School Seminars and Colloquia

Debiased estimation of proportions in group testing

Statistics Seminar

by Dr. Graham Hepworth

Institution: The University of Melbourne
Date: Thu 13th September 2007
Time: 12:00 PM
Location: Room 213, Richard Berry Building, University of Melbourne

Abstract: Group testing occurs when units from a population are pooled together and
tested as a group for the presence of an attribute, such as a disease. A
positive result indicates the presence of at least one positive unit in the
group. This work concerns the estimation of the proportion p of the
population possessing the attribute. Our interest in group testing arose
from an assessment of virus levels in carnations grown in glasshouses. In
that study a sequential three-stage procedure was used, with progressively
smaller group sizes at each stage.

The MLE of p is biased, as are other estimators. Analytical correction of
the bias works well when all groups have the same number of units. It does
not work so well, however, for most sequential procedures because of the
rather unpredictable bias patterns. We present a numerical bias-correction
technique, which begins with the MLE, and removes a substantial amount of
the remaining bias with each iteration. The result is an almost unbiased,
or "debiased" estimator. This technique can be used with any group testing
procedure, and has more general applicability.

For More Information: Dr. Owen Jones