|
Activity Number:
|
29
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Sunday, July 29, 2007 : 2:00 PM to 3:50 PM
|
|
Sponsor:
|
Section on Physical and Engineering Sciences
|
| Abstract - #308411 |
|
Title:
|
Identification of Dispersion Effects in Replicated Experiments
|
|
Author(s):
|
Cheryl Dingus*+ and Angela Dean and Bruce Ankenman
|
|
Companies:
|
Battelle and The Ohio State University and Northwestern University
|
|
Address:
|
505 King Ave, Columbus, OH, 43201,
|
|
Keywords:
|
Dispersion ; Replicated Experiment
|
|
Abstract:
|
A factor is said to have a dispersion effect if the variability of the response is different at the different factor level settings. In replicated experiments, replicate observations for a treatment are used as a measure of the variability of response. Using traditional methods, a summary statistic is used as the analysis variable in an F-test. The disadvantage of traditional analysis methods is the loss of power to identify effects. This talk will present an alternative method of analysis that transforms each observation into an "individual measure of the variability", preserving all the original degrees of freedom and thereby increasing power over the traditional method. Several functions that have been examined for this alternative analysis will be presented. Results will show that the new methodology is better able to detect dispersion of a single factor than the traditional method.
|
