.png){kind=icon}
.png){kind=icon}
Return
Copula-based models in compositional data analysis
.png){kind=icon}
PRE
AI
DOI：10.1016/j.jspi.2026.106416.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Compositional data, which are multivariate fractional data with unit-sum constraints, often occur in various fields. Both the log-ratio method and the Dirichlet regression rely on assumptions of the same type of marginal distribution. We propose a distribution framework for this class of data by incorporating the Sigmoid transformation of log-ratio coordinates and utilizing Beta, proportional inverse Gaussian and simplex marginal distribution with the Gaussian copula function. Simultaneously, we propose an unified model to construct regression models specifically designed for compositional data. This framework unifies a large class of new and existing log-ratio type regressions and covers a very flexible class of regression models. The properties of the proposed estimator are assessed numerically through a simulation study and real-data analysis.
Keywords:
Gaussian copula
Compositional data
Logratio coordinates
Regression analysis
Journal
J
IF:
0.8
Papers:
38
Citations:
4.5K