The multivariate Poisson-log normal distribution

From MaRDI portal





For the statistical analysis of multivariate counts, as they may arise in ecology as the numbers of different species in geographical areas, a rich parametric class of multivariate count distributions would be useful. After a review of the corresponding literature the paper gives a study of the multivariate Poisson-log normal class of distributions defined as follows: Let g(\(\theta\) \(| \mu,\Sigma)\) denote the density function of the d- dimensional log normal distribution \(\Delta^ d(\mu,\Sigma)\), so that \(g(\theta | \mu,\Sigma)=(2\pi)^{-d}(\theta_ 1,...,\theta_ d)^{-1}.\) Then the multivariate Poisson-log normal distribution is the \(\Delta^ d(\mu,\Sigma)\) mixture of d independent univariate Poisson distributions \(Po(\theta_ i)\) with density function \[ p(x| \mu,\Sigma)=\int \prod^{d}_{i=1}f(x_ i| \theta_ i)g(\theta | \mu,\Sigma)d\theta \quad (x_ 1,...,x_ d=0,1,...), \] where the integration has to be taken over the positive orthant of the d- dimensional real space. The first two moments and inequalities between them are given and the parameter estimation is described. The model of the count distribution p(x\(| \mu,\Sigma)\) is compared with other models and an application to trivariate bacterial counts is given.




Cited in
(95)








This page was built for publication: The multivariate Poisson-log normal distribution

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q4729179)