100 workstation running ULTRIX, which communicates with procedures running on a Sequent 581 with 10 processors, running DYNIX via RPC. The design decisions and implementation details of our system are discussed at length along with a detailed example of the system at work. We critically evaluate the approach we have taken and show why it can scale to a very large class of scientific problems. We conclude that this distributed environment should be a representative of future scientific problem solving environments. / Master of Science