arXiv subjects

Rafael B. Stern

Publications and source records attributed to Rafael B. Stern.

15 recordsLinked to original sources

arXiv2025

Conformal Calibration of Statistical Confidence Sets

Constructing valid confidence sets is a crucial task in statistical inference, yet traditional methods often face challenges when dealing with complex models or limited observed sample sizes. These challenges are frequently encountered in modern applications, such as Likelihood-Free Inference (LFI). In these settings, confidence sets may fail to maintain a confidence level close to the nominal value. In this paper, we introduce two novel methods, TRUST and TRUST++, for calibrating confidence sets to achieve distribution-free conditional coverage. These methods rely entirely on simulated data from the statistical model to perform calibration. Leveraging insights from conformal prediction techniques adapted to the statistical inference context, our methods ensure both finite-sample local coverage and asymptotic conditional coverage as the number of simulations increases, even if n is small. They effectively handle nuisance parameters and provide computationally efficient uncertainty quantification for the estimated confidence sets. This allows users to assess whether additional simulations are necessary for robust inference. Through theoretical analysis and experiments on models with both tractable and intractable likelihoods, we demonstrate that our methods outperform existing approaches, particularly in small-sample regimes. This work bridges the gap between conformal prediction and statistical inference, offering practical tools for constructing valid confidence sets in complex models.

stat.ME↗

arXiv2025

Covariate balance under Bayesian decision theory

We study optimal sample allocation between treatment and control groups under Bayesian linear models. We derive an analytic expression for the Bayes risk, which depends jointly on sample size and covariate mean balance across groups. Under a flat conditional prior, the covariate mean balance term simplifies to the Mahalanobis distance. Our results reveal that the optimal allocation does not always correspond to equal sample sizes, and we provide sufficient conditions under which equal allocation is optimal. Finally, we extend the analysis to sequential settings with groups of patients arriving over time.

math.ST↗

arXiv2025

Nonparametric FBST for Validating Linear Models

The Full Bayesian Significance Test (FBST) possesses many desirable aspects, such as dismissing the need for hypotheses to have positive prior probability and providing a measure of evidence against $H_0$. Still, few attempts have been made to bring the FBST to nonparametric settings, with the main drawback being the need to obtain the highest posterior density (HPD) in a function space. In this work, we use a Gaussian processes prior to derive the FBST for hypotheses of the type $$ H_0: g(\boldsymbol{x}) = \boldsymbol{b}(\boldsymbol{x})\boldsymbolβ, \quad \forall \boldsymbol{x} \in \mathcal{X}, \quad \boldsymbolβ \in \mathbb{R}^k, $$ where $g(\cdot)$ is the regression function, $\boldsymbol{b}(\cdot)$ is a vector of linearly independent linear functions -- such as $\boldsymbol{b}(\boldsymbol{x}) = \boldsymbol{x}'$ -- and $\mathcal{X}$ is the covariates' domain. We also make use of pragmatic hypotheses to verify if the data might be compatible with a linear model when factors such as measurement errors or utility judgments are accounted for. This contribution extends the theory of the FBST, allowing its application in nonparametric settings and providing a procedure that easily tests if linear models are adequate for the data and that can automatically perform variable selection.

stat.ME↗

arXiv2025

REACT to NHST: Sensible conclusions to meaningful hypotheses

While Null Hypothesis Significance Testing (NHST) remains a widely used statistical tool, it suffers from several shortcomings in its common usage, such as conflating statistical and practical significance, the formulation of inappropriate null hypotheses, and the inability to distinguish between accepting the null hypothesis and failing to reject it. Recent efforts have focused on developing alternatives that address these issues. Despite these efforts, conventional NHST remains dominant in scientific research due to its procedural simplicity and mistakenly presumed ease of interpretation. Our work presents an intuitive alternative to conventional NHST designed to bridge the gap between the expectations of researchers and the actual outcomes of hypothesis tests: $\texttt{REACT}$. $\texttt{REACT}$ not only tackles shortcomings of conventional NHST but also offers additional advantages over existing alternatives. For instance, $\texttt{REACT}$ applies to multiparametric hypotheses and does not require stringent significance-level corrections when conducting multiple tests. We illustrate the practical utility of $\texttt{REACT}$ through real-world data examples.

stat.ME↗

arXiv2024

PROTEST: Nonparametric Testing of Hypotheses Enhanced by Experts' Utility Judgements

Instead of testing solely a precise hypothesis, it is often useful to enlarge it with alternatives that are deemed to differ from it negligibly. For instance, in a bioequivalence study one might consider the hypothesis that the concentration of an ingredient is exactly the same in two drugs. In such a context, it might be more relevant to test the enlarged hypothesis that the difference in concentration between the drugs is of no practical significance. While this concept is not alien to Bayesian statistics, applications remain confined to parametric settings and strategies on how to effectively harness experts' intuitions are often scarce or nonexistent. To resolve both issues, we introduce PROTEST, an accessible nonparametric testing framework that seamlessly integrates with Markov Chain Monte Carlo (MCMC) methods. We develop expanded versions of the model adherence, goodness-of-fit, quantile and two-sample tests. To demonstrate how PROTEST operates, we make use of examples, simulated studies - such as testing link functions in a binary regression setting, as well as a comparison between the performance of PROTEST and the PTtest (Holmes et al., 2015) - and an application with data on neuron spikes. Furthermore, we address the crucial issue of selecting the threshold - which controls how much a hypothesis is to be expanded - even when intuitions are limited or challenging to quantify.

stat.ME↗

arXiv2024

Regression Trees for Fast and Adaptive Prediction Intervals

Predictive models make mistakes. Hence, there is a need to quantify the uncertainty associated with their predictions. Conformal inference has emerged as a powerful tool to create statistically valid prediction regions around point predictions, but its naive application to regression problems yields non-adaptive regions. New conformal scores, often relying upon quantile regressors or conditional density estimators, aim to address this limitation. Although they are useful for creating prediction bands, these scores are detached from the original goal of quantifying the uncertainty around an arbitrary predictive model. This paper presents a new, model-agnostic family of methods to calibrate prediction intervals for regression problems with local coverage guarantees. Our approach is based on pursuing the coarsest partition of the feature space that approximates conditional coverage. We create this partition by training regression trees and Random Forests on conformity scores. Our proposal is versatile, as it applies to various conformity scores and prediction settings and demonstrates superior scalability and performance compared to established baselines in simulated and real-world datasets. We provide a Python package clover that implements our methods using the standard scikit-learn interface.

stat.ML↗

arXiv2023

Hierarchical clustering: visualization, feature importance and model selection

We propose methods for the analysis of hierarchical clustering that fully use the multi-resolution structure provided by a dendrogram. Specifically, we propose a loss for choosing between clustering methods, a feature importance score and a graphical tool for visualizing the segmentation of features in a dendrogram. Current approaches to these tasks lead to loss of information since they require the user to generate a single partition of the instances by cutting the dendrogram at a specified level. Our proposed methods, instead, use the full structure of the dendrogram. The key insight behind the proposed methods is to view a dendrogram as a phylogeny. This analogy permits the assignment of a feature value to each internal node of a tree through an evolutionary model. Real and simulated datasets provide evidence that our proposed framework has desirable outcomes and gives more insights than state-of-art approaches. We provide an R package that implements our methods.

stat.ME↗

arXiv2023

Flexible conditional density estimation for time series

This paper introduces FlexCodeTS, a new conditional density estimator for time series. FlexCodeTS is a flexible nonparametric conditional density estimator, which can be based on an arbitrary regression method. It is shown that FlexCodeTS inherits the rate of convergence of the chosen regression method. Hence, FlexCodeTS can adapt its convergence by employing the regression method that best fits the structure of data. From an empirical perspective, FlexCodeTS is compared to NNKCDE and GARCH in both simulated and real data. FlexCodeTS is shown to generally obtain the best performance among the selected methods according to either the CDE loss or the pinball loss.

stat.ME↗

arXiv2022

Logical coherence in Bayesian simultaneous three-way hypothesis tests

This paper studies whether Bayesian simultaneous three-way hypothesis tests can be logically coherent. Two types of results are obtained. First, under the standard error-wise constant loss, only for a limited set of models can a Bayes simultaneous test be logically coherent. Second, if more general loss functions are used, then it is possible to obtain Bayes simultaneous tests that are always logically coherent. An explicit example of such a loss function is provided.

math.ST↗

arXiv2021

CD-split and HPD-split: efficient conformal regions in high dimensions

Conformal methods create prediction bands that control average coverage assuming solely i.i.d. data. Although the literature has mostly focused on prediction intervals, more general regions can often better represent uncertainty. For instance, a bimodal target is better represented by the union of two intervals. Such prediction regions are obtained by CD-split , which combines the split method and a data-driven partition of the feature space which scales to high dimensions. CD-split however contains many tuning parameters, and their role is not clear. In this paper, we provide new insights on CD-split by exploring its theoretical properties. In particular, we show that CD-split converges asymptotically to the oracle highest predictive density set and satisfies local and asymptotic conditional validity. We also present simulations that show how to tune CD-split. Finally, we introduce HPD-split, a variation of CD-split that requires less tuning, and show that it shares the same theoretical guarantees as CD-split. In a wide variety of our simulations, CD-split and HPD-split have better conditional coverage and yield smaller prediction regions than other methods.

stat.ML↗

arXiv2019

Flexible distribution-free conditional predictive bands using density estimators

Conformal methods create prediction bands that control average coverage under no assumptions besides i.i.d. data. Besides average coverage, one might also desire to control conditional coverage, that is, coverage for every new testing point. However, without strong assumptions, conditional coverage is unachievable. Given this limitation, the literature has focused on methods with asymptotical conditional coverage. In order to obtain this property, these methods require strong conditions on the dependence between the target variable and the features. We introduce two conformal methods based on conditional density estimators that do not depend on this type of assumption to obtain asymptotic conditional coverage: Dist-split and CD-split. While Dist-split asymptotically obtains optimal intervals, which are easier to interpret than general regions, CD-split obtains optimal size regions, which are smaller than intervals. CD-split also obtains local coverage by creating a data-driven partition of the feature space that scales to high-dimensional settings and by generating prediction bands locally on the partition elements. In a wide variety of simulated scenarios, our methods have a better control of conditional coverage and have smaller length than previously proposed methods.

stat.ME↗

arXiv2019

Teaching decision theory proof strategies using a crowdsourcing problem

Teaching how to derive minimax decision rules can be challenging because of the lack of examples that are simple enough to be used in the classroom. Motivated by this challenge, we provide a new example that illustrates the use of standard techniques in the derivation of optimal decision rules under the Bayes and minimax approaches. We discuss how to predict the value of an unknown quantity, $θ\! \in \! \{0,1\}$, given the opinions of $n$ experts. An important example of such crowdsourcing problem occurs in modern cosmology, where $θ$ indicates whether a given galaxy is merging or not, and $Y_1, \ldots, Y_n$ are the opinions from $n$ astronomers regarding $θ$. We use the obtained prediction rules to discuss advantages and disadvantages of the Bayes and minimax approaches to decision theory. The material presented here is intended to be taught to first-year graduate students.

math.ST↗

arXiv2019

Logically-consistent hypothesis testing in the hexagon of oppositions

Although logical consistency is desirable in scientific research, standard statistical hypothesis tests are typically logically inconsistent. In order to address this issue, previous work introduced agnostic hypothesis tests and proved that they can be logically consistent while retaining statistical optimality properties. This paper characterizes the credal modalities in agnostic hypothesis tests and uses the hexagon of oppositions to explain the logical relations between these modalities. Geometric solids that are composed of hexagons of oppositions illustrate the conditions for these modalities to be logically consistent. Prisms composed of hexagons of oppositions show how the credal modalities obtained from two agnostic tests vary according to their threshold values. Nested hexagons of oppositions summarize logical relations between the credal modalities in these tests and prove new relations.

math.LO↗

arXiv2018

Pragmatic hypotheses in the evolution of science

This paper introduces pragmatic hypotheses and relates this concept to the spiral of scientific evolution. Previous works determined a characterization of logically consistent statistical hypothesis tests and showed that the modal operators obtained from this test can be represented in the hexagon of oppositions. However, despite the importance of precise hypothesis in science, they cannot be accepted by logically consistent tests. Here, we show that this dilemma can be overcome by the use of pragmatic versions of precise hypotheses. These pragmatic versions allow a level of imprecision in the hypothesis that is small relative to other experimental conditions. The introduction of pragmatic hypotheses allows the evolution of scientific theories based on statistical hypothesis testing to be interpreted using the narratological structure of hexagonal spirals, as defined by Pierre Gallais.

stat.OT↗

arXiv2014

Compensating for the loss of a chance

Civil liability for a lost chance applies to legal cases in which a tortious action changes the probabilities of the outcomes that can be obtained by the victim. A central point in the application of this type of liability is the valuation of damages. Despite the practical importance of the valuation of lost chances, the legal restrictions that guide it have rarely been discussed explicitly. In order to discuss these restrictions, we propose an abstract description of a lost chance case in which there are multiple possible outcomes and the victim can make a choice that affects these outcomes. Given this description, we propose six conceptual questions to guide the valuation of lost chances. We discuss alternative answers to these questions and present the formulas that derive from them. More specifically, we show that the main formulas that have been proposed for medical misdiagnosis cases are particular instances of the alternatives we discuss.

stat.AP↗