Seminar: Cluster Randomized Ranked Designs

SEMINAR 

DEPARTMENT OF INDUSTRIAL ENGINEERING 

Cluster Randomized Ranked Designs

by

Omer Ozturk

Department of Statistics, The Ohio State University, USA

Abstract:

In this presentation, we investigate the intricacies of constructing a design of experiment focused on assessing the impact of treatments or interventions on a broader scale—specifically, the effects on a system, methodology, procedure, or policy, rather than on individual workers, participants, or singular units. Our focus is on the innovative approach known as a Cluster Randomized Design (CRD), wherein treatments are randomly assigned not to individual subjects, but rather to cluster units.

When the target of the intervention is a collective entity or system rather than a particular person, such as a worker, patient or student, CRD is an appealing procedure. For example, CRD is better able to evaluate whether a new standard of production, guideline recommendation, or other practice-wide, company-wide, or system-wide change is affecting outcomes. In education research, the CRD would be appealing to see if a new training (or intervention) program is more effective than a traditional training program at schools or industrial complex. In this setup, CRDs provide logistical flexibility, a real-world view of research and a reduction in the total cost of the experiment, while minimizing the possibility of treatment contamination between units randomized to different intervention programs.

Cluster randomized designs (CRD) provide a rigorous development for randomization principles for studies where treatments are allocated to cluster units rather than the individual subjects within clusters. It is known that CRDs are less efficient than completely randomized designs since the randomization of treatment allocation is applied the cluster units. To mitigate this problem, we embed ranked set sampling design from survey sampling studies into CRD for the selection of cluster and subsampling units. We show that ranking groups in ranked set sampling act like a covariate, reducing the expected mean squared cluster error and increasing the precision of the sampling design. We provide an optimality result to determine the sample sizes at cluster and sub-sample level. We apply the proposed sampling design to a longitudinal study from an education intervention program. We also provide an example where order restricted cluster randomized design would be appropriate in an industrial experiment.

Bio: 

Omer Ozturk is a professor of statistics. He currently serves as an associate editor for Environmental and Ecological Statistics, Statistics and Probability Letters, Communications in Statistics- Data Analysis and Simulation, and Communications in Statistics- Theory and Methods. His research was founded by NSA, NSF at USA and Grain Research Development Company (GRDC) of Australia. He visited University of Canterbury, New Zealand, University of Adeliade, Australia as research associates. He was frequently invited to US Census Bureau as a Summer at Census Scholar. He served as publication officer in the section of nonparametric statistics in ASA. He was Elected fellow of ASA.

Date: Friday, May 31, 2024

Time: 15:00-16:00

Place: Engineering Building, South Campus, VYKM 4