Dan Kowal

Dan Kowal, PhD

Dobelman Family Assistant Professor

Department of Statistics

Rice University

Research overview:

My research program contains three main themes: 

1. Bayesian models and algorithms for large and dependent data: In modern applications, dependencies in the data are often unavoidable and appear concurrently (e.g., multivariate, time series, functional, and/or spatial data). A focal point of our research is the careful development of Bayesian models and algorithms for reliable and scalable inference with dependent data, including cases with restricted or nonstandard support (e.g., discrete, mixed, or spherical data).

2. Modeling, synthesis, imputation of mixed data: Mixed datasets, including continuous, count, ordinal, and unordered categorical variables, are abundant and challenging to model jointly—especially in the presence of informative missingness. Our research innovations include nonparametric and semiparametric modeling frameworks that provide correct support, substantial joint modeling flexibility with theoretical guarantees, and highly convenient and scalable computing. This work is particularly useful for prediction, data synthesis for privacy protection, and imputation of missing data.

3. Predictive inference for more interpretable uncertainty quantification: Almost universally, statistical measures of uncertainty are linked to unobservable parameters in complex models, which inhibits clear communication to both scientific and public audiences. Our work introduces a unified, posterior predictive framework for versatile, observation-driven uncertainty quantification. This framework is widely applicable to classical statistical problems as well as complex Bayesian models, and includes important special cases such as subset selection, detection of critical windows of susceptibility (e.g., to adverse environmental exposures), and inference with missing data.

This research is motivated by urgent and open questions in public health, epidemiology and environmental justice, physical activity data, economics, and finance.

My research has been recognized with a Young Investigator Award (Army Research Office), the inaugural Blackwell-Rosenbluth award (International Society of Bayesian Analysis) for “junior researchers in different areas of Bayesian statistics”, and several paper and dissertation awards.

News:

Install and load using the following code:

library(devtools)

devtools::install_github("drkowal/fosr")

library(fosr)

Uses include:

  1. Bayesian estimation and inference for function-on-scalar regression: fosr(...) 

  2. Decision-theoretic approach to variable selection in functional regression: fosr_select()

  3. Additional tools for plotting, simulations, and evaluation of model fit

Install and load using the following code:

library(devtools)

devtools::install_github("drkowal/dsp")

library(dsp)

Uses include:

  1. Curve-fitting of irregular data via Bayesian trend filtering: btf(...)

  2. An adaptive time-varying parameter regression model: btf_reg(...)

  3. Curve-fitting of irregular data with unequally-spaced observations: btf_bpsline(...)