# ENBIS: European Network for Business and Industrial Statistics

You are here:

## Free ENBIS Webinar by Marina Vives-Mestres: "Introduction to Compositional Data Analysis with Applications to Customer Survey Analysis"

17 February 2016; 12:45 – 13:30; Webinar

Marina Vives-Mestres will give an introduction to CoDa based on her talk during ENBIS-15. The webinar will be moderated by Jacqueline Asscher.

Compositional Data (CoDa) consists of multivariate data with strictly positive components of constant sum. Natural examples consist of chemical formulations or results from a survey. We first introduce the definition of CoDa through simple examples and then show the need of an analysis based on log ratios of components, also called coordinates. We then describe the principles of working with coordinates and graphically compare classical analysis with a log ratio analysis in the cases of: principal component analysis, cluster analysis, linear discriminant analysis and linear regression models. We also discuss the problem of zeros and how to deal with them in CoDa.

We finish this introductory webinar with an example of applications in the field of customer survey analysis. Specifically, we analyse the annual customer satisfaction survey of the ABC Company presented and analysed in detail in the book edited by Kenett and Salini (2011). The questionnaire consists of an assessment of overall satisfaction evaluated on a five-point anchored scale, so that it can be analysed from a CoDa perspective, and almost 50 statements with two types of scores: an evaluation score and a measure of item importance. Other questions such as repurchasing intentions and descriptive variables for each customer are used in analysing the ABC dataset.

We show how CoDa methods can contribute to provide a map of customer’s opinion, improve decision making, identify improvement areas or weak points, set service level targets and help improve the questionnaire itself. We also compare the findings to several statistical models presented in Kenett and Salini (2011) such as PLS, hierarchical models, fuzzy sets, log-linear models and control charts. Graphical tools to communicate CoDa results are also proposed. The general idea is that one can increase the information quality of a customer survey analysis by combining more than one technique.

References

Kenett, RS., Salini, S. (2011). Modern Analysis of Customer Satisfaction Surveys: with applications using R. Chichester: UK. JohnWiley and Sons.