Multiple discrimination between Dirichlet populations
Daan de Waal, Sean van der Merwe · South African Statistical Journal · 2014
This paper shows applications of the discriminant function of the Dirichlet distribution in the multivariate and matrix cases. The first is on discriminating between unknown Dirichlet populations with an application on the famous Fisher's Iris data with four variables and three populations. This is a case where the Dirichlet parameters have to be estimated for each population. The second application is on constructing rejection regions for matrix Dirichlet variables with known parameters. The correspondence between the likelihood ratio test statistic for testing equality of covariance matrices and the discriminant function between matrix Dirichlet populations is discussed. The application is on volatility of the Southern Oscillation Index (SOI) for specific months among high, medium and low groupings.