Archived Content

The Office of Disease Prevention (ODP) archives materials that are more than 3 years old and no longer being updated. Over time, links and other information may have changed. We cannot guarantee that all of the links in these materials will be current or accurate.

Methods: Mind the Gap

Webinar Series

Causal Inference With Instrumental Variables

October 4, 2018
Stuart Baker, Sc.D.
Stuart G. Baker, Sc.D.

Mathematical Statistician
Division of Cancer Prevention
National Cancer Institute

View the Webinar

About the Webinar

In this webinar, Dr. Stuart G. Baker focuses on the method of latent class instrumental variables (often called the LATE or CACE approach), which was developed independently by Imbens and Angrist (1994) and Baker and Lindeman (1994). 

Dr. Baker discusses how to estimate the effect of treatment received (in the complier latent class) in a randomized trial with all-or-none compliance. One example involves estimating the effect of breast cancer screening on breast cancer mortality in a randomized trial in which some participants refuse screening. He also discusses the paired availability design that he (along with K. S. Lindeman) developed in 1994, which estimates the effect of treatment received (among the consistent-receiver latent class) in a set of before-and-after studies with changes in availability of treatment over time. 

Dr. Baker’s main application of the paired availability design is a health promotion study involving the effect of epidural analgesia on the probability of Cesarean section. He contrasts the results from the paired availability design with those from a multivariate analysis that likely omitted an important confounder. The emphasis of the presentation is on assumptions and concepts, combined with some simple formulas.

About Stuart G. Baker

Dr. Baker was the first recipient of the distinguished alumni award from the Department of Biostatistics at the Harvard School of Public Health. He is also a fellow of the American Statistical Association. 

Dr. Baker's statistical research contributions include: the paired availability design for historical controls, latent class instrumental variables, relative utility curves for evaluating prognostic markers, benefit functions for evaluating predictive markers, identifying biologically relevant models in time series microarray data, a biomarker pipeline to evaluate markers for the early detection of cancer, estimation of cumulative false positive rates in cancer screening, five criteria to evaluate surrogate endpoints, estimating overdiagnosis rates in cancer screening, and the latent class twin method for estimating genetic susceptibility. He is also interested in paradoxes in cancer biology and recently proposed the detached pericyte hypothesis for tumorigenesis.