2025 Roeling Conference titles and abstracts
2025 Lloyd Roeling UL Lafayette Statistics Conference
and Fall 2025 Louisiana ASA Chapter Meeting
Friday and Saturday 24-25 October 2025
Maxim Doucet Hall room 211
University of Louisiana at Lafayette
Lafayette, Louisiana
Titles and Abstracts
(last updated 12 October 2025)
Fiducial Generative Models
Jan Hannig
Kenan Distinguished Professor and Chair
Department of Statistics & Operation Research
The University of North Carolina at Chapel Hill
jan.hannig@unc.edu
Jan Hannig's web page
While generalized fiducial inference (GFI) and its variants have yielded many theoretical and practical results to parametric inference and uncertainty quantification, applying it to generative models remains challenging. We identify three key issues misspecification, metric choices, and over-parameterization hinder the direct application of the GFI to generative models. In this paper, we propose a novel method based on the framework of generalized fiducial inference, designed to construct distributional estimates over the parameter space given observed data, while also enabling uncertainty quantification for generative models. We employ a truncation-based approach and further provide a theoretical analysis of its behavior under varying truncation parameters. Both theoretical results and empirical evidence suggest that, with an appropriately chosen truncation parameter, the truncated distribution derived from generalized fiducial inference achieves valid coverage of the true parameter and leads to improved generalization performance.
Joint Work with Zijie Tian, T. C. M. Lee (UC Davis)
Estimating Multiple Missing Observations in Factorial Experiments
Kumer P. Das
Associate Vice President for Research and Innovation
University of Louisiana at Lafayette
kumer.das@louisiana.edu
Estimating multiple missing observations in factorial experiments is crucial for preserving the integrity of the analysis, ensuring an accurate interpretation of interactions and effects, and maintaining the validity of experimental conclusions. This study introduces a novel methodology for estimating multiple missing observations in factorial experiment data, specifically addressing three distinct missingness scenarios. Through analytic solutions derived from minimizing the squared error loss (L2 norm), the proposed approach ensures robust and accurate estimation of missing observations. The performance of the method was evaluated through simulations with varying numbers of replications, where the results demonstrated consistent reductions in bias, variance, mean absolute error (MAE), and mean square error (MSE) as m increased. Furthermore, comparisons of missing value scenarios highlighted the influence of specific missingness patterns on estimator performance. The findings underscore the efficacy and adaptability of the proposed methodology, providing a foundation for future extensions to replicated and higher-dimensional factorial designs.
Joint work with Aaron C. Marshall (Virgina Tech.) and Shaha A. Patwary (Butler Univ.)
Newer Innovations in Randomized Response Technique (RRT) Research
Sat Gupta
Professor Emeritus (Statistics)
Mathematics and Statistics
UNC Greensboro
Greensboro, NC
sngupta@uncg.edu
In this talk, we will discuss various new RRT innovations that have taken place in the last 20 years such as optionality, mixture models, trust enhancement, unified measure of model quality, measurement errors in the binary domain, and adverse effect of the use of auxiliary information in the RRT domain.
Optimal Prediction and Tolerance Intervals for the Ratio of Dependent Normal Random Variables
Gboyega David Adepoju
Assistant Professor of Statistics
Department of Mathematics
Nova Southeastern University
Fort Lauderdale, Florida
gadepoju@nova.edu
A simple exact method is proposed for computing prediction intervals and tolerance intervals for the distribution of the ratio X1/X2 when (X1,X2) follows a bivariate normal distribution. The methodology uses the factors available for computing one-sample prediction intervals and tolerance intervals for a univariate normal distribution. Both one-sided and two-sided intervals are constructed, and the two-sided tolerance intervals are obtained with and without imposing the equal-tail requirement. The results are illustrated using a practical application that calls for the computation of prediction intervals and tolerance intervals for the distribution of the ratio X1/X2. The application is on the cost-effectiveness of a new drug compared to a standard drug.
Joint work with K. Krishnamoorthy (University of Louisiana at Lafayette).
A simulation approach to the statistical analysis of interval-valued data
William H. Woodall
Professor Emeritus
Virginia Tech
bwoodall@vt.edu
In this presentation I will discuss the statistical analysis of interval-valued data. Many papers have been written on the neutrosophic approach, which will be shown to be nonsensical. A proposed simulation-based approach provides useful information without the need of complicated interval statistics or interval analysis. Examples will be given involving regression, designed experimentation, and statistical process monitoring.
References:
Haq, A. and Woodall, W. H. (2025), A Critique of Neutrosophic Statistical Analysis Illustrated with Interval Data from Designed Experiments, Journal of Quality Technology 57(3), 257-264.
Steiner, S. H. and Woodall, W. H. (2025), Control Charting with Interval-Valued Data, resubmitted to Quality Engineering
Woodall, W. H., Driscoll, A. R., and Montgomery, D. C. (2022), A Review and Perspective on Neutrosophic Statistical Process Monitoring Methods, IEEE Access 10, 100456 – 100462.
Woodall, W. H., King, C., Driscoll, A. R., and Montgomery, D. C. (2025), A Critical Assessment of Neutrosophic Statistical Methods, to appear in Quality Engineering.
An energy statistic-based test for exponentiality
Nurudeen Ajadi
Department of Mathematics
UL Lafayette
nurudeen.ajadi1@louisiana.edu
We propose a novel test for exponentiality in the context where the parameter is estimated using energy statistics. The test is affine invariant and consistent against all fixed alternatives. Monte Carlo simulations demonstrate that the proposed test effectively controls the Type I error rate. Furthermore, power comparisons show that our method performs competitively with existing approaches. The utility of the test is further illustrated through applications to two real-world datasets.
Simple Closed-Form Confidence Intervals for Stress-Strength Reliability in Normal Distributions and Comparisons
Ibrahim Adenekan
Department of Mathematics
UL Lafayette
ibrahim.adenekan1@louisiana.edu
In this talk, we address the problem of interval estimation of the reliability parameter R = P(X1 > X2), where X1 and X2 are independent normal random variables with unknown means and variances. We propose closed-form confidence intervals for R using the fiducial approach and the parametric bootstrap approach. The new confidence intervals and other available closed-form confidence intervals are compared for accuracy and precision. The study also extends to estimation of P(X1 - X2 > t) where t is a specified number. Based on the comparison study, some recommendations are made for applications. An example with practical data is provided to illustrate the methods.
BayBiMR: a Bayesian bi-directional Mendelian Randomization method accounting for correlated and uncorrelated pleiotropy
Siyi Chen
Department of Biostatistics and Data Science
Louisiana State University Health Science Center
New Orleans, Louisiana
sche11@lsuhsc.edu
Mendelian randomization (MR) uses genetic variants as instruments to infer causal relationships between traits. Standard approaches are sensitive to horizontal pleiotropy and often assume the pleiotropic component is independent of instrument strength, especially in the case when bi-directional causal relationships are present. We propose BayBiMR, a Bayesian bi-directional MR model that jointly estimates causal effects for both directions while allowing both (i) horizontal pleiotropy and (ii) latent correlated pleiotropy via correlation between direct genetic effects and pleiotropic effects. This model employs spike-and-slab shrinkage with flexible priors on inclusion rates, effect variances, and correlation parameters. Posterior inference is performed using a blocked Gibbs sampler with Metropolis-Hastings updates for correlation coefficients. Simulations demonstrate improved performance in the presence of correlated pleiotropy relative to existing methods. We also provide real-data analyses that illustrate the method on GWAS summary statistics.
Fiducial Inference in Inverse Sampling
Bao Anh Maddux
Department of Mathematics
Winston-Salem State University
Winston-Salem, NC
Inverse sampling (negative binomial sampling) arises when sampling continues until a specified number of successes is observed—common in epidemiological studies of rare diseases, quality control, and low-response surveys. Despite practical importance, inference methods for inverse sampling remain limited. This talk presents fiducial-based approaches for three problems: (1) confidence intervals for proportions and their functions (differences, ratios, odds ratios), (2) prediction intervals for future sample sizes, and (3) tolerance intervals. We develop simple, closed-form approximate intervals that are easy to implement. Extensive numerical studies show that exact methods are overly conservative while large-sample approximations are often liberal. Our fiducial and score-based intervals achieve coverage probabilities near nominal levels with superior precision. The methodology is illustrated through clinical examples including maternal congenital heart disease studies and the American Community Survey, demonstrating practical utility in real applications.
Confidence Intervals for Normal Quantiles: One- and Two-Sample
Justin Dunnam
Department of Mathematics
University of Louisiana at Lafayette
We consider the problem of constructing confidence intervals (CIs) for quantiles of normal distributions. We first show that the CI based on the UMVUE (Chakraborti and Li, 2007, The American Statistician) and the classical one based on the noncentral t distribution are the same. We also provide a simple closed-form CI on the basis of a normal approximation to the noncentral t distribution. This approximate CI is comparable with the classical one, and enables us to find a fiducial distribution for a normal quantile. This fiducial approach for the one-sample problem can be easily extended to two-sample problems of constructing CIs for the difference/ratio of quantiles. The proposed CIs are evaluated in terms of coverage probability and precision, and their performance is compared with existing CIs. An application to wood industries is provided to illustrate the methods.
Limitations of Transformations for Count Data: Evidence from Simulation and Implications for Health Research
Achraf Cherkaoui
Biostatistics and Data Science Program
LSU Heath Science Center
New Orleans, Louisiana
Count data are among the most fundamental outcomes in the health sciences, yet they violate the assumptions of commonly applied parametric methods, such as linear regression, due to their non-normal distribution. A longstanding practice, particularly among non-statisticians, has been to transform count variables to approximate normality; however, this approach has notable limitations, including difficulties in handling zero observations and an elevated risk of false-negative results. In this study, we conduct a simulation experiment using data generated from Poisson distribution to compare the performance of linear regression applied to transformed counts (log, square root) with generalized linear models (GLMs) fitted directly to untransformed data. The results demonstrate that transformation-based methods perform poorly except under limited conditions, whereas GLM approaches consistently yield accurate and unbiased estimates. These findings underscore the superiority of GLMs and related frameworks for analyzing count data. Importantly, in health research contexts, reliance on transformation-based methods can obscure meaningful associations and lead to misleading inferences, particularly in epidemiological and clinical studies where policy and intervention decisions may follow. We recommend that researchers move away from transformation-based approaches and instead adopt Poisson- and negative binomial–based modeling strategies to ensure more reliable, valid, and interpretable analyses of count outcomes.
Joint work with Evrim Oral (LSUHSC).
Robust Alternatives to Conventional AR(q) Models: Simulation Results and Empirical Evidence from COVID-19 Data
Mohamed Mohamed
Biostatistics and Data Science Program
LSU Heath Science Center
New Orleans, Louisiana
In time series analysis, the assumption of normally distributed innovations is often imposed, yet it may be overly restrictive in empirical applications. This study reassesses the applicability of the robust estimation method developed by Tiku, Wong, and Bian for autoregressive AR(q) models, which is designed to provide reliable estimation under deviations from normality. Through extensive simulation experiments, we demonstrate that conventional AR(q) models can yield biased and inefficient estimates when innovations arise from skewed distributions, whereas the robust method achieves substantial improvements in estimation accuracy. The methodological utility is further illustrated through an application to daily COVID-19 case data compiled from Louisiana state and local health agencies during the pandemic.
Joint work with Evrim Oral (LSUHSC).
Latent Structure of Mixed Neuropathology: A Cross-Cohort Factor Analytic Brain Pathology Score
Shubhabrata Mukherjee
Research Associate Professor
Department of Medicine
University of Washington
Seatle, Washington
smukherj@uw.edu
Many individuals with clinically diagnosed Alzheimer’s disease (AD) dementia are found to have multiple comorbid neuropathologies at autopsy. To better capture this complexity, we developed and harmonized a composite Brain Pathology Score (BPS) across four large autopsy cohorts. Using confirmatory factor analysis, we derived a continuous latent score from standard neuropathological variables, treating overlapping items across cohorts as anchors. A bifactor model with residual correlations between Thal phase and CERAD score, and between hippocampal sclerosis and LATE staging, provided the best fit. We evaluated the BPS against existing gold-standard summary measures—the AD Neuropathologic Change (ADNC) score and the Global Pathology (GPATH) score. Across cohorts, the harmonized BPS outperformed ADNC and GPATH in its associations with the last known clinical diagnosis of dementia and AD, and showed stronger associations with cognitive scores (memory, executive functioning, and language) at their final visit. This work demonstrates a psychometrically rigorous approach to constructing a harmonized composite measure of mixed neuropathology. The BPS places autopsy cohorts on a common scale, enabling more robust cross-cohort comparisons and future meta-analyses.
Dependency Tests for Mixed Graphical Models using Chatterjee's Correlation Coefficient Penalized by L1 Norm
Christian Hacker
Louisiana Tech University
Chatterjee's rank-based correlation coefficient provides a powerful tool for measuring the strength of nonlinear relationships between random variables and for testing whether two variables are independent. An important feature of this coefficient is that it is not necessarily symmetric when the input variables are interchanged. We exploit these properties to construct Partial Ancestral Graphs (PAGs) under the assumption of latent variables, which leads to the weaker notion of d-separation known as m-separation. To discover the underlying causal graph, we first define a pair of Markov Random Field (MRF) models to estimate Chatterjee's coefficient for every ordered pair of variables. We then impose an L1-norm penalty and employ the Resilient Backpropagation (Rprop) algorithm to optimize the model parameters. This enables a score-based approach to construct a PAG, in contrast to more popular constraint-based methods such as the Fast Causal Inference (FCI) algorithm.
Joint work with Xiyuan Liu (Louisiana Tech University).
Generalized Exponentially-Modified Gaussian (EMG) Regression and Parametric Conditional Quantile Curve Estimation
Derek S. Young
Dr. Bing and Mrs. Rachel Zhang Professor of Statistics
Dr. Bing Zhang Department of Statistics
University of Kentucky
derek.young@uky.edu
Derek Young’s web page
The exponentially-modified Gaussian (EMG) distribution arises as the sum of independent Gaussian and exponential random variables. It is, thus, characterized by three parameters: the Gaussian component’s mean and variance parameters and the exponential component’s rate parameter. EMG-based models have been shown to be effective in problems spanning chromatography, neurology, and stochastic frontier analysis. This paper introduces a generalized EMG regression model that allows each parameter in the univariate EMG distribution to (possibly) be modeled as a function of covariates using link functions as in a generalized linear models framework. We further introduce a way to estimate conditional quantile curves using a bootstrap. The performance of the estimating algorithm is assessed with a numerical study. The methods discussed are then applied to data on reaction times.
Joint work with Yanxi Li (MSU Denver).
Bayesian Analyses and Design of Aggregated Group Sequential N-of-1 Clinical Trials
Md Abdullah Al-Mamun
Biostatistics and Data Science Program
LSU Heath Science Center
New Orleans, Louisiana
N-of-1 trials offer a personalized approach to clinical research, allowing for the evaluation of individualized treatments through repeated crossover designs. While aggregating multiple N-of-1 trials increases statistical power, existing methods often fail to account for treatment heterogeneity across individuals. Current Bayesian approaches primarily focus on hierarchical models, which assume a common distribution of treatment effects and may overlook distinct patient subgroups. To address heterogeneity among patient subgroups, we propose a Bayesian mixed modeling approach in N-of-1 trials that identifies subgroups of patients with similar treatment responses while allowing for individual variation. Our clustering approach dynamically groups patients based on treatment effects, while the mixed approach integrates hierarchical and clustering structures to enhance flexibility. We implement adaptive Markov Chain Monte Carlo methods, including Metropolis-Hastings and Gibbs sampling, for efficient posterior inference. To validate our methods, we propose to conduct extensive simulation studies under varying treatment effect scenarios. The hypothesis is that compared to the hierarchical method, the clustering and mixed approaches can improve the estimation of treatment effects by accurately detecting patient subgroups with similar responses. Adjusting grouping thresholds affects clustering accuracy, with the mixed approach consistently achieving the best balance between reducing bias and identifying subgroups. By addressing limitations in existing Bayesian N-of-1 trial models, this research can advance statistical methodologies for personalized treatment evaluation and provide a scalable framework for precision medicine.
Joint work with Evrim Oral (LSUHSC).
Tests and Confidence Intervals for the Mean of a Zero-Inflated Poisson Distribution
Meesook Lee
Department of Mathematical Sciences
McNeese State University
Lake Charles, Louisiana
The zero-inflated Poisson (ZIP) model is often postulated for count data that include excessive zeros. This ZIP distribution can be regarded as the mixture of two distributions, one that degenerate at zero and another that is Poisson. Unlike the Poisson mean, the mean of the ZIP distribution is product of the mixture parameter and the Poisson parameter, and it is not simple to make inference on the ZIP mean. In this talk, the problem of making inference on the mean of a ZIP distribution is addressed. Confidence intervals based on the likelihood approach, and bootstrap approach are provided. A signed likelihood ratio test for one-sided hypotheses is also developed. Proposed methods are evaluated for their properties by Monte Carlo simulation. Finally, the methods are illustrated using two examples.
Joint work with Dustin Waguespack and K. Krishnamoorthy (UL Lafayette).
Dimension Reduction for Spatially correlated data
Hossein Moradi Rekabdarkolaee
Bowling Green State University
Dimension reduction provides a useful tool for statistical data analysis with high-dimensional data. In this study, we develop a new dimension reduction method for multivariate regression with spatially correlated data. The efficacy of this new solution is illustrated through simulation studies and a real data analysis.
Addressing Unknown Heterogeneity in Microbiome Sequencing Studies
Ni Zhao
Department of Biostatistics
Johns Hopkins University
Microbiome data, akin to other high-throughput data, suffer from technical heterogeneity stemming from differential experimental designs and processing. In addition to measured artifacts such as batch effects, there are heterogeneity due to unknown or unmeasured factors, which lead to spurious conclusions if unaccounted for. With the advent of large-scale multi-center microbiome studies and the increasing availability of public datasets, the issue become more pronounced. Current approaches for addressing unmeasured heterogeneity were primarily developed for microarray and/or RNA sequencing data. They cannot accommodate the unique characteristics of microbiome data such as sparsity and over-dispersion. Here, we introduce Quantile Thresholding (QuanT), a novel non-parametric approach for identifying unmeasured heterogeneity tailored to microbiome data. QuanT applies quantile regression across multiple quantile levels to threshold the microbiome abundance data and uncovers latent heterogeneity using the thresholded binary residual matrices. We validated QuanT using both synthetic and real microbiome datasets, demonstrating its superiority in capturing and mitigating heterogeneity, improving the accuracy of downstream analyses, such as prediction analysis, differential abundance tests, and community-level diversity evaluations.
Two-time scale maximum likelihood estimate using different tail perturbation
Dao Nguyen
Mathematics Department
The University of Mississippi
In this paper, we generalize Stein’s identity for the normal distribution to the p-generalized Gaussian distribution, enabling more flexible perturbations that account for diverse tail behaviors. Leveraging these gradient approximations, we develop a novel two-time-scale maximum likelihood estimation method based on stochastic approximation. This algorithm is further incorporated into the iterated filtering framework, thereby relaxing the conventional assumption of bounded variance in two-time-scale stochastic approximations. Finally, we demonstrate the effectiveness of the proposed approach in fitting both linear and nonlinear complex scientific models.