Bayesian statistics investigates the probability of an event based on prior information and new data that might, or might not, be related to the event within the model. The method is used to address various challenges across science, engineering and finance.
Over the past 25 years, Marina Vannucci
, the Noah Harding Professor of Statistics at Rice University, has become a leading expert in the use of Bayesian variable selection methods in linear settings, mixture models and graphical models, and related computational algorithms.
Vannucci and Mahlet Tadesse
, professor and chair of mathematics and statistics at Georgetown University, are co-editors of a textbook in the field of Bayesian statistics. The Handbook of Bayesian Variable Selection
, which will be published on Dec. 20 by Chapman and Hall/CRC, is available for pre-order. The researchers also contributed chapters covering their research.
Throughout her career, Vannucci has witnessed substantial developments in the use of Bayesian variable selection. “More so in recent years, especially with the proliferation of large data sets. Identifying relevant variables to include in a model makes for simpler interpretation, avoids overfitting and multicollinearity, and offers insights into the mechanisms underlying observed phenomena.”
Vannucci’s book chapter covers the theory and practice of discrete spike-and-slab priors, a modern variable selection technique that considers additional components, such as varying coefficient terms smooth interactions, spatial effects, and cluster-specific random effects.
Vannucci has a solid history of scientific collaborations and is particularly interested in applications of Bayesian inference to high-throughput genomics, neuroscience, and neuroimaging.
In a 2013 paper
, co-authored with Christian Brownlees, an associate professor in the Department of Economics and Business at the Universitat Pompeu Fabra, a Bayesian approach was used to capture daily heterogeneity in intra-daily financial-duration time series.
In this work, the researchers introduced a novel specification able to capture long-range dependence. The method involved the use of a three-component model that consisted of an autoregressive daily random effect, a semiparametric time-of-day effect, and an intra-daily dynamic component: the Mixed Autoregressive Conditional Duration (Mixed ACD) Model.
The empirical application to a set of widely traded U.S. tickers showed that the model is able of capturing low-frequency dependence in duration time-series. The researchers also found that the degree of dependence and dispersion of low-frequency dynamics were higher in periods of higher financial distress.
Vannucci joined the Rice faculty in 2007 after serving on the faculty at Texas A&M for nine years. Her research interests include Bayesian statistics, variable selection, graphical models, statistical computing and wavelets, with applications in genomics, neuroimaging and engineering.
“Variable selection is especially important,” she said, “when the number of potential predictors is substantially larger than the sample size and sparsity can reasonably be assumed.”
Vannucci earned her Ph.D. in statistics from the University of Florence, Italy in 1996. She served as Rice’s chair of the Department of Statistics from 2014 to 2019 and holds an adjunct appointment in biostatistics at the MD Anderson Cancer Center.
She served as editor-in-chief of the International Society of Bayesian Analysis (ISBA) journal Bayesian Analysis and is co-editor of the Journal of the American Statistical Association--Theory and Methods.
She is an elected fellow of the ISBA, the American Statistical Association, the Institute of Mathematical Statistics and the American Association for the Advancement of Science. Last spring, Vannucci was awarded the 2020 Zellner Medal from ISBA for exceptional service with long-lasting impact.