Title: | Learning Covariance Structures of Multidimensional Linear Mixed Models |
Speaker: | Professor Yuedong Wang, Department of Statistics and Applied Probability, University of California, Santa Barbara, USA |
Time/Place: | 14:00 - 15:00 FSC1217, Fong Shu Chuen Library, HSH Campus, Hong Kong Baptist University |
Abstract: | We propose a computationally efficient framework for the estimation of multidimensional response mixed-effects models. The key idea is to formulate a moment error function to measure the error of covariance estimates. Minimizing this loss entails solving a nonlinear semidefinite programming problem, which can be solved efficiently by a proximal gradient algorithm. The proposed method is much faster than some of the existing likelihood-based methods while maintaining comparable accuracy across a range of simulations. We provide non-asymptotic analysis to study various concentration properties of the covariance estimators, which have not previously been investigated in the studies of mixed-effects models. |
Title: | Multi-state Model and Structural Selection for the Analysis of Depressive Symptom Dynamics in Middle-aged and Older Adults |
Speaker: | Dr. Chuoxin Ma, Department of Statistics and Data Science, BNU-HKBU United International College, China |
Time/Place: | 15:00 - 16:00 FSC1217, Fong Shu Chuen Library, HSH Campus, Hong Kong Baptist University |
Abstract: | Depressive symptoms are increasingly common in middle-aged and older adults. People may experience transitions across different underlying states due to symptom severity fluctuation over a course of many years. Characterising the dynamics of depression and identifying key factors associated with different depressive states is important for the development of effective interventions. We proposed a multistate modeling framework and developed model structure selection procedure to identify covariates with time-varying coefficients, time-independent coefficients, and null effects. The mental health status of Chinese residents aged 45 and older were analysed based on the China Health and Retirement Longitudinal Study (CHARLS). |
Title: | On Application of Semiparametric Methods |
Speaker: | Dr. Zhijian Li, Department of Statistics and Data Science, BNU-HKBU United International College, China |
Time/Place: | 16:00 - 17:00 FSC1217, Fong Shu Chuen Library, HSH Campus, Hong Kong Baptist University |
Abstract: | Generally, in the semiparametric model, we are only interested in the finite-dimensional parameters while taking the infinite-dimensional nonparametric components as nuisance parameters. The semiparametric theory provides a novel insight and deeper understanding of nonparametric models and gives rise to efficient estimators of the finite-dimensional parameters in these models. In this talk, we introduce the semiparametric theory and its application in nonparametric models and partially linear models. |
Title: | PhD Oral Defense: Meta-analysis with Rare Events or Non-normal Effects |
Speaker: | Mr WEI Jiajin, Department of Mathematics, Hong Kong Baptist University, Hong Kong |
Time/Place: | 10:00 - 12:00 FSC1217, or Zoom (Meeting ID: 994 4134 3729 Password: 635797) |
Abstract: | Meta-analysis is a statistical method to synthesize evidence from multiple studies that address the same scientific question. The common-effect model and the random-effects model are two commonly used models for meta-analysis to combine the observed effect sizes from each study. In clinical studies, experimental data commonly involve binary outcomes, where the effect sizes, e.g., the odds ratio and the relative risk, are reported to quantify the treatment effect between two groups. When the rare events, in particular the zero event, occur in one or both groups, there will be calculation problems in the estimation of the effect sizes and the subsequent meta-analysis. Moreover, the inverse binomial proportion, another important statistical parameter related to the binomial distribution, may also suffer from the zero-event problem in its interval estimation. The first part of the thesis focuses on the meta-analysis of relative risk and the interval estimation of the inverse binomial proportion, with a concentration on solving the zero-event problem. In addition, in the random-effects model, conventional meta-analysis assumes the random effects or the true effect sizes of each study are independent and identically distributed random variables from a normal distribution for ease of statistical analyses. In practice, however, the distribution of random effects may be non-normal, skewed or heavy-tailed. Motivated by this, the second part of the thesis is proposed to loosen the assumption of normal random effects through semiparametric approaches. In Chapter 2, we revisit a recent meta-analysis of coronavirus disease 2019 data with the relative risk being the effect size. We conduct a comparative study involving four continuity correction methods and another method without the continuity correction, namely the generalized linear mixed models, all of which can overcome the zero-event problem. To further advance the literature, we also introduce a new method of the continuity correction for estimating the relative risk. Simulation studies are carried out to compare the performance of the methods when there are few studies. In addition, we also illustrate the usefulness of those methods by reanalyzing the coronavirus disease 2019 (COVID-19) data, which investigates the effectiveness of physical distancing on the infection risk. In Chapter 3, we introduce another important statistical parameter from the binomial distribution, namely the inverse binomial proportion, which has wide applications in sampling techniques and clinical trials. In the last two decades, some point estimators have been proposed in the existing literature. However, to the best of our knowledge, there is little work on the interval estimation of the inverse binomial proportion. In this chapter, we study its interval estimation from two perspectives. One is to construct the confidence intervals by inverting those for the binomial proportion, and the other is to directly derive the confidence intervals by the Wald, score, variance stabilizing transformation, and Bayesian methods. We further study the statistical properties of the proposed confidence intervals and compare their finite sample performance through numerical studies. Lastly, we also apply the confidence intervals to a meta-analysis on the prevalence of heart failure among COVID-19 patients with mortality. In Chapter 4, we propose a new framework for the semiparametric estimation of the random-effects model in meta-analysis. In contrast to the conventional model, our method does not assume a normal distribution for the random effects. The parameters of the semiparametric model are made up of the finite-dimensional parametric components of interest and the infinite-dimensional nonparametric component corresponding to the unknown distribution for the random effects. The new semiparametric estimator is derived by solving an estimating equation composed of efficient score functions. The consistency, efficiency and asymptotic normality of our new estimator are further derived. Finally, we conduct simulation studies to evaluate the finite sample performance of our new estimator, showing that the performance of the new estimator is at least comparable to or more robust than the existing estimators in most settings. |
Title: | Simulation-based Sensitivity Analysis for Models with Non-ignorable Missing Data |
Speaker: | Professor Jian Qing SHI, Department of Statistics and Data Science, Center for Biostatistics, Southern University of Science and Technology, Shenzhen, China |
Time/Place: | 14:00 - 15:00 FSC1217, Fong Shu Chuen Library, HSH Campus, Hong Kong Baptist University |
Abstract: | Sensitivity analysis is widely used to address missing data problems, especially in cases with non-ignorable missing data. In these situations, traditional full-likelihood method may not be suitable. Sensitivity analysis assesses how the conclusions of a study may vary on assumptions or parameters related to missing data mechanism (MDM). In this talk, a new approach will be discussed, focusing on investigating different assumptions about MDM by comparing simulated datasets generated from various models of non-ignorable missing data with the observed data. This method is versatile and can be applied to a range of scenarios. Examples from meta-analysis and causal inference will be presented to demonstrate the effectiveness of this approach. |
Title: | On the Backward Error Incurred by the Compact Rational Krylov Linearization |
Speaker: | Dr. Hongjia Chen, Nanchang University, China |
Time/Place: | 10:00 - 11:00 FSC1217, Fong Shu Chuen Library, HSH Campus, Hong Kong Baptist University |
Title: | Bayesian Methods for Random-Effects Meta-Analysis of Rare Binary Events in Biomedical Research |
Speaker: | Professor Xinlei Sherry Wang, Department of Mathematics, University of Texas at Arlington, USA |
Time/Place: | 10:00 - 11:00 FSC1217, Fong Shu Chuen Library, HSH Campus, Hong Kong Baptist University |
Abstract: | Meta-analysis of rare binary events in biomedical research is often hampered by low statistical power in individual studies and limitations of traditional methods. This paper introduces novel Bayesian procedures that address these challenges. Our approach employs a flexible random-effects model, eliminating the need for pre-specified directions of variability, and incorporates Pólya-Gamma augmentation for efficient computation. Simulations and real-world applications (56 studies on rosiglitazone, 41 studies on stomach ulcers) demonstrate less biased and more stable estimates compared to conventional methods. Further, we propose a Bayesian goodness-of-fit test within a binomial-normal hierarchical model. This test leverages pivotal quantities and the Cauchy combination method for dependent p-values, effectively utilizing all data, including double zeros without artificial correction. Simulations and real data applications confirm well-controlled Type I error rates and increased power for detecting model misspecification. |
Title: | Original Studies on the Prevention and Treatment of Diabetic Panvascular Disease by Integrating Chinese and Western Medicine |
Speaker: | Ms. Wenting Wang, China Center for Evidence-Based Medicine of TCM, China Academy of Chinese Medical Sciences, China |
Time/Place: | 11:00 - 12:00 FSC1217, Fong Shu Chuen Library, HSH Campus, Hong Kong Baptist University |
Abstract: | Diabetic panvascular diseases (DPDs) has become a major public health problem, and the search for preventive and curative drugs with panvascular benefits and the elucidation of their mechanisms have become a major focus and challenge in the field of Cardiometabolic disease. From a systemic and holistic perspective, Traditional Chinese medicine (TCM), with its multi-component, multi-target and systemic regulation, has unique advantages and development potential in the prevention and treatment of DPDs, and can work synergistically with Western medicine to improve patients' clinical symptoms and poor prognosis. However, the unclear biological substance of TCM symptoms and the unclear mechanism of action of TCM formulas have seriously restricted its clinical application. Focusing on the above issues, our team has conducted a series of original studies on diabetic coronary artery disease, diabetic atherosclerosis and vascular calcification, and has obtained a number of research results, which provide evidence to support the construction of Chinese and Western medicine diagnostic and treatment strategies for diabetic DPDs. |
We organize conferences and workshops every year. Hope we can see you in future.
Learn MoreProf. M. Cheng, Dr. Y. S. Hon, Dr. K. F. Lam, Prof. L. Ling, Dr. T. Tong and Prof. L. Zhu have been awarded research grants by Hong Kong Research Grant Council (RGC) — congratulations!
Learn MoreFollow HKBU Math