Session 36:  Advances in Nonparametric Statistics and their Applications

Session title: Advances in Nonparametric Statistics and their Applications
Organizer: Narisetty, Naveen (UIUC)
Chair: Fei Xue (UIUC)
Time: June 6th, 8:30am – 10:00am
Location: VEC 902/903

Speech 1: Fly-By-Night Life Insurance and the NPMLE for Weibull Frailty Models
Speaker:  Roger Koenker (UIUC and UCL) 
Abstract: Classical parametric statistical models are often quite fragile when confronted with data that fails to conform precisely to their idealized conditions.  In survival analysis more flexible models have been constructed as mixtures, yielding estimators with improved robustness properties.  These frailty models can themselves be parametric, but recent work has stressed non-parametric mixtures.  This talk will describe an application of the latter approach to the study of mortality rates for a large sample of mediterranean fruit flies.  In addition to revealing some rather surprising biological findings, the methods illustrate a general approach to estimation based on the Kiefer Wolfowitz non-parametric MLE for mixture models  and its use in a somewhat fanciful compound decision problem.

Speech 2: Learning from Dr. Martin Luther King Jr: Text analysis and statistical approaches for civil rights
Speaker: Christopher Kinson (UIUC ) 
Abstract: The recent shifts in the political landscape have heightened the importance of education and advancement for minorities, especially those underrepresented in STEM disciplines.  Minority students have been battling for equal access and rights to education at institutions of higher learning and the resources therein throughout the history of the US.  We as a scientific community must boldly step up to improve the outcomes of underrepresented students. The objective of this project is to educate and recruit Black and African students into the fields of statistics and data science. We create this research cohort and lab environment for the students to contribute to their own knowledge and to the body of knowledge within data science and the digital humanities. This lab is a creative and interdisciplinary space where the subject is one of the most important figures in Black history, Rev. Dr. Martin Luther King Jr. Specifically, we employ text analysis of King’s writings and speeches as well as articles written about him. Additionally, we educate students about statistics and introduce them to statistical programming in R.  The project tackles several challenges in broadening the participation of underrepresented students in STEM and celebrates the achievements made in that process.

Speech 3: Inference on the dependence structure of time series extremes
Speaker: Stanislav Volgushev (U Toronto) 
Abstract: Many natural phenomena such as extreme precipitation or heat waves can be described as maxima over blocks of time series. If the dependence structure of such extremes is of interest, component-wise maxima of vector-valued time series need to be considered. Under suitable assumptions on a vector-valued time series, properly standardized component-wise maxima are known to converge to a multivariate extreme-value distribution. In the present talk, we discuss functional central limit theorems and resulting inference procedures for the dependence structure of this limiting distribution. We propose several improvements over existing approaches which allow to reduce both bias and variance. We also contrast our approach (which is based on taking block maxima) with the peak-over-threshold approach which is a popular tool in the analysis of extremes.