|
Activity Number:
|
270
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Tuesday, August 4, 2009 : 8:30 AM to 10:20 AM
|
|
Sponsor:
|
Section on Statistical Computing
|
| Abstract - #303810 |
|
Title:
|
Two Graphical Methods for Outlier Detection in Discrete Distributions
|
|
Author(s):
|
Fiona McElduff*+
|
|
Companies:
|
University College London
|
|
Address:
|
Institute of Child Health, W5.09, Epidemiology and Biostatistics, London, WC1N 1EH, United Kingdom
|
|
Keywords:
|
outliers ; discrete distributions ; empirical probability generating function ; surprise index
|
|
Abstract:
|
Discrete distributions are often long tailed with a few observations with high values that may be crucial in statistical analyses. Such values may either be due to characteristics of the underlying distribution or may potentially be outliers. We present two graphical methods for outlier detection in discrete distributions. The empirical probability generating function (epgf) provides a smooth projection of observed discrete data and can be plotted with the probability generating function (pgf) of the theoretical probability distribution to reveal potential outlying observations. The Surprise Index is an empirical measure of how unexpected an observed value is with respect to a probability model. We apply these methods to model data from epidemiological and clinical studies. The analyses and graphical presentations discussed are implemented in the R framework for Statistical Computing.
|
