Abstract
Changes in statistical thinking and improvements in methodology of designing experiments and analyzing data in the past quarter-century are highlighted in this article. Following a few introductory and historical comments are two major sections, one devoted to fundamental issues and controversies, the other to procedural developments. Future-oriented as scientists are prone to be, they should refrain from associating the merit of an idea with its propinquity. Perhaps more important to valid assessment of experimental results than any of the developments of the past two or three decades are the fundamental concepts of Fisher regarding the nature of experimental units, their replication and intimate association with experimental error, and the crucial role of randomization. To be abhorred are the propensities of scientists 1) to calculate standard errors as if variances were unequal when conclusions have been based on the assumption that they are equal, 2) to rely almost exclusiveiy on "all-pairs" techniques for specific comparisions of means, and 3) to join the cult of "P<.05 or perish". Recent advances in analyses for categorical data, regression, mixed-model factorials, and repeated measurements are reviewed. In the future, additional emphasis is likely to be given to 1) response surface designs, 2) examination of data for validity of assumptions (via interactive computer systems), and 3) multivariate analysis of experiments with multiple responses to treatment.

This publication has 189 references indexed in Scilit: