Graduate Students Seminar

Speaker 1: Yuxin Zhang

Title

Optimizing Sample Allocation: Comparing Neyman, Exact Optimal, and Power Allocation Methods for Subnational Area Precision

Abstract

Effective sample allocation is essential for achieving precise, cost-effective, and representative data in both regional and national surveys. This presentation examines three key sample allocation strategies: Neyman allocation, exact optimal sample allocation, and power allocation.  Starting with Neyman allocation, we explore its traditional approach of minimizing variance by allocating samples in proportion to population size and variability across areas. Building on this, we look at exact optimal sample allocation, which further refines sampling to boost precision without increasing the total sample size, enhancing efficiency beyond Neyman’s approach. Lastly, we address power allocation, a strategy designed specifically for subnational areas that incorporate statistical power requirements to ensure reliable estimates across regions with varying population characteristics. Through a comparison of these methods, we highlight how each one uniquely addresses sampling challenges, supporting more effective survey designs in geographically diverse contexts and improving data quality overall.

Speaker 2: Maliha Tasmiah Noushin

Title

Sobol Sensitivity Analysis in Systems Biology and their Monte Carlo Estimates

Abstract

Sensitivity analysis is a method to determine how different values of an input in a model will impact the outputs. By performing a sensitivity analysis of a model, we can determine which input has the greatest effect on the outputs or a particular model output. The Sobol method is a model-independent general sensitivity analysis technique that is based on variance analysis. This method can be used for nonlinear and non-uniform functions  and models. The Sobol sampling scheme is one of the most efficient sampling schemes to calculate the sensitivity indices of the input parameters, which is very easy to implement and inexpensive. It covers the entire parameter space more uniformly than other sampling schemes and is also reproducible. Using the SALib library in Python, we can calculate the first and total order indices of the parameters of a model and determine the critical parameters to the model output.