| Abstract: | Popularized scientific accounts have publicized the use of the computer for solving combinatorial problems associated with the construction of specific kinds of experiment designs, such as sets of orthogonal Latin squares and Hadamard matrices. Historically, the regular combinatorial properties associated with such designs have been required to obviate extremely tedious calculations required for design analysis. Now computers are generally available for performing these computations, so that interest in regular designs is waning. The experiment designer is now free to select a design for its appropriateness to the experiment and its statistical properties. Once again, he interacts with the computer, using it to evaluate the statistical properties of designs, even before the experiment is carried out. Based on such calculations, he may make a selection from a number of ad hoc designs constructed for a given experiment without regard for combinatorial properties. Although the experiment designer will always be a frequent computer user, the most important interaction between the two will probably be in the application of experiment design to computer simulation. Parametric studies of comprehensive system models that have been programmed for computers can be conducted efficiently and economically with experiment design and the associated analysis techniques. A typical example is given. |
