Estimating Means Using Empirical Likelihood in Large Datasets
Open Access
Author:
Ghatak, Maitraya
Graduate Program:
Statistics
Degree:
Master of Science
Document Type:
Master Thesis
Date of Defense:
June 29, 2021
Committee Members:
Ephraim Mont Hanks, Program Head/Chair Nicole Alana Lazar, Thesis Advisor/Co-Advisor G. Jogesh Babu, Committee Member
Keywords:
Empirical Likelihood Algorithms for Mean Estimation
Abstract:
Empirical Likelihood does not suffer from model misspecification and provides robust estimates of means. However, computational inefficiency arises in large datasets, where multiple points must be tested on a grid at minute intervals. I identify how these computational inefficiencies can affect mean estimates and show that partitioning, subsampling, and certain other algorithmic methods can be used to obtain reliable estimates of the mean. I build three algorithms that rely on a two-part approach, which involves narrowing down the space of possible mean estimates in subsequent steps. I find that using this approach while also taking advantage of parallel computing can provide reliable estimates which converge very quickly. I discuss further modifications that can be used to extend these algorithms to other contexts.
