Analysis of Compositional Time Series from Repeated Surveys

摘要

A compositional time series is a multivariate time series in which each of the series has values bounded between zero and one and the sum of the series equals one at each time point. Data with such characteristics are observed in repeated surveys when a survey variable has a multinomial response but interest lies in the proportion of units classified in each of its categories. The main approach to analyzing Compositional Time Series data has been based on the application of an initial transform to break the unit sum constraint. Box-Cox transformation originally was envisioned as a panacea for simultaneously correcting normality, linearity and homoscedasticity. However, one thing is clear; that seldom does this transformation fulfill the basic assumptions as originally suggested. This paper aims at reviewing works relating to these transformations with some modifications and illustrative example as would be applicable to the analysis of compositional time series data.