Geometric stratification is an extremely simple way of stratifying skewed populations, taking the boundaries in geometric progression. Researchers using the method have recently highlighted some implementation difficulties which give rise to unfeasible solutions, in particular with strata which are too small or even empty.
In this paper we suggest a modification of geometric stratification, adding empirical rules for determining outliers, take-all strata, and end points, in order to improve the efficiency and ensure a feasible set of boundaries. After a brief overview of geometric stratification we will describe the proposed adjustments, and compare the efficiency of the estimators obtained by the modified method with those of Lavallee and Hidiroglou (1988) and Kozak (2004). These comparisons are implemented using the R package on Stratification divised by Rivest and Baillargeon (2007).
Kosak M. (2004) Optimal stratification using random search methods in agricultural surveys, Statistics in Transition, 6, 5, 797-806.
Lavallee, P. and Hidiroglou M. (1988) On the stratification of skewed populations, Survey Methodology, 14, 33-43.
Rivest, L.P. and Baillargeon S.(2007) Univariate Stratification of Survey Populations available on the crann website at http://www.r-project.org/
Keywords: Coefficients of variation; Skewed data; Efficiency; Outliers
Biography: Jane Horgan is Professor of Statistics in the School of Computing at Dublin City University. She has also lectured at the Autonomous University of Mexico, University of Dar-es-salaam, Dublin Institute of Technology, University College Cork and the London School of Economics. A fellow of the Institute of Statisticians, Professor Horgan has published extensively in the areas of statistical sampling and estimation. Her research interests include applications to both financial and rare incidence populations and statistical computing.