Practical Options for the Detection and Management of Sample Data Outliers

• Activity Number: 547
• Type: Contributed
• Date/Time: Wednesday, August 7, 2013 : 10:30 AM to 12:20 PM
• Sponsor: Biopharmaceutical Section
• Author(s): Alfred Barron*+
• Companies: Janssen Research & Development
• Keywords: outliers ; robust ; m-estimation ; median-MAD ; simulation

Abstract

To help provide for more accurate assessment of experimental results, statistical methods have been developed for research investigators to identify and address the impact of extreme or outlying sample values, often the source of highly data variability.

Classical statistical methods based on the first two moments such as linear regression or ANOVA can be adversely affected by outliers, often providing a misleading fit of the data. Alternative robust statistical methods exist that are more resistant to extreme values and thus, more reliable in providing a stable inference about the research treatments in question. Moreover, these methods provide working options for research investigators which help to preserver data integrity, thus helping to meet some of the goals of quality compliance standards.

Methods for identifying outliers based on the boxplot and a median-median absolute deviation (MAD) argument will be addressed in addition to a method of robust data transformation that down weighs the impact of outliers using M-estimation.