Online Program Home
My Program

Abstract Details

Activity Number: 564 - Analysis of Left-Censored Data (E.G., Below Detection): Real-World Problems in Need of Statisticians
Type: Topic Contributed
Date/Time: Wednesday, July 31, 2019 : 2:00 PM to 3:50 PM
Sponsor: Biometrics Section
Abstract #301776 Presentation
Title: Quantifying Information in Left-Censored Data: Why the Percent Censored Is a Misleading Metric
Author(s): Brenda W Gillespie*
Companies: University of Michigan
Keywords: left-censored; detection limit; limit of detection; information

Left-censored data often occur as values below a detection limit in toxicology, chemistry and environmental science. Analysis guidelines from governmental agencies in the U.S. and Europe are based on the number of observed values and the percent of censored values, with a high percent censored considered a liability. While the number of observed values is a useful measure, the percent censored fails to convey the range of information inherent in a censored value: A value left-censored at infinity brings no information (the true value could be anywhere); a value left-censored near zero brings almost complete information (the true value is very small). Improved instrumentation has reduced detection limits, giving more precise information on values below detection. An environmental initiative that reduces levels of toxins (which is good) will result in more left-censored values that, unfortunately, are considered a liability. A metric is proposed to replace 'percent censored' with a measure of information based on the number of observed values plus a fraction of each censored value, depending on its detection limit.

Authors who are presenting talks have a * after their name.

Back to the full JSM 2019 program