Multiple Imputation for Directional Data.
Conference
64th ISI World Statistics Congress - Ottawa, Canada
Format: CPS Abstract
Keywords: imputation, simulation
Abstract
Observations consisting of directions or angles are found across many areas of science, including ecology, earth sciences, environmental science, and medicine. Examples of such data are the angular movements of an animal relative to a food source or other attractor, wind directions, diurnal measurements of admission times to an intensive care unit, and departure directions of birds after release. Circular data arise whenever directions are measured, and are usually expressed as angles relative to some fixed reference point, such as Due North. Time data measured on a 24 h clock may also be converted to angular measurements, with 0:00 corresponding to 0◦ and 24:00 to 360◦. Like other datasets, missing values is a major problem in the case of circular data as well. In this paper, we discuss some imputation methods for missing values in circular data by using the technique of multiple imputation via chained equations. This ensures that, the relationships among these variables can be used for better imputation of missing values. We restrict our attention to bivariate datasets consisting of at least one circular variables. The performance of the method is assessed via comparison of the distribution of the original and imputed datasets through an extensive simulation study.
Keywords: Directional data, Missing data, Multiple imputation via chained equations, Simulations