You are here:
METHODS OF DEALING WITH VALUES BELOW THE LIMIT OF DETECTION USING SAS
Croghan, C AND P P. Egeghy. METHODS OF DEALING WITH VALUES BELOW THE LIMIT OF DETECTION USING SAS. Presented at Southeastern SAS User Group, St. Petersburg, FL, September 22-24, 2003.
o Develop publically accessible databases
o Disseminate human exposure data bases and data tools in an appropriate manner to improve human exposure research and information.
o Ensure that human exposure data sets generated by HEASD and its contractors or collaborators are useful for data analysis and for human exposure modeling, by providing formats and guidance on data base creation, documentation, storage, and retrieval.
o Ensure that data sets produced in HEASD conform with ORD and EPA requirements.
o Compile and translate existing human exposure data sets into useful formats for use by HEASD scientists and modelers.
o Make fully documented human exposure data sets available for researchers with collaborating external organizations.
Due to limitations of chemical analysis procedures, small concentrations cannot be precisely measured. These concentrations are said to be below the limit of detection (LOD). In statistical analyses, these values are often censored and substituted with a constant value, such as half the LOD, the LOD divided by the square root of 2, or zero. These methods for handling below-detection values two distributions, a uniform distribution for those values below the LOD, and the true distribution. As a result, this can produce questionable descriptive statistics depending upon the percentage of values below the LOD. An alternative method uses the characteristics of the distribution of the values above the LOD to estimate the values below the LOD. This can be done with an extrapolation technique or maximum likelihood estimation. An example program using the same data is presented calculating the mean, standard deviation, t-test, and relative difference in the means for various methods and compares the results. The extrapolation and maximum likelihood estimate techniques have smaller error rate than all the standard replacement techniques. Although more computational, these methods produce more reliable descriptive statistics.
This paper has been reviewed in accordance with the United States Environmental Protection Agency's peer and administrative review policies and approved for presentation and publication.