Annual Review of Statistics and Its Application - Current Issue

Due to dependent happenings, vaccines can have different effects in populations. In addition to direct protective effects in the vaccinated, vaccination in a population can have indirect effects in the unvaccinated individuals. Vaccination can also reduce person-to-person transmission to vaccinated individuals or from vaccinated individuals compared with unvaccinated individuals. Design of vaccine studies has a history extending back over a century. Emerging infectious diseases, such as the SARS-CoV-2 pandemic and the Ebola outbreak in West Africa, have stimulated new interest in vaccine studies. We focus on some recent developments, such as target trial emulation, test-negative design, and regression discontinuity design. Methods for evaluating durability of vaccine effects were developed in the context of both blinded and unblinded placebo crossover studies. The case-ascertained design is used to assess the transmission effects of vaccines. The novel ring vaccination trial design was first used in the Ebola outbreak in West Africa.

Infectious diseases pose a persistent challenge to public health worldwide. Recent global health crises, such as the COVID-19 pandemic and Ebola outbreaks, have underscored the vital role of infectious disease modeling in guiding public health policy and response. Infectious disease modeling is a critical tool for society, informing risk mitigation measures, prompting timely interventions, and aiding preparedness for healthcare delivery systems. This article synthesizes the current landscape of infectious disease modeling, emphasizing the integration of statistical methods in understanding and predicting the spread of infectious diseases. We begin by examining the historical context and the foundational models that have shaped the field, such as the SIR (susceptible, infectious, recovered) and SEIR (susceptible, exposed, infectious, recovered) models. Subsequently, we delve into the methodological innovations that have arisen, including stochastic modeling, network-based approaches, and the use of big data analytics. We also explore the integration of machine learning techniques in enhancing model accuracy and responsiveness. The review identifies the challenges of parameter estimation, model validation, and the incorporation of real-time data streams. Moreover, we discuss the ethical implications of modeling, such as privacy concerns and the communication of risk. The article concludes by discussing future directions for research, highlighting the need for data integration and interdisciplinary collaboration for advancing infectious disease modeling.

Estimating the mortality associated with a specific mortality crisis event (for example, a pandemic, natural disaster, or conflict) is clearly an important public health undertaking. In many situations, deaths may be directly or indirectly attributable to the mortality crisis event, and both contributions may be of interest. The totality of the mortality impact on the population (direct and indirect deaths) includes the knock-on effects of the event, such as a breakdown of the health care system, or increased mortality due to shortages of resources. Unfortunately, estimating the deaths directly attributable to the event is frequently problematic. Hence, the excess mortality, defined as the difference between the observed mortality and that which would have occurred in the absence of the crisis event, is an estimation target. If the region of interest contains a functioning vital registration system, so that the mortality is fully observed and reliable, then the only modeling required is to produce the expected deaths counts, but this is a nontrivial exercise. In low- and middle-income countries it is common for there to be incomplete (or nonexistent) mortality data, and one must then use additional data and/or modeling, including predicting mortality using auxiliary variables. We describe and review each of these aspects, give examples of excess mortality studies, and provide a case study on excess mortality across states of the United States during the COVID-19 pandemic.

Points of Significance is an ongoing series of short articles about statistics in Nature Methods that started in 2013. Its aim is to provide clear explanations of essential concepts in statistics for a nonspecialist audience. The articles favor heuristic explanations and make extensive use of simulated examples and graphical explanations, while maintaining mathematical rigor. Topics range from basic, but often misunderstood, such as uncertainty and p-values, to relatively advanced, but often neglected, such as the error-in-variables problem and the curse of dimensionality. More recent articles have focused on timely topics such as modeling of epidemics, machine learning, and neural networks. In this article, we discuss the evolution of topics and details behind some of the story arcs, our approach to crafting statistical explanations and narratives, and our use of figures and numerical simulations as props for building understanding.

Difficulties in reproducing results from scientific studies have lately been referred to as a reproducibility crisis. Scientific practice depends heavily on scientific training. What gets taught in the classroom is often practiced in labs, fieldwork, and data analysis. The importance of reproducibility in the classroom has gained momentum in statistics education in recent years. In this article, we review the existing literature on reproducibility education. We delve into the relationship between computing tools and reproducibility through visiting historical developments in this area. We share examples for teaching reproducibility and reproducible teaching while discussing the pedagogical opportunities created by these examples as well as challenges that the instructors should be aware of. We detail the use of teaching reproducibility and reproducible teaching practices in an introductory data science course. Lastly, we provide recommendations on reproducibility education for instructors, administrators, and other members of the scientific community.

Health policy evidence-building requires data sources such as health care claims, electronic health records, probability and nonprobability survey data, epidemiological surveillance databases, administrative data, and more, all of which have strengths and limitations for a given policy analysis. Data integration techniques leverage the relative strengths of input sources to obtain a blended source that is richer, more informative, and more fit for use than any single input component. This review notes the expansion of opportunities to use data integration for health policy analyses, reviews key methodological approaches to expand the number of variables in a data set or to increase the precision of estimates, and provides directions for future research. As data quality improvement motivates data integration, key data quality frameworks are provided to structure assessments of candidate input data sources.

Phonetics is the scientific field concerned with the study of how speech is produced, heard, and perceived. It abounds with data, such as acoustic speech recordings, neuroimaging data, and articulatory data. In this article, we provide an introduction to different areas of phonetics (acoustic phonetics, sociophonetics, speech perception, articulatory phonetics, speech inversion, sound change, and speech technology), an overview of the statistical methods for analyzing their data, and an introduction to the signal processing methods commonly applied to speech recordings. A major transition in the statistical modeling of phonetic data has been the shift from fixed effects to random effects regression models, the modeling of curve data (for instance, via generalized additive mixed models or functional data analysis methods), and the use of Bayesian methods. This shift has been driven in part by the increased focus on large speech corpora in phonetics, which has arisen from machine learning methods such as forced alignment. We conclude by identifying opportunities for future research.

Differential privacy is widely considered the formal privacy for privacy-preserving data analysis due to its robust and rigorous guarantees, with increasingly broad adoption in public services, academia, and industry. Although differential privacy originated in the cryptographic context, in this review we argue that, fundamentally, it can be considered a pure statistical concept. We leverage Blackwell's informativeness theorem and focus on demonstrating based on prior work that all definitions of differential privacy can be formally motivated from a hypothesis testing perspective, thereby showing that hypothesis testing is not merely convenient but also the right language for reasoning about differential privacy. This insight leads to the definition of f-differential privacy, which extends other differential privacy definitions through a representation theorem. We review techniques that render f-differential privacy a unified framework for analyzing privacy bounds in data analysis and machine learning. Applications of this differential privacy definition to private deep learning, private convex optimization, shuffled mechanisms, and US Census data are discussed to highlight the benefits of analyzing privacy bounds under this framework compared with existing alternatives.

In this article, we review the literature on statistical theories of neural networks from three perspectives: approximation, training dynamics, and generative models. In the first part, results on excess risks for neural networks are reviewed in the nonparametric framework of regression. These results rely on explicit constructions of neural networks, leading to fast convergence rates of excess risks. Nonetheless, their underlying analysis only applies to the global minimizer in the highly nonconvex landscape of deep neural networks. This motivates us to review the training dynamics of neural networks in the second part. Specifically, we review articles that attempt to answer the question of how a neural network trained via gradient-based methods finds a solution that can generalize well on unseen data. In particular, two well-known paradigms are reviewed: the neural tangent kernel and mean-field paradigms. Last, we review the most recent theoretical advancements in generative models, including generative adversarial networks, diffusion models, and in-context learning in large language models from two of the same perspectives, approximation and training dynamics.

In recent years, there has been a growing trend of applying reinforcement learning (RL) in financial applications. This approach has shown great potential for decision-making tasks in finance. In this review, we present a comprehensive study of the applications of RL in finance and conduct a series of meta-analyses to investigate the common themes in the literature, such as the factors that most significantly affect RL's performance compared with traditional methods. Moreover, we identify challenges, including explainability, Markov decision process modeling, and robustness, that hinder the broader utilization of RL in the financial industry and discuss recent advancements in overcoming these challenges. Finally, we propose future research directions, such as benchmarking, contextual RL, multi-agent RL, and model-based RL, to address these challenges and to further enhance the implementation of RL in finance.

The Hawkes process is a model for counting the number of arrivals to a system that exhibits the self-exciting property—that one arrival creates a heightened chance of further arrivals in the near future. The model and its generalizations have been applied in a plethora of disparate domains, though two particularly developed applications are in seismology and in finance. As the original model is elegantly simple, generalizations have been proposed that track marks for each arrival, are multivariate, have a spatial component, are driven by renewal processes, treat time as discrete, and so on. This article creates a cohesive review of the traditional Hawkes model and the modern generalizations, providing details on their construction and simulation algorithms, and giving key references to the appropriate literature for a detailed treatment.

One of the most important tasks in sports analytics is the development of binary response models for head-to-head game outcomes to estimate team and player strength. We discuss commonly used probability models for game outcomes, including the Bradley–Terry and Thurstone–Mosteller models, as well as extensions to ties as a third outcome and to the inclusion of a home-field advantage. We consider dynamic extensions to these models to account for the evolution of competitor strengths over time. Full likelihood-based analyses of these time-varying models can be simplified into rating systems, such as the Elo and Glicko rating systems. We present other modern rating systems, including popular methods for online gaming, and novel systems that have been implemented for online chess and Go. The discussion of the analytic methods are accompanied by examples of where these approaches have been implemented for various gaming organizations, as well as a detailed application to National Basketball Association game outcomes.

The emergence of functional magnetic resonance imaging (fMRI) marked a significant technological breakthrough in the real-time measurement of the functioning human brain in vivo. In part because of their 4D nature (three spatial dimensions and time), fMRI data have inspired a great deal of statistical development in the past couple of decades to address their unique spatiotemporal properties. This article provides an overview of the current landscape in functional brain measurement, with a particular focus on fMRI, highlighting key developments in the past decade. Furthermore, it looks ahead to the future, discussing unresolved research questions in the community and outlining potential research topics for the future.

Simulation-based methods for statistical inference have evolved dramatically over the past 50 years, keeping pace with technological advancements. The field is undergoing a new revolution as it embraces the representational capacity of neural networks, optimization libraries, and graphics processing units for learning complex mappings between data and inferential targets. The resulting tools are amortized, in the sense that, after an initial setup cost, they allow rapid inference through fast feed-forward operations. In this article we review recent progress in the context of point estimation, approximate Bayesian inference, summary-statistic construction, and likelihood approximation. We also cover software and include a simple illustration to showcase the wide array of tools available for amortized inference and the benefits they offer over Markov chain Monte Carlo methods. The article concludes with an overview of relevant topics and an outlook on future research directions.

Causal mediation analysis provides an attractive framework for integrating diverse types of exposure, genomic, and phenotype data. Recently, this field has seen a surge of interest, largely driven by the increasing need for causal mediation analyses in health and social sciences. This article aims to provide a review of recent developments in mediation analysis, encompassing mediation analysis of a single mediator and a large number of mediators, as well as mediation analysis with multiple exposures and mediators. Our review focuses on the recent advancements in statistical inference for causal mediation analysis, especially in the context of high-dimensional mediation analysis. We delve into the complexities of testing mediation effects, especially addressing the challenge of testing a large number of composite null hypotheses. Through extensive simulation studies, we compare the existing methods across a range of scenarios. We also include an analysis of data from the Normative Aging Study, which examines DNA methylation CpG sites as potential mediators of the effect of smoking status on lung function. We discuss the pros and cons of these methods and future research directions.

Beyond the main genetic and environmental effects, gene–environment (G–E) interactions have been demonstrated to significantly contribute to the development and progression of complex diseases. Published analyses of G–E interactions have primarily used a supervised framework to model both low-dimensional environmental factors and high-dimensional genetic factors in relation to disease outcomes. In this article, we aim to provide a selective review of methodological developments in G–E interaction analysis from a statistical perspective. The three main families of techniques are hypothesis testing, variable selection, and dimension reduction, which lead to three general frameworks: testing-based, estimation-based, and prediction-based. Linear- and nonlinear-effects analysis, fixed- and random-effects analysis, marginal and joint analysis, and Bayesian and frequentist analysis are reviewed to facilitate the conduct of interaction analysis in a wide range of situations with various assumptions and objectives. Statistical properties, computations, applications, and future directions are also discussed.

Instrumental variables (IVs) are widely used to study the causal effect of an exposure on an outcome in the presence of unmeasured confounding. IVs require an instrument, a variable that (a) is associated with the exposure, (b) has no direct effect on the outcome except through the exposure, and (c) is not related to unmeasured confounders. Unfortunately, finding variables that satisfy conditions b or c can be challenging in practice. This article reviews works where instruments may not satisfy conditions b or c, which we refer to as invalid instruments. We review identification and inference under different violations of b or c, specifically under linear models, nonlinear models, and heteroskedastic models. We conclude with an empirical comparison of various methods by reanalyzing the effect of body mass index on systolic blood pressure from the UK Biobank.

Adaptive experiments such as multi-armed bandits adapt the treatment-allocation policy and/or the decision to stop the experiment to the data observed so far. This has the potential to improve outcomes for study participants within the experiment, to improve the chance of identifying the best treatments after the experiment, and to avoid wasting data. As an experiment (and not just a continually optimizing system), it is still desirable to draw statistical inferences with frequentist guarantees. The concentration inequalities and union bounds that generally underlie adaptive experimentation algorithms can yield overly conservative inferences, but at the same time, the asymptotic normality we would usually appeal to in nonadaptive settings can be imperiled by adaptivity. In this article we aim to explain why, how, and when adaptivity is in fact an issue for inference and, when it is, to understand the various ways to fix it: reweighting to stabilize variances and recover asymptotic normality, using always-valid inference based on joint normality of an asymptotic limiting sequence, and characterizing and inverting the nonnormal distributions induced by adaptivity.

Functional data analysis (FDA) studies data that include infinite-dimensional functions or objects, generalizing traditional univariate or multivariate observations from each study unit. Among inferential approaches without parametric assumptions, empirical likelihood (EL) offers a principled method in that it extends the framework of parametric likelihood ratio–based inference via the nonparametric likelihood. There has been increasing use of EL in FDA due to its many favorable properties, including self-normalization and the data-driven shape of confidence regions. This article presents a review of EL approaches in FDA, starting with finite-dimensional features, then covering infinite-dimensional features. We contrast smooth and nonsmooth frameworks in FDA and show how EL has been incorporated into both of them. The article concludes with a discussion of some future research directions, including the possibility of applying EL to conformal inference.

In medical studies, time-to-event outcomes such as time to death or relapse of a disease are routinely recorded along with longitudinal data that are observed intermittently during the follow-up period. For various reasons, marginal approaches to model the event time, corresponding to separate approaches for survival data/longitudinal data, tend to induce bias and lose efficiency. Instead, a joint modeling approach that brings the two types of data together can reduce or eliminate the bias and yield a more efficient estimation procedure. A well-established avenue for joint modeling is the joint likelihood approach that often produces semiparametric efficient estimators for the finite-dimensional parameter vectors in both models. Through a transformation survival model with an unspecified baseline hazard function, this review introduces joint modeling that accommodates both baseline covariates and time-varying covariates. The focus is on the major challenges faced by joint modeling and how they can be overcome. A review of available software implementations and a brief discussion of future directions of the field are also included.