Research
The state I want is not observed. The measurements I get are imperfect. I work on whether the target is identified from a given design, how to estimate it, and how to push that uncertainty through the number a decision actually uses.
Dissertation (in progress). Advisor: Valen E. Johnson.
Chapter I — Prevalence without a gold standard.
Prevalence and diagnostic accuracy are estimated from imperfect tests and heterogeneous study designs. I characterize when those parameters are identified, then estimate them with Bayesian latent-class and latent-severity models, checked by parameter-recovery simulations.
Chapter II — Cost burden.
Prevalence is an intermediate. The target is a distribution over economic loss, obtained by propagating the Chapter I posterior together with uncertainty in population and cost inputs.
Application: diagnostic testing and disease burden. The methods are not specific to that application.
Selected work
Experience
- Advisor: Valen E. Johnson.
- Identifiability and Bayesian estimation for latent disease status without a gold standard.
- Uncertainty propagation from prevalence into a distribution over cost burden.
- Undergraduate probability, estimation, testing, regression, and computation in R.
- Credible intervals for predictive models, so uncertainty was attached to the forecast rather than reported separately.
- Repeatability and reproducibility studies to quantify measurement variability for product validation.
- Predictive models from chemical composition to transmission-fluid performance.
- High-dimensional regression and clustering on DNA methylation data.
- NHANES analysis of dietary risk factors for hypertension in pregnancy.
- Analysis contributed to peer-reviewed papers listed below.
Publications
Collaborative papers. Role on each paper is listed after the citation. Dissertation chapters are in preparation and are not listed as publications.