Don Edwards
Department of Statistics
University of South Carolina
Exact, Flexible Sequential Acceptance Sampling for Attributes,
With Applications to Medicare Fraud Investigations
We consider the problem of sampling without replacement from a finite lot of
N items, where each item may be identified simply as tainted (in error,
defective, etc) or not. Our goal is to make inference on the total number
of items in error NE over the entire lot. Modern computing power renders
sequential and two-stage approaches to this problem completely tractable
without resorting to binomial approximations, which will be unreliable in
the important cases when NE is near 0 or N. We discuss the theory and
computational issues for general elimination boundaries. We show that the
solution to the fully sequential problem reduces to predictable
finite-sample or multistage approaches using strategic choices for
elimination boundaries. We discuss issues in choosing these boundaries to
minimize costs in an important auditing application: random sampling of paid
Medicare claims for investigation of suspicious billing practices by
supposed health care providers. Billions of dollars are lost each year to
Medicare fraud.
This is ongoing joint work with:
Iliana Ignatova
Dept. of Statistics
Cal Poly San Luis Obispo
Roland Deutsch
Dept. of Mathematics
University of North Carolina Greensboro
Back to Colloquium Series