The New England Journal of Statistics in Data Science

Modern Statistics is made of the sensible combination of direct evidence (the data directly relevant or the “individual data”) and indirect evidence (the data and knowledge indirectly relevant or the “group data”). The admissible procedures combine the two sources of information, and technology advances make indirect evidence more substantial and ubiquitous. It has been pointed out, however, that an important problem of Statistics when “borrowing strength” is to treat in a fundamentally different way exceptional cases that do not adapt to the central “aurea mediocritas”. This is what has been coined as “the Clemente problem” in honor of R. Clemente, an exceptional batter [6]. In this article, we argue that the problem is caused by the simultaneous use of square loss function and conjugate (light-tailed) priors, which is the usual procedure. We propose in their place to use robust penalties, in the form of losses that penalize more severely huge errors, or (equivalently) priors of heavy tails which make the exceptional more probable. Using heavy-tailed priors, we can reproduce, in a Bayesian structured way, Efron and Morris’ “limited translated estimators” (with Double Exponential Priors) and “discarding priors estimators” (with Cauchy-like priors), which discard the prior in the presence of prior-likelihood conflict. We show that both Empirical Bayes and Full Bayes approaches can alleviate the Clemente problem and beat the James-Stein estimator in terms of smaller square errors, for sensible Robust Bayes priors. We model in parallel Empirical Bayes and Fully Bayesian hierarchical models, illustrating that the differences among sensible versions of both are relatively small, as compared with the effect due to the robust assumptions. We follow [16] in using a heavy-tailed (scaled) Beta2 distribution for (squared) scales that arise naturally as an alternative to the usual Inverted-Gamma distribution. The combination of a Cauchy Prior for location and Scaled Beta2 for square scales yields a novel closed-form prior for location, extremely suitable for Objective Robust Bayesian Analysis. Finally, we calculate the predictive intervals and the Robust models covers the Clemente average at $80\% $ of probability, which the conjugate do not. Our approach has connections with [25], which employs a completely different methodology. This is an instance of the convergence of the best frequentist and Bayesian analyses [see 3].

Natural language processing (NLP) algorithms have demonstrated significant capabilities in understanding responses to open-ended questions in survey data. However, the reliability and uncertainty of these methods on this task still need to be thoroughly investigated. To address this issue, this paper presents a comprehensive comparative analysis of various NLP methods for detecting fine-grained emotions in student responses about their mental health during the COVID-19 pandemic. The evaluated models include a Lexicon-based approach, the bag-of-words (BoW) model, Term Frequency-Inverse Document Frequency (TF-IDF), a pre-trained Bidirectional Encoder Representations from Transformers (BERT) model, MentalBERT, and OpenAI’s GPT-3.5. We carefully assess the efficacy of these models in accurately classifying emotions into predetermined categories using performance metrics such as accuracy and F1 score. Furthermore, model stability and distinguishing ability were quantified through repetitive cross-validation and the Area Under the Receiver Operating Characteristic Curve (AUC). The consistency of emotion detection across different models is also evaluated. The study highlights that the effectiveness of employing NLP methods for mental health analysis may vary depending on the emotions being analyzed, and their stability and uncertainty require thorough examination. Our work can provide valuable guidance for data scientists on applying NLP methods to survey data, particularly for understanding survey respondents’ emotions.

We consider the class of inverse probability weight (IPW) estimators, including the popular Horvitz–Thompson and Hájek estimators used routinely in survey sampling, causal inference and for Bayesian computation. We focus on the ‘weak paradoxes’ for these estimators due to two counterexamples by Basu (1988) and Wasserman (2004) and investigate the two natural Bayesian answers to this problem: one based on binning and smoothing: a ‘Bayesian sieve’ and the other based on a conjugate hierarchical model that allows borrowing information via exchangeability. We compare the mean squared errors for the two Bayesian estimators with the IPW estimators for Wasserman’s example via simulation studies on a broad range of parameter configurations. We also prove posterior consistency for the Bayes estimators under missing-completely-at-random assumption and show that it requires fewer assumptions on the inclusion probabilities. We also revisit the connection between the different problems where improved or adaptive IPW estimators will be useful, including survey sampling, evidence estimation strategies such as Conditional Monte Carlo, Riemannian sum, Trapezoidal rules and vertical likelihood, as well as average treatment effect estimation in causal inference.

Efficacy testing is a cornerstone of clinical trials, ensuring that medical interventions achieve their intended therapeutic effects. Over the decades, a wide range of statistical methodologies have been developed to address the complexities of clinical trial data, including parametric, nonparametric, Bayesian, and machine learning approaches. Parametric methods, such as t-tests, ANOVA, and LMMs, have traditionally been the foundation of efficacy testing due to their efficiency under well-defined assumptions. Nonparametric techniques, including the Friedman test, Brunner-Munzel test, and modern extensions like nparLD, have emerged as robust alternatives, particularly for skewed, ordinal, or non-normal data. Bayesian methodologies have enabled the incorporation of prior information and uncertainty quantification, while machine learning techniques, such as deep learning and reinforcement learning, are revolutionizing trial designs and outcome predictions. Despite these advancements, significant gaps remain, including challenges in handling high-dimensional data, missingness, and ensuring equitable efficacy testing across diverse populations. This review provides a comprehensive overview of these statistical methods, highlighting their applications, strengths, limitations, and future directions. By bridging traditional statistical frameworks with modern computational techniques, the field can continue to advance toward more reliable and personalized clinical trial methodologies.

Time-to-event (TTE) endpoints are widely used in drug development and biomedical research. Traditional statistical models, for example the Cox regression model, have been used to predict TTE outcomes. Recent studies have also employed flexible machine learning (ML) methods, for example, tree models, to obtain superior prediction performance. In addition, post-baseline time-varying predictors have recently been reported to improve prediction using ML methods. In this study, we applied the Cox model and ML methods to predict the onset of TTE with both baseline and post-baseline predictors. We evaluated the predictive performance of these models using various metrics, including the time-dependent area under the receiver operating characteristic curve (AUC), the concordance index (C-index), and integrated Brier scores. We also used these metrics as criteria to guide the selection of predictors in the predictive models. Our findings indicate that the Cox model remains a robust choice, often comparable to ML methods in moderate sample sizes, provided the proportional hazards assumption holds. However, tree-based methods demonstrate superior performance in capturing complex, nonlinear interactions, albeit requiring larger sample sizes to stabilize predictions.

The timing of longitudinal measurements may depend upon outcome or disease severity. In biomedical studies relying on clinical encounter data, patients often have dense, irregular collections of visit data when suffering a worse health condition. In parallel, the longitudinal measurements may be impacted by the period of irregular visiting. Ignoring the impact of the outcome-dependent visiting process when constructing a longitudinal disease progression model can produce biased results. We propose a Bayesian joint model linking a mixed-effects model for the longitudinal marker and Weibull proportional hazards model with a log frailty for the visiting process, adjusting both longitudinal marker and event processes with covariates. We examine different random effect structures and performance characterizing disease trajectory. Motivated by clinical data on cystic fibrosis lung disease, we estimate the longitudinal process for lung function decline. Individuals with lower lung function tend to have more frequent clinical visits than those with higher lung function. Simulation studies suggest that incorporating a time-dependent Gaussian process is more important for model fit than adding the survival model via joint modeling; the random intercepts model exhibits maximum bias, especially when there is an outcome-dependent visiting process.

Social networks primarily focus on the phenomenon of the contagion effect when examining behavior patterns within specific social groups. However, the impact of peer effects is characterized by the tendency to imitate the behaviors of friends and the selection process, where individuals tend to affiliate with others sharing similar traits, significantly contributing to shaping social behaviors that are frequently interconnecting. This article presents a Bayesian approach that uses latent-space estimation methods to detect and examine contagion effects, considering the impact of social selection. The research provides a methodological explanation, followed by a sequence of simulation trials designed to explore operational functionalities and possible real-world applications. To illustrate the potential correlation between changes in alcohol use and the influence of social networks, this study concludes by presenting an example of adolescent drinking behavior.

This study explores the application of the t-link model in high-dimensional variable selection for binary regression. The t-link model provides flexibility in binary modeling and offers robust inference in the presence of outliers, making it a preferable alternative to the commonly used probit and logit links. To address the computational challenges posed by a large number of covariates, the skinny Gibbs algorithm is employed, and the consistency of variable selection under this approximate algorithm is established. These advancements in both computational and theoretical perspectives enhance the practicality and ease of implementing the t-link model. The performance of the t-link model, with a specified degrees of freedom, is compared to logit link and the probit link through simulation studies and an application to PCR data. The results demonstrate the robustness and computational efficiency of the proposed method.

In extreme value analysis, the impact of rounding in data, a form of quantization, on statistical inferences beyond point estimation has not been comprehensively studied. This paper addresses these challenges by considering rounded data as interval-censored. The maximum likelihood estimators of the model parameters tailored to account for interval censoring are asymptotically unbiased and efficient. Further, we adapt classic goodness-of-fit tests, such as the Anderson-Darling test, for rounded data based on the maximum likelihood estimator. The resulting tests have appropriate sizes and considerable power. One application of such tests is threshold selection for the peak over threshold approach in extreme value analysis. The efficacy of our estimation approach and the goodness-of-fit tests are demonstrated through a simulation study involving data rounded from generalized Pareto distributions. Applying this method to precipitation data from 18 stations in eastern Washington, an area with typically low precipitation and expecting a significant rounding effect, we observe narrower interval estimates of return levels.

Clinical trial design for rare diseases can be challenging due to limited data, heterogeneous clinical manifestations and progression, and a frequent lack of adequate knowledge about the disease. Multiple endpoints are usually used to collectively assess the effectiveness of the investigational drug on multiple aspects of the disease. Here we propose an adaptive design based on the promising zone framework, allowing for sample size re-estimation (SSR) using interim data for a clinical trial involving multiple endpoints. The proposed SSR procedure incorporates two global tests: the ordinary least squares (OLS) test and the nonparametric permutation test. We consider two SSR approaches: one is based on power (SSR-Power) and the other on conditional power (SSR-CP). Simulation results show that the adaptive design achieves type I error control and satisfactory power. Compared with the permutation test, the OLS test has improved type I error control when the sample size is small and the timing of the interim analysis is early; while the permutation test achieves slightly higher power in most scenarios. Regarding the SSR methods, SSR-CP consistently achieves higher power than SSR-Power but often requires a larger sample size and more frequently reaches the maximum allowable sample size. The proposed design is particularly useful when the trial has a small initial sample size and has opportunity to adjust the sample size at an interim analysis to achieve adequate power.