MSc in Mathematical and Computational Sciences Thesis Defence--Aadesh Nunkoo

Title of the thesis:

“Optimal Two-Phase Sampling Within Coarsened Strata: A General Framework”

Two-phase designs are cost-effective sampling strategies in epidemiological research when complete data collection is infeasible. While optimal designs are well-studied for naturally discrete phase-I data, deriving optimal phase-II sampling probabilities for strata formed by coarsening mechanisms lacks systematic treatment. This thesis develops a general framework for deriving optimal phase-II sampling probabilities within coarsened strata under inverse probability weighting, extending existing optimal design theory to accommodate any coarsening mechanism. A key result is that deterministic coarsening avoids additional modeling assumptions beyond those required in discrete settings. We implement practical two-stage adaptive schemes that estimate required components without a pilot study. Simulation studies with 6- and 12-strata designs show that the optimal and adaptive designs outperform traditional proportional and balanced sampling. These designs also possess minimax properties across IPW, CML, and CEML and remain robust under model misspecification. An application to NHANES survey data demonstrates strong efficiency gains achieved by our proposed designs. This work provides implementable, efficient strategies for two-phase studies under coarsened phase-I data.

Date/Time/Location: December 8, 1:00 pm via web conference

If you wish to attend the public presentation, please contact the Graduate Studies Coordinator at gsc@upei.ca to receive the link.

EVERYONE IS WELCOME.