Much of the art of biostatistics is a pragmatic balance, combining rigorous-but-abstract concepts from theory with application-based insight about what might work in practice. In this talk we explore how a single framework – using “balanced loss functions” – can motivate methods through explicit tradeoffs, rather than relying on analysts’ pragmatism or artistry. Our first example develops a celebrated bias-correction method due to Zhong & Prentice (2008, Biostatistics), that corrects for regression to the mean – and shows its close connection to shrinkage estimates due to Stein and others, that trade off accuracy (proximity to the truth) for simplicity (defined as proximity to zero). A second example explores how the tradeoff that determines statements of uncertainty for standard parametric inference can be modified based on apparent violations of model assumptions – leading to a new way to view quasi-likelihood, and use of robust standard errors. Extensions of both examples will be considered.

This is joint work with Chloe Krakauer and Kendrick Li.

About Ross L. Prentice Endowed Lecture:

The Prentice Endowed Professorship was established in 1998 by the University of Washington and the Fred Hutchinson Cancer Research Center to honor the continuing service of Dr. Ross Prentice to both institutions. The professorship is generally awarded on an annual basis to a biostatistician who works collaboratively between the UW Department of Biostatistics the Division of Public Health Services at Fred Hutch. The recipient delivers the annual Prentice Lecture and teaches a UW course on a topic related to their research.