Ensuring balance through optimal allocation of experimental units with known categorical covariates into two treatments. Issue 3 (3rd May 2020)

• Record Type: Journal Article
• Authors: Hore, Samrat; Dewanji, Anup; Chatterjee, Aditya
• Is Part Of: Statistics. Volume 54:Issue 3(2020)
• Page Start: 559
• Page End: 576
• Publication Date: 2020-05-03
• Subjects: Optimal allocation -- linear model -- covariate mean balance -- variable neighbourhood search -- simulation; 62K05 -- 62J10 -- 37M05; Mathematical statistics -- Periodicals; 519.505
• DOI: 10.1080/02331888.2020.1740998
• Languages: English
• ISSNs: 0233-1888

Abstract

ABSTRACT: The balanced allocation of experimental units with regard to various known covariates among several treatment groups, before the physical experiment takes place, is often considered to be the most reasonable allocation scheme in all intervention studies and clinical trials. It is well-known that covariate mean balance over various treatment groups ensures widely used D - and A -optimality. However, it is not well-understood if the reverse proposition holds or not. For continuous covariates, it has been observed previously by the same authors, through a computationally intensive method, that covariate mean balance is nearly achieved for an optimal design. In the present paper, it has been analytically established that, with categorical covariates, the covariate mean balance can be ensured through D - and A -optimality.