Humboldt-Universität zu Berlin - Statistics

Humboldt-Universität zu Berlin | School of Business and Economics | Statistics | Teaching | Courses | Winter term 2019/20 | Research Seminar in Statistics: Compositional Data Analysis (SE)

Research Seminar in Statistics: Compositional Data Analysis (SE)



Course Description

Across various fields, such as Economics, Biosciences, Political Sciences, Geosciences and Computer Vision, there are numerous examples of data sets, where we want to analyze the shares of parts of a whole. Say, we want to investigate voter shares among different parties, how consumers allocate their budget among available commodities, or we want to analyse the cell-type compositions of different tissues. Another example is time-use data, which not only looks at labor supply but at the complete picture of how men and women distribute their time between work, family care, household and other activities.
With the interest lying in the relative counts or amounts, this can be considered a special kind of multivariate data with elements constrained to be positive and summing up to 1. This induces negative correlations between the different shares and the data structure has to be taken into account in the analysis to avoid spurious results. This has motivated a vivid branch of statistical research, known as Compositional Data Analysis, to develop methodology for this special data structure.

In this seminar, the students will learn the foundations of Compositional Data Analysis and the so-called Aitchison Geometry, as well as selected state-of-the-art methodology with topics ranging from explorative data analysis, principle component analysis, Bayesian analysis and regression for compositional data to geostatistics and time series of compositonal data and extensions to density functions. 


Good knowledge of Statistics is recommended for this course.

Course Learning Objectives

Course Structure

For details please see Moodle.