|
Activity Number:
|
359
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Wednesday, August 6, 2008 : 8:30 AM to 10:20 AM
|
|
Sponsor:
|
Section on Quality and Productivity
|
| Abstract - #302251 |
|
Title:
|
Nonparametric Test Families for Highly Censored and/or Skewed Data
|
|
Author(s):
|
Thomas J. Bzik*+ and Robert Brill
|
|
Companies:
|
Air Products and Chemicals, Inc. and ICL Performance Products
|
|
Address:
|
5613 Tavern Circle, Macungie, PA, 18062,
|
|
Keywords:
|
nonparametric ; method detection limit ; quantile ; percentile ; SEMI C64-0308 ; ship to control
|
|
Abstract:
|
A new industry standard, SEMI C64-0308 - SEMI Statistical Guidelines for Ship To Control, uses nonparametric methodologies to determine when underlying distributions have changed. These data distributions may be both highly skewed as well as highly censored to a method detection limit. A methodology for detecting distributional change by testing whether an upper or lower tail quantile has changed is provided. The 90th and 10th percentile case will be developed and the methodology can be extended to other percentiles of interest. An approximating equation that provides very reliable critical values for n1 and n2>=30 for the proposed test statistic is provided. A performance comparison is made to one and two-sided versions of Tukey's Quick Test. The new methodology provides a more powerful test.
|
