Ítem


Graphing and Communicating Compositional Data in High Dimensions

Visualization of data becomes more challenging as the dimensionality of the data increases,impacting not only the display of the data itself but also the modeling results.This paper discusses common visualization techniques for compositional data. None of themseem to be well suited for changes in compositions that depend on either a metric covariate or afactor. The clr-deviation chart as a chart with a factor or covariate as abscissa and all centered logratio-transformed component values superimposed on the ordinate axis is then introduced jointlywith the clr-component chart. The clr-deviation chart takes advantage of the sum-equals-zeroproperty of clr-transformed compositional data. It has some theoretical and practical advantagesover alternatives and one major disadvantage – an arbitrarily scaled ordinate axis; its propertiesare discussed.The usefulness of the methods are illustrated using an example analyzing the changes of proportionsof the different diseases treated by hospitalization over a period of 13 years in Germany

Universitat de Girona. Departament d’Informàtica i Matemàtica Aplicada

Altres contribucions: Universitat de Girona. Departament d’Informàtica i Matemàtica Aplicada
Autor: Ulbrich, H.F.
Resum: Visualization of data becomes more challenging as the dimensionality of the data increases,impacting not only the display of the data itself but also the modeling results.This paper discusses common visualization techniques for compositional data. None of themseem to be well suited for changes in compositions that depend on either a metric covariate or afactor. The clr-deviation chart as a chart with a factor or covariate as abscissa and all centered logratio-transformed component values superimposed on the ordinate axis is then introduced jointlywith the clr-component chart. The clr-deviation chart takes advantage of the sum-equals-zeroproperty of clr-transformed compositional data. It has some theoretical and practical advantagesover alternatives and one major disadvantage – an arbitrarily scaled ordinate axis; its propertiesare discussed.The usefulness of the methods are illustrated using an example analyzing the changes of proportionsof the different diseases treated by hospitalization over a period of 13 years in Germany
Accés al document: http://hdl.handle.net/2072/273425
Llenguatge: eng
Editor: Universitat de Girona. Departament d’Informàtica i Matemàtica Aplicada
Drets: Tots els drets reservats
Matèria: Anàlisi multivariable -- Congressos
Multivariate analysis -- Congresses
Estadística matemàtica -- Congressos
Mathematical statistics -- Congresses
Títol: Graphing and Communicating Compositional Data in High Dimensions
Tipus: info:eu-repo/semantics/conferenceObject
Repositori: Recercat

Matèries

Autors