Abstract:
|
Screening experiments involve several factors and, often, no replications. In a 1959 article, Cuthbert Daniel proposed the use of half-normal plots of estimated effects as a way to help identify which ones are the big players. He provided scales and decision limits for use with graph paper for various common effect counts. Since that time, statistical practice has moved away from hand calculations and graph paper. The use of ordinary normal plots rather than half-normal ones has become common for identifying active effects, and additionally, various diagnostic uses for these plots have developed.
In general, plots are a very good thing for displaying statistical results. But some plots are better suited for a purpose than others, and it is also important to consider who is constructing the plot, as well as who it is presented to. In this talk, we examine what characteristics of effects are most important to display when trying to identify active effects, and what plots are best for those purposes. Special emphasis is placed on what should be taught in short courses, where we have limited time to expose non-experts to important statistical concepts and methods.
|