Networking meta-analysis (NMA) is a statistical tool for combining independent studies or clinical trials and contrasting multiple treatment effects simultaneously in a systematic review. Currently, two types of approaches are used for network meta-analysis. The first is contrast-based, in which a baseline treatment is specified in each study and the analysis focuses on modeling treatment contrasts (relative treatment effect, typically log odds ratios) within each study. Another approach is arm-based, which focuses on describing population-averaged absolute effect sizes for each treatment arm, treatment-specific parameters, such as absolute risks, are modeled and estimated. Bayesian methods are well-suited for NMA where models can be complex, and hence are widely applied recently in this area. This talk shows how to conduct Bayesian network meta-analysis, to estimate both absolute and relative effects, and to handle binary, continuous, and count outcomes. Many R packages and SAS’s Bayesian procedures provide convenient access with high performance and efficient sampling. I will present worked-out examples to illustrate how to use them for estimation, inference, and prediction.

As sources of data become more plentiful and massive data sets are easy to acquire, new ethical issues arise involving data quality and privacy, and the analysis, interpretation and dissemination of data-driven decisions. There are numerous anecdotes involving abuses of complex data analyses and algorithms, and the impact they have had on society. Past statements of professional ethics for statisticians were limited to roles traditionally held by statisticians. But as the lines blur between those roles and recently expanded roles in data science, we have both the obligation and the opportunity to expand the discussion of data science ethics as well. This talk will address what statisticians and data scientists can do to help enhance data science ethics in practice, with a focus on how statisticians can help raise awareness and encourage implementation of ethical best practices.

Mixture models are one of the most widely used statistical techniques, when dealing with data from heterogeneous populations. Generally speaking, the Bayesian and Machine Learning literature have been divided between finite mixture (possibly with random number of components) and infinite mixtures. The main contribution of this work is to bring the two modelling strategies together. To this end, following a Bayesian nonparametric perspective, we introduce a new class of prior process: the Normalized Independent Point Process. We investigate the probabilistic properties of this new class in terms of the induced partition of the observations and provide posterior characterization of the process. Moreover, we discuss important special cases. One of the main advantages of this new construction is that it allows designing a conditional algorithm for finite mixture models with random number of components which overcome the challenges associated with the Reversible Jump Markov chain Monte Carlo scheme and the recently proposed marginal algorithms. We illustrate the new class of prior processes on the popular Galaxy data example and on a relevant application in population genetics.

High dimensional categorical data are routinely collected in biomedical and social sciences. It is of great importance to build interpretable models that perform dimension reduction and uncover meaningful latent structures from such discrete data. Identifiability is a fundamental requirement for valid modeling and inference in such scenarios, yet is challenging to address when there are complex latent structures. In this article, we propose a class of interpretable discrete latent structure models for discrete data and develop a general identifiability theory. Our theory is applicable to various types of latent structures, ranging from a single latent variable to deep layers of latent variables organized in a sparse graph (termed a Bayesian pyramid). The proposed identifiability conditions can ensure Bayesian posterior consistency under suitable priors. As an illustration, we consider the two-latent-layer model and propose a Bayesian shrinkage estimation approach. Simulation results for this model corroborate identifiability and estimability of the model parameters. Applications of the methodology to DNA nucleotide sequence data uncover discrete latent features that are both interpretable and highly predictive of sequence types. The proposed framework provides a recipe for interpretable unsupervised learning of discrete data, and can be a useful alternative to popular machine learning methods.
(Joint work with Yuqi Gu, Department of Statistics, Columbia University)

Hierarchical models present a powerful tool in statistics, machine learning and numerous applied fields. They allow for complex statistical dependence to be effectively expressed via latent variables. They also enable the "borrowing of strength" — a heuristic intuition hinting at the benefit of combining multiple data sets by making them share some latent variables in common. To understand this phenomenon, one needs to study the convergence behavior of latent variables that arise in hierarchical models. In this talk I will describe recent progress in the study of such questions, by reviewing some recent results, highlighting a few primary techniques and outlining several open questions. The main focus is on geometric properties of latent variable models which are examined via various notions of optimal transport distances and singularity structures. Modeling objects of interest include canonical mixture models, Bayesian mixture of product distributions, as well as nonparametric hierarchical Bayesian models such as the hierarchical Dirichlet processes.

In this talk, I will provide an overview of data science methods, including methods for Bayesian analysis, causal inference, and machine learning, to inform environmental policy. This is based on my work analyzing a data platform of unprecedented size and representativeness. The platform includes more than 500 million observations on the health experience of over 95% of the US population older than 65 years old linked to air pollution exposure and several confounders. Finally, I provide an overview of studies on air pollution exposure, environmental racism, wildfires, and how they also can exacerbate the vulnerability to COVID-19.